A comparison of sex, morphology, physiology and behavior of black-capped chickadees trapped using two common capture methods

Study system

We sampled black-capped chickadees at 10 different sites on properties of the Queen’s University Biological Station (QUBS; near Elgin, Ontario; 44°34′N, 76°19′W) from October through December of 2014. Black-capped chickadees form flocks generally comprising 3–12 individuals and occupy a relatively stable winter territory (Foote et al., 2010), permitting sampling of multiple individuals at one site. Chickadees readily come in to human-provided food sources, such as bird feeders, and so can be captured with food-baited traps. Finally, black-capped chickadees and their relatives in the family Paridae (e.g., blue tits and great tits) are a common subject of a diverse array of biological studies involving capture and sampling.

Capture and sampling

We captured birds between 07:00 and 13:00, starting no earlier than 1 h after sunrise to minimize the influence of diel variation in circulating hormone concentrations. We trapped at a given site for no more than 112 min within this time window (mean time from capture of first to last bird at a given site = 86 min). We used both capture methods at each site: walk-in (Potter) traps baited with black oil sunflower seeds and mist nets paired with playback of an audio recording of chickadees mobbing an eastern screech-owl (Megascops asio). Eastern screech-owls are a predator of adult black-capped chickadees, and their presence elicits an anti-predator behavioral response (Shedd, 1983; Foote et al., 2010). We put baited walk-in traps on platforms (~1.5 m off the ground) out at sites with their doors secured open for several days before sampling (mean = 13 days; range = 3–24 days) to allow birds to locate the seed and become accustomed to entering the traps to forage. This acclimation period should reduce the role of individual variation in neophobia in capture by this method. We sampled at nine sites on the same day, alternating the order of which capture method we used first between sites, and waiting a minimum of 30 minutes after capture of the last individual by the first method before beginning use of the second technique at the same site. At one site, we mist netted on the day after trapping via walk-in trap because of deteriorating weather conditions on the first day. Results do not differ if data from this site are excluded. When mist netting, we immediately turned off playback once an individual became entangled in the net. Playback was resumed only when one person could watch the net again to stop playback and extract the next captured bird. Audio playback never extended beyond 10 min of uninterrupted play (mean = 198 s; range = 10–600 s). Because chickadee flocks typically consist of 6–8 individuals (Smith, 1991), we limited sampling to the first three birds captured by each method, to ensure both methods could be used on each flock. This approach should also maximize our chance of detecting differences among birds caught by the two methods, if they exist, based on the assumption that the first few birds caught by each method are those that are most prone to capture by the method (if individuals vary in their trappability by these methods). Trapping sites were spaced to minimize flock territory overlap (mean distance = 8.8 km; range = 285 m to 21 km). We did not capture any individuals at more than one trapping site. We also did not capture any individual by more than one method, within each site.

Upon capture, we collected an initial blood sample into a heparinized microcapillary tube via puncture of the brachial vein (maximum 70 μL) with a 26-gauge needle as quickly as possible after the bird entered either capture device (i.e., the time when the door closed on the walk-in trap or the bird became entangled in the mist net). Three minutes post-capture is commonly used as a cut-off for measurement of baseline corticosterone (the primary glucocorticoid in birds) (Romero & Reed, 2005); however, we found a linear increase in ln-transformed corticosterone with sampling time (Fig. S1; linear model: N = 52, β = 0.006 ± 0.001, p < 0.001). Because of this relationship with sampling time, and differences in sampling time between methods (mean sampling time: walk-in trap = 135.2 s, mist net = 162.1 s; t-test: t = 2.27, df = 44.11, p = 0.03), we used the residuals of the linear regression of ln-transformed corticosterone on sampling time including samples collected past the 3-min time period (mean sample time = 148.12 s, range = 55–275 s) in subsequent analyses. Results do not differ if the 10 samples collected beyond 3 min post-capture are excluded from analyses, or if raw corticosterone concentrations, uncorrected for the relationship with sampling time, are used. The amount of time elapsed from initiating the capture attempt to the time the bird was captured was unrelated to corticosterone (linear model: β = −7.34 × 10⁻⁵ ± 7.79 × 10⁻⁵, p = 0.35), suggesting no effect of duration of disturbance due to our presence or audio playback at the trapping site.

We collected morphological measurements from each bird, including flattened wing chord using a wing ruler (±0.5 mm), tarsus length with calipers (±0.1 mm), and body mass with a Pesola spring scale (±0.5 g). We estimated a fat score for each individual by observing the amount of visible fat deposits in the furculum (inter-clavicular depression) using a scale from 0 (no visible fat) to 5 (excessive fat deposits) (Krementz & Pendleton, 1990). We calculated a scaled mass index (SMI) to estimate body condition of each individual using the slope of the least-squares regression of ln-transformed body mass (grams) on ln-transformed wing length (mm). We used wing length instead of tarsus length because it was more strongly correlated with body mass. We followed the Thorpe-Lleonart equation to estimate SMI: individual body mass × (population mean wing length/ individual wing length)^slope (Peig & Green, 2009). In this equation, the “slope” exponent is the scaled major axis slope, which is estimated as the slope from the least-squares regression divided by the correlation between body mass and wing length (Peig & Green, 2009). In our data, the correlation between body mass and wing length was 0.70, and the SMA slope was 2.12. The population mean wing length was 66.16mm.

