Heart rate and swimming activity as indicators of post-surgical recovery time of Atlantic salmon (Salmo salar)

Background: Fish telemetry using electronic transmitter or data storage tags has become a common method for studying free-swimming sh both in the wild and in aquaculture. However, sh used in telemetry studies must be handled, anaesthetised and often subjected to surgical procedures to be equipped with tags, processes that will shift the sh from their normal physiological and behavioural states. In many projects, information is needed on when the sh has recovered after handling and tagging so that only the data recorded after the sh has fully recovered are used in analyses. We aimed to establish recovery times of adult Atlantic salmon (Salmo salar) after an intraperitoneal tagging procedure featuring handling, anaesthesia and surgery. Results: Based on ECG and accelerometer data collected with telemetry from nine individual Atlantic salmon during the rst period after tagging, we found that heart rate was initially elevated in all sh, and that it took an average of ≈ 4 days and a maximum of 6 days for heart rate to return to an assumed baseline level. One activity tag showed no consistent decline in activity, and two others did not show strong evidence of complete recovery by the end of the experiment: baseline levels of the remaining tags were on average reached after ≈ 3.3 days. Conclusion: Our ndings showed that the Atlantic salmon used in this study required an average of ≈ 4 days, with a maximum interval of 6 days, of recovery after tagging before tag data could be considered valid. Moreover, the differences between recovery times for heart rate and activity imply that recovery time recommendations should be developed based on a combination of indicators and not just on e.g. behavioural observations.


Background
Fish telemetry/biologging is a method of monitoring free-swimming sh where individual animals are equipped with electronic tags that often contain sensors for collecting data on the conditions within or near the sh (Cooke et al., 2011;Thorstad et al., 2013). Such tags may either be transmitter tags transferring data wirelessly to the user (see Føre et al., 2011 for details on the structure of an electronic transmitter tag) or data storage/archival tags (DSTs) that store data in internal storage mediums accessible only after the sh (and tag) has been recaptured (Thorstad et al., 2013). Irrespective of tag type, most studies using such methods aim to assess the status of wild sh in ecological settings (e.g. Welsh et al., 2013;Taylor et al., 2017), to evaluate how sh communities respond to man-made structures (e.g. Cooke et al., 2004), or as a tool to provide knowledge for sheries management (reviewed by Crossin et al., 2017). The interest in using this approach in aquaculture is also increasing, both because ongoing technological advances are rapidly expanding the possibilities (Hussey et al., 2015), and because new production philosophies such as Precision Fish Farming promote monitoring at an individual level (Føre et al., 2018a). Example uses of telemetry/biologging in aquaculture include studies to assess sh responses during welfare-critical operations such as crowding (e.g. Føre et al., 2018b) and transport (e.g. Brijs et al., 2018), and responses to environmental variability such as temperature variations (e.g. Johansson et al., 2009).
In animal monitoring, it is essential to ensure that the observed animals are representative of the targeted population. When using telemetry, the sh selected for tagging must therefore be representative both before and after the tags are deployed. Ideally, this means that the selection of sh should be truly random and representative, and that the tags do not in uence physiology or behaviour in such a way that the tagged sh differ signi cantly from untagged sh (e.g. Wright et al., 2018). In addition, tagging procedures include several steps (e.g. handling, anaesthesia and surgical procedures) that may induce stress, that in turn may lead to physiological and/or behavioural changes in the sh (Thoreau and Baras, 1997;Jepsen et al., 2001;Connors et al., 2002;Campbell et al., 2005;Thorstad et al., 2013). Acute (short term) followed by chronic (long term) stress in farmed sh may lead to undesirable effects such as reduced disease resistance, reduced growth rates, impaired health, and increased mortality (Wedemeyer, 1997;Pickering, 1998;Schreck, 2000;Ellis et al., 2002). Stress responses in sh are described by primary responses that include the release of stress hormones such as catecholamines and cortisol into the circulation system, followed by secondary responses such as changes in glucose levels, electrolyte balance and heart rate and, nally tertiary (whole animal) responses. If the sh is unable to acclimate to the stressor at this stage, effects such as behavioural changes, decreased reproductive capacity and growth may occur, sometimes even resulting in that the animal dies (see Iwama et al., 2006 and references therein). If such changes are chronic, the sh cannot be considered representative of the population and should be excluded from further analyses (Mulcahy, 2003;Cooke et al., 2011). Conversely, if the changes are transient, the sh may be considered fully recovered once the response patterns return to those expected from an untagged sh. This means that tagged sh can be used in analyses if the data from the period of recovery are excluded. However, this also raises the question: