 Research Article
 Open Access
 Published:
Control of endemic swine flu persistence in farrowtofinish pig farms: a stochastic metapopulation modeling assessment
Veterinary Research volume 48, Article number: 58 (2017)
Abstract
Swine influenza viruses (swIAVs) are known to persist endemically in farrowtofinish pig farms, leading to repeated swine flu outbreaks in successive batches of pigs at a similar age (mostly around 8 weeks of age). This persistence in European swine herds involves swIAVs from European lineages including H1_{av}N1, H1_{hu}N2, H3N2, the 2009 H1N1 pandemic virus and their reassortants. The specific population dynamics of farrowtofinish pig farms, the immune status of the animals at infectiontime, the cocirculation of distinct subtypes leading to consecutive or concomitant infections have been evidenced as factors favouring swIAV persistence within herds. We developed a stochastic metapopulation model representing the cocirculation of two distinct swIAVs within a typical farrowtofinish pig herd to evaluate the risk of reassortant viruses generation due to coinfection events. Control strategies related to herd management and/or vaccination schemes (batchtobatch or mass vaccination of the sow herd and vaccination of growing pigs) were implemented to assess their relative efficacy regarding viral persistence. The overall probability of a coinfection event for France, possibly leading to reassortment, was evaluated to 16.8%. The export of consecutive piglets batches was identified as the most efficient measure facilitating swIAV infection fadeout. Although some vaccination schemes (batchtobatch vaccination) had a beneficial effect in breeding sows by reducing the persistence of swIAVs within this subpopulation, none of vaccination strategies achieved swIAVs fadeout within the entire farrowtofinish pig herd.
Introduction
Swine influenza A viruses (swIAVs) are polymorphic enveloped singlestranded RNA viruses from the Orthomyxoviridae family widespread in pigproduction units throughout the world [1]. These viruses are of great economic importance for the swine industry because of their involvement as a major cofactor of porcine respiratory disease complex [2, 3] and being also responsible in some cases for severe pulmonary distress leading to growth retardations. In the past, mainly sporadic outbreaks, affecting a large part of the herd population in a relatively short timeinterval but with shortterm consequences at the herd scale, were reported. However, in recent years endemic forms of influenza infections have been increasingly documented with a global persistence of swIAV viruses at the herd scale, systematically affecting successive batches of growing pigs [4, 5]. As such, the burden of respiratory diseases in growing pigs due to bacterial coinfections is increasing and causes alarming use of antibiotics.
Three main subtypes have been circulating in swine populations worldwide: H1N1, H1N2 and H3N2 [6,7,8]. Those subtypes permanently evolve, ending in different lineages containing genetic components derived from both avian and human influenza A strains. To date, endemic swIAVs persistence in European swine herds includes H1_{av}N1, H1_{hu}N2, H3N2, the 2009 H1N1 pandemic virus and their reassortants. Weak crossimmunity between subtypes and the rapid spread of swIAVs within herds might cause multiple, and possibly concomitant, infections in one animal [4, 8, 9]. Coinfection events can lead to the emergence of reassortant viruses, potentially more pathogenic for animals and/or transmissible to humans, and are therefore recognized as a main threat for veterinary and public health [10,11,12,13,14,15].
Although the understanding of swIAVs transmission in pig populations is clearly pivotal to manage the risk of spillover to humans, Dorjee and collaborators highlighted the gap of knowledge on influenza dynamics at the pig farmlevel [16]. Combining epidemiological, viral and immunological characteristics of swIAVs, mathematical models are comprehensive tools to analyze how the population dynamics and immunity influence the dynamics of influenza A virus infections [17,18,19,20]. To date, four modeling studies focusing on swIAV dynamics in swine production units have been published. Reynolds et al. [19] developed a deterministic model to assess the impact of vaccination strategies (e.g. mass and prefarrowing vaccination) on the spread of influenza A infection in breeding and finishing herds based on parameter estimations from experimental conditions. Unless the vaccine and spreading strains were fully homologous, no vaccination strategy was able to eradicate the virus. This result is consistent with field observations where vaccination has only limited effect on the infection dynamics at the herd scale. Vaccination is commonly performed in breeding animals with two main objectives: the reduction of clinical expression in gestating sows and further delivery of maternallyderived antibodies (MDAs) for clinical protection of the offspring. Although the first objective is globally achieved, a recent study by Cador et al. [17] highlighted an ambiguous role of MDAs on the transmission dynamics. The presence of maternal immunity in young piglets was evidenced to extend the duration of the epidemics within batches, which in turn favored the transmission of the infection from batch to batch increasing therefore the persistence of swIAV at the population level. Pitzer et al. [18] used a stochastic model to analyze critical herd sizes for swIAV persistence according to herd type and management practices (variation of the betweenbirth interval and between two introductions of finishers) and showed that the swIAV was able to persist in relatively small populations. White et al. [20] recently proposed a stochastic model representing the infection dynamics of swIAV in a typical US farrowtowean production unit. The authors evaluated intervention strategies based on different vaccination schemes, biosecurity measures and management options and confirmed the role of piglets in swIAV persistence in breeding herds. These four modeling studies highlight the complex relationship between population dynamics, immunity and transmission dynamics of swIAV among a pig population, considering only one subtype.
Cocirculation of different swIAV subtypes within one swine herd is regularly observed in field conditions [4, 5] but has not yet been covered in modeling studies to date. Such phenomenon could increase the likelihood of virus persistence at the herd level triggering the risk of coinfection, possibly leading to reassortant viruses. In the present study we extended the stochastic eventbased metapopulation model from Cador et al. [17] to represent cocirculation of two swIAVs in different farrowtofinish pig farm settings in order to identify drivers responsible for viruses persistence at the herd level. The model was also used to assess the risk of coinfection events and to evaluate the impact of control strategies on the transmission dynamics and persistence of swIAVs at the herd scale based on (1) the implementation of different vaccination schemes and (2) the concurrent export of weaning piglet batches.