We fit each bird with a uniquely numbered Canadian Wildlife Service aluminum band, a plastic colored leg band (Avinet, Dryden, NY, USA), and a passive integrative transponder (PIT) leg band (Eccel Technology, Leicester, UK). After we had obtained all measurements and banded birds, we placed them in an opaque, breathable cloth bag until 30 minutes after capture. We then collected a second blood sample for measurement of capture stress-induced corticosterone levels using the same collection method as described above. We failed to collect sufficient blood to measure initial or stress-induced corticosterone for a few birds (three initial, two stress-induced). We kept all blood samples cool by storing them on ice for transport to the laboratory at Queen’s University. All capture and handling methods were approved by the Queen’s University Animal Care Committee (protocol 2013-057) and banding was conducted under a Canadian Wildlife Service banding permit (permit 10771).

Behavioral assays

We completed all behavioral trials during December 2014, between 08:00 and 12:00, a minimum of 22 days after capture (mean = 35 days; range = 22–49 days). We used radio frequency identification (RFID) readers mounted to bird feeders (Bird Feeder Reader Meter, IB Technology, Buckinghamshire, UK) at the feeding stations that had been established at the time of capture at each site. Birds at each site were allowed to habituate to the RFID feeder readers for a minimum of two full days (mean = 6; range = 2–8). We used RFID readers to measure activity at the feeder for three separate segments of the behavioral trials (details below). We set readers to record individual PIT tag numbers, time and date twice per second. We carried out trials over three consecutive days, with different segments of data collection occurring on separate days: a pre-trial control period, response to a predator model, and response to a novel object, with the type of stimulus (model predator or novel object) presented on the second or third days, alternating order between sites. Each trial lasted 2 h and all trials occurred at the same time of day within sites, allowing for direct comparison of behavior among trials. To assay risk-taking in response to a predation threat, we exposed each flock of chickadees to a small owl model paired with playback of an audio recording of an eastern screech-owl’s call. We used five different screech-owl recordings, all obtained from the Macaulay Library at the Cornell Lab of Ornithology. We played each audio recording from a FoxPro Scorpion X1B speaker (FOXPRO Inc., Lewistown, PA, USA) programed to play back the audio stimulus (5 min) interspersed with segments of silence of varying duration (range 3–13min). Each site received the same program of audio stimulus and silence, but we varied which of the five recordings was used. We checked volumes of each call with a sound meter to ensure all were played at 80–90 decibels at a distance of 1 m. This range reflects natural variation of screech-owl vocalizations as captured within the recordings we used, and playback volume was similarly varied across all sites. At each site, we placed the model predator and speaker in a tree 1–1.5 m from the RFID feeder reader. We then left the area for 2 h and used the RFID log to determine number of visits to the feeder in the presence of the predator stimulus. To assay willingness to visit the feeder in the presence of a novel object, we placed a red plastic drinking cup directly on top of the RFID feeders and left it in place for 2 h, again recording visits from the RFID log. Some of the initially captured birds were not recorded at RFID feeders during behavioral trials. Sample sizes are provided for each measurement (Table 1).

We assessed response to a predator and a novel object by tallying the total number of trips to the feeder and the mean duration of time spent at the feeder during each trip (total duration of time detected at the feeder in seconds, divided by number of trips) for each individual during each 2-h trial (pre-trial control, predator trial, novel object trial). If an individual was detected at the feeder during some but not all of the trials, we assigned them a value of zero for both behaviors (number and mean duration of visits) for the trials they when they were not detected, but we excluded individuals that were never detected at the feeder during any of the trials from the analyses. We analyzed foraging behavior using both raw values for each individual (number and mean duration of visits to the feeder during each trial) and by calculating the change in behavior relative to the control trial (i.e., (number or duration of visits during predator or novel object exposure—number or duration of visits during pre-trial control)/number or duration of visits during pre-trial control). With this calculation, a negative value represents a reduction of trips to or duration of time spent at the feeder, a positive value represents an increase, and a zero reflects no change. The magnitude of the value reflects the scale of the change relative to behavior in the absence of the stimuli. Importantly, with this calculation of change in behavior, values cannot be lower than −1, which represents individuals that were present during the pre-trial control but did not visit the feeder during the predator or novel object trial (number/duration of feeder visits = 0). One individual that visited the feeders during the novel object trial was excluded from these analyses because it did not visit the feeder during the pre-trial control (or predator trial), and so a scaled change in behavior could not be calculated. Results do not differ if the response to the stimulus is calculated as a simple difference without scaling by the control value (e.g., stimulus behavior–pre-trial control behavior). We did not assess repeatability of behaviors within individuals and so do not assume that foraging behaviors during the various trials reflect individual variation in personality. Previous estimates suggest that activity, response to novel objects, and exploratory behaviors have low to moderate repeatability in black-capped chickadees (Devost et al., 2016) and other bird species (Van Oers et al., 2004; Bell, Hankison & Laskowski, 2009; Garamszegi et al., 2015; Audet, Ducatez & Lefebvre, 2016; Holtmann, Lagisz & Nakagawa, 2017).