how can we de ne when a sh is properly recovered after a tagging procedure? Jepsen et al. (2001) sought to identify the duration of post-surgery recovery for Chinook salmon (Oncorhynchus tshawytscha) by studying changes in commonly used blood indicators of the primary (cortisol) and secondary (glucose and lactate) stress responses in teleosts. The authors found that all measured parameters decreased from initially elevated levels to within normal ranges within 7 days post-surgery, with glucose and lactate (substrate and by-product, respectively, of elevated anaerobic metabolism) normalising during the rst 24 h, a recovery time resembling that seen in several studies (e.g. Martinelli et al., 1998;Bridger and Booth, 2003). Coping with stress is also an energy-demanding process (Barton and Schreck, 1987) and one of the most common indicators of metabolic effects due to stress is the increase in plasma glucose concentration (Iwama et al., 2006). Such changes have recently been shown to lead to increased heart rates also in sh (Svendsen et al., 2020). Other studies have aimed to evaluate post-surgery recovery by comparing the behaviour of the tagged sh to their behaviour before surgery or in untagged cohabitant sh. This method has for instance been applied in laboratory experiments with tilapia (Tilapia sp.) who appeared fully recovered 24 h post-surgery after displaying loss of equilibrium and reduced swimming activity and feeding just after tagging (Thoreau and Baras, 1997). Swimming activity was then assessed by measuring the posture of the sh, and presented as the percentage of the time the sh was resting (assuming an oblique angle with the snout towards the surface) or actively swimming (horizontal orientation or snout pointing toward the bottom).
Recovery after tagging may also be studied with sensor telemetry. The information conveyed by the tag must then re ect the state of the sh, and typical sensor values for unstressed sh should be available as a baseline for comparison. Previous studies using this approach include using heart rate tags to compare tagging methods for black cod (Paranotothenia angustata, Campbell et al., 2005), and more recently to study post-surgery stress-responses (Brijs et al., 2019b) and potential effects of antibiotics on postsurgical recovery (Hjelmstedt et al., 2020) in rainbow trout (O. mykiss). While Brijs et al. (2019b) implied a recovery from surgical implantation >72h, Hjelmstedt et al. (2020) demonstrated a decrease in heart rate to within baseline levels 72-96 h after anaesthesia and surgery. Other sensor measurements that could potentially be used in this way include tri-axial accelerometers, as previous studies have identi ed links between accelerometer based activity proxies that are particularly sensitive to tail beat frequency and amplitude and orientation changes, and stress in salmon (Kolarevic et al., 2016;Føre et al., 2018).
Although Atlantic salmon (Salmo salar) has been frequently studied using telemetry, there is still a lack of detailed quantitative information on the post-surgery recovery of this species. We therefore sought to identify the recovery time of Atlantic salmon after intraperitoneal tagging. This was done using heart rate and acceleration data collected using intraperitoneally implanted electronic tags, meaning that data could be collected without introducing the additional handling stress that would accompany other methods such as blood sampling. The parameters were chosen because they have previously been found to be linked with stress (e.g. Brijs et al., 2019a;Brijs et al., 2019b;Føre et al., 2019) and welfare (Hvas et al., 2020a) in salmonids and are commercially available in archival and telemetry tags. The data were collected in a controlled experiment in tanks studying how stress responses in Atlantic salmon can be measured using state-of-the-art technology. The stress response part of this experiment is described in greater detail by Svendsen et al. (2020).

Experimental site and sh
The experiments were conducted at the NINA Ims Research Station near Stavanger, Norway, between January and March 2019, using 60 hatchery reared adult Atlantic salmon of the Aqua Gen strain (mean 55.5 ± stdev 5.7 cm fork length, mean weight 2100 g). The experiment started on January 28 th by stocking four square tanks (tank 1-4, 215 cm side, 122 cm depth, 5600 l) with seven sh each. The sh were then allowed to habituate to the tanks for a period of 21 days until February 18 th when three sh in each of tanks 1-4 were selected at random and equipped with tags, resulting in 12 tagged sh in total (Table 1). The tanks were set up with ow-through con guration, with ltered freshwater from the nearby Imsa river mixed with small amounts (3-6 ppt, average 5 ppt) of seawater supplied from seawater inlets at 30 m depth to ensure a stable and homogeneous water quality and avoid the introduction of parasites and pathogens to the tanks. Consequently, tank water properties followed the ambient conditions in the river, temperatures increasing from 3.9 to 5.0 ˚C and with DO varying between 93.8 and 101.2 % between the start and end of the experiment (March 15 th ). Oxygen sensors and oxygenation were also used to prevent unfavourable DO levels. The sh were fed once per day between 08:00 and 10:00 in the morning throughout the entire experimental period, with each meal consisting of 2 dl tank -1 (Skretting Røye Vitalis 600-60A 7 mm pellets). The sh were not subjected to any fasting during the experiment period.
