Sequential infection can decrease virulence in a fish–bacterium–fluke interaction: Implications for aquaculture disease management

Abstract Hosts are typically infected with multiple strains or genotypes of one or several parasite species. These infections can take place simultaneously, but also at different times, i.e. sequentially, when one of the parasites establishes first. Sequential parasite dynamics are common in nature, but also in intensive farming units such as aquaculture. However, knowledge of effects of previous exposures on virulence of current infections in intensive farming is very limited. This is critical as consecutive epidemics and infection history of a host could underlie failures in management practices and medical intervention of diseases. Here, we explored effects of timing of multiple infections on virulence in two common aquaculture parasites, the bacterium Flavobacterium columnare and the fluke Diplostomum pseudospathaceum. We exposed fish hosts first to flukes and then to bacteria in two separate experiments, altering timing between the infections from few hours to several weeks. We found that both short‐term and long‐term differences in timing of the two infections resulted in significant, genotype‐specific decrease in bacterial virulence. Second, we developed a mathematical model, parameterized from our experimental results, to predict the implications of sequential infections for epidemiological progression of the disease, and levels of fish population suppression, in an aquaculture setting. Predictions of the model showed that sequential exposure of hosts can decrease the population‐level impact of the bacterial epidemic, primarily through the increased recovery rate of sequentially infected hosts, thereby substantially protecting the population from the detrimental impact of infection. However, these effects depended on bacterial strain–fluke genotype combinations, suggesting the genetic composition of the parasite populations can greatly influence the degree of host suppression. Overall, these results suggest that host infection history can have significant consequences for the impact of infection at host population level, potentially shaping parasite epidemiology, disease dynamics and evolution of virulence in farming environments.

Infections from multiple parasites are common also in intensive production environments, where high densities of susceptible hosts favour the spread of virulent pathogens (Kennedy et al., 2016;Pulkkinen et al., 2010). Infections can cause significant economic loss by impairing quality, condition and growth of crops and farmed animals. For example, in aquaculture, parasitic infections are considered one of the most important threats for development of the industry. Similar to natural conditions, parasitic epidemics in aquaculture are typically consecutive with different parasites infecting their hosts in varying timescales. Disease epidemics typically sweep through aquaculture units at different times in response to variation in pathogen ecology and host susceptibility (e.g. cohorts of varying age) (Karvonen, Rintamäki, Jokela, & Valtonen, 2010;Rintamäki-Kinnunen & Valtonen, 1997). This creates favourable conditions for development of cumulative infection history of hosts that can affect virulence in subsequent disease outbreaks. Earlier studies in fish have shown that a prior parasite exposure can influence the outcome of simultaneous re-exposure of the host by multiple parasite genotypes (Klemme et al., 2016), alter associations between parasite species (Karvonen, Seppälä, & Valtonen, 2009), and influence parasite community composition of the host (Benesh & Kalbe, 2016).
However, knowledge of the implications of sequential exposure of hosts to multiple parasites in farming environments is still very limited. Consequently, most infections occurring in intensive farming units are commonly treated instantaneously with very little consideration of previous or existing other infections, which, among aquaculture fish, can range from viruses and bacteria (Mohanty & Sahoo, 2007;Pulkkinen et al., 2010;Skall, Olesen, & Mellergaard, 2005;Tobback, Decostere, Hermans, Haesebrouck, & Chiers, 2007) to protozoans and metazoans (Hakalahti & Valtonen, 2003;Karvonen, Savolainen, Seppälä, & Valtonen, 2006;Rintamäki-Kinnunen & Valtonen, 1997). Regardless, infection history of a host population could influence ongoing epidemics and potentially underlie failures in management practices and medical intervention of diseases.
