Populations
Field data were collected from 2016 to 2018 in four different populations of tree swallows breeding in nest-boxes. Populations were located in Chattanooga, Tennessee (TN) (35.1°N, 85.2°W, 206 m elevation, established in 2006, 2018: n = 73), Ithaca, New York (NY) (42.5°N, 76.5°W, 340 m elevation, established in 1986, 2016: n = 42), Burgess Junction, Wyoming (WY) (44.5°N, 107.3°W, 2451 m elevation established in 1995, 2018, n = 77) and in McCarthy, Alaska (AK) (61.4°N, 143.3°W, 445 m elevation, established in 1993, 2016: n = 28, 2017: n = 41). Tree swallows are widely distributed across North America and breed in a variety of environments. These populations were chosen to allow for comparisons among populations breeding in environments with different degrees of environmental predictability and reproductive value. Populations at higher latitude (Alaska) or elevation (Wyoming) are expected to experience cooler and more unpredictable weather conditions and a shorter breeding season starting later in the year (late May). In Tennessee, near the Southern edge of the breeding distribution, tree swallows are expected to experience warmer and more predictable weather conditions and a long breeding season with the first egg usually laid earlier in the season (early to mid-April). In New York, environmental conditions and breeding season length are expected to be intermediate, with the first egg usually laid in early May.
General field methods and stress manipulation
In the four populations, nests were monitored every 1–2 days throughout the breeding season from the initiation of activity at each site to fledging, except for the last week of nestlings’ development (to avoid inducing premature fledging). For every active nest, we recorded clutch initiation date and completion dates, clutch size, hatch date, brood size, and the number of nestlings fledged. Nestling fates were determined by checking boxes 22–24 days after hatching as fledging usually occurs around day 20. We installed radio frequency identification (RFID) units on each box on the fourth day of incubation (see below). Birds were captured at their nest boxes by hand or using a manually activated trap. All birds were captured and sampled on specific days of life history substages, and during a set time of day, to reduce the variation in circulating glucocorticoid hormones resulting from circadian rhythms. Adult females were captured between 0700 and 1000 h in NY, TN and WY and between or 0600 and 0900 h in AK to compensate for the earlier start of activity due to the increased day length compared to the other populations.
Females were initially captured 6 or 7 days after clutch completion (capture number 1). At this capture, we took a first blood sample within 3 min of initial disturbance to measure baseline circulating corticosterone levels. A second blood sample was taken after 30 min of restraint in a cloth bag to measure stress-induced corticosterone levels. Immediately after this sample was taken, females were injected with dexamethasone (dex) (0.5 μl g−1, Dexamethasone Sodium Phosphate, Mylan Institutional LLC), a synthetic glucocorticoid that binds to receptors within the HPA axis, in order to assess negative feedback indepently of the stress-induced corticosterone level29. A final blood sample was taken 30 min after dex injection to measure the degree of down-regulation in circulating corticosterone (a measure of negative feedback). Between samples, we weighed the females, and measured the length of their skull from the back of the head to the bill tip (head-bill) and flattened wing length. Non-banded individuals received USGS leg bands and a celluloid colour band with attached passive integrated transponder (PIT) tag encoding a 10-digit hexadecimal string (Cyntag, Cynthiana, KY). Female age was determined based on plumage coloration and characterized as second year (SY) or after second year (ASY)57.
As part of a separate study, at their first capture, adult females were allocated to one of the experimental groups: control, feather restraint (in which three primaries were reversibly attached to alter flight ability and thereby increasing the cost of foraging), or predator exposure (see29 for details on both experimental treatments). Treatments started after the first capture and lasted for 5–6 days. As these treatments had no effects on HPA axis regulation (Table S1, S2, S3), data from all individuals were included in analyses.
Females were then recaptured 5–6 days later (on incubation day 12 or 13; capture 2), at the end of the experimental treatments (see above). At this capture, we only took a baseline blood sample and weighed the bird before release. Finally, we recaptured females again 6–8 days after eggs hatched (capture 3). We followed the same procedure as in capture 1, taking a baseline, restraint stress-induced, and post-dex blood samples, and again weighed each female.
Twelve days after eggs hatched, each nestling received an USGS leg band, was weighed and had head-bill and flat wing length measured.
All blood samples were collected from the alar vein, in heparinised microhematocrit capillary tubes. Blood samples were then transferred to microcentrifuge tubes, and kept on ice until centrifugation (within 4 h). After separation, the plasma was stored at − 20 °C in the field and then at − 80 °C in the lab until analysis. All methods were approved by Cornell IACUC and conducted with appropriate state and federal permits. We followed the guidelines to the use of wild birds in research of the Onithological Council for the care and use of animals.