Biotelemetry/logging systems and surgical procedures All 12 tagged sh (Table 1) were equipped with one of three different types of heart rate monitoring Data Storage Tags (DSTs, Star Oddi Ltd.): 4 x DST milli-HRT (39.5 x 13 mm, 11.8 g in air); 4 x DST centi-HRT (46 x 15 mm, 19 g); 4 x DST centi-HRT ACT (46 x 15 mm, 19 g). Using different DST types rather than equipping all sh with the same tag types allowed us to also investigate whether all three tag varieties would be suitable for experiments with Atlantic salmon, which is relevant because this is one of the rst applications of this technology on this species. Furthermore, since all three tag types were from the same provider, contained the same type of heart rate sensor and comparable sampling frequencies (80 Hz over 7.5 s per HR sample point for the centi tags and 100 Hz over 15 s per HR sample point for the milli tags), and applied the same post-processing methods to the resulting data, they provided heart rate data sets that were comparable among tags. The milli-HRT type was set with a higher sample storage interval (10 min) than the others (5 min) as they used more of their internal storage medium for raw ECG traces. All data were timestamped using the tag internal clocks to facilitate comparison, and eventual clock drift between individual clocks was negligible compared to the time scale of the experiment. One tag type (DST centi-HRT ACT) also measured activity using an embedded tri-axial accelerometer (1 Hz sampling rate).
In addition to the DSTs that were applied, a total of 4 tagged sh (two sh each from tanks 1 and 2, Table  1) were tted with acoustic tags (A MP-9, 24.4 x 9 mm, 3.6 g; Thelma Biotel AS) that contained tri-axial accelerometers (5 Hz sampling rate) and transmitted an activity proxy derived from the accelerometer measurements every 40 s. These tags compute the proxy by rst high pass ltering the accelerations from all three axes using a cutoff frequency of 0.2 Hz to remove low frequency acceleration components due to gravity and body orientation. The remaining high frequency components then mainly contain accelerations caused by features related to bodily movement that are of interest when evaluating activity levels, such as tail beats (frequency and amplitude) and rapid changes in attitude/orientation. The Euclidian norm of the three high pass ltered accelerometer axes is then computed to yield the magnitude of the total high pass ltered 3D acceleration sensed by the accelerometer. Although Føre et al. (2018) used the same activity proxy with a maximum value of 3.465 m s -2 , we chose to limit the proxy to 0-2.1 m s -2 in our study as this gave us a higher resolution and hence precision for the activity measures.
Moreover, Føre et al. (2018) observed very few activity values above 2 ms -2 in Atlantic salmon during stressing, implying that using a lower range would not compromise the ability to capture the dynamics associated with salmon swimming activity. To be comparable with the data from the acoustic tags, the activity data from the centi-HRT ACT DSTs were analysed similarly by applying ltering and computing the Euclidian norm as explained for the acoustic tags (see Svendsen et al., 2020 for more details). Adding the acoustic tags thus allowed us to compare their activity proxies with those based on the acceleration data from the DSTs and resulted in that the experiment produced 12 data sets on heart rate, and 8 data sets on swimming activity. With mean sh weight being 2100 g and a maximum total tag weight carried by an individual at 22.6 g (DST centi-HRT + A MP-9) the tag vs. sh weight ratio of all sh were well within the informal rule of thumb of 2 % for maximum tag mass relative to sh mass (Thorstad et al., 2013).
Each tag implantation was started by capturing a random sh from an experiment tank using a knotless dip net and immediately transferring it to an anaesthetic bath (Benzoak Vet, 70 mg/L) where the sh was kept until it lost its equilibrium and stage III anaesthesia (Coyle et al., 2004) was reached (average time 7.7 min). The sh was then carefully placed with its ventral side up on a specialised surgical table with a v-shaped mid-section designed such that the head of the sh was immersed in water throughout the whole procedure. A hose circulating anaesthetic (Benzoak Vet, 35 mg/L) through the orobranchial cavity of the sh was inserted into its mouth and the head was covered by a moist cloth (Figure 1).