Here, we studied effects of host sequential exposure on parasite virulence in an interaction between two widely distributed aquaculture parasites, the bacterium Flavobacterium columnare and the fluke Diplostomum pseudospathaceum. Bacterium F. columnare, the causative agent of the columnaris disease, is an opportunistic pathogen and currently considered as one of the most severe disease threats in fish farming . The disease can cause considerable losses if not treated with antibiotics Pulkkinen et al., 2010;Wagner, Wise, Khoo, & Terhune, 2002), which in many cases has resulted in emergence of antibiotic-resistant bacterial strains (Declercq, Boyen, et al., 2013). The trematode D. pseudospathaceum causes local, but significant aquaculture problems by blinding fish (Karvonen, 2012). Unlike F. columnare, infections of D. pseudospathaceum are not transmitted directly between fish, but the life cycle includes three hosts (snail, fish and fish-eating bird) and fish become infected when in contact with the parasite larvae (cercariae) released from infected snails. Infections of F. columnare and D. pseudospathaceum can co-occur in aquaculture fish (Karvonen et al., 2006;Sundberg et al., 2016). They also interact in genotype-specific manner when infecting the host at the same time, which can result in higher morbidity of fish, that is virulence, and higher infection success of the fluke (Louhi, Sundberg, Jokela, & Karvonen, 2015). We first exposed rainbow trout (family Salmonidae) hosts to both parasites in two experiments manipulating the timing between the infections from few hours to several weeks and monitoring the disease-related morbidity of the fish. Based on our previous results on simultaneous infections of the two parasites (Louhi et al., 2015), we expected that both the short-term and the long-term sequences between the infections would result in lower bacterial virulence, possibly depending on the strain-genotype combinations of the parasites.
Similarly, we expected sequential infection to change infection success of the fluke. Second, we developed a compartmental mathematical model capturing the disease dynamics in a host population to explore how sequential exposure of hosts to the two parasites could influence disease-related mortality, and total host population size, of farmed fish. Overall, our results suggest that host infection history can potentially shape parasite virulence over a long time period, which may have implications for evolution of virulence as well as for disease prevention strategies in intensive farming systems.

| Sampling and genotyping of flukes
Lymnaea stagnalis snails (n = 42), intermediate hosts for D. pseudospathaceum, shedding clonal fluke cercariae were collected from Lake Vuojärvi (62°24′54″N, 25°56′14″E), Finland. Fifteen cercariae were collected from each snail and stored individually in Eppendorf tubes in 15 μl of lake water and frozen in −20°C for subsequent microsatellite analysis to identify snails that were infected with one fluke genotype (Louhi, Karvonen, Rellstab, & Jokela, 2010;Reusch, Rauch, & Kalbe, 2004) (Table S2). Parasite DNA was extracted according to Criscione and Blouin (2004). Snails infected with one genotype were stored individually in 1 L of water at 6°C and fed ad libitum with lettuce until the beginning of the experiment. Note that all parasite genotypes are produced sexually in the avian definitive host, which is why all infections in the snails are unique and individual genotypes persist in a host population only for one complete round of the parasite life cycle.

| Experimental exposure 1
Naïve, uninfected juvenile rainbow trout (Oncorhynchus mykiss; age 2.5 months, average length ± SE = 38.23 ± 0.2 mm) were obtained from a hatchery in Central Finland. Fish were maintained in aerated ground water with continuous water flow (17°C) for 4 weeks before the experiments and fed daily with commercial fish food pellets.
Prior to the exposures, the water temperature was raised slowly to 25°C (2°C every second day) to allow fish acclimation to experimental conditions. Three freshly grown strains of F. columnare (1-3, see Table S1) and clonally produced cercaria larvae of D. pseudospathaceum from three single-genotype-infected snails (A-C; see Table S2) were used in the fish exposures. Three hours prior to the exposures, the snails were placed individually in 2.5 dl of water (20°C) and allowed to produce cercariae. Cercarial density from each snail (fluke genotype) was estimated by counting ten times 1 ml subsamples from each container.