Materials and methods
Model structure
Population dynamics
Two subpopulations—breeding sows and growing pigs—are considered. Animals are subdivided into batches according to their physiological state (breeding sows) or age (growing pigs). The breeding sows iterate through three physiological states (service [32 days], gestation [82 days], lactation [26 days]), while growing pigs pass through three stages (lactation [21 days], nursery [51 days] and fattening [105 days]). Batches of animals are managed independently and housed in facilities according to their physiological or growing stage. Breeding sows’ facilities are divided in rooms containing batches having the same physiological stage. Growing pigs’ facilities are divided in rooms containing one single batch of animals with movements occurring at fixed times according to allinallout management policy at the room level. Direct physical contacts between batches of sows and growing pigs only occur in the farrowing room, during the lactating stage. More information and a flow diagram of the farrowtofinish pig farm can be found in Cador et al. [17]. The model was implemented in Matlab (MATLAB 2012b, TheMathWorks, Inc., Natick, Massachusetts, United States).
The original model considered one batchrearing (denominated BR thereafter) management with seven batches (7BR system) and was extended to account for several BR systems commonly used in the field [21]. Each BR system leads to specific population dynamics and herd structure with different numbers of batches and corresponding rooms in the different facilities of the herd (Table 1). Thus, in the BR systems with short betweenbatch intervals (10 and 20BR systems; 2 or 1 weekinterval, respectively), a high number of batches are housed in the same facility at the same time.
Epidemiological model
Starting from a susceptibleinfectedrecoveredsusceptible (SIRS) model, the epidemiological model in this study also accounts for specific features such as maternallyderived antibodies (MDAs) in neonates (lower susceptibility to infection) and sequential or concurrent infections with two different subtypes. The infectious process is ruled by the Gillespie’s direct algorithm [22], where each random event corresponds to a health transition of a single animal. Due to the fastacting transmission between individuals, the batch was selected as epidemiological unit. No efficient crossprotection after sequential infections by two different subtypes has been evidenced neither in field nor experimental conditions [4, 8]. Therefore, no crossprotection has been implemented in the model allowing the animals to be coinfected by the two viral subtypes simultaneously or consecutively. swIAV infection states and health transitions are presented in Figure 1, denoting the swIAV subtypes by subscripts 1 and 2. Subsequent infections are explicitly included in the model to represent realistic durations of shedding periods and the development of strainspecific immunity. Animals can be infected by the second subtype while still shedding the first subtype (I _{12} or I _{21} classes) or after recovery from the first subtype given the absence of crossimmunity (Y _{2} or Y _{1} classes). Strainspecific immunity durations were assumed gammadistributed with a mean duration of 180 days [17]. Sows infected by a second subtype (Y _{2} or Y _{1} classes) and recovered from the first one (R _{1} or R _{2} classes) are assumed to develop an immune response regarding the second strain (R _{12} or R _{21}). The duration of dual immunity was restricted to 90 days after which immunity to the first subtype waned while specific immunity to the second strain persisted for 90 days (total immunity duration of 180 days). In consequence, breeding animals can be reinfected by the same subtype after waning immunity (Figure 1). Given the short lifespan of growing pigs (180 days), the loss of immunity towards a specific subtype with a subsequent reinfection by the same virus was deemed extremely unlikely. Hence, growing pigs were assumed to experiment only one infection for each subtype and develop specific immunity lasting for their economic lifespan. Waning of active and maternal immunity is represented by the stage approach [23,24,25]. The piglets’ immune status after colostrum intake is determined by the health states of the farrowing sows (denoted with superscript i). Recovered sows (R ^{i}_{1} , R ^{i}_{2} , R ^{i}_{12} , R ^{i}_{21} and R ^{i}) give birth to M ^{i} piglets; Y _{1} and Y _{2} sows to M ^{i=1} piglets and S, I _{1}, I _{2}, I _{12} and I _{21} sows to fully susceptible piglets S. Piglets with MDAs (M ^{i} classes) were assumed to be partially protected against infection through a reduced susceptibility [26]. This protective factor was considered strainspecific based on the proportion of farrowing sows immune to each subtype.
Merging this epidemiological model with the population dynamics resulted in an eventdriven epidemiological stochastic model embedded in specific population dynamics.
Implementation of vaccination in the swIAV infectious model
The trivalent vaccine used in European countries covers the majority of the circulating strains in Europe [27]. In our model, vaccination provides partial protection against the two circulating subtypes and is ineffective when applied to actively shedding animals (I and Y classes) because the immunity onset is established after 7 days at least after primary vaccination (RESPIPORC FLU3, summary of product characteristics, [28]). Vaccinated susceptible and immune animals (S and R classes, respectively) develop vaccineinduced immunity against both viral subtypes and are represented by V classes. Waning of vaccineinduced immunity was assumed gammadistributed and modeled using 7exponential classes (stage approach). Vaccinated pigs have a reduced swIAV transmission probability and a lower shedding period [29]. In addition, piglets with high antibody levels (M ^{i=1} to M ^{i=3} stages) were assumed not to react to vaccination due to interference between passive and vaccineinduced immunity in young piglets [30]. Conversely, the vaccination of piglets with lower antibody levels (M ^{i=4} to M ^{i=7} stages) induced an increased duration of the vaccineinduced immunity according to the antibody decay of the piglets. Vaccinated animals are assumed to experience at most one infection by either subtype after which a longlasting immunity to both subtypes is established until slaughter age [31].