A 2-3 cm incision was made along the sagittal plane starting slightly more than one tag length (i.e. the length of the tag to be implanted) posterior from the transverse pericardial septum.
A nger was inserted through the incision to locate the transverse pericardial septum. While retaining the nger inside the peritoneal cavity for support, a needle was positioned in the skin just posterior to the transverse septum and slightly laterally from the sagittal plane. The nger was withdrawn, and a smooth plastic spoon inserted through the incision until it was just below the needle insertion point. The needle was then pushed through the peritoneal wall while simultaneously withdrawing the spoon to extract the needle out through the incision while protecting the viscera. One end of a suture threaded through the end of the tag was inserted into the tip of the needle. The needle was then withdrawn to pull the suture out through the needle's entry point. This procedure was then repeated on the other side of the sagittal plane. The tag was then inserted through the incision and anchored anteriorly in the peritoneal cavity using the suture and an (external) surgical knot. For the four sh also equipped with separate acoustic tags, the second tag was inserted into the peritoneal cavity through the same incision. Finally, the incision was closed using interrupted sutures. The sh was then transferred to a recovery tank with circulating seawater where it was kept until it regained consciousness, upon which it was transferred back into the tank it was collected from. See Table 1 for anaesthesia bath and surgery durations for all tagged sh.

Timeline and experimental design
Since the present study focused on investigating the post-tagging recovery, the analyses only included data from the two weeks following tagging. To avoid inducing other stress effects that could disturb their recovery, the sh were sheltered from all potential stress factors except those necessary to feed and provide for the sh in this period.
None of the sh exhibited signs of adverse health after tagging or during the trials, and all sh were euthanised after the conclusion of the experiment. Posthumous pathology of all remaining experimental sh at the end of the experiment (19 female, 23 male) revealed that about one third of these sh (14 in total, 8 F, 6 M) exhibited signs of sexual maturation through the experimental period, including 5 of the tagged individuals (Table 1). Although this appeared to have little direct impact on the sh in three of the tanks, the data from the sh in one of the tanks (tank 3) were excluded from the statistical data analyses due to perpetual inter-individual aggression between two matured males in that tank throughout the experimental period. This left nine sh tagged with DSTs measuring heart rate, six of which also measured activity. Since two of these sh contained both a DST and an acoustic tag measuring activity, this resulted in a total of eight time-series of activity.

Data processing and statistics
Heart rate data were used as downloaded from the DSTs. Outliers were removed using the Median Absolute Deviation (MAD) approach (Leys et al., 2013), using a MAD decision criterion of 3, which is a conservative value (see Miller, 1991). The MAD decision criterion denotes the standard deviation from the dataset's sample average above which samples are rejected. The MAD decision criterion typically ranges from 2 (poorly conservative) to 3 (very conservative). In this study, the choice of 3 is justi ed by the measured heart ranges compared to typical heart rates published in literature (15 < HR < 80) for Atlantic salmon and comparable species (Lucas, 1994, Brijs et al., 2019. Activity data from the DST centi-HRT ACT tags were downloaded as raw acceleration values along all three axes, and then subjected to similar post processing as that used to compute the activity proxy in the A MP-9 acoustic transmitter tags to yield a comparable measure of activity between the two tag types. In a non-decomposed time-series, circadian variation (that between day and night) and irregular variation (that other than circadian of long-term) had the potential to obscure long-term trends in heart rate and activity. Time-series of heart rate and activity were therefore rst decomposed into circadian, long-term trend, and irregular components. Decomposition, and subsequent removal of the circadian and irregular components of the time-series, leaving a long-term component (that showed the long-term growth or decline of the time-series values over the temporal extent of the series), allowed for examination of the form of the long-term trends towards recovery. To decompose each time-series, it was rst binned into 15 min intervals (each 15 min interval showing a mean heart rate or intensity over that interval), and then converted into a time-series object (R function ts {stats}; Becker et al (1988)). Time-series objects were then decomposed using the Seasonal Decomposition of Time Series by Loess R function stl {stats} (B.D. Ripley; Fortran code by Cleveland et al (1990) from "netlib"). Long-term trend components were then analysed for a systematic change in heart rate or activity that could be indicative of a post-surgery recovery by rst modelling the temporal relationship and then compartmentalising this into pre-and postrecovery phases.
