Quantification and characterisation of commensal wild birds and their interactions with domestic ducks on a free-range farm in southwest France

The role of commensal birds in the epidemiology of pathogens in poultry farms remains unclear. Our study aimed to identify potential key species for interactions with domestic ducks on one free-range duck farm in southwest France. Methods combined direct individual observations on duck outdoor foraging areas, network analysis, and general linear mixed models of abundances. Results showed a wide diversity of wild bird species visiting foraging areas, heavily dominated in frequency by White wagtails (Motacilla alba) and Sparrows (Passer domesticus and Passer montanus). These also were the only species seen entering duck premises or perching on drinkers in the presence of ducks. Moreover, White wagtails were the species most frequently observed on the ground and in close proximity to ducks. Network analysis suggested the role of White wagtails and Sparrows in linking ducks to other wild birds on the farm. The abundance of White wagtails was positively associated with open vegetation, with the presence of ducks and particularly in the afternoon, while the abundance of Sparrows was positively associated only with the fall-winter season. By precisely characterising interactions, the study was able to identify few wild bird species which should be prioritized in infectious investigations at the interface with poultry.


Methods
Study site. The study site is a typical duck farm in the department of Gers, southwest France, which was affected by HPAI during the 2016-2017 winter epizooty (Fig. 1a). The surroundings of the farm are an undulating landscape of small mixed crop and stock farms with woods on hills (Fig. 1b). Although not a wetland area, many artificial water dams of various sizes are dispersed in the surroundings for agricultural use. The farm breeds mule ducks under the 'Canard à Foie Gras du Sud-Ouest' label, which requires at least 14 weeks of unlimited outdoor access. The farm is composed of two small 0.5 ha outdoor foraging areas for ducklings (1 day to 1 month of age), and eight large 1.5 ha foraging areas for growing ducks (1 month to 14-16 weeks of age) (Fig. 1c). Like an increasing number of poultry farms in the region, an agroforestry program is implemented on the farm, so trees for wood are planted on all outdoor foraging areas, and hedges of fruit trees are planted around some of the areas (Fig. 1b,c).
Field observations of wild birds. From July 2020 to June 2021, standardised observations of wild birds were implemented on the farm's outdoor foraging areas. An average of 8.7 (median of 8) observation sessions (each lasting one hour) were conducted each month 4 weeks apart (except in February and May due to the regulatory stamping out and cleaning for HPAI that took place on the farm) over two consecutive days. Generally, four sessions were distributed evenly over the day: half an hour after sunrise, mid-morning, mid-afternoon, and one hour before sunset. In total, 87 observation sessions were carried out during 20 days distributed over 10 months (see description in Supplementary Table S1). For each month, three of the 10 foraging areas were randomly selected to meet the following criteria: one with ducks present on it, one with no ducks for at least 2 weeks, and one with a recent change in use (installation or removal of ducks in the preceding 2 weeks). The three selected areas were then visited on an alternating basis over the 2 days of observation, so that each had at least one morning and one afternoon observation when possible, or at least two different times of day (see details in Supplementary Table S1). The same observation points were used for each foraging area, with only a regular adaptation to the direction of sunlight between morning and afternoon (Fig. 1c). Due to regulatory restrictions on duck movements between farms and on outdoor access during the high-risk period of HPAI introduction in the winter of 2020-2021, no area with a recent change in use was present in January 2021, and no area with presence of ducks in March, April and June 2021 (see Supplementary Table S1). Each observation session was conducted by the same person and consisted of 30 continuous left-to-right screenings of the area using binoculars to record individual observations of each wild bird present on a vertical projection of up to 10 m of the area including fences. For each bird, the following information was recorded (see details in Supplementary Table S2): species, behaviour and direct environment (location, type of perch or ground, proximity of less than 1 m from a domestic duck). If a bird could not be identified to the species level, all birds of the same genus or family were grouped together in the database. The observer also recorded the following environmental information: date, time, weather, presence and number of domestic ducks.