Force of infections
As swIAV transmission occurs mainly through pigtopig contact and the exposure to contaminated aerosols, the force of infection applied to each animal includes (1) the direct transmission from infected animals within the room and (2) the indirect transmission through airborne route from infected animals within the entire facility. Subtypespecific forces of infection are calculated combining direct and indirect transmission routes. Parameter definitions and values are provided in Table 2.
The withinroom direct transmission force of infection λ ^{direct}_{ i } (t, r) for subtype i, at time t in room r is defined as:
I _{ i } represents the number of animals infected by subtype i only, I _{ ij } and I _{ ji } the number of animals infected by both subtypes simultaneously (accounting for the sequence of infections) and Y _{ i } the number of animals infected by subtype i while immune against j. I ^{v}_{ i } corresponds to the number of animals infected by subtype i while vaccinated. β and β _{ v } denote the related transmission rates per day for nonvaccinated and vaccinated pigs, respectively. Transmission rates for both subtypes are assumed to be the same.
The betweenroom airborne force of infection λ ^{indirect}_{ i } (t, r) for subtype i, at time t in room r is expressed from the total prevalence of infected animals by subtype i at time t in neighbouring rooms \(r^{\prime}\):
Here, β _{ air } denotes the transmission rate by airborne route and N _{ r’} the total number of pigs in the other rooms \(r^{\prime}\) of the same facility. This airborne force of infection is also applied to susceptible animals during transfer from one facility to another [17].
Therefore, the global force of infection λ (t, r) for a given subtype i at time t in room r is:
The force of infection for animals with MDAs is assumed to be \(\varepsilon *\lambda_{i} \left( {t,r} \right)\) using a reduced susceptibility factor ɛ [26].
Initialisation and study design
Subtypes were introduced separately with a lagtime of 20 weeks. The first subtype was introduced in a fully susceptible and demographically stable herd by the importation of a shedding gilt during the replacement process in the first batch between the farrowing and service room. Assuming no crossimmunity between the two subtypes, the population remained fully susceptible to the second subtype. Introduction of the second subtype was performed as described for the first subtype. Simulations were run for 5 years after the second virus introduction. For each scenario, 200 simulations are performed to capture the variability induced by stochastic processes while keeping a reasonable simulation time.
Impact of the BR system on swIAV persistence
The impact of the different BR systems has been evaluated regarding the time to swIAVs fadeout and the probability of coinfection events. The latter risk was approximated as the proportion of days with coinfections on the total number of days with infected animals for each BR system (daily probability of coinfection). Finally, the global probability of coinfection events was assessed by combining the proportion of simulations with coinfections with the daily probability of coinfection occurrence, weighted by the relative proportion of each BR system in the population in France [21].
Implementation of control strategies
We tested 13 different combinations of vaccination schemes and piglet batch export. Each scenario has been simulated for two extreme BR systems in terms of batch populationsize and time interval between batches (5 and 20BR).
Vaccination
Vaccination is implemented 3 months after the introduction of the second subtype.
In the present study, three vaccination schemes were considered:

Batchtobatch vaccination of the breeding sows. Vaccination is implemented in the gestating room 15 days before farrowing on all animals from the batch expecting to farrow. Hence vaccination time is ruled by the physiological status of the sows and the different batches are desynchronized in terms of boost vaccine immunity. This vaccine strategy aims at inducing a high antibody level in colostrum (prefarrowing vaccination) and further transfer to piglets.

Mass vaccination of the breeding sows. Vaccination is implemented every 3 or 4 months for all breeding sows present in service, gestating and farrowing rooms at the same time, in order to reduce infection pressure in breeding sow facilities.

Batchtobatch vaccination of the breeding sows and growing pigs. In addition to batchto batch vaccination of the sows, growing pig vaccination is implemented in the five first batches entering the nursery from the beginning of the vaccination program to reduce the infection pressure in growing pig facilities. The same scenario was also tested on the five first batches entering the finishing rooms.
Export of batches of weaned piglets
Batch export is implemented 3 months after the introduction of the second subtype. We tested the export of one batch of weaned piglets at a regular interval (every 24 weeks). The time interval between 2 exports was chosen to represent the export of a whole batch of weaned piglets to an external weantofinish site reared on an allin allout principle. The export of consecutive batches (2 or 4 batches according to the BR system) has also been tested.
Statistical analysis of scenarios outputs
The efficiency of the different control strategies was evaluated as regards the probability of swIAVs fadeout within the herd. Time to swIAVs fadeout was studied using survival analysis comparing survival curves corresponding to different strategies using logrank test. When conditions of proportional hazards assumption were met, a Coxproportional hazard model was used to estimate Hazard ratios (HR) and compare control scenarios to the baseline (no measure implemented).
Results
Description of simulations after introduction of the two subtypes
Virus introduction via an infectious gilt (on D0 and D140) caused an initial peak in the number of infected sows due to the fully susceptible population (Figure 2). Transmission events to growing pigs occurred in the farrowing site, triggering the virus spread into the nursery and the finishing facilities. After an initial large outbreak in growing pigs and breeding animals, virus persistence was observed at the herd level due to the constant introduction of susceptible animals and immunity decay. However, sporadic fadeout periods were alternatively observed in growing pig and breeding sow subpopulations. These periods remained of relatively short durations due to virus transfer from one subpopulation to the other during betweenfacility movements (example at D780 in Figure 2).