The relationship between the long-term trend component of heart rate or activity (y) and time posttagging (t) was modelled using an exponential decay model: where α de nes the decay constant from y 0 (at time zero) to y p , the model plateau. Models were tted with the nls {stats} R function (D.M. Bates and S. DebRoy: D.M. Gay for the Fortran code used by algorithm = "port"), using the self-starting asymptotic regression function SSasymp {stats} (J. Pinheiro and D.M. Bates). Most trend components followed an exponentially decaying pattern, ensuring model convergence, but some included parts that were inconsistent with an exponential decay. Firstly, some tags (three heart-rate tags and four activity tags) showed a short initial post-surgery increase in registered values at the beginning of the experiment. Secondly, some tags (one heart rate and two activity tags) showed an increase in registered values after ≈ 5-6 d. This late increase in activity or heart-rate was likely a result of a separate, post-recovery change in behaviour of these individuals. To ensure model convergence, these parts of the long-term trend components were removed prior to model tting. That is, the exponential model was only tted to parts of the long-term trend component that were consistent with a post-surgery exponential decline. One activity tag ( sh F4 in tank 8) did not show an exponential decline with time and was thus not tted with a model.
Identi cation of breakpoints between pre-and post-recovery phases was done on an individual basis. The breakpoint between pre-and post-recovery for each tag was set where the heart rate or activity reached a recovery threshold, de ned as the heart rate or activity level delimiting those pre-and post-recovery. A recovery threshold was de ned for each tag as the mean +2SD of the long-term trend component values calculated from the nal three days of the tted series. Inspection of the tags showed that trend components were approaching asymptotes in the nal three days, so it was reasonable to assume that values from these days represented post-recovery signature. Thresholds were established on an individual basis to allow for post-recovery heart rate or activity to change according to individuals.

Post-surgery recovery
Daily heart rate signi cantly declined from a mean of 36.0 bpm (range = 24.6 -45.6, SD = 5.6, n = 9) on the day of surgery to a mean of 22.3 bpm (range = 17.5 -26.6, SD = 2.6, n = 9) 13 days later (one sided  (Figure 2 B). However, individual variation in activity was high (Figure 2 B). Both heart rate and activity displayed circadian variation. Heart rate was greater during daytime (mean = 25.8 bpm, range = 22.2 -26.7, SD = 1.9, n = 9) than during night (mean = 22.7 bpm, range = 19.6 -24.9, SD = 1.9, n = 9) ( The heart rate trend component showed a decline that could be modelled with an exponential decay function (  Figure 3). However, the trend component still showed considerable temporal variation, depending on the tagged individual. For example, the trend component for sh F4 showed a sharp decline during the rst day after tagging, but this then uctuated for the remainder of the two-week post-tagging period. The activity trend component also showed a pattern consistent with an exponential decay (Figure 4), except for one sh ( sh F8) where an exponential decay model could not be tted due to the activity trend component peaking ≈ 7 d after tagging. Two sh ( sh F1 and F2) showed an exponential decline in activity but did not reach a plateau during the study period, suggesting that these sh has not fully recovered in terms of activity.
Time to recovery (as de ned by the location of the breakpoint between pre-and post-recovery phases) varied between individuals, and the metric used (heart rate or activity, Figure 3, Figure 4, Table 2). The mean threshold value for heart rate in a 'recovered' individual was 23.8 bpm (range = 21.2 -26.0, SD = 1.18, n = 9). The mean time to reach this threshold (i.e. breakpoint between pre-recovery and post-recovery) was 4.1 d (range = 1.3 -5.8, SD = 1.7, n = 9). The threshold for activity recovery was greater for the acoustic tags (mean = 0.44 m s -2 , n = 2) than the DSTs (mean = 0.29 m s -2 , n = 3), re ecting the higher activity values registered by the acoustic tags. For the activity tags where there was evidence of recovery, the mean time taken to reach the threshold was similar to that for the heart rate tags (mean 3.3 d, range = 2.1 -5.7, SD = 0.09). For the two individuals that were each tagged with two activity tags, the identi ed breakpoints between the parts of the time series classi ed as pre-and post-recovery depended on the tag: in both individuals, the threshold to reach post-recovery occurred later for the acoustic tag than the DST.