Description of wild bird population diversity and behaviours. For each observation session, the maximum number of individuals observed per species during a single screening (i.e. at the same time) was recorded. This number was used as an estimation of the minimum number of individuals per species and per session. All sessions of the same month were then grouped together to calculate the monthly means of minimum number of individuals per species on the farm. The Shannon and Simpson diversity indices, and Piélou evenness index 32,33 were calculated based on the monthly means rounded to the upper integer, using the "diversity" www.nature.com/scientificreports/ function of the "vegan" package on R software version 4.0.5 34 . The three indices were defined by the following formulae: p i = 1 . According to these definitions, the Shannon index represents the entropy in the population, i.e. the uncertainty in identifying the species of a random individual in the population. The Simpson index represents the diversity, i.e. the probability for two random individuals to belong to the same species. This second diversity index is less sensitive to rare species than the first one, so it is more suitable to compare populations from different contexts 32 . The Piélou index represents the equitability of species abundances, that is the evenness of distribution of individuals among the different species 32,33 .
The data subset of observations in presence of ducks on the area (for either more than 2 weeks or recently introduced, counting for 34 out of the 87 sessions) were used to quantify behaviours allowing contacts (direct or indirect) between wild birds and domestic ducks. Recordings of whether each wild bird was at a distance of less than one meter from a duck were used to assess direct contacts 35 , including droplet emissions (allowing airborne viral transmissions based on approximate droplet emission distance 36 and virus infectivity in droplets 37 ) or even physical contacts. Indirect contacts via biological matter (allowing indirect viral transmissions such as faeces, surfaces, water, or airborne particles) were assessed via recordings of bird locations on duck foraging areas, in particular around aggregation spots, such as duck premises, feeders, drinkers, and on trampled wet ground. The cumulative numbers of individuals showing each type of interaction with ducks (direct or indirect), or each type of behaviour (entering premises, on the ground or perched) were then compared between species or group of species.

Network of between-species co-occurrences.
To analyse the co-occurrence patterns (as a proxy for indirect contacts) between wild bird species and domestic ducks, data of observations on the 87 sessions with or without ducks on the area were used. An undirected weighted network was built, considering bird species (wild bird species or groups of species, and domestic ducks) as nodes, and two different bird species recorded during the same observation session (i.e., co-occurrence) as an edge. Edges were then weighted by the frequency at which a given pair of bird species was recorded. Each node was characterised by its degree, representing the number of other nodes (i.e., species) connected to it. Communities of more densely connected species 38 were defined by a walktrap algorithm based on a random walk through the edges 39 using the "cluster_walktrap" function of "igraph" package 40 , with a four-step walk selected as giving the most synthetic and biologically accurate result 39 . The network analysis was conducted using the "igraph" package 40 on R software version 4.0.5 34 .

Influence of environmental factors on wild species abundance. Environmental factors influencing
the abundance of commensal wild birds on foraging areas were investigated, focusing on the species most at risk for interactions at the wild-domestic interface. To do so, counts of the most abundant species in presence of ducks were analysed according to environmental variables. Species selected for the analysis were the most frequently observed in presence of ducks, i.e. present on more than 25% of sessions with ducks in the area: White wagtails and Sparrows. This ensured also to exclude species with observation counts too low for analysis, i.e. too low for models to converge or to be interpretable. The following explanatory categorical variables were considered: time of day (morning or afternoon), weather (sun, clouds, or rain), season (spring-summer or fall-winter), type of surrounding vegetation of the area (woods and hedges, or open field) and presence of ducks on the area (either absent, or present in the last 2 weeks). All further statistical work was conducted using R software version 4.0.5 34 . As univariate distribution of abundances for both species showed no significant effect of weather (p-values over 0.7 for