Louse‐induced mortality thresholds in Atlantic salmon of wild‐origin

Wild salmonid fish are faced with the challenge of infections from the ectoparasitic salmon louse, which thus dictates the management of salmon aquaculture in areas where farms and wild populations co‐exist. The mortality risk of wild Atlantic salmon is used as an indicator of threat levels when monitoring wild populations, and therefore the aim of this study was to robustly investigate the effect of infection intensity on probability of mortality in salmon of wild origin. To assess critical infection levels, recently‐smoltified Atlantic salmon were infected with intensities ranging from 0 to 2.5 lice g−1 and exposed to a 48‐h simulated migration. The first mortalities occurred when preadults appeared and ceased as lice became adults. Infection intensity negatively affected mortality probability and survival time, whereas the simulated migration had no influence on the probability of mortality. Individual‐level effects of infection on host mortality is difficult to quantify in the field; however, the correlation between louse load and mortality probability obtained in this study can facilitate indirect estimates of mortality risk of wild salmon populations and contribute to conservation efforts associated with this parasite.


| INTRODUCTION
The salmon louse (Lepeophtheirus salmonis) is a marine ectoparasitic copepod, commonly found on wild salmonids in the North Atlantic including Atlantic salmon (Salmo salar), sea trout (Salmo trutta) and Arctic char (Salvelinus alpinus).Wild anadromous Atlantic salmon spawn in freshwater and spend the first part of their lives in lakes and rivers, where they undergo smoltification which prepares them for the marine environment before migration to the open sea.In Norway, the body weight of recently-smoltified Atlantic salmon ranges from 10 to 80 g (Thorstad et al., 2011).Smoltified fish migrate through the river toward the open ocean which offers increased opportunity to feed (Klemetsen et al., 2003), but also exposes the fish to increased rates of predation.Predation of smolts has been shown to be especially high in river mouths and estuaries (Thorstad et al., 2012), coinciding with the physiological cost of smoltification that creates an overall high-risk period.The out-migrating Atlantic salmon are largely exposed to parasite infections during their movement through the fjords or coastal waters, as salmon farms that act as reservoirs for the parasite are mainly located here.In the northern Atlantic countries, farming of Atlantic salmon and rainbow trout (Oncorhynchus mykiss) has increased host density and availability, facilitating the dramatic proliferation of salmon louse populations and thus significantly increased infection pressure in coastal areas (Dempster et al., 2021;Krkošek, 2010;Krkošek et al., 2005).
The lifecycle of the salmon louse encompasses eight stages, of which the third stage, the copepodid, is the infective stage.After infecting their host, the copepodid starts to feed and molts through two sessile stages, chalimus 1 and 2, followed by three mobile stages, preadult 1, 2 and the adult stage (Hamre et al., 2013;Johnson & Albright, 1991a).The salmon louse attaches to the fish surface (skin or fins) by a filament throughout the sessile stages and during every molt.The latter mobile stages use their cephalothorax as a suction cup and can move around the surface of the fish.The louse feeds on mucus, epidermis, and in later stages, blood of its salmonid host (Grimnes & Jakobsen, 1996;Heggland et al., 2019).The fish can suffer significant physiological and pathological consequences that are largely dependent on louse abundance and developmental stage.Grazing of the epidermis can cause wounds leading to secondary infections and osmoregulatory issues, and louse infections may affect the immunological capabilities of the fish (Barker et al., 2019;Llewellyn et al., 2017;Øvergård et al., 2023;Ugelvik & Dalvin, 2022;Wootten et al., 1982).Salmonids display a range of responses toward this parasite but many, including all the Atlantic species, have a limited immunological response toward the parasite resulting in low rates of clearance (Braden et al., 2017;Dalvin et al., 2020;Fast et al., 2002;Jones, 2011;Ugelvik et al., 2022).Meanwhile, the effects of sublethal infections, which have implications especially for the fitness of wild individuals, such as reduced growth and condition (Fjelldal et al., 2022;Susdorf, Salama, Todd, et al., 2018), indirect delayed maturation (Vollset et al., 2014), reduced growth of reproduction organs in mature males (Fjelldal et al., 2022), reduced swimming performance (Bui et al., 2016;Wagner et al., 2003), stress tolerance, and maturation timing remains under-scrutinized.
Parasitic infections in fish are expected to pose an energetic burden to the host (Hvas & Bui, 2022) and thus from an ecological point of view, this implies that migrating recently-smoltified Atlantic salmon are most vulnerable.A range of laboratory experiments have been undertaken to quantify the impact of salmon louse infections on survival of wild salmonids.Although these experiments are not directly comparable to field studies, it is generally agreed upon that louse-induced mortality is mainly associated with fish with high infection intensity and the development of lice to the mobile preadult phase (Dawson et al., 1999;Fjelldal et al., 2020;Grimnes & Jakobsen, 1996;Wagner et al., 2008).Multiple field studies suggest that salmon lice are a key driver of Atlantic salmon post-smolt mortality (Bøhn et al., 2020).However, the exclusive effect of louse infections is difficult to tease apart from the many interacting factors that influence survival during the early phase of migration; reduced survival is probably due to both a direct mortality of fish with severe infections, increased risk of predation, and lowered fitness due to diminished growth.Furthermore, a reduced swimming performance (Bui et al., 2016) during the migration could both increase the risk of being infected and predated upon.
The impact of salmon lice on wild salmonid populations due to aquaculture of Atlantic salmon is currently perceived as one of the largest obstacles to sustainable expansion of the industry (Vollset et al., 2018).As the global leader in salmon production, the Norwegian government has implemented a management system ("the traffic light system") to regulate sustainable industry growth and fulfill its conservation responsibilities (Myksvoll et al., 2018).There, monitoring of infection levels in wild salmonids is a key indicator for assessment of industry impact on population fitness (Sandvik et al., 2020).Determination of thresholds inducing physiological effects and mortality in Atlantic salmon is essential for management of this protected species.In Norway, the management of salmon lice in fish farms is based upon numerous indicators of the health of wild salmonid populations, including observed infection levels on wild salmon.The mortality risk of local populations is defined by host tolerance of infections, using a louse index with no assumed negative effects on the host if the louse infection is less than 0.1 lice g À1 , and 100% mortality risk if above 0.3 lice g À1 (Taranger et al., 2015).This index includes mortality and secondary effects as lost growth, which then translates to increased risk of predation, delayed maturation and reduced reproductive output.The threshold for mortality in wild salmon is complicated to determine, as all fish caught in nature are "survivors" as potential louse-induced reduction in the ability to swim or feed will cause infected fish to be quickly eliminated in a natural environment.
The evidence used in creating this risk index is largely based on laboratory experiments using cultivated postsmolts (Taranger et al., 2015), although translating lethal thresholds determined in an experimental tank trial to wild salmonids is challenging.In the sea, salmon must spend energy catching prey while escaping predators and actively swim to migrate to the open ocean, whereas in a tank environment, feed is provided ad libitum and there are no predators present.However, to isolate the lethal effect of louse infection levels, the tank environment offers the possibility to carefully control the infection levels on individual fish and closely monitor the outcome.
In the present study, the goal was to determine mortality threshold levels of louse infection by measuring the effects of different infection intensities and stages of lice.Furthermore, the study was performed on recentlysmoltified salmon from wild parents and included a migration swim trial to better simulate ecologically relevant conditions.