Although raw values of mean heart rate on the day of anaesthesia and surgery (mean = 36.0 bpm, range = 24.6 -45.6, SD = 5.6, n = 9) varied more than the recovery threshold (mean = 23.8 bpm, range = 21.2 -26.0, SD = 1.8, n = 9, Table 2), there was a clear declining trend for all tagged individuals. With the exception of one individual (F8, Figure 4), there was a similar trend for activity: day of anaesthesia and surgery, mean = 0.64 m s -2 , range = 0.39 -0.92, SD = 0.20, n = 7; recovery threshold, mean = 0.36 m s -2 , range = 0.28 -0.43 SD = 0.07, n =7. For both heart rate and activity, raw values pre-recovery were signi cantly greater than those post-recovery (one sided Wilcoxon signed rank test: heart rate, V = 45, p = 0.002, n = 9; activity, V = 28, p = 0.008, n = 7). Table 2: Recovery based on heart rate and activity sensors. Activity sensors with a * suffix indicate acoustic tags. . "No fit" indicates that the long-term component of the time-series did not follow an exponential decline and that an exponential model could not be fitted; "No rec" indicates that it was possible to fit an exponential model to the timeseries but that recovery thresholds and times were not assigned because the fitted exponential model did plateau.

Discussion
The current study showed plateauing of most time-series, indicative of recovery, within the 14 d of the experiment. Two activity tags, F1(Aco) and F2(Aco), however did not show plateauing, suggesting that the tagged sh had not fully recovered in terms of activity during this period. Other time-series showed gentle gradients even after the recovery breakpoint (for example, the F6 heart rate tag) so the de nition of the point of recovery of some individuals as having fully recovered is less robust. However, identi ed breakpoints generally corresponded with systematic changes in the time-series. For example, the breakpoint on the F6 heart rate tag occurred in a trough separating the sharp initial decline over the rst 5.75 d with the gentle gradient afterwards, so it is reasonable to infer that the identi ed breakpoint corresponded to the transition to post-recovery. The modelling approach used here allowed for a consistent method for establishing the time until recovery among a group of time-series. It should be noted however that estimated times until recovery are dependent on modelling approach used. For instance, tting an exponential model to raw-rather than detrended timeseries, or using a different method to establish a breakpoint between pre-and post-recovery parts of the time-series, would yield different estimates. The exponential model used in this study is a well-validated method for modelling physiological recovery (Bartels-Ferreira et al. 2016) but alternative approaches may also be considered (e.g. Svendsen et al. 2020). The sample size of sh in this study was small (N = 9); a larger sample size would allow a better quanti cation of the range of behaviour during recovery and allow better selection of the modelling approach.
The heart rate data suggest that the tagged Atlantic salmon in our study could only be considered fully recovered from the anaesthesia and surgical procedure of intraperitoneal tag implantation after an average of ≈ 4 and up to a maximum of 6 days post-surgery. While some studies have indicated longer recovery times post-tagging (Hvas et al., 2020a), our observations concur with several previous studies that have reported similar lengths of recovery post-tagging as our study (Martinelli et al., 1998;Jepsen et al., 2001;Bridger and Booth, 2003;Brijs et al., 2018;Brijs et al., 2019b). Although some data series from the tagged sh in our study may visually appear to continue declining after ful lling the recovery threshold criteria, these changes were not found to be statistically signi cant. Recovery results based on activity data varied more in the recovery threshold criteria and time to recovery than heart rate, suggesting it might be a less consistent indicator of recovery between individuals. Moreover, both the temporal patterns and absolute values changed less for activity than heart rate between post-tagging and postrecovery periods, implying a lower ratio between the baseline pattern (i.e. circadian variations) and the changes in activity caused by the tagging procedure. Together, these factors suggest that activity may be a less consistent indicator of post-tagging recovery than heart rate, and that heart rate might be a generally more sensitive indicator than activity, especially for post-tagging recovery.
It is also important to note that there were individual variations in the recovery time assessed from heart rate. Although inter-individual variation in recovery time might be an inherent effect one should expect when tagging A. salmon, we did nd that mature sh had a lower heart-rate recovery time than immature sh. However, the low sample size did not provide enough statistical power to robustly test in uences on recovery time, so we recommend further studies with larger sample sizes to increase power in analyses of potential in uences.
Based on these results, we urge caution on using telemetry data collected after anaesthesia and surgery without rst ensuring that the sh are fully recovered (Mulcahy, 2003;Cooke et al., 2011). Biosensors that measure heart rate and/or activity can be potent tools in such evaluations, as they provide quantitative, high resolution data that will be both more consistent, precise and objective in capturing the full postanaesthesia/surgery effects than e.g. comparing behavioural observations of tagged vs. untagged sh.