Hormone assay

We centrifuged blood samples for 5 min at 6,000 rpm within 6 h of collection, which is well within the timing during which hormone concentrations remain stable at cool temperatures (Khonmee et al., 2019). We then preserved plasma and red blood cells separately at −20 °C until further processing. We measured total plasma corticosterone using an enzyme immunoassay (EIA) with a detection limit of approximately 30 pg/mL (Lot No. 0457099; Cayman Chemical Co., Ann Arbor, MI, USA). No samples fell below this detection limit. Prior to assaying experimental samples, we validated the assay for use in this species by assaying a serial dilution of pooled plasma collected from multiple black-capped chickadees, sampled during the same life history stage, as well as a serial dilution of pooled plasma that was spiked with corticosterone to increase concentration by 10 ng/ml. Both serial dilution curves ran parallel to the assay standard curve. We assayed 7.5 μL of each plasma sample from focal birds in duplicate after dilution (1:16, using steroid buffer) on three plates following supplier instructions. We ran eight known-concentration standards, also in duplicate, on each plate to determine intra and inter-plate variation. The coefficient of variation based on these replicate standards was 5.3% between plates and averaged 3.0% within plates. Baseline corticosterone concentrations in this population during the non-breeding season have been estimated to be somewhat repeatable (r = 0.26; Montreuil-Spencer et al., 2019). We have not estimated repeatability of stress-induced corticosterone concentrations for this population, but they are generally more repeatable than baseline concentrations (Schoenemann & Bonier, 2018). As such, individual variation in corticosterone concentrations is expected to reflect both flexible and consistent differences among individuals.

Molecular sexing

To differentiate male and female black-capped chickadees, which are not highly dimorphic in morphology or plumage, particularly outside of breeding, we extracted DNA from red blood cells using Qiagen DNeasy Blood and Tissue Kits following the supplier’s protocol for nucleated blood cells (Qiagen Inc., Toronto, ON, Canada). We used molecular sexing techniques as described by Griffiths et al. (1998). After separation of PCR products by electrophoresis, we identified females by the presence of two bands (WZ) and males by the presence of a single band (ZZ). As quality control measures, we included samples from birds of known sex as well as negative controls in all PCRs, which always produced expected results.

Statistical analysis

We completed all analyses using R (version 4.0.0), and all raw data and code for the main analysis have been uploaded to Open Science Framework (available at: https://osf.io/zsjme/). To determine if individuals captured by the two methods differ for any of the traits we measured, we used conditional inference tree analysis using the ctree function in the package partykit (Hothorn & Zeileis, 2015). Conditional inference tree analyses are a form of non-parametric regression tree analysis that uses recursive partitioning to identify factors that predict variation in a response variable (Hothorn, Hornik & Zeileis, 2006; Hothorn & Zeileis, 2015). We included all of the trait measurements as candidate predictor variables (Table 1) and method of capture (walk-in trap or mist net) as the response variable. Results do not differ if we instead analyze each trait separately as a response variable in a linear mixed-effect model with capture method as the sole fixed effect and capture site, order of use of capture methods and/or order of stimulus presentation (for behavioral data) as random effects. Conditional inference tree and other decision tree approaches are increasingly being used in ecological research (Johnstone, Lill & Reina, 2014; Maslo et al., 2016; Burke et al., 2020; d’Entremont et al., 2020), and can be preferable to traditional regression methods as they can allow for a more conservative, refined and holistic approach to identifying differences among groups (Cutler et al., 2007; Müller, Schröder & Müller, 2009; Blank & Blaustein, 2014). Conditional inference trees are also powerful because of their ability to simultaneously assess categorical and continuous variables (including data with non-normal distributions) and to accommodate missing values (Hothorn, Hornik & Zeileis, 2006).

We conducted a follow-up analysis on a subset of our data to determine if the birds that were most readily caught by each method differed from each other for any of the traits that we measured. For this analysis, we repeated the conditional inference tree analyses as described above, but limited to the first three individuals caught at each site, based on the assumption that these first-caught birds were the most trappable. This follow-up analysis suffers from the limitations of a reduced sample size (N = 30) and a lack of pairing of methods within sites, but could be informative, particularly if results differed from those of our main analysis.

Some capture methods might select for a subset of the individuals captured by other methods with a narrower range of trait values, which could result in similar means but different variances among groups. As such, in addition to analyses of mean trait values, we also compared variances among individuals caught by the two capture methods using separate modified Levene’s tests of homogeneity of variances from the trait median, which is robust to departures from normality (Carroll & Schneider, 1985).