| MATERIALS AND METHODS
The experimental procedure was designed to replicate a typical series of events involving recently-smoltified Atlantic salmon migrating to the sea while experiencing infection with salmon lice.The experiment was carried out in accordance with the regulations governed by the Norwegian Food Safety Authority (application ID 21864 and 12935).All fish were monitored closely throughout the experimental period to detect any development of moribund fish.

| Fish husbandry
Atlantic salmon originating from wild brood stock (Etne River, Norway) were reared at the Matre Research Station, Norway.At the parr stage (December 16, 2019), 600 fish were PIT-tagged (Glass tag 2, 12 mm, TrackID AS, Stavanger, Norway) into their abdominal cavity.The fish were smoltified by increasing temperature and photoperiod (Table 1).The fish were transferred to the experimental tanks (1 Â 1 Â 0.42 m) prior to the infection, which took place 1 week after introduction to full strength seawater.Fish were fed size-appropriate pellets according to standard husbandry procedures.However, a reduced feed regime (40% less feed provided than standard aquaculture protocol) was used to better represent a wild phenotype.
To accommodate sampling logistics, fish were divided into five groups (Groups A-E), with 40 fish per tank (Appendix 1).Each group consisted of three tanks each undergoing an infection procedure with different numbers of lice including sham infections of the control group.All tanks underwent the same timeline for smoltification, infection, and migration simulation; however, the treatment groups were staggered by 7 days.