Alternative parameters that could be used to assess post-tagging recovery in individual sh include blood glucose, lactate or pulse oximetry/ppg. These could provide a more direct assessment of stress levels in salmon, but we are not aware of any commercial electronic tags able to sense such parameters in live sh. Other techniques based on measuring cortisol in faecal matter (Cao et al., 2017) or bioelectric eld monitoring akin to that used by sharks (Kalmijn, 1972) could potentially result in future solutions that could be used evaluate recovery in a less invasive and independent manner, where the sh are monitored before, during and after the procedure. However, these methods are still to be developed to a stage where they can be applied to free swimming sh, at least in large groups under commercial production conditions, and would only be able to provide information on a group level.
All three DST types tested in this experiment appeared to be suitable for applications on Atlantic salmon as all tagged sh provided valid heart rate data. Moreover, the activity proxy computed from the DSTs containing accelerometers were found to be comparable to those measured by the acoustic tags (see Svendsen et al., 2020 for details on this comparison). The lower absolute amplitude of the activity proxies computed from the DST data was probably caused by them sampling at a lower frequency (1 Hz) than the acoustic tags (5 Hz), thereby capturing fewer high frequency components. The surgical procedure used to implant the heart rate tags was much simpler than the procedure needed for multivariate implants recently used in rainbow trout by Brijs et al. (2019a), but was more comprehensive and invasive than that used for conventional intraperitoneal tag placement. It is likely that less complex surgical procedures would lead to shorter recovery times in Atlantic salmon, as previously found for rainbow trout (Altimiras and Larsen, 2000;Gräns et al., 2014). However, it is probably reasonable to be conservative with respect to recovery times, especially if the data are to be used e.g. as a management tool in aquaculture applications or to evaluate stress effects on sh in conjunction with ecological studies. Using data from sh that are still recovering from post-anaesthesia/surgery effects in such applications could result in sub-optimal management decisions or erroneous conclusions that could have rami cations beyond the study itself.
The sh included in the analyses exhibited heart rates that gradually stabilised at daily means between 21 and 26 bpm (daily variations between 15 and 30 bpm, similar to that observed by for adult A. salmon of mean fork length of 62.3 cm at 4˚ C by Lucas, 1994). Due to the similarities across tanks and individuals, this range in heart rate may be typical for Atlantic salmon of this size and with the prevailing temperatures. Moreover, all individuals in tanks 1, 2 and 4 had similar circadian rhythms (higher heart rates during daytime than at night) and gradual post-surgery declines in mean daily heart rate (from more than 30 bpm after surgery, to 21-26 bpm after up to six days). This implies a regularity across individuals that increases the likelihood that heart rate may function as a consistent stress indicator in Atlantic salmon that may be used to assess sh recovery after tagging. The tagged sh in tank 3 were excluded from the study due to inter-individual aggression. These individuals demonstrated measured heart rates that differed from the others both in individual and aggregate values . Although these sh also showed signs of circadian variation in heart rate, the mean value did not appear to decline over the days following tagging, an effect that was attributed to inter-individual aggression, all else being held equal. This may indicate that the stress induced by the aggression between the two males in this tank overrode the stress response due to recovery. A potential interpretation of this is that the aggressive encounters caused chronically elevated stress levels that masked the recovery stress caused by handling, anaesthesia and surgery. This could further mean that recovery stress can be di cult to monitor if the sh are simultaneously in uenced by independent external events, such as individual interactions due to dominance hierarchies (Sloman et al., 2001;Cubitt et al., 2008).
Based on established knowledge on how salmon swimming speeds are affected by variations in light intensity (Oppedal et al., 2011), as well as previous telemetry studies applying similar activity proxies on salmon in sea-cages (e.g. Føre et al., 2018), we expected to see a circadian rhythm in activity where activity was higher during day than at night in the present study. In contrast to these expectations, the circadian trends in the activity of our sh were on average higher during night-time than during day. A similar "inverse circadian" rhythm was observed in salmon reared in sh tanks during the period after tagging by Kolarevic et al. (2016) and could imply that a "normal circadian" activity rhythm may arise only after the salmon have recovered after tagging in tanks. Conversely, the circadian rhythm in heart rate was more like expected (higher during daytime), meaning that the sh displayed generally higher heart rates when measured activity was low than when activity was high. This may seem counter-intuitive as one would expect more active sh to display higher heart rates since salmon tend to display increased heart rates with increased swimming activity (Hvas et al., 2020b). However, it is possible that the higher heart rates during daytime were caused by effects such as feeding activity (Eliason et al., 2008;Gräns et al., 2009) or perceived increased predation risk due to higher light levels (Johnsson et al., 2001). These results are unexpected and very interesting, but further extrapolations and discussions on this matter would probably require further experiments with more data.