Kruskal-Wallis rank sum tests), this variable was excluded from model selections. Association between pairs of categorical explanatory variables was examined using chi-squared tests, considered significant if their p-value was greater than 0.05. The strength of this association was estimated by Pearson's Phi using "phi" function from package "effectsize" 41 , and considered too large to keep both variables if their value was greater than 0.5 42 . Since species' counts showed a negative binomial distribution, and to account for groups of sessions from the same 2-day visits of each month, general linear mixed-effects models using a log link negative binomial error distribution were used with the month as the random variable, using the "glmer.nb" function from "lme4" package 43,44 . Variable selection within the multivariate analysis was based on automated selection based on AIC using the function "dredge" from package "MuMIn" 45 , starting from a full model with all interactions between explanatory variables. Among minimal ranked models in a range of two units above the lowest AIC, the most parsimonious one was selected as the final model. The regression coefficients for the final model were expressed as odds ratios (OR) with 95% confidence intervals (CI), and its goodness of fit was estimated by the trigamma conditional R 246 using the "rsquared" function from package "piecewiseSEM" 47,48 . Using data of observations in presence of ducks, the direct environment of individual birds were analysed to quantify behaviours allowing potential direct or indirect interactions between wild birds and domestic ducks. Nine of the 23 species (8 of the 22 groups of species) observed in the presence of domestic ducks were recorded at least once at less than one meter from ducks (Fig. 3a). The frequency of these direct contacts represented 131 observations in the 953 total observations with presence of ducks (13.7%), implying a cumulative count of 294 individuals out of the 2446 cumulative individual birds. The vast majority of direct contacts implied MOTALB with 106 observations of 245 individuals, accounting for 83% of all birds showing direct contact with ducks and for 23% of all interactions with ducks shown by the species (Fig. 3a). This was followed by PAS spp, with 14 observations of 32 individuals, accounting for 11% of birds in direct contact with ducks and for 3% of interactions with ducks by the species (Fig. 3a). This was followed by Turdus merula TURMER and Garrulus glandarius GARGLA with each one 5 individuals (each one accounting for 1.7% of birds in direct contact, and respectively 19% and 36% of interactions by each species), followed by Aegithalos caudatus AEGCAU (3 individuals, 1.0% of birds in direct contact and 100% of interactions by the species), Streptopelia decaocto STRCTO (2 individuals, 0.7% of birds in direct contact and 40% of interactions by the species), and Sturnus vulgaris STUVUL and Phoenicurus ochruros PHOOCH (each one with 1 individual, 0.3% of birds in direct contact, and respectively 50% and 4% of interactions by each species) (Fig. 3a). Recordings of bird behaviour and locations on foraging areas revealed situations of proximity between wild birds and duck aggregation areas such as feeders, drinkers or inside duck premises (Fig. 3b). These observations (    www.nature.com/scientificreports/ duck premises, by 3 and 28 individuals (on 3 and 9 occasions), respectively (Fig. 3b). Contrary to other species, MOTALB were most of the time observed on the ground of duck foraging areas, on 272 observations (52% of 523 for the species) implying a cumulative number of 623 individuals out of 1079 (58%). Of these, 114 observations of 243 individuals (22%) were precisely on wet ground trampled by ducks (Fig. 3b).

Description of wild bird population diversity and behaviours.
Network of between-species co-occurrences. The undirected weighted network of between-species co-occurrences on the farm over 1 year was analysed. The network consisted of 36 nodes, one for domestic ducks and the other 35 for wild bird species (including 4 groups of species and 31 individual species) (Fig. 4). Nodes were connected by 323 edges representing the aggregation of 1217 observed co-occurrences of species pairs on the 87 selected observation sessions. Frequencies of co-occurrences (i.e., weight of edges) were very heterogeneous, with MOTALB and PAS spp paired on 74% of sessions as these species were observed on 85% and 80% of sessions, respectively. In contrast, other pairs of wild species were observed on less than 25% of all observation   (Fig. 4). Three species were not grouped with others by the algorithm. The three communities were representative of habitat preference of birds.  Table 2). The fitted model was much less predictive with 9.8% of explained variance according to the trigamma conditional R 2 .