| Infection protocol
Depending on the intended infection intensity of the tank, the number of lice introduced to the tank was calculated by the intended intensity (in lice g À1 ), assuming 30% infection success and taking into account mean fish weight in the tank.Thus, the five treatment groups were subjected to infection pressures to achieve louse levels ranging from 0 to 2.5 lice g À1 , categorized as control (no lice), low (0.06 lice g À1 ), medium (0.24 lice g À1 ), high (0.48 lice g À1 ), or very high (>0.75 lice g À1 ) (Appendix 1).For example, a Medium infection level tank with 35 fish of 34 g would have 952 copepodids used for challenging the whole tank.
Lice for infection were sourced from offspring of adult lice collected from salmon farms in Austevoll and Matre, Norway.Prior to an infection challenge, egg strings were collected and incubated at 12 C with continuous flow of seawater (Hamre et al., 2009) and hatching date was monitored daily.Two days after molting into the stage, copepodids were enumerated and used for infection challenges.Infections were performed simultaneously for all three tanks in the Group.Water level was lowered to approximately 50%, and in-flow reduced to $6 L min À1 .Copepodids were then added, and left under those conditions for 45 minutes, until the water reached the original level and in-flow was reestablished to preinfection level.Oxygen was monitored throughout the process, and stationary fish were occasionally encouraged to move by waving a hand above the tank.

| Simulated ocean migration
A large custom-made swim tunnel setup (Remen et al., 2016) was used to simulate the seaward migration distance in groups of Atlantic salmon with varying salmon louse infection levels.The chamber of the swim tunnel was cylindrical in shape with controlled current velocities.Propeller speed (rpm), which correlated to desired current velocity (cm s À1 ), was set by an external controller.Current velocities were verified using a handheld velocimeter (Flowtherm NT, Hoentzsch GmbH, Waiblingen, Germany) when the swim chamber was absent of fish.
The system was provided with water of the same quality as the experimental tanks (12 C, 34 ppt), and a constant inflow into the tunnel.Observation of fish in the chamber was accomplished via a camera installed behind the rear grid of the swim section or through the plexiglass top.
The "migration" speed of 50 cm s À1 used for the simulation was derived from the minimum cost of transport (Brett, 1995;Weihs, 1973) as measured on similar sized fish in a pilot trial.Approximately 75 fish (3 tanks) were tested simultaneously; as each group included tanks of different infection pressure levels (Appendix 1), swim tunnel runs contained a common garden group of fish with varying infection levels.As lice were at the late copepodid or chalimus 1 stage, host transfer was not expected during the common garden period.
One day prior to the migration simulation (Table 1), the fish were sedated, identified by their PIT, measured, and louse abundance recorded, and transferred by hand to the swim tunnel.From each tank, 25 fish were then transferred to the swim tunnel (migrators), and the remaining 10 fish were returned to their experimental tank (non-migrators).Fish were left in the tunnel overnight to acclimate at a minimal current speed (6 cm s À1 ).
When the migration simulation began, the velocity was increased to 15 cm s À1 , then a further 10-15 cm s À1 at 5 min increments until the objective speed of 50 cm s À1 was reached.The simulated migration lasted for 48 h, corresponding to a total distance of 86.4 km, where conditions were held continuous throughout the period and fish were monitored closely.
At the conclusion of the simulation, a very light sedative (0.001 g L À1 metomidate hydrochloride) was added to the water to assist with collection of the fish from the chamber and transfer back to their original tanks.For fish that did not undergo the simulated migration, a similar procedural control treatment was applied to replicate the handling process.

| Assessments of louse infection
Abundance, determination of louse stage, and louse sex (at the preadult and adult stages) were recorded for each fish three times (Figure 1): at 3 dpi (COP sampling), 11 dpi (CH2 sampling), and either if an individual survived (final sampling; 25-35 dpi) or at death (mortality sampling; 15-30 dpi).Loss of lice and host switching is frequently observed in laboratory experiments (Bui et al., 2018) and hence the purpose of the extensive monitoring was to obtain information about the number of lice on individual fish enabling assessment of the total parasite burden throughout the experimental period.Infection assessments were conducted with fish in shallow white plastic tubs holding sedated water, with additional light sources by experienced louse enumerators.
Infection level was expressed as lice per gram of host weight.Thus, the reported measure of infection in this study is infection intensity, calculated by Ln Fw À1 where Ln is louse number (of any stage) and Fw is weight (g) of the fish at the time of counting.Retainment of lice throughout the experiment period was calculated for individual fish.

| Monitoring moribund status and mortality
The higher levels of infection were expected to induce mortality in salmon, and thus fish were monitored closely to minimize suffering and to determine the time of death as accurately as possible.Due to ethical concerns, fish were euthanized when severe wounds were evident, or when salmon became moribund, defined as abnormal swimming behavior, change in color, or loss of equilibrium.These humane thresholds indicate that the fish would have died even without intervention.Thus, the date of euthanasia or when an individual was found dead (i.e., rapid deterioration not captured by monitoring efforts) was considered the "mortality dpi."Moribund or severely wounded fish were netted and euthanized with an overdose of sedation.