In growing pigs and at the batch level, swIAVs infections of piglets occurred on successive batches at a similar age (i.e. around weaning). Consecutive batches showed similar outbreaks (Figure 3, three consecutive batches reared in a 10BR system) however different patterns could be observed regarding the cocirculation of subtypes: (a) the infection by the subtype i closely followed by the infection by subtype j (or vice versa), allowing coinfections of the piglets during a short overlapping timeinterval (Figure 3A); (b) the strict concomitant infections by the subtypes i and j, inducing a moderate number of infections by each subtype separately but a great number of coinfected piglets (Figure 3B); (c) the infection by the subtype i in young piglets followed by the infection of subtype j a few months later in finishing rooms, inducing two distinct outbreaks (Figure 3C). In each scenario, between 60 and 75% of the piglets were infected at the epidemic peak.
Impact of the BR system on swIAVs dynamics
Impact of the BR system on the global withinherd persistence
Survival analysis of swIAV persistence at the herd level showed a low probability of infection fadeout up to 5 years after introduction (Figure 4). In the absence of external reintroduction and for all BR systems, at least one swIAV subtype was found to persist for more than 3 years with a probability of 60% whatever the BR system. Differences between BR systems were however significant (p < 0.001, logrank test). For BR systems with short intervals between batches (10 and 20 batches with a 14 and 7days interval respectively), we observed a systematic endemic persistence. For these BR systems, only 8 out of 400 simulations showed stochastic fadeout before virus transmission while the infectious process lasted up to 5 years (simulation time) in all the other simulations (Table 3). Coinfections in growing pigs occurred in 84% of the 200 simulations in herds managed according to 10 and 20 batchrearing systems.
The two BR systems with the largest betweenbatch intervals (4 and 5BR with a 35 and 28days interval respectively) showed similar behavior with a 10% fadeout probability in the first months after introduction followed by a slow decay of persistence probability throughout the simulationtime. While displaying a higher fadeout probability than the other BR systems in the two first months after introduction due to the longest inbetween batches intervals, the average probability of persistence after 5 years was evaluated to 61 and 86% for herd managed according to 4 and 5BR systems, respectively. Fortyfive to 46% of simulations resulted in coinfection events in growing pigs, reducing the probability of coinfection by 1.8 when compared to intensive batchrearing systems (10BR and 20BR). The cocirculation of both subtypes was nevertheless still more frequent than the circulation of a unique subtype in these BR systems. The 7BR system (21day interval) showed an intermediate behavior regarding the coinfections events at the herd level, occurring in 49% of simulations (Table 3). Although swIAVs were more likely to persist during the first year after introduction compared to the 4 and 5BR systems, the 7BR system showed a continuous decay over time reaching the lowest probability of swIAV persistence after 5 years postintroduction (35% on average; data not shown). The order of subtype introduction did not have any effect with similar proportions of the simulations showing the circulation of only the first or the second subtype, irrespectively of the BR system (Table 3).
Assessment of the frequency of coinfection events according to the BR system
Cocirculation of both subtypes can lead to coinfections (green lines, Figure 2). The BR system had a significant impact on the presence of cocirculations at the herd level and coinfection at the individual level (Table 3). Likewise, the probability of coinfection events was significantly different between the BR systems (Kruskal–Wallis test, p < 0.001) (Figure 5). As such, the median occurrence of coinfections was 5.4, 8.1 and 16.4% for the 4, 5 and 7BR systems, respectively, compared to 58.8 and 91.9% in the 10 and 20BR systems. The 10BR system presented the highest dispersion in the occurrence of coinfections. When accounting for the relative proportion of each BR system in the French pig herd populations, the overall probability of a coinfection event possibly leading to reassortment was 16.8%.
Evaluation of control measures
Impact of vaccination on swIAVs persistence in breeding sows
The implementation of vaccination 3 months after the introduction of the second subtype induced a significant rise of fadeout probability in breeding sows in the 5BR system while having no impact on the persistence in breeding sows reared in 20BR systems (see Additional file 1).
In the 5BR system, all vaccination schemes significantly increased swIAV fadeout probability (HR = 4.3 [3.1–5.9] for batchtobatch vaccination, HR = 1.6 [1.2–2.1] for mass vaccination, Cox proportional hazard model, p values < 0.05). No significant difference was observed in regards with mass vaccination schedule (3 or 4 monthinterval) and outputs from these two vaccination schemes have been merged for further analyses. Batchtobatch vaccination led to a rapid decrease of swIAV persistence probability a few months after implementation while the mass vaccination had a limited impact at that time. Mass vaccination reduced swIAV persistence probability at regular intervals but was still less efficient in sows 4 years after introduction compared to the batchtobatch vaccination.
Impact of vaccination on swIAV persistence at the herd level
Although vaccination increased swIAV fadeout probability in breeding sows, the effect on the global persistence of swine flu was not reflected at the herd level. In the 5BR system, although a global difference was found between the tested scenarios (Figure 6), when each vaccination strategies was compared to the reference ‘No control measures’, no significant differences were found (Cox proportional hazard model, p > 0.05). Vaccination strategies had no impact on global swIAV persistence in the 20BR system (data not shown).
Impact of the association of vaccination and batch export on swIAV persistence at the herd level
In the 5BR system, the export of piglet batches in vaccinated herds increased the probability of swIAV fadeout compared to vaccination alone (Figure 7A). The concurrent export of two successive batches led to a higher probability of fadeout than the export of two batches 24 weeks apart, e.g. within the batchtobatch vaccination scheme: HR = 13.0 [7.6–22.0] vs. HR = 6.2 [3.65–10.5], respectively, taking the “No measure” scenario as the baseline (Cox proportional hazard model, p values < 0.05). No significant difference was observed between the vaccination schemes when combined with export of weaned piglets (Figure 7A). In the 20BR system, the simultaneous export of four consecutive batches was required to achieve a probability of swIAV withinherd fadeout of 14% (Figure 7B).