A pairwise infection design was then applied to test virulence and intensity of infection across the combinations (Table S3). In the experiment, 20 rainbow trout were exposed individually to single bacterial strains (5 × 10 3 colony-forming units/ml; 3 × 20 fish), single fluke genotypes (50 cercariae/fish; 3 × 20 fish) or co-exposed to both bacteria and flukes in nine different combinations (9 × 20 fish), totalling 300 fish. To explore the effect of short-term sequential infection on virulence, the co-exposure matrix (9 × 20 fish) was replicated together with the simultaneous infections so that each of the nine co-exposure combinations received the fluke first and the bacterium 4 hr later. Bacterial single infections (3 × 20 fish) were also repeated at this time with freshly grown strains to control for possible changes in bacterial virulence. A negative control group of 30 fish receiving pure culture medium and/or water instead of bacteria or flukes, respectively, was also established. Overall, the setup totalled 570 fish (Table S3). The infection doses corresponded to those in natural conditions. For example, fish infected with F. columnare can emit bacterial concentrations that are orders of magnitude higher than those used here (Kunttu, Valtonen, Jokinen, & Suomalainen, 2009) and one infected snail can release thousands of D. pseudospathaceum cercariae per day (Karvonen, Kirsi, Hudson, & Valtonen, 2004;Karvonen, Rellstab, Louhi, & Jokela, 2012). All fish were haphazardly assigned to the different treatment groups (single exposure to F. columnare, single exposure to D. pseudospathaceum, exposure to both parasites) in the simultaneous and sequential exposures.
The exposures and the subsequent monitoring took place in small containers with 500 ml preaerated ground water (25°C). The fish were checked every hour for disease symptoms and morbidity.
Morbid fish that had lost their natural swimming buoyancy and did not respond to external stimuli were considered dead and were euthanized using MS-222 anaesthetic every hour. This gave an accurate estimate of time of death (see Louhi et al., 2015). The fish were immediately sampled for presence of F. columnare on the skin and gills (by culture on Shieh containing tobramycin [Decostere et al., 1997]), and dissected for D. pseudospathaceum in the eye lenses. The establishment of D. pseudospathaceum in the eye lenses takes place within few hours from exposure (Louhi et al., 2015). The dissection protocol was used to determine the exact shape of the time-establishment relationship used in estimation of differences in fluke abundance among the treatment groups (see below). The experiment was terminated at 28 hr post exposure when 87.5% of the fish exposed to the bacterium had died. All surviving fish were subsequently euthanized (MS-222) and examined for bacterial and fluke infection as described above. Bacterial cultures confirming F. columnare infection were incubated at 22°C for 2 days and checked for presence of bacterial colonies.
Data on fish survival were analysed using Cox regression with sequential infection, F. columnare strain and D. pseudospathaceum genotype as fixed covariates, and fish length as a continuous covariate.
Since the bacterial virulence changed slightly during the 4 hr interval between the infections (see results), fish groups exposed only to the bacterium at 0 hr and 4 hr were used as reference categories in the analysis. Thus, the effect of sequential infection on virulence would be seen as a significant three-way interaction between the fixed factors. In addition, we applied analysis of covariance (ANCOVA) to data on fluke numbers in fish eyes to identify factors that affected infection intensity in exposures together with the bacterial strains.
To correct for variation in fluke exposure and establishment time between fish individuals showing different time of survival, we used the residuals of the nonlinear asymptotic regression predicting infection intensity as function of survival time as the response variable (see Louhi et al., 2015).