| Data handling and statistical analyses
At every sample, louse abundance, stage, and sex (mobile stages only) and fish weight, pit-ID was recorded.For all analyses, data on individual fish were used to capture the variability in infection intensity, with tank included as a factor.Salmon were categorized by their endpoint: Control groups were fish that did not experience infection, survivor groups were those individuals that remained alive until the termination of the experiment, and mortality groups were those that died or were euthanized due to severe infection.Across groups, control, and survivor samples were collected on equivalent respective days of the timeline; however, mortality final samples were collected at various times according to their individual survival duration (Figure 1).
Infection intensity values were transformed to produce more consistent dependency between the transformed variable and survival rates.In particular, the intensity variable "lice g À1 " was transformed using multiple approaches (see Appendix 2); the transformation log (√(1 + lice_intensity)) exhibited the most evenly distributed survival rates among all transformations and was used in all relevant analyses.For simplicity, transformed values will be still referred to as infection intensity.
Salmon that underwent the simulated migration were termed migrators, and those that were procedural controls were non-migrators.To determine whether the simulated migration influenced mortality risk of infection salmon, endpoint status (i.e., survivor, mortality) was compared with migration status (i.e., control-migrator, migrator) using a binomial-distribution GLMM model (Model A, Table 2).
Notably, we approached the question of survival probability using two different models to generate a doseresponse curve with infection; one that used infection intensity as measured at 11 dpi (Model B), and another that used an average infection intensity that incorporated the three louse counts, to better reflect the total louse burden as loss of lice occur during the infection (Model C, Table 2; Appendix 2).The dataset was analyzed using the two approaches due to the robustness of the 11 dpi chalimus counts in that all fish were assessed at the same dpi, whereas the last sample occurred at different time points (time of death or at the end of the experimental period) with uncertainty associated with the attrition of mobile F I G U R E 1 Louse development and experimental events.The x-axes illustrate days postinfection (dpi) and predicted louse development is indicated by the boxes; stages are copepodid (COP), chalimus 1 (CH1) and 2 (CH2), preadult 1 males (PAM1) and females (PAF1), preadult 2 males (PAM2) and females (PAF2), and adult males (AM) and females (AF).As development in a cohort of lice is slightly asynchronous, stages overlap indicating that more than one stage can be found on a fish.lice over time.Mortality is likely influenced by the individual's infection "history," and so the second approach used a GAMM that fits a nonlinear smoothing term to capture this infection level over time (Appendices 2 and 5).
The relationship between infection intensity and survival time (i.e., dpi until death) was restricted to the Mortality group and thus was analyzed separately from survival probability using a truncated negative-binomial distribution (Model D, Table 2, Appendix 2).
All analyses were conducted in the R software environment (R Development Core Team, 2021) using the packages coxme, survival, ggplot2, ggfortify, mgcv, and gamlss.tr.An overview of the statistical analysis performed is summarized in Table 2, and described in detail in Appendix 2. Akaike information criterion values were used in model selections to determine best fit.

| Infection levels over time and across groups
Initial infection with L. salmonis across all fish resulted in a comprehensive spread of infection intensities from 0 to 2.5 lice g À1 at 3 dpi (Figure 2).At 11 dpi, louse abundances were highly correlated to individuals' 3 dpi levels (linear model: slope = 0.82, R 2 = .91),with negligible number of lice lost through the simulated migration process.Maximum intensity at 11 dpi was 2.4 lice g À1 , with only four individuals harboring more than 2.0 lice g À1 .
As lice progressed to mobile stages, fish with high parasite loads reached moribund status or died (mortality group).Infection intensities recorded at 15-30 dpi ranged from 0.06 to 2.44 mobile lice g À1 .An increase in the number of lice after 11 dpi occurred in 22% of fish from the mortality group, indicating transfer of mobile lice from other fish, while 76% of the group lost lice during this period.For the survivor group, there was a larger loss of lice from the 11 dpi count to the final count, with a maximum intensity of less than 0.5 lice g À1 at the final time point (Figure 2).These infection intensities translated to maximum abundances of 128 copepodids (3 dpi), 110 chalimi (11 dpi), and 88 (mortality group) or 22 (survivor group) mobile lice per fish.Louse intensities at last count were strongly correlated with the levels recorded at 3 and 11 dpi (slope 0.72, R 2 = .78and slope 0.84, R 2 = .80,respectively).There were relatively equal ratios of male and female lice on the mortality fish at time of death (Appendix 3).