Discussion
Detailed counts and individual observations of wild birds visiting outdoor foraging areas of a typical free-range duck farm made it possible to characterise the species diversity of the avian commensal population visiting duck facilities, as well as to quantify different types of direct and indirect interactions with duck flocks. This study of wild bird interactions with free-range poultry is to our knowledge the most detailed one after few previous studies [50][51][52][53][54] , with the combination of year-long series of data, species-level identification and individual behaviour quantification. The originality of this study lies in the multi-species contact network analysis applied to the context of a wildlife-livestock interface 55 , in addition with the focus on a free-range duck farm that represents a specific but least-studied type of poultry production. The farm studied is not directly connected to any wetland or littoral areas, which makes it theoretically unattractive for wild aquatic birds such as Anseriformes, Charadriiformes or Ardeidae that are considered as main reservoirs for AIV 8,9 . Indeed, results showed that the vast majority of visiting wild birds were non-aquatic species (except for one Grey heron, Ardea cinerea). Most of them were passerines, which allowed to specifically focus this www.nature.com/scientificreports/ study on terrestrial non-aquatic avian communities of such farm environments. The observed species richness and diversity on foraging areas is slightly lower than periurban areas of France 20 years ago 56 , or comparable to most types of agricultural lands in France 29 . However, the actual diversity of the farm is certainly greater than what was observed in this study, as farm areas outside duck foraging areas were specifically excluded, and only visual counts were applied. Indeed, the surrounding vegetation (woods, riparian vegetation, hedges) and habitats (various crops, duck premises, barns, old houses and gardens) of the whole farm are attractive for a vast diversity of wild birds throughout the year, as suggested by a study in the Netherlands 50 and as seen on visits for other purposes and on nest captures on the study farm (20-30 species, H = 36.6-54.6, unpublished data). Interestingly, the profile of bird species observed on our study farm was quite different from other studies. Columbidae, Corvidae and Sturnidae were observed on less than 21% of observation sessions, and mostly in small numbers, while they composed a large part of observations in studies on other world regions and were thus identified as important species 7,27,51,53,54 . This difference in visiting bird species highlights the importance of studying wilddomestic bird interface in each environmental context of interest. Thus, infectious risk analyses can be focused on species that are the most abundant and in close interaction with domestic birds, such as White wagtails and Sparrows in the context of this study. Co-occurrence and behaviour analyses together revealed the key role of White wagtails and House and Tree sparrows in the multi-species interaction network surrounding free-range domestic ducks. These three species were by far the most frequent and abundant species either when ducks were present or not, all year long, and with a resulting high number of species in indirect contact (co-occurring in the same area). It appears that White wagtails and Sparrows can play a role as a link (at least by indirect contacts on duck foraging areas) between ducks and other wild birds that may less frequently visit ducks, thus intensifying the global connection between the wild and domestic compartments of the local interface. The three species, and to a lesser extent other species of the duck-centred community (number 1), seem to intensify local connections between communities of birds that are usually observed in distinct habitats such as woods or open lands. Moreover, White wagtails were observed in co-occurrence with the only aquatic bird (Grey heron) observed during this study on an area with no ducks on it. As White wagtails are very mobile and omnipresent on the farm, it is highly probable that on the same day (or few surrounding days if considering AIV incubation similar to Sparrows 10-13 ) the same individuals went in direct or indirect contact with other species of wild or domestic birds of the farm. White wagtails may then help connect domestic ducks and the community of commensal birds of the farm to more distant and less frequent species that may be more susceptible to infections. Such epidemiological connection between White wagtails and European starlings, another commensal bird species of poultry farms, has already been suggested by a previous study in the same region 57 . In addition to their abundance, White wagtails and Sparrows showed bold behaviours towards ducks and their facilities, frequently approaching ducks at a close distance, perching on their drinkers and feeders, flying into their premises, and foraging on trampled grounds rich in duck faeces. The one-meter distance chosen here was an estimation of potential direct airborne transmission via droplets applied to various respiratory infectious agents, including AIV 35 . It was based on approximate droplet emission distance 36 and virus infectivity in droplets 37 . Although observations of wild birds in close proximity to ducks were overall infrequent, they constituted the only evidence of possible direct infectious transmission between wild birds and ducks, as physical contacts were never observed in this study, and have rarely been observed on poultry farms in other studies 51,54 . In addition to direct contact occasions, the presence of White wagtails and Sparrows on drinkers, feeders, inside duck premises or on wet trampled ground made indirect transmission of potential infectious agents from and to ducks highly possible. This risk has to be further evaluated for these particular species of wild birds. While Sparrows are commonly observed in most other studies of poultry farms 7,25,27,54 , and are known for their susceptibility to AIV infection [10][11][12][13] , they are sedentary and probably purely commensal hosts that may only transmit infectious agents at a very local scale. White wagtails are rarely considered in epidemiological studies around poultry farms 6,54,58,59 ; however, the species is much more mobile than Sparrows, both on a daily basis 60 and seasonally, with partial migratory behaviour and winter aggregations in the region of this study 61 . Besides farms, the species is detected in a large diversity of habitats, and it is particularly attracted by wetlands, foraging on open areas along river or lake banks where wild aquatic birds can be present (data from French Breeding Bird Survey 62 ). White wagtails are thus excellent candidates as bridge hosts 16 for pathogens shared between domestic ducks and water birds, such as AIV, as suggested in a previous multi-site study 7 . This role on the study farm, which serves as a model of farms in southwest France, needs to be further evaluated through local ecological and epidemiological studies.