Discussion
In this study, we extended a validated modeling framework to investigate swIAV transmission dynamics of two subtypes within a farrowtofinish pig herd and to evaluate control strategies to prevent endemic persistence. Our model simulations showed an almostmechanic repetition of swIAV outbreaks with infections in successive batches of piglets at a similar age. This pattern has also been shown in field conditions [4] by studying three endemically infected herds. Cocirculation of two distinct subtypes (H1_{av}N1 and H1_{hu}N2) was observed at the batch level and/or at the individual level, resulting in two distinct outbreaks or episodes with coinfections and reassortant viruses. They also identified a faster spread of the virus in pigs in finishing rooms compared to nursery rooms. Simulations from our model are consistent with this behaviour with a sharper epidemic peak in the three outbreaks occurring in older animals although the transmission characteristics between both subtypes were the same in our model. When two influenza outbreaks affect animals consecutively (slight overlap), the order of subtype infection changed from batch to batch. This phenomenon has not been shown in field conditions to date but could contribute to the understanding of repeated infections from batch to batch.
The present model highlighted the difficulty in containing transmission once a virus is introduced within the farm, consistently with White et al. [20]. Indeed, independently from the batchrearing system, at least one of the two introduced viruses was still circulating 5 years after a unique introduction in 78% of the simulations. However, some differences were observed according to the BR systems. The 10 and 20BR systems, characterized by short betweenbatch intervals and a large herd size (430 and 620 sows, respectively), inducing a huge number of animals in nursery and fattening facilities, showed a systematic persistence of swIAVs after introduction. However, both rearing systems displayed different occurrence of coinfections. Although both subtypes were circulating in the majority of the cases, a higher variability was associated with the 10BR system due to an important number of simulations with a long persistence of a unique subtype. Moreover, the occurrence of coinfection events appeared sporadic as compared to 20BR system when both subtypes were circulating within the herd (median duration of the presence of both subtypes: 767 vs. 1831 days in the 20BR system) with temporary switch between subtypes. However, because of the size of the population and the frequent introduction of susceptible animals (every 2 weeks) in this BR system, the persistence of a unique subtype for 5 years occurred more frequently than in BR systems with larger betweenbatch intervals.
The batchrearing systems with the largest betweenbatch intervals (4 and 5 batches) have intrinsic specific characteristics that favour swIAV fadeout on top of the interval duration between batches. Fadeout of the virus in 10% of the simulations in the month following the introduction was due to the particular structure of the pig herds reared with the 4BR system. Indeed, there is a unique batch in service room in herds reared with this system. Therefore, as sows are housed in this facility during 33 days and the virus was introduced at the entrance of the animals within the room, the epidemic outbreak could resume before the end of this period. Thus, recovered sows entering in gestating rooms couldn’t initiate the infectious process in other batches. The fadeout of the virus in around 10% of the simulations 1 month later in pig herds reared in the 5BR system was also likely due to the structure of the herd. Although two batches are housed in the service room at the same time helping the transmission of the virus in the gestation room, the duration until the first entrance of pregnant sows in farrowing room is 23 days (only 9 days in the 4BR system), yielding to the possible termination of the infectious process in the gestation room with a less likely spread to the farrowing rooms.
Once the virus introduced in growing pigs, a higher persistence was observed in the 4 and 5BR compared to the 7BR system probably due to the batch size (48 and 49 vs. 29 sows per batch respectively) and a different weaning age (21 vs. 28 daysold), which could favour fadeout in farrowing rooms before nursery entrance. The persistence could also be enhanced in the 5BR system by the larger number of animals in the herd (245 total sows vs. 192 and 203 in the 4 and 7BR systems). A modeling study focusing on the circulation of one influenza strain corroborates this effect of population size on swIAV persistence [18]. The choice was made to represent each BR system with the corresponding average herd size from statistics available at the country level [21]. Another option could have been to control for herd size and to evaluate only the impact of the herd organization and the timeinterval between batches. However, this would have led to unrealistic situations as the choice of the BR system is governed partly by the herd size (large herds are generally reared using a 10 or 20BR system). Moreover, representing realistic situations showed that the timeinterval between batches is not the only determinant of swIAV persistence but the batch size and possibly the weaningage also have an impact on infection dynamics.
The development of a model representing cocirculation of two swine influenza strains within pig herds had not been carried out before while models on the cocirculation of influenza A virus strains in humans exist [32, 33]. The model developed by Moghadas et al. [32] allowed consecutive infections of individuals assuming a partial crossimmunity after a first infection. In our model, no crossimmunity between subtypes was included as all the available experimental [34, 35] or field data [4] suggested an extremely limited crossreactivity between the main subtypes known to endemically coinfect swine herds in the EU context (H1_{av}N1 and H1_{hu}N2). Zhang et al. [33] modelled the possibility of coinfections and the generation of a reassortant strain at a rate function of the days with coinfections. Only coinfection events were represented in our model as data were lacking to parameterize the likelihood of reassortant generation in case of coinfection in the swine context. Experimental in vitro, ex vivo and in vivo studies would be required to further parameterize this phenomenon. However, using our coinfections event as a proxy and combining the proportion of simulations with coinfections with the daily probability of occurrence, weighted by the relative proportion of each BR system in the population, the probability of coinfection events (16.8%) was found higher than the actual proportion of reassortant strains reported by the national surveillance system of swine influenza infections in French herds (3–5% according to years; [36]). The difference might be due to the probability of successful virus reassortment, which is not accounted in our analysis.