| Effect of migratory experience on mortality
There were no mortalities during or in the days after the simulated migration (4-6 dpi).Fish that experienced the simulated migration exhibited moderate fin erosion (on the dorsal and pectoral fins) compared with the fish that remained in the tank, but no other phenotypic or behavioral differences were observed.Then, 71% of the experimental fish underwent the simulated migration, and there was no effect of this experience on risk of mortality (Model A: χ 2 (1) = 0.806, p = .37;Appendix 4).Further, the migration variable was not included after model selection in the GLMM survival analyses (Model B-see the following section on survival; p = .87).

| Mortality analysis: Probability of survival with infection intensity
No mortalities occurred in the control fish, and no mortalities occurred in the infected salmon while lice were still in the sessile stages (0-14 dpi).As such, all mortalities were associated with the development of lice to mobile preadult stages.During the observation period after the simulated migration, fish were removed from the experiment for three reasons: death, moribund status, or due to severe wounds.The latter category included fish that developed lethal wounds without concomitant development to moribund status or death.Wounds developed somewhat differently in groups of fish with low louse load, which were characterized by wounds in the head region, whereas fish with higher louse load typically harbored severe wounds on their dorsal-anterior region and peduncles.Nearly all fish with 0.4-0.5 and >0.5 lice g À1 (96 and 100%, respectively) died, while 0.3-0.4lice g À1 caused mortality in 77% of fish.On the contrary, 85% of fish with less than 0.3 lice g À1 survived (Figure 3a).To describe the relation between lice numbers and the survival of salmon, probability models of mortality to predict the effect of louse infection were derived.Two models were created, one GLMM (Model B) based on the number of lice recorded at the chalimus stage (11 dpi), and one GAMM (Model C) which encompassed the lice load throughout the experimental period using both the 11 dpi and the final count.Both approaches described a strong negative relationship between louse intensity and survival probability.

| Model B. Survival based on chalimus stage intensity (infection intensity at 11 dpi)
In the final model, only louse intensity significantly affected survival of the fish (SE = 3.09, Z = 3.38, p < .001),whereby the correlation between survival and infection intensity was: The regression would then predict >99% mortality risk for salmon infected with more than 1.6 lice g À1 (based on the CH2 counts).The corresponding GLMMbased prediction plot of survival with infection intensity is shown in Figure 3b.

| Model C. Survival based on infection intensity over the entire trial period
Similar to the GLMM, louse intensity increased risk of mortality (edf = 3.30, Ref.df = 4.15, χ 2 = 64.68,p < .0001),but not among Groups tested (edf = < 0.001, Ref.df = 4.00, χ 2 = 0.00, p = .563).Estimates from the GAMM predicted >99% mortality risk in fish hosting more than 0.38 lice g À1 ; the corresponding model-based prediction plot shows that the decay of survival appears to occur on average at lower (transformed) louse intensities than when based on 11 dpi counts (Figure 3c, compared with Figure 3b representing the GLMM plot).The GAMM results also confirmed that the probability of mortality is linearly related to the louse infection intensity, validating the transformation of the infection intensity variable.

| Model D. Survival duration with infection intensity
The first fish death was recorded at 15 dpi when the female lice had developed to the preadult 1 stage and ceased by 30 dpi when female lice had become adults (Figures 1  and 4).At 15 and 16 dpi, male lice were approximately equally distributed between the preadult 1 and 2 stage, after which male lice were predominantly in the preadult 2 stage until 19 dpi when the first adults were observed (Figure 1).Mortalities occurred first in fish with high infection intensities: between 15 and 18 dpi (corresponding to preadult 1 stage of females) deaths were observed almost exclusively in fish with parasite loads of ≥1.0 lice g À1 at time of death (Figure 4).After 20 dpi, all Mortality fish with ≥1.0 lice g À1 had died, and fish with lower louse intensities <1.0 lice g À1 died only from 17 to 30 dpi (Figure 4).
The final model, based on mortality fish data (i.e., excluding all surviving fish), included only the predictor louse intensity which strongly influenced survival rate (SE = 0.02, Z = À6.62,p < .001),with the regression indicating that with every 0.22 lice g À1 increase of louse intensity, the survival time (in days) was reduced by 10% (95% confidence interval: 6.85 and 12.24%; Figure 4).