Analysis of environmental factors favouring abundance of Sparrows showed a higher concentration of birds during the fall and winter seasons. Either by population increase (young birds of the year from the local breeding population), or by aggregation on foraging areas rather than other environments, this element certainly increases their role in intensifying local connections between communities of birds. This intensification of connections is of particular concern for future epidemiological studies as it happens at a time of year when persistence of infectious agents in the environment is favoured by low temperatures and high humidity [63][64][65] , and when highly pathogenic AIV are regularly at risk of being introduced in poultry farms of the region by migratory water birds 19,66 . At the scale of the study farm, Sparrows seemed to have no preference in terms of weather, time of day or surrounding vegetation. This reveals their constant presence under all conditions around duck areas, probably due to their granivore feeding behaviour associated with their sedentary behaviour around farm buildings where they build colony shelters. In contrast, White wagtails were slightly more abundant in the afternoon and consistently more abundant when ducks were present or on areas that were not surrounded by hedges or woods. The species therefore seems to require more specific types of environments than Sparrows, and probably also chooses the best time of day to forage on duck areas when arthropods are more active. Indeed, most of the time they were observed on the ground hunting for arthropods that are numerous around duck manure and litter 67  www.nature.com/scientificreports/ White wagtails when they are not observed on the farm. However, duck presence and open-vegetation areas showed a weak positive association that might increase the effect of one variable or the other in the model. Moreover, duck presence showed a moderate positive association with fall-winter season, which also needs to be taken into account in conclusions. Therefore, abundance of White wagtail may be positively influenced by fall-winter season in parallel with duck presence, and abundance of Sparrows may be positively influenced by duck presence in parallel with fall-winter season. Practices of the studied farm are very likely representative of most regions of southwest France, thus allowing to extrapolate a good part of the conclusions, provided that local bird communities are similar. In particular, the farm has been applying daily preventive measures that are recommended by sanitary authorities for several years: covering feeders and drinkers, repelling wildlife on the farm with electric fences and rodent poisoning, checking for and removing dead domestic ducks every day. However, it should be noted that from January 2021 the presence of ducks in the outdoor areas was disturbed as the farm was under regulatory restrictions due to regional control measures for the HPAI H5N8 epizooty that occurred in the winter of 2020-2021. Consequently, no recent introduction or removal of ducks happened on the farm in January, then preventive culling in February (no observation) led to the absence of ducks during the months which followed in the spring of 2021, allowing vegetation to grow on all foraging areas. These changes in surface use and activity on the farm may have impacted the presence of commensal birds. This idea seems to be supported by the lowest diversity and evenness indices in March, as abundances were dominated by few species (mostly groups of CARLIS, and few MOTALB and PAS spp) that comply with the cleaning activities that were occurring on most parts of the farm. With growing vegetation and less activity during spring, these changes possibly reverted in favour of less farm-dependent species along with the ascending migratory visitors, making the diversity indices rise in April.