Although this study underlines the importance of critical evaluation with regards to recovery from anaesthesia and surgery when using telemetry, the data collected also highlight the importance of telemetry as a method for studying free swimming sh. The heart rate and activity values for all tagged sh eventually plateaued, possibly indicating that they all recovered from the anaesthesia/surgery, and posthumous pathology revealed no in ammations or other apparent morphological signs of reduced welfare due to the surgical procedures. Even though the low water temperatures during the experiment may have led to handling and surgery having less impact on the sh, the tagging procedure used here was more complex than conventional intraperitoneal tagging. It is thus reasonable to conclude that sh carrying telemetry tags can be considered representative members of the group they were selected from once they are fully recovered from anaesthesia and surgery, provided that they were a representative selection to begin with. However, this also requires that the recommendations on ratio between tag size and sh size are not exceeded (e.g. "the 2% rule", Thorstad et al., 2013). Since we worked with adult salmon with a mean weight of 2100 g, and the maximum tag weight carried by the sh was 22.6 g (around 1% of the sh body mass) this was not a challenge in our study.

Future research and potential technological improvements
Since this study only focused on Atlantic salmon exposed to one set of environmental conditions, it is di cult to assess if these concerns are also relevant for other species, and/or sh under different conditions. Similar studies on rainbow trout using the same tag type found that they recovered 72-96 h after surgery (Brijs et al., 2019b), which was shorter than the Atlantic salmon in the present study. Moreover, wounds in Atlantic salmon are known to heal faster in warmer temperatures than in cold water (Jensen et al., 2015), suggesting that the low water temperatures in the present study may have contributed to longer recovery periods. These elements suggest that species speci c effects or differences in external environmental conditions are important to consider when studying recovery times.
Future studies on the relationship between heart rate and post anaesthesia/surgery recovery time should therefore be conducted for other species of interest, across relevant temperature ranges, to obtain a more complete picture of this relationship.
In the present experiment, the sh were kept in groups in small tanks. To investigate how recovery time is affected by eventual scaling effects and social/inter-individual effects arising due to group dynamics, future studies addressing post-tagging effects should be done with a larger number of tagged sh at larger spatial scales. This would also enable a deeper scrutiny into individual variations in recovery, as a higher number of tagged sh would provide a good foundation for nding statistical relationships on the individual level. Although our present results imply that inter-individual variations are a prominent feature in the recovery time of tagged salmon, a larger sample number will be necessary to properly conclude upon the nature of such variations. To increase the relevance of a larger follow-up study, it could be done in sh cages in the marine environment, perhaps rst by using meso-scale size cages containing fewer sh than a commercial cage but at similar densities, and then moving to full-scale studies to cover all steps in the transition from lab to industrial scale.

Conclusion
The main conclusion from this study is that the Atlantic salmon in these experiments required an average of ≈ 4 and up to a maximum interval of 6 days of recovery after anaesthesia and surgery before their heart rates returned to assumed baseline routine values. Moreover, although observation of behaviour and/or activity may alone be insu cient to assess that the sh has physiologically recovered, activity measurements indicated similar recovery periods to those based on heart rate, although there was a longer maximum period of 10 days. We therefore urge caution when using data collected after surgery and anaesthesia in studies using biologging/telemetry tags. Assuming that we want all individuals to be recovered, our study thus implies that only data collected after 6 days recovery time should be used for further analyses. However, this recommendation would only be applicable to studies featuring Atlantic salmon reared in similar experimental conditions as we used. Since recovery time will vary with factors such as sh species, water temperature, invasiveness of the surgery, anaesthesia time, sh density and physical scale, it is di cult to make general recommendations on when one can assume the sh to be recovered from tagging, and the data to be safe for use in biological analyses. However, by conducting experiments similar to the present study where these parameters are varied, a more complete picture of how we need to account for sh recovery after tagging in telemetry studies may be obtained.

Declarations
Ethics approval and consent to participate All sh handling and surgery were made in compliance with the Norwegian animal welfare act and were approved by the Norwegian Animal Research Authority (permit no. 18/18431).

Consent for publication
Not applicable.
Availability of data and materials