Control strategies were evaluated based on two BR systems selected among the five tested. According to the global spread of the virus, two subsets of BR systems were distinguished with similar behaviors obtained with the 4, 7 and 5BR systems, and with the 10 and 20BR systems, respectively. To assess the impact of control measure, the 5 and 20BR systems resulting in the longest swIAV persistence has been selected within each subset.
A key finding of this study was the selection of the export of piglet batches as the most effective measure favoring infection fadeout. The number of batches to export to obtain a significant effect depended on the BR system. Indeed, the export of four consecutive batches was required in the 20BR system while the export of a single batch in the 5BR system was sufficient to observe a significant decrease of swIAV persistence. The theoretical export of four consecutive batches appears difficult to set up in field conditions but highlights the necessity to insert “gaps” in the growing pigs facilities in order to block the swIAV infection dynamics. In this type of large herds with short betweenbatch intervals, a solution could be to breed the minimum of batches on the main breeding site and to export, if possible, other batches to another weantofinish site. Indeed, lower swIAV persistence has been observed in specialized growing sites compared to farrowtofinish pig herds [18, 37]. The consecutive export of piglet batches also led to an increased virus extinction compared to a regular export (24 weeks apart), suggesting a synergic effect on the infection disruption of consecutive exports. This is probably due to the longer period without introduction of a new cohort of potentially susceptible piglets in nursery.
Our model indicates that none of current vaccination strategies were sufficient to eliminate influenza in farrowtofinish pig herds although it is a frequent measure implemented for swIAV control in pig herds [38]. Similarly to other modeling studies [19, 20], mass vaccination as well as batchtobatch vaccination alone did not significantly reduce swIAV withinherd persistence. Although vaccination was found to favour swIAV fadeout in breeding sows in large betweenbatch interval BR systems, it had no impact on the global swIAV persistence. Indeed, even when sows were vaccinated, the viruses could still circulate and be maintained in the growing pig subpopulation and be further reintroduced in the breeding part of the herd when sows and piglets have contact (e.g. weaning stage). Usually carried out at the end of gestation, the aim of the batchtobatch vaccination is to protect sows against reproductive disorders and to deliver maternal immunity to neonatal piglets. However, the adverse effect of maternal immunity previously highlighted [17] suggested thinking flu vaccination in swine operations differently. Incidentally, the batchtobatch vaccination of sows, although allowing a 10% infection fadeout probability after implementation compared to the control scenario, did not affect the probability of fadeout afterwards, probably because the impact was counterbalanced by the adverse effect of maternal immunity in piglets. The mass vaccination was expected to confer a herd immunity to the sows [39] but the model showed a better efficiency of the batchtobatch vaccination in breeding sows. This might be due to the systematic boost of immunity in sows a couple of weeks before farrowing preventing the relaunch of the infectious process in the breeding herd by growing pigs. White et al. [20] showed that early weaning of piglets after 0–7 days of age reduced endemic prevalence in farrowtowean units. Because such earlyweaning is not allowed in Europe (Council Directive 91/630/EEC), we evaluated the export of 3weekold piglets aiming at breaking the infectious process in growing pig population in combination with vaccination schemes. When the export of batches was implemented, the mass vaccination appeared more helpful to decrease swIAV withinherd persistence. Conversely, no differences between vaccination schemes were observed in the case of the concurrent export of consecutive batches, probably because of the limited effect of vaccination within the large impact of the export. The additional vaccination of five consecutive batches (deemed as the maximum affordable) of growing pigs or finishers did not significantly increase swIAV fadeout and could not mimic the effect of growingpigs batches exports.
In the present model, the vaccination of the animals induced a tenfold lower transmission rate and a twofold reduction of the duration of the shedding period compared to fullysusceptible animals [29]. To the best of our knowledge, no quantitative data on the amount of virus shed by vaccinated animals have been published to date. Thus, this reduction of transmission could be due to a reduction of susceptibility in piglets having vaccineinduced immunity. A tenfold lower susceptibility has been tested with the present model but model outcomes as regards vaccine impact were not modified (data not shown). However, parameterization of the vaccine effect was made using the only data available and corresponding to a US vaccine evaluated in front of a US strain challenge. Further data should be collected on vaccine efficacy regarding the reduction in transmission and/or susceptibility to infection in the EU context to consolidate our conclusions.
swIAV endemic persistence in farrowtofinish herds was shown in this study to be determined by multiple characteristics, which are not independent. Hence the choice of a BR system involves a specific herd structure, subpopulation sizes and betweenbatch time intervals. They all participate to a different degree to the persistence of the infectious process. As such, the advantage of long intervals between batches is possibly counterbalanced by large subpopulations. The observed chronic persistence of swIAVs at the herd level in field conditions and the difficulty to eradicate the infection once introduced even using different vaccination programs can be understood in the light of the present study. Control and progressive eradication of the infection requires combined vaccination programs adapted to the BR system in association with rearing practices aiming at introducing gaps in the growing part of the herd and an adequate separation between the breeding and the growing part of the herd to prevent reactivation.
References
 1.
Van Reeth K, Brown IH, Olsen CW (2012) Influenza virus. In: Zimmerman JJ, Karriker LA, Ramirez A, Schwartz KJ, Stevenson GW (eds) Diseases of Swine 10^{th}. WileyBlackwell, Iowa, pp 557–571
 2.