| Experimental exposure 2
To explore the longer-term effect of sequential exposure, 960 rainbow trout from the same lot as in Experiment 1 were divided into 6 tanks, each with 160 fish and 70 L of water (16°C). Three of the groups were exposed to a total of 480 D. pseudospathaceum cercariae per tank, three cercariae per fish, while the other three groups served as unexposed controls. A low-dose exposure was used to keep the number of parasites establishing in eye lenses low so that the parasite would not influence subsequent fish growth (Karvonen, 2012). Parasite cercariae were produced from two L. stagnalis snails as described above and pooled for the exposure (different genotypes to those used in the re-exposure below (D-F), or in Experiment 1 [A-C]). During the exposure, the incoming water was turned off and was turned back on after 1 hr. As the cercarial infective lifespan is less than 30 hr (Karvonen, Paukku, Valtonen, & Hudson, 2003), parasites that failed to locate a fish, if any, were eventually lost from the tanks.
Fish were then maintained for 5 weeks and fed daily with fish pellets. Possible acquired host responses against D. pseudospathaceum are cross-reactive across parasite genotypes (Rellstab, Karvonen, Louhi, & Jokela, 2013), which minimized genotype-specific responses, if any, between the first infection and the re-exposure (see below). Water temperature was then raised slowly to 25°C as described above. Fish with and without the previous fluke infection were exposed either to single Flavobacterium (strains 1-3, Table S1; 2 × 3 × 20 fish), single Diplostomum (genotypes D-F, note different genotypes to Experiment 1 because of mortality among the snails carrying genotypes A-C, Table S2; 2 × 3 × 20 fish), or to pairwise combinations of the two (nine different combinations; 2 × 9 × 20 fish) ( Table S4). Doses of the bacteria (5 × 10 3 colony-forming units/ ml) and flukes (50 cercariae/fish) at re-exposure were the same as in the first experiment. Each treatment had 20 replicate fish taken randomly from groups of previously unexposed and exposed fish (the three replicate tanks were pooled). A negative control group of 30 fish was also established. The entire setup totalled 630 fish (Table S4). Again, fish were maintained individually and followed for disease symptoms until 28 hr postexposure as described above. All fish that died or survived the experiment were sampled for bacterial presence of the skin and gills, and dissected for the number of flukes.
Flukes originating from the first (three cercariae per fish) and the second (50 cercariae per fish) exposure were separated according to their size.
Data were analysed using Cox regression with initial fluke infection, F. columnare strain and D. pseudospathaceum genotype as fixed covariates, and fish length as a continuous covariate. Analysis of covariance (ANCOVA) was applied to residual fluke numbers as described above. All analyses were conducted in SPSS 24 statistical package. The experiments were approved by Finnish Regional State Administrative Agency (licence number ESAVI/1375/04.10.03/2012), and they conformed to the animal care legislation of Finland. there was no bacterial epidemic, or (b) there was a bacterial epidemic, but no fluke infection. All models were run in R 3.5.1.
The residual number of flukes was significantly different between the fluke genotypes, with genotype explaining large part of the variation ( Figure S1, Table S5). Sequential infection did not affect the residual parasite numbers overall, but had an effect depending on the bacterial strain, fluke genotype and their interaction ( Figure   S1, Table S5). This suggests that the administration of the bacterium four hours later also affected the fluke numbers in strain-genotypespecific manner.
Again, the effect depended on the bacterial strain and was evident with strains 2 and 3 when co-exposed with the fluke (Figure 2, Table 1).
However, in single bacterial exposures, the effect of decreased virulence with the prior exposure to flukes was consistent across all strains (Wald = 5.472, p = .019; Figure 2). Single fluke infection did not cause significant mortality (two fish out of 120 exposed only to flukes died during the experiment). No mortality was observed among the 30 un-  (Table S5).

| Predicted population-level effects of fluke infection on the impact of a bacterial epidemic
Our model showed that the presence of fluke infections can protect the host population from the detrimental effect of a bacterial epidemic, most notably when using parameter values from Experiment 1, which assumed short-term sequential infections ( Figure 3).
However, the magnitude of this protective effect varied considerably across bacterial strain and fluke genotype combinations; bacterial strain 1 (Figure 3, top row) appeared to be the most easily overcome strain, with increasing cercarial force of infection (Γ) leading to progressively higher levels of end-of-season fish abundance.