| Probability of mortality and intensity thresholds
In this study, we characterized the correlation between infection intensity and probability of mortality in  recently-smoltified wild-type Atlantic salmon subjected to a simulated seaward migration, after having been infected with L. salmonis with infections ranging from 0 to 2.4 lice g À1 .The survival chance of Atlantic salmon was 50% when infected with 0.43 sessile lice g À1 , while mortality probability approached 100% when infection was above 1.6 sessile lice g À1 based on 11 dpi counts, and 0.38 lice g À1 based on average number of lice over the entire experimental trial.The lethal thresholds reported here are generally in agreement with previous studies when considering variations in fish sizes and discriminating between sessile and mobile louse numbers.Grimnes and Jakobsen (1996) reported a lower lethal threshold in 40 g fish (0.75 preadult lice g À1 ), but when backcalculating their threshold value to the likely intensity of chalimus stages in the same fish (as per their own calculations of louse loss; also see Wagner et al., 2008), the thresholds become similar at $1.7 sessile lice g À1 .In slightly larger 60 g fish, Finstad et al. (2000) found infection intensities of 1.6 preadult lice g À1 in mortality fish compared with 0.58 preadult lice g À1 in surviving fish at 25 dpi (at 8-10 C).Dawson et al. (1999) observed no mortality in fish with approximately 0.2 lice g À1 and noted that at the last sampling when lice were adult, physiological parameters and skin erosion indicated that the fish had recovered.Nolan et al. (1999) performed experiments on much larger fish and applied preadult and adult lice directly: they found no mortality but physiological changes at 0.04 lice g À1 .More recently, Fjelldal et al. (2020) reported mortality in salmon infected with 0.6 preadult lice g À1 , which aligns with our observations of 100% mortality in salmon with >0.5 lice g À1 based on the integrated louse intensity over the experimental period.
The migration simulation did not affect mortality risks, which shows that infected Atlantic salmon are well equipped to perform their migratory swim even when enduring additional challenges.However, and more importantly, the elaborate experimental design used here also provided novel and detailed insights into infectionlevel dependent mortality risks as well as effects on growth and physiology in Atlantic salmon of wild origin.The distance of the simulated migration (≈80 km) was based on the natural migration route of Atlantic salmon from the Etne River, which runs into the Hardangerfjord situated in southern Norway.Test speed was derived from the minimum cost of transport, which is the optimum migratory speed that also coincides with observed swimming speeds of wild salmonids (Brett, 1995;Drenner et al., 2012;Weihs, 1973).As such, exposing young Atlantic salmon to a realistic physical challenge should allow for an ecologically relevant assessment of how lice impact wild populations during this critical stage of their lifecycle (Thorstad et al., 2012).