Deblanc C, Robert F, Pinard T, Gorin S, Quéguiner S, GautierBouchardon AV, Ferré S, Garraud JM, Cariolet R, Brack M, Simon G (2013) Preinfection of pigs with Mycoplasma hyopneumoniae induces oxidative stress that influences outcomes of a subsequent infection with a swine influenza virus of H1N1 subtype. Vet Microbiol 162:643–651
 3.
Fablet C, MaroisCréhan C, Simon G, Grasland B, Jestin A, Kobisch M, Madec F, Rose N (2012) Infectious agents associated with respiratory diseases in 125 farrowtofinish pig herds: a crosssectional study. Vet Microbiol 157:152–163
 4.
Rose N, Hervé S, Eveno E, Barbier N, Eono F, Dorenlor V, Andraud M, Camsusou C, Madec F, Simon G (2013) Dynamics of influenza a virus infections in permanently infected pig farms: evidence of recurrent infections, circulation of several swine influenza viruses and reassortment events. Vet Res 44:72
 5.
SimonGrifé M, MartinValls GE, Vilar MJ, Busquets N, MoraSalvatierra M, Bestebroer TM, Fouchier RA, Martín M, Mateu E, Casal J (2012) Swine influenza virus infection dynamics in two pig farms; results of a longitudinal assessment. Vet Res 43:24
 6.
Brown IH (2000) The epidemiology and evolution of influenza viruses in pigs. Vet Microbiol 74:29–46
 7.
Kyriakis CS, Rose N, Foni E, Maldonado J, Loeffen WLA, Madec F, Simon G, Van Reeth K (2013) Influenza A virus infection dynamics in swine farms in Belgium, France, Italy and Spain, 2006–2008. Vet Microbiol 162:543–550
 8.
Van Reeth K, Brown IH, Durrwald R, Foni E, Labarque G, Lenihan P, Maldonado J, MarkowskaDaniel I, Pensaert M, Pospisil Z, Koch G (2008) Seroprevalence of H1N1, H3N2 and H1N2 influenza viruses in pigs in seven European countries in 2002–2003. Influenza Other Respir Viruses 2:99–105
 9.
SimonGrifé M, MartínValls GE, Vilar MJ, GarcíaBocanegra I, Mora M, Martín M, Mateu E, Casal J (2011) Seroprevalence and risk factors of swine influenza in Spain. Vet Microbiol 149:56–63
 10.
Hiromoto Y, Parchariyanon S, Ketusing N, Netrabukkana P, Hayashi T, Kobayashi T, Takemae N, Saito T (2012) Isolation of the pandemic (H1N1) 2009 virus and its reassortant with an H3N2 swine influenza virus from healthy weaning pigs in Thailand in 2011. Virus Res 169:175–181
 11.
Howard WA, Essen SC, Strugnell BW, Russell C, Barass L, Reid SM, Brown IH (2011) Reassortant pandemic (H1N1) 2009 virus in pigs, United Kingdom. Emerg Infect Dis 17:1049–1052
 12.
Ma W, Lager KM, Vincent AL, Janke BH, Gramer MR, Richt JA (2009) The role of swine in the generation of novel influenza viruses. Zoonoses Public Health 56:326–337
 13.
Kitikoon P, Vincent AL, Gauger PC, Schlink SN, Bayles DO, Gramer MR, Darnell D, Webby RJ, Lager KM, Swenson SL, Klimov A (2012) Pathogenicity and transmission in pigs of the novel A (H3N2)v influenza virus isolated from humans and characterization of swine H3N2 viruses isolated in 2010–2011. J Virol 86:6804–6814
 14.
Liu Q, Ma J, Liu H, Qi W, Anderson J, Henry SC, Hesse RA, Richt JA, Ma W (2012) Emergence of novel reassortant H3N2 swine influenza viruses with the 2009 pandemic H1N1 genes in the United States. Arch Virol 157:555–562
 15.
Vincent A, Awada L, Brown I, Chen H, Claes F, Dauphin G, Donis R, Culhane M, Hamilton K, Lewis N, Mumford E, Nguyen T, Parchariyanon S, Pasick J, Pavade G, Pereda A, Peiris M, Saito T, Swenson S, Van Reeth K, Webby R, Wong F, CiacciZanella J (2013) Review of influenza A virus in swine Worldwide: a call for increased surveillance and research. Zoonoses Public Health 61:4–17
 16.
Dorjee S, Revie CW, Poljak Z, McNab WB, Sanchez J (2016) Onehealth simulation modelling: a case study of influenza spread between human and swine populations using NAADSM. Transbound Emerg Dis 63:36–55
 17.
Cador C, Rose N, Willem L, Andraud M (2016) Maternally derived immunity extends swine influenza A virus persistence within farrowtofinish pig farms: insights from a stochastic eventdriven metapopulation model. PLoS One 11:e0163672
 18.
Pitzer VE, Aguas R, Riley S, Loeffen WLA, Wood JL, Grenfell BT (2016) High turnover drives prolonged persistence of influenza in managed pig herds. J R Soc Interface 13(119):20160138
 19.
Reynolds JJ, Torremorell M, Craft ME (2014) Mathematical modeling of influenza A virus dynamics within swine farms and the effects of vaccination. PLoS One 9:e106177
 20.
White LA, Torremorell M, Craft ME (2017) Influenza A virus in swine breeding herds: combination of vaccination and biosecurity practices can reduce likelihood of endemic piglet reservoir. Prev Vet Med 138:55–69
 21.