The magnitude of these effects varied with fluke genotype, from around 40% at the highest levels of Γ examined for fluke genotype A, up to ~90% protection for fluke genotype C. However, population-level protection from the other bacterial strains was negligible, regardless of the fluke genotype. Where population-level protection was observed, (e.g. for bacterial strain 1), this was driven primarily by the increased recovery rate from bacterial infection of sequentially infected fish, as switching this component off (i.e. assuming recovery rates were the same regardless of the individual's prior fluke infection history) resulted in the loss of population-level protection (Figure 4). Running the model using parameter values from Experiment 2, assuming longer-term sequential infections, revealed low levels of population-level protection, and now most commonly observed for fluke genotype D and for bacterial strain 3 ( Figure S2). TA B L E 1 Results of stepwise Cox regression analyses on mortality of rainbow trout co-exposed to three strains of the bacterium Flavobacterium columnare and three genotypes of the fluke Diplostomum pseudospathaceum in all possible combinations in Experiments 1 and 2 Previously, we have shown that the virulence of simultaneous infections of F. columnare and D. pseudospathaceum in fish is determined by complex genotype-specific interactions (Louhi et al., 2015).

| D ISCUSS I ON
Our present results, suggesting both short-term and long-term influences of sequential infections, add yet another dimension to these G × G interactions. Mechanistically, the lower virulence in sequential compared to simultaneous infections could be related, for example, to reduction in the rate of bacterial invasion to the host's body (Louhi et al., 2015), or to higher efficiency of the host's immune system to F I G U R E 2 Mean survival times (±SE) of rainbow trout previously unexposed (Unexp., open boxes) or exposed to Diplostomum pseudospathaceum (Exposed, grey boxes) when reexposed to three strains of the bacterium Flavobacterium columnare Overall, our results add significantly to earlier studies on sequential infections between different parasite taxa, majority of which have used single genotypes/strains (Ben-Ami et al., 2011;Clay, Dhir, et al., 2019;Doublet et al., 2015;Hoverman et al., 2013;Lohr et al., 2010;Marchetto & Power, 2018;Natsopoulou et al., 2015), by emphasizing the importance of variation in infection outcomes depending on the specific G × G parasite combinations. Indeed, combining G × G interactions in multiple parasites with host infection history makes estimation of virulence and virulence evolution increasingly challenging (Karvonen, Jokela, & Laine, 2019). Nevertheless, such interactions  (Karvonen et al., 2006;Stables & Chappell, 1986), even a low-level fluke infection could potentially decrease morbidity and mortality associated with an imminent bacterial exposure. This is consistent with earlier results of less-virulent parasites providing host with at least some degree of protection against later arriving virulent strains/species in plants (Adame-Alvarez, Mendiola-Soto, & Heil, 2014;Seifi, Nonomura, Matsuda, Toyoda, & Bai, 2012;Tollenaere, Susi, & Laine, 2016) and invertebrates (Ben-Ami, Mouton, et al., 2008;Clay, Cortez, et al., 2019;Wuerthner, Hua, & Hoverman, 2017). Similarly, the model on sequential infections 5 weeks apart predicted recovery in some of the parasite combinations, although the magnitude of this protective effect was clearly lower. While we could not compare the two experiments directly because they used different fluke genotypes, the experimental results and the model predictions suggest that the protective effect of sequential fluke infection against the bacterial outbreak might decay with time. If true, this could reflect operation of different short-term and long-term host-level mechanisms such as reduction in parasite facilitation (Louhi et al., 2015) or activation of host immune system (Klemme et al., 2016) (see above). These are promising leads for future studies on the detailed mechanisms underlying the effects of sequential infections on disease epidemiology in this system. Our model assumes single strain-genotype combinations between the parasites whereas in reality aquaculture fish would likely to be exposed to mixed genotypes of both flukes (Rauch, Kalbe, & Reusch, 2005) and bacteria (Kunttu et al., 2012;Sundberg et al., 2016), significantly increasing the number of possible genotype combinations and outcomes of the disease. Thus, the model predictions on effects of sequential infection in specific strain-genotype combinations can be considered as extremes, ranging from no effect to nearly total recovery of the host population. Given the likelihood of multiple strain-genotype infections, the reality is likely to lie somewhere in between these extremes, depending on the genetic composition of the bacterial and fluke populations in the water. Further, the model predictions were driven mainly by host recovery while our experimental data showed that sequential infection between the parasites also prolonged the lifetime of the hosts. While this effect was small in these experimental conditions, where the disease progression from exposure to morbidity is very fast (most fish had died within 28 hr), it could be speculated that the effect of increase in lifetime in aquaculture conditions, where the length of an untreated epidemic is typically several days or even weeks (Räihä, Sundberg, Ashrafi, Hyvärinen, & Karvonen, 2019), could be stronger.