As the aim of this study was to describe commensal birds, the observation protocol had to focus on foraging areas only. If observation biases hid some elusive or hardly identifiable species, these could be considered as occasional or distant visitors rather than truly commensal. It is certain that the most important species targeted in terms of numbers and duck-centred behaviours were certainly not missed by the protocol implemented in this study, although the sanitary disturbances described above might have lowered their numbers in the spring of 2021. However, in order to be more exhaustive regarding bird species that visit the farm, a study at night time could be useful as domestic ducks have access to foraging areas during the night and nocturnal birds (owls and snipes for instance) were detected on some occasions during visits for other purposes. To detect a broader range of species and complete the assessment of targets for future studies, the observation method implemented in this study could then be complemented by other methods like nocturnal cameras 51,54 or identification of bird DNA accumulated in environmental samples on domestic duck areas 68 . These methods could provide large benefits and should be considered in further research.
In a context of declining farmland bird populations in France and Europe 28,69 , it is interesting to highlight the presence of particularly threatened species in our study site. According to evaluation of French breeding bird populations 49,69 , two of the species observed on foraging areas of the farm are declining steeply and are classified as endangered on the national IUCN Red List: Emberiza schoeniclus (− 50% 2001-2019 decline), and Passer montanus (− 60%). Five other species also are declining steeply but continue to be classified as vulnerable on this list: Anthus pratensis (− 66%), Chloris chloris (− 50%), Carduelis carduelis (− 31%), Dryobates minor (− 31%), and Emberiza citrinella (− 54%). It therefore appears important to preserve in particular the consequent nesting population of Passer montanus thriving on this farm thanks to shelters in old buildings and rich sources of food brought by diverse agricultural activities and natural habitats.
The overwhelming presence of White wagtails and Sparrows observed throughout the year on the farm, and their frequent and close proximity with domestic ducks, raises questions concerning the efficacy of contactrestricting measures. Indeed, the ability of these commensal species to adapt to complex environments driven by human activities makes them virtually impossible to strictly keep away from free-range poultry farms. In addition, in traditional free-range systems such as the one studied here, sustainable measures have to preserve the freedom of domestic ducks as much as possible, but also preserve endangered farmland birds by having as little unwanted impact as possible. Some wildlife-preventing systems with variable efficacy and costs are being tested on poultry farms and other agricultural facilities 59,70 : nets, laser 71 , balloons, kites 72 , propane canons, and scaring alarm calls 73 . While they may keep the most fearful and larger species away (particularly aquatic birds), the boldest ones might still eventually find a way to access the attractive places on farms where animals live, either by habituation 71 or by taking advantage of inherent flaws or defects in the system. To avoid habituation, costly strict wildlife-prevention measures should be used for short periods of time and when absolutely needed. Moreover, most contact-restriction measures are aimed at preventing poultry from being infected by wildlife, even though it has been proven to be a minor introduction risk in previous HPAI epizooties 17,74 . Contact-restriction measures often do not consider the environmental transmission risk from poultry to wildlife, but waste water, feathers, dust and aerosols coming from thousands of infected birds on a farm can represent a massive contamination source for commensal wild birds living in the same environment [75][76][77] . After characterising what and how wild-domestic interactions occur and identifying a few species that could represent ecological bridges, it is necessary to assess whether these bridges are epidemiologically functional as well 78 . In a second phase, considered commensal species have to be characterised with regard to their infectious susceptibility and excretion capacity for the pathogen of interest, such as HPAI. Effectively infected commensal wild birds could then spread the infectious agent to other farms, or to a wild maintenance host. Without complete knowledge of the bridge host potential of every commensal species for every infectious agent, and as a precautionary measure, overall biosecurity measures on poultry farms stay useful and need to consist of multiple complementary measures. They should prevent wild birds from coming into contact with poultry or their facilities, but also poultry-contaminated materials (litter, manure, waste water, dust) from being accessible to wild birds and vice versa (natural watercourse or water body, water run-off, litter and feeding materials). It is of first importance to take into account all possible dispersion