Agriculture chamber of Brittany Region (2017) Résultats des élevages de porcs en Bretagne–2014. Gestion Technique des Troupeaux de TruiesChambre d’agriculture Bretagne. http://www.bretagne.synagri.com/ca1/PJ.nsf/TECHPJPARCLEF/25099/$File/RésultatsPorcs2014VF.pdf?OpenElement
 22.
Gillespie DT (1977) Exact stochastic simulation of coupled chemical reactions. J Phys Chem 81:2340–2361
 23.
Lloyd AL (2001) Realistic distributions of infectious periods in epidemic models: changing patterns of persistence and dynamics. Theor Popul Biol 60:59–71
 24.
Vergu E, Busson H, Ezanno P (2010) Impact of the infection period distribution on the epidemic spread in a metapopulation model. PLoS One 5:e9371
 25.
Wearing HJ, Rohani P, Keeling MJ (2005) Appropriate models for the management of infectious diseases. PLoS Med 2:e174
 26.
Cador C, Hervé S, Andraud M, Gorin S, Paboeuf F, Barbier N, Quèguiner S, Deblanc C, Simon G, Rose N (2016) Maternallyderived antibodies do not prevent transmission of swine influenza A virus between pigs. Vet Res 47:86
 27.
Simon G, Larsen LE, Dürrwald R, Foni E, Harder T, Van Reeth K, MarkowskaDaniel I, Reid SM, Dan A, Maldonado J, Huovilainen A, Billinis C, Davidson I, Agüero M, Vila T, Herve S, Breum SØ, Chiapponi C, Urbaniak K, Kyriakis CS, Brown IH, Loeffen W (2014) European surveillance network for influenza in pigs: surveillance programs, diagnostic tools and Swine influenza virus subtypes identified in 14 European countries from 2010 to 2013. PLoS One 9:e115815
 28.
RESPIPORC FLU3 (2009) Annex 1. Summary of product characteristics. http://www.ema.europa.eu/docs/en_GB/document_library/EPAR__Product_Information/veterinary/000153/WC500067628.pdf. pp 1–21
 29.
Romagosa A, Allerson M, Gramer M, Joo H, Deen J, Detmer S, Torremorell M (2011) Vaccination of influenza a virus decreases transmission rates in pigs. Vet Res 42:120
 30.
Vincent AL, Ma W, Lager KM, Richt JA, Janke BH, Sandbulte MR, Gauger PC, Loving CL, Webby RJ, GarcíaSastre A (2012) Live attenuated influenza vaccine provides superior protection from heterologous infection in pigs with maternal antibodies without inducing vaccineassociated enhanced respiratory disease. J Virol 86:10597–10605
 31.
Van Reeth K, Labarque G, Pensaert M (2006) Serological profiles after consecutive experimental infections of pigs with European H1N1, H3N2, and H1N2 swine influenza viruses. Viral Immunol 19:373–382
 32.
Moghadas S, Bowman CS, Arino J (2009) Competitive interference between influenza viral strains. Can Appl Math Q 17:309–316
 33.
Zhang XS, De Angelis D, White PJ, Charlett A, Pebody RG, McCauley J (2013) Cocirculation of influenza A virus strains and emergence of pandemic via reassortment: the role of crossimmunity. Epidemics 5:20–33
 34.
Trebbien R, Bragstad K, Larsen LE, Nielsen J, Bøtner A, Heegaard PM, Fomsgaard A, Viuff B, Hjulsager CK (2013) Genetic and biological characterisation of an avianlike H1N2 swine influenza virus generated by reassortment of circulating avianlike H1N1 and H3N2 subtypes in Denmark. Virol J 10:290
 35.
Van Reeth K, Brown I, Essen S, Pensaert M (2004) Genetic relationships, serological crossreaction and crossprotection between H1N2 and other influenza a virus subtypes endemic in European pigs. Virus Res 103:115–124
 36.
Simon G, Hervé S, Rose N (2013) Epidemiosurveillance of swine influenza in France from 2005 to 2012: programs, viruses and associated epidemiological data. Bull Epidémiol Santé Animale Aliment 56:17–22 (in French)
 37.
Loeffen WL, Hunneman WA, Quak J, Verheijden JH, Stegeman JA (2009) Population dynamics of swine influenza virus in farrowtofinish and specialised finishing herds in the Netherlands. Vet Microbiol 137:45–50
 38.
Hervé S, Garin E, Rose N, Marcé C, Simon G (2014) French network for the surveillance of influenza viruses in pigs (Résavip): results of the first three years of operation. Bulletin Epidémiol Santé Animale Aliment AnsesDGAl 63:10–14
 39.
Corzo CA, Gramer M, Kuhn M et al (2012) Observations regarding influenza A virus shedding in a swine breeding farm after mass vaccination. J Swine Health Prod 20:283–289
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
CC and MA developed the mathematical model and participated in data analyses and interpretations. CC drafted the manuscript; LW optimized the mathematical model; NR coordinated the study and participated in data analyses and interpretations. All the coauthors revised the manuscript. All authors read and approved the final manuscript.
Acknowledgements
LW is supported by the Research Foundation Flanders (FWO, G043815N) and the Antwerp Study Centre for Infectious Diseases (ASCID) at the University of Antwerp. The authors thank the CRP Regional Pig Committees for Bretagne, Pays de la Loire and Normandie, the INAPORC National Pork Council and the Brittany Region for their financial support. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Author information
Affiliations
Corresponding author
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
About this article
Cite this article
Cador, C., Andraud, M., Willem, L. et al. Control of endemic swine flu persistence in farrowtofinish pig farms: a stochastic metapopulation modeling assessment. Vet Res 48, 58 (2017). https://doi.org/10.1186/s1356701704621
Received:
Accepted:
Published: