To assess the unbiased variation in the prevalence of ant broods in bird nests, we conducted the study in one of the last fragments of temperate primeval forest in lowland Europe. Such old-growth stands have been preserved in the extensive Białowieża Forest (c. 1500 km2) which straddles the Polish-Belarusian border. The regional climate is subcontinental with annual mean temperatures during May–July of 13–18 °C, and mean annual precipitation ranging between 426 and 940 mm [31, 32]. The altitude ranges from 134–140 m to 200 m a.s.l. .
The best-preserved stands are strictly protected within the Białowieża National Park (hereafter BNP; coordinates of Białowieża village: 52°42′ N, 23°52′ E), where species richness is high and the communities’ structures, interspecific interactions and natural processes have been little affected by direct human activity. Conducting the study in this forest offered a unique opportunity to observe the behaviour of birds and ants under conditions that likely prevailed across lowland Europe before widespread deforestation and forest exploitation by humans [26, 27, 31].
We collected data mainly in BNP, with a few additional observations from adjacent managed forest. We used three permanent study plots in BNP (denoted as MS, N, W) totalling c. 130 ha [26, 33] and also other fragments of primeval oak-lime-hornbeam Tilio-Carpinetum or mixed Pino-Quercetum stands, which are the main habitats of Wood Warblers [34, 35]. The stands are a fine-grained mosaic of microhabitats within the broad habitat types that cover large areas of the forest [31, 36]. The multi-layered stands are composed of various tree species of diverse sizes, aged up to several hundred years old, dominated by hornbeam Carpinus betulus, lime Tilia cordata, oak Quercus robur, spruce Picea abies, maple Acer platanoides and pine Pinus sylvestris, which occur in varying proportions between oak-lime-hornbeam and mixed stands. Fallen and standing dead wood is moderately common or abundant [31, 37].
We focused on a ground-nesting songbird, the Wood Warbler Phylloscopus sibilatrix, and mainly Myrmica ant species that also raise their broods on the forest floor. Wood Warblers are small (c. 10 g) migratory songbirds that winter in equatorial Africa and breed in temperate European forests . The birds build dome-shaped nests of woven grass, tree leaves and moss, lined with animal hair. The nests are usually well-hidden among leaf litter and sparse vegetation on the ground .
A previous study in BNP found that c. 30% of Wood Warbler nests contained mostly Myrmica ruginodis and M. rubra broods, which were located within the nest walls . Both Myrmica species are abundant ant species in many parts of Eurasia [40, 41]. Their colonies contain from tens to thousands of workers, and can be found on the forest floor. The densities of ant colonies can be limited by the availability of warm nest locations that depend on exposure to the sun in cool, temperate woodlands [21, 42, 43]. Therefore, access to active nests of birds, heated from within by the owners, could be particularly important for these insects, which require nest temperatures above 15 °C for the development of their broods [16, 17, 29, 30].
Placement of bird nests
We searched for Wood Warbler nests on a daily basis from late April until mid-July in 2018–2020 by following birds, mainly during nest-building. The nesting period of Wood Warblers largely overlapped with the peak of brood-rearing by Myrmica (and Lasius) ants, preceding the ants’ nuptial flights in July–September [17, 44, 45]. We inspected each bird nest every 1–6 days to establish the dates of egg-laying commencement (when the 1st egg was laid), hatching (assigned as day 0 of the nestlings’ age), nestlings vacating the nest (fledging), or nest failure. The breeding attempts of Wood Warblers usually lasted for 32 days from the first egg being laid until all young fledged, and included 5–7 days of egg-laying, 13 days of egg incubation and 12–13 days of chick-rearing [34, 38]. Nest failure was primarily due to predation, which is the main cause of the Wood Warbler nest losses in BNP [34, 39, 46].