Importantly, these increases in infected host lifetime, though beneficial to the individual host, could have counter-productive effects on the host population as a whole. Similar to host "tolerance" responses to infection (Ayres & Schneider, 2012;Medzhitov, Schneider, & Soares, 2012;Råberg, Graham, & Read, 2009), by keeping infected hosts alive, this prolongs the window of opportunity for infection to other hosts in the population, increasing overall infection prevalence, and resulting in a net decrease in host population abundance (e.g. Vale, Fenton, & Brown, 2014), and, as most clearly seen here, in the absence of any effect of fluke infection on host recovery from bacterial infection (Figure 4). However, the detailed epidemiological consequences of the prolonged lifetime are currently unknown and require further work. Finally, competition/interactions between the bacterial strains (Kinnula, Mappes, & Sundberg, 2017;Sundberg et al., 2016), in interaction with the flukes infecting the hosts simultaneously (Louhi et al., 2015) or sequentially (this study), could also shape the evolution of virulence. For example, as the order of parasite arrival to host can significantly alter the outcome of virulence (Alizon, de Roode, & Michalakis, 2013), factors such as G × G interactions between parasites could be important in determining which virulence genotypes are favoured by selection under each infection scenario (Karvonen, Jokela, et al., 2019). However, while our results support G × G variation in infection outcomes, a formal study of implications of sequential infections on evolution of virulence is beyond the scope of the present model/study.
To conclude, flavobacteria are currently considered among the most important bacterial fish pathogens worldwide. Interactions between increasingly virulent strains of F. columnare (Kinnula et al., 2017;Pulkkinen et al., 2010;Sundberg et al., 2016) and those between the bacterium and other co-occurring parasite species (Bandilla, Valtonen, Suomalainen, Aphalo, & Hakalahti, 2006;Louhi et al., 2015;Xu, Shoemaker, & LaFrentz, 2014) can significantly alter the disease outcomes and influence predictions on virulence of infection. Our results, suggesting protective effect of sequential infections from a less-virulent parasite against a highly virulent one (see also King et al., 2016), provide an interesting viewpoint into how epidemics could be altered by the host infection history. While the epidemiological effects of consecutive parasite outbreaks on subsequent disease occurrence are generally poorly known, results from our model suggest that changes in disease epidemiology with sequential exposure can also have important implications for the need and success of disease management practices and medication protocols. In a wider perspective, possible reduction in use of medication could also constrain environmental discharge of medical residues and pathogen evolution for antibiotic resistance (Cabello, 2006;Martinez, 2009).

ACK N OWLED G EM ENTS
The authors would like to thank Jenni Marjakangas for assistance in

CO N FLI C T O F I NTE R E S T
We have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are archived in Dryad Digital Repository. https ://doi. org/10.5061/dryad.8f57ck3 (Karvonen, Fenton, & Sundberg, 2019).