| Effect of infection intensity and louse stages on survival time
Mortality probability and rate was directly influenced by infection intensity, whereby the first mortalities occurring at 15 dpi in the most infected individuals, and ceased after 30 dpi with fish harboring lower parasite abundances.Infection intensity becomes more complex to interpret over time owing to natural attrition of lice once they become mobile on the host as preadults.Intensity levels of mobile lice are thus not fixed during the mortality period (15-30 dpi) and cannot be used to compare between fish that died at different times.The analyses therefore used the counts conducted at 11 dpi, when all fish were assessed and still had the same louse stage, to predict survival times.The resulting model indicated a $10% reduction in survival days with increasing intensities of 0.2 lice g À1 .This study was performed at 12 C to represent a realistic temperature for an out-migrating Atlantic salmon in spring-early summer (Godwin et al., 2020).However, survival days will likely decrease with increasing temperature as louse developmental time accelerates in warmer waters (Hamre et al., 2019).
Survival probability of sea trout has recently been estimated in a field study by releasing infected fish into a fjord and monitoring their fate via acoustic tracking (Serra-Llinares et al., 2020).Also, other parasites in salmonids have been studied using truncation in macroparasite distributions to identify the threshold of parasite intensity, above which coho salmon (Oncorhynchus kisutch) suffer significant mortality (Ferguson et al., 2011).Otherwise, apart from reporting time of death or percentage mortalities, no previous work has correlated mortality rate with linearly increasing salmon louse infections in a salmonid.Our results therefore fill a significant knowledge gap for this hostparasite system as highlighted by Wagner et al. (2008).
Louse-induced fish mortality encompassed three subcategories in this study: death, moribund, or euthanasia due to severe wounds.This may have affected the recorded survival time as moribund or wounded fish might have died at a later day if not removed from the experiment.However, ethical obligations restricted any extended suffering in these animals.Still, all mortality fish would presumably have survived slightly longer in their controlled tank environment with sterilized seawater than in nature, where fish in the wild are continuously exposed to pathogens and are hence at risk of secondary infection.
Because of the experimental design and also the host-parasite system, we cannot tease apart the effects of the louse stages and survival time, and thus the intensity-dependent effect of stages.Further, linked to the progression through stages is the natural attrition of lice on the host, resulting in decreasing infection intensities over time within an individual; for those fish that had initially moderate infection intensities but survived, their parasite loads were greatly reduced.Hence, for survivors, we cannot determine whether survival was due to the reduction in intensity over time.Louse loss occurs on salmon when lice reach the mobile stages (Fjelldal et al., 2020), but in experimental settings it can be exacerbated through handling or procedures (Bui et al., 2020).In addition, there may be slight differences in virulence between male or female lice through their size and behavior.Mortality was mainly observed as all females became preadults at 16 dpi, but whether this relates to females being more virulent, or the fact that development of preadult females coincides with a larger number of preadults on the fish as the population leaves the chalimus phase, is not possible to determine (Hamre et al., 2019).Here, there was a general conversion of sex ratio to 50% on mortality fish (Appendix 3) after initial mortalities, suggesting it is the infection intensity that is important.However, in the field, the larger adult females are more likely to remain on the host while males are more likely to change hosts to optimize reproductive opportunity (Connors et al., 2011;Johnson & Albright, 1991b;Stephenson, 2012).
No mortalities were recorded when lice were at the sessile chalimus stages in this study, corroborating that the mortality-inducing effects are not incurred before lice molt to the preadult stage (Bjørn & Finstad, 1998;Finstad et al., 2000).However, as mentioned above, because of the intrinsic shift between stages over time, it is unknown whether the preadult stage themselves cause this lethal effect or whether the total duration of infection might incur long-term responses that leads to the observed mortality.Also, fish mortality may be an indirect result of changes in exocrine excretions from the louse that may modulate host response (Øvergård et al., 2016).A future study might clarify the cause of fish mortality by applying infection intensities of >1 preadult louse g À1 to small uninfected salmon, or maintaining long durations of only chalimi stages (i.e., repeated infections and removal of mobile lice as they appear) to determine stage-dependent impacts of severe infections.
After 30 dpi, no further mortalities were recorded, and surviving fish retained a maximum of $0.4 sessile lice g À1 at this time.Of the survivors, the maximum intensity at 11 dpi was 1.7 lice g À1 , well within the higher thresholds of mortality risk range, showing that some individuals were able to tolerate this level of infection and eventually recover.Although we cannot exclude any potential louse-induced mortality that could have occurred beyond the experimental period, a second wave of mortality appeared unlikely, through the number of remaining lice and welfare condition of the fish.This is supported by a similar pattern that was observed by Dawson et al. (1999) where reductions in appetite, skin pathology, and disturbance in blood parameters subsided after 30 dpi even though fish were still infected.The potential for natural recovery from louse infections is an encouraging finding and will be crucial for success during the sea-faring phase of Atlantic salmon.However, surviving fish may still suffer lasting effects of louse infection such as an increasing basal energetic burden (Hvas & Bui, 2022), which may affect growth and ultimately impair fitness (Birkeland, 1996;Skilbrei et al., 2013;Susdorf, Salama, Todd, et al., 2018).

| Knowledge contribution to conservation management
The specific thresholds for louse-induced mortality were generated to be able to assess mortality risk of wild salmon populations (Taranger et al., 2015), and are used with substantial implications to management strategies by government authorities (Myksvoll et al., 2018).However, the small number of empirical studies on lice pathogenicity that were used to estimate mortality thresholds by Taranger et al. (2015) are known to have inadequacies in their estimates.Previous reports of threshold levels did not use experimental designs from which a survival curve could be produced.Here, we generated a range of infection intensities that covered sublethal and lethal levels to determine thresholds for parasite-induced mortality, under laboratory conditions.The resulting GLMM and GAMM regressions are analogous to dose-response curves, which allow for calculations of predicted survival chance with infection intensity, thus providing the first complete estimate of mortality risk based on sessile or total louse infection intensities in Atlantic salmon.The index aligns surprisingly well with predicted mortality risk from this study when using overall infection ("total lice") calculations from Model C, for each risk level (Table 3).Similarly, a recent meta-analysis calculated the 50% mortality risk at 0.24 lice g À1 , whereby most studies included only recorded infection intensity of preadult and/or adult lice (Ives et al., 2023).
In practice, when developing predictions in an applied context, the choice of which model to use from this study-that which was based off chalimus infection intensities (Model B), or overall infection (Model C)needs to consider the nature of host infections and potential disparity in lice loss over time in the wild compared to in tanks.Specifically, will the fish only have a single louse stage (where the chalimus-based GLMM model would be appropriate), or is there the potential for a varying infection demographic (where then the GAMM model is more suitable)?There should be caution in interpreting mortality risks estimated by these two models as the former intrinsically produces a higher threshold.For example, if the traffic light system were to adopt the outputs of the chalimus-based Model B, then risk assessments of wild populations would be vastly different, as the upper threshold of 100% mortality would change from 0.3 lice g À1 (Taranger et al., 2015) to 1.6 lice g À1 (Table 3).Thus, the nuances around louse development and attrition on wild fish and the nature of monitoring programs, as well as other factors such as sea temperature and fish size, must be taken into account when utilizing these results.Furthermore, these models were generated from laboratory conditions where fish had high louse attrition and did not experience challenges that would otherwise reduce fitness in wild individuals, thus our results are likely an overestimation of survival probability and duration.
The predictive value of our survival models will significantly benefit conservation management decisions that require information on the population-level impact of salmon lice on wild populations.Detecting parasiteinduced mortality in nature is notoriously difficult and relies on an assumption of the hosts' tolerance to low levels of salmon lice.These data will help strengthen conservation efforts by providing relevant authorities robust scientific grounds for defining sustainable infection pressures in coastal environments subjected to intensive sea cage aquaculture.In the Norwegian traffic light system, several methods are used to estimate the salmon louse levels on out-migrating post-smolts, such as virtual smolt models (Johnsen et al., 2021;Kristoffersen et al., 2018) and trawl surveys, and are in addition informed by estimates of louse density as by use of sentinel cages and a coupled biological-hydrodynamic model (Sandvik et al., 2020).This empirical study can be used in combination with field surveys and modeling approaches to estimate the parasite-induced mortality rate in wild populations (Johnsen et al., 2021;Susdorf, Salama, & Lusseau, 2018).
This study provides evidence for the expected risk of mortality and survival duration due exclusively to salmon louse infections in Atlantic salmon, with size relevance to wild out-migrating salmon.Combined with modeling efforts and indirect estimates of population-level infection rates (through field surveys), these data can facilitate the assessment of population health of wild salmonids through estimations parasite-induced mortality rates.While exceedingly significant for the authorities managing sea lice epidemics in salmon farming countries, this type of experiment provides data that can contribute to understanding population-and landscape-level effects of a parasite on host mortality in a significant wildlife system (i.e., Wilber et al., 2020).
AUTHOR CONTRIBUTIONS Samantha Bui and Sussie Dalvin conceptualized the study and were assisted by Per Gunnar Fjelldal, Malthe Hvas, and Ørjan Karlsen who participated in design and methodological planning.Samantha Bui, Malthe Hvas, and Sussie Dalvin performed the experimental investigation.Samantha Bui performed the formal analyses of the experimental data with input from Per Gunnar Fjelldal, Malthe Hvas, Ørjan Karlsen, and Sussie Dalvin.Samantha Bui and Sussie Dalvin drafted the manuscript, which was revised by Per Gunnar Fjelldal, Malthe Hvas, and Ørjan Karlsen.All authors approved the final manuscript.
T A B L E 3 Comparison of mortality risk of salmon louse infections as calculated from this study (of chalimus or total louse infections, from Models B and C, respectively) and the previous risk index values (as presented by Taranger et al., 2015).

F
I G U R E 2 Louse infection intensities (lice g À1 ) for individual fish, each represented by a gray line, over the experimental period (DPI; days post infection).Blue dotted lines indicate dpi when all fish were sampled (at COP and CH2 stages).Survivors (left panel) (N = 212) were sampled at 3, 11, and 25-35 dpi while Mortalities (right panel) (N = 152) were sampled at 3, 11, and 15-30 dpi depending on their survival duration.

F
I G U R E 4 Survival time (days postinfection, dpi) of fish with transformed louse infections intensities (lice g À1 ), with the predicted survival curve shown, as calculated from Model D. Note that this dataset only includes individuals that died during the trial period (mortalities).

F
I G U R E 3 (a) Proportion of survival of the cohort in relation to their louse intensity at last sample, plotted from raw data.Dark gray sections indicate Survivors (1), while light gray indicates Mortalities (0).Shown on the secondary y-axis is the proportion of the lice intensity group that survived.Bin widths are proportional to the N fish for each density group.(b) GLMM-based estimation of the survival depending on transformed louse intensity (at 11 dpi) = the GLMM prediction model curve (including 95% confidence intervals).(c) GAMM plot of predicted survival depending on transformed louse intensity per day.Transformation of louse load data: log(√(1 + lice_intensity)).
Outline of actions in the experimental design and the number of days between each activity.
Summary of statistical analyses performed.