To check if Wood Warblers preferred to nest near specific habitat features that might potentially hold ant colonies, we took descriptions of 187 Wood Warbler nest locations and compared them to 187 controls. The controls were points chosen haphazardly c. 30 m from the bird nests. The direction of each control was appointed by an observer turning around and stopping on a command from a second observer, who had no eye-contact with the partner. The distance of 30 m was measured in strides, always by the same observer. On a few occasions, a control fell on a road or in a meadow, which were unrepresentative of Wood Warbler nest-locations. Therefore, the procedure was repeated in such cases, and a new control was selected.
The descriptions of the Wood Warbler nests and the controls included the presence of the nearest tussock of vegetation (fern, grass or sedge), fallen tree branch and/or larger tree log, which were the distinctive features on the forest floor that might also potentially hold ant colonies [40, 41, 47]. Each nest could have multiple features recorded. The minimum diameter of the recorded deadwood branches was 1 cm, which was the minimum branch size that contained an ant colony in this study area (M. Maziarz, pers. obs.). Other fallen deadwood with a minimum diameter of 5 cm was defined as ‘tree log’. We used the two categories of deadwood to consider their potential differences in ‘quality’ as nest locations for ants due to the potentially varying microclimates . We measured the distance to all nearest features (tussock, branch and/or log) that were present within three metres of the rim of a bird nest or from the control, i.e. within a reachable distance for Myrmica (and Lasius) ants (M. Maziarz, pers. obs. [44, 49]).
The density and placement of ant colonies
To test whether Wood Warblers selected forest patches with higher densities of ant colonies, and to establish the placement of ant colonies, in 2018–2020 we searched for ant colonies on the forest floor. We defined an ant colony as a group of ants, including workers and/or a queen that were accompanying larvae or pupae, occupying a ‘nest’ structure other than a bird nest .
We conducted searches for ant colonies within 133 pairs of 3 × 3 m sample squares, with one of each pair centred on a Wood Warbler nest and the other on the haphazardly allocated control point, located c. 30 m from the nest (see above). To avoid disturbing Wood Warblers and exposing them to nest predators, we searched for ant colonies only after the Wood Warbler chicks had fledged or the birds’ breeding attempts had failed naturally.
The surveys entailed careful inspection of the forest litter to find all ant colonies within the squares. We treated colonies as present if a brood (larvae/pupae) could be seen above the ground surface, enabling the precise location of each ant colony. First, we marked all colonies found within a plot, and then we measured the distances between them. We took descriptions of the location of all ant colonies found within the plots, and collected specimens of ant workers for later identification. Where any ant broods occurred less than 55 cm apart, we treated these as one ant colony to avoid potential multiple counts of the same colony. In those situations, we used the description of the brood location that was found first.
Inspection of the Wood Warbler nests for ant broods
To establish the presence of ant broods within the walls of Wood Warbler nests, in 2018–2020 we collected 260 bird nests from the field after the chicks had fledged or the breeding attempt had failed, but only if the nest structure remained intact. We placed each nest into a sealed and labelled plastic bag, which contained information on the collection date and the nest identification number. To ascertain the presence or absence of ant broods in the bird nests, for those nests collected in 2018, we carefully pulled the nesting material apart and searched for ant larvae or pupae amongst it . If an ant brood was present, we collected five to ten ant workers from each bird nest into labelled tubes filled with alcohol, for later species identification.
For bird nests collected in 2019–2020, we automated nest examination by first extracting invertebrates from them using a Berlese–Tullgren funnel. Each Wood Warbler nest was covered with fine metal mesh and placed c. 15 cm under the heat of a 40 W electric lamp. All specimens, including ants, were caught in 100 ml plastic bottles containing 30 ml of 80% ethanol, installed under each funnel. Specimen extraction with the Berlese–Tullgren funnel usually took three days per nest. Next, we checked the nesting material as described above to ensure that no specimens remained. All ant specimens were then separated from other invertebrates caught in tubes and identified to species level.
Observed and expected frequency of ant broods within bird nests
To test for non-random presence of ant broods in the Wood Warbler nests, we compared the observed and expected proportions of bird nests containing ant broods. The observed proportions were calculated separately for each year in 2018–2020.
To obtain the expected proportions we performed simulations of the number of cases when an individual random point fell within an 8 cm radius of the centre of a 3 × 3 m square, representing an ant brood within a typical Wood Warbler nest, while a random point outside of the 8 cm radius represented an ant colony outside of a bird nest. The limit of 8 cm corresponded to the average radius of a Wood Warbler nest (M. Maziarz, pers. obs.). The simulations were based on a uniform distribution function that generated random deviates. We calculated an expected mean proportion of bird nests with ant broods and the 95% confidence intervals using bootstrapping (40,000 replications) in the ‘boot’ package in R [50, 51].
We repeated these simulations a further five times with an incrementally increasing number of random points allocated to the survey square in each replication, up to a maximum of six hypothetical ant colonies (the maximum number recorded within a real sample plot in BNP; Additional file 1: Table S1). To match the classification of a single ant colony in the field (see above), we set a minimum allowable distance of 55 cm between simulated ant colonies.
We multiplied the derived expected and observed proportions by 100 to obtain percentages.
Nest-site selection by the birds
According to our preliminary analyses, a tussock of vegetation, a fallen branch, or a tree log was present within three metres of 80–98% of the Wood Warbler nests and 79–98% of controls. Therefore, to provide a more sensitive test of the birds’ preference for nesting close to any of these habitat features, we considered them as present only if they were within 10 cm from the edge of a typical Wood Warbler nest of an 8 cm radius (M. Maziarz, pers. obs.), i.e. within 18 cm from the centre of a bird nest or a control.
To test if birds selected nest-sites close to a tussock of vegetation, a fallen branch and/or a tree log, we compared the frequency of any of these three features at 187 bird nests and 187 controls, using Chi2-tests with Yates' continuity correction. We did the comparisons separately for each of the three categories of habitat features, and included pooled samples from all years (2018–2020), as separate annual calculations were prevented by small sample sizes of bird nests or controls located at tree logs.
To check if Wood Warblers preferred to nest in areas more densely populated by ants, we compared the number of ant colonies recorded on the sample plots (3 × 3 m) centred on 133 Wood Warbler nests with the number of ant colonies found on 133 control plots. We tested the differences using a generalised linear model (GLM) with a Poisson error distribution and log-link function. The model contained the number of ant colonies as a response variable, and fixed covariates of year and the type of sample plot (bird nest vs control), with both covariates set as factors. An interaction term between year and plot type was insignificant in an initial model (AIC = 821.5), so it was removed from the final approach.
In all analyses, we treated the nest site choice of birds as independent each year because Wood Warblers show a low return rate to their breeding grounds in Continental Europe (up to 5% in Białowieża Forest; ), so it was unlikely that the nests found in different years could belong to the same birds.
Ant colony placement
To assess which features on the forest floor were used by ants for raising their broods, we calculated the annual proportions of ant colonies recorded in the most frequent categories of: fallen branch (≥ 1 and < 5 cm diameter), tree log (≥ 5 cm diameter), tussock of vegetation, and additionally deciduous tree-leaves, moss, bird nest, and ‘other’ uncommon locations, such as fallen spruce bark, standing tree or stump, soil, tree root, bracket fungus, spruce cone, or molehill. Colony frequency in each category was calculated separately for each year and for the two types of sample plots (centred on Wood Warbler nests and control locations). If an ant colony was located under two or more of the different features, the record was divided between the categories. For example, if an ant colony was found under a fallen branch lying on a tussock of vegetation, or under moss on a branch or log, each feature category received a score of 0.5. The sample sizes of ant colonies found in the sample plots at bird nests and control locations were respectively 73 and 53 in 2018, 64 and 58 in 2019, and 110 and 96 in 2020.
Prevalence of ant broods within bird nests in relation to bird nest placement
To test whether bird nests that were situated close to potential locations of ant colonies, such as tussock of vegetation, fallen branch or tree log, contained ant broods more often than the nests placed away from these features, we compared the frequencies of ant broods in these nests using Chi2- tests with Yates' continuity correction.
To check if the likelihood of an ant brood occurring in a Wood Warbler nest was higher in plots containing a greater number of ant colonies, we used a GLM with binomial error distribution and ‘logit’ link function. The model contained the presence or absence of an ant brood in a bird nest (respectively n = 41 and 88 bird nests) as a response variable, and a fixed covariate of the number of ant colonies on a sample plot. Preliminary analysis showed an insignificant effect of an interaction term between year (set as a factor) and the number of ant colonies (AIC = 167.1), or the fixed effect of year alone (AIC = 163.3), so both terms were dropped from the final model.
In the analyses, we assumed that the sample sizes included independent ant colonies each year because both the Wood Warbler nests and control locations always fell in different forest localities, which determined the searches for ant colonies each year. As such, it was unlikely that the same ant colonies would be tested repeatedly between years.
‘Thermal’ factors influencing the occurrence of ant broods in bird nests
To test which of the multiple variables related to the thermal activity of the birds during their breeding cycle progression, or weather conditions, may influence the likelihood of ant colonies occurring in Wood Warbler nests, we performed model selection based on the AICc criterion .
The variables included in the models were: the nest stage (egg-laying, incubation or early nestling stage vs late nestling stage, when chicks were 5 days or older until fledging or failure), the mean daily ambient temperature (5-day average) and the daily sum of rainfall (5-day sum) preceding the Wood Warbler nest failure or chicks’ fledging, the delay of bird nest collection from the field (the number of days following fledging of the chicks or nest failure until the nest was collected), and year to account for the annual variation in all variables (Additional file 1: Table S2; for detailed description of the variables see Additional file 1: Table S3).
To assess weather conditions, we extracted mean daily ambient temperatures and daily sums of rainfall from the meteorological station in Białowieża village, situated approximately 1–6 km from the study areas. For each nest, we calculated the 5-day average temperature and the 5-day sum of rainfall preceding the date of the nest failure or chicks’ fledging. The date of nest failure or chicks’ fledging was a mid-date between the last visit of an observer when the nest was still active and the next visit, when nest failure or chick fledging had occurred (Additional file 1: Table S3). We used the 5-day periods to test the effect of weather conditions on the ants’ colonisation because we were unable to define the exact date of ants relocating their broods into active nests of birds (currently occupied by the nest owners). Confirming the presence of ant broods within the structure of bird nests was possible only after dissecting the nesting material , and so it was unethical until after birds had vacated the nests. We presumed that the 5-day period preceding nest failure or chicks’ fledging would be the most sensitive for determining the weather impact on ant colonisation: long enough for ant workers to respond to weather conditions and relocate their broods into bird nests [16, 54], or to stay within the bird nests if ant colonisation had already taken place.
For model selection, we created a global model, which was a GLM with a binomial error distribution and ‘logit’ link function. The model contained a response variable of the presence or absence of an ant brood in a bird nest (respectively n = 56 and 204 nests), and fixed covariates of nest stage and year set as factors, the delay of bird nest collection from the field, and also the main effects and the interaction term between the mean daily ambient temperature and the daily sum of rainfall. The remaining interactions between year and the nest stage, and between year or the nest stage and other continuous covariates, were insignificant in prior tests (AIC ≥ 262.4), so they were dropped from the model selection. We performed model selection based on the global model using the dredge function in the MuMIn package in R , where the null model contained only the intercept.
The coefficients and 95% confidence intervals (CI) of the variables were assessed from model averaging in the MuMIn package across the top candidate models with Δ AICc < 2.
We performed all statistical analyses in R version 4. 0. 2 .