Provisioning behaviour
Number of feeding trips for females from nestling ages 1–18 was recorded using radio-frequency identification (RFID) devices (Cellular Tracking Technologies; Rio Grande, NJ, USA)58. RFID units were installed on each active box on day 4 of incubation. Antennae were fastened around each entrance hole so that birds had to pass directly through an antenna to enter or exit the box. We programmed our RFID units to sample for PIT tags every second between 0500 and 2200 h each day as tree swallows are not very active at night. Poll time was set at 500, and cycle time at 1,000. The delay time (minimum period of time between successive tag recordings) was set to 1 s. RFID boards were powered by 12 V 5Ah (PS-1250, PowerSonic, San Diego, CA, USA) batteries that were replaced every five days. At the first capture, each bird was fitted with a PIT tag attached to a colour band. Each PIT tag encoded a unique 10-digit hexadecimal string that was recorded, along with a time stamp, when birds passed through or perched on the antenna (see59 for more details). From the raw RFID records, we determined the number of daily feeding trips for each female through 18 days of age for the brood using an algorithm validated in the New York population59.
Corticosterone assay
Steroids were extracted from plasma samples using a triple ethyl acetate extraction and then corticosterone levels were determined using an enzyme immunoassay kit (DetectX Corticosterone, Arbor Assays: K014-H5) previously validated for tree swallows60. Samples were run in duplicate and all samples from an individual were run on the same plate. In total we ran 47 assays with an average extraction efficiency of 92.8% and a detection limit of 0.47 ng ml−1 (calculated as described in60). The intra-assay variation based on duplicate samples was 8.88% and the inter-assay variation based on plasma pool run across plates was 11.1%.
Data analysis
To characterise the degree of environmental unpredictability at the different field sites we calculated the unpredictability of temperature variables. We obtained historical weather data for each site over as long a yearly range as possible. For New York we obtained data from the North East Climate Center (https://www.nrcc.cornell.edu/) for the Game Farm road weather station (from 1983, located about 7 km from field sites) and from the Western Regional Climate Center (https://wrcc.dri.edu/) for the Prentice Cooper State Forest station in Tennessee (from 2003, located about 16 km from field sites), the May Creek station in Alaska (from 1990 located about 19 km from field sites) and the Burgess station in Wyoming (from 1992, located about 5 km from field sites). From these data, we extracted the average daily temperature, the daytime average daily temperature (between 0600 and 2200) which is the period when the swallows are the most active and the daily maximum temperature, which is known to affect flying insects’ activity and therefore food availability61.
We quantified the unpredictability of these temperatures variables during the breeding season: from April to June in Tennessee and New York and from May to July in Alaska and Wyoming. We calculated unpredictability using a general additive model (GAM) following the methods described in Franch-Gras et al.43. For each site, these variables were divided by their mean to normalise them before analysis43. This model (one for each site) considers the dispersion of the data in the time series around a typical curve of the normalised variable. For each weather variable and site, the typical curve was fitted in a GAM model in relation to the day of the year using the gam function in the mgcv package in R 3.5.3 (R Core Team, 2019). As suggested by Franch-Gras and collaborators43, we fitted the GAMs using cubic splines as smoothing function to not a priori constrain the shape of the curve. The standard deviation of the residuals of the fitted model (SDres) represents an index of unpredictability for each variable43. This index gives an overall measure of unpredictability using the historical records and is not intended to indicate variation in weather in the particular years of study at each site.
All data come from the first nesting attempt of each female. We compared females’ corticosterone levels by fitting a generalised linear mixed model (GLMM) with a gamma distribution that included population, life history substage, sample (baseline, stress-induced and post-dex), female age, treatment and their interactions (2-, 3-, 4- and 5-ways that include sample) as fixed factors, relative clutch initiation date as a covariate, female identity and experimental year as random intercepts, and population as a random slope. We further characterised females’ corticosterone regulation by calculating their glucocorticoid stress response as the difference between stress-induced and baseline corticosterone levels and their negative feedback as the difference between post-dex and stress-induced corticosterone levels. As it is not clear yet which trait is more important for the regulation of the phenotype—the raw hormone level or the response (the derived value)—we ran similar models on stress-induced and post-dex corticosterone values with baseline and stress-induced corticosterone levels as covariate respectively. For negative feedback, we also ran a model on the percentage change in corticosterone between the stress-induced and post-dex sampling periods. In order to directly compare the strength of the relationship between temperature unpredictability or reproductive value and females’ corticosterone regulation, we ran the same models but we replaced the fixed effect of population with either average temperature unpredictability or total breeding season length as a continuous variable. Then we used an information theory approach by comparing models’ Akaike Information Criterion (AIC) scores to determine which model, and thus which of these two variables, better predicts the observed variation in corticosterone regulation. We compared the magnitude of females’ acute stress response and negative feedback strength using GLMMs fit with a normal distribution including population, capture number, female age and their interactions as fixed factors, relative clutch initiation date as a covariate, female identity and experimental year as random intercepts, and population as a random slope. Population was added as a random slope in order to allow changes in HPA axis regulation to be different between populations. Within each population, we determined whether corticosterone level at each time point, and stress response and negative feedback were correlated using Pearson correlations.
We calculated the total breeding season length during the experimental year as the number of days between the first clutch initiation and the last day nestlings fledged at each site. We compared populations using a GLM fitted with a Poisson distribution. We also compared clutch size, brood size, hatching success and fledging success between populations. Population, female age and their interactions were added as fixed factors, and relative clutch initiation day as a covariate. The model for relative clutch initiation date was fitted with a normal distribution, models for clutch size and brood size with a Poisson distribution and models for hatching success and fledging success with a binomial distribution.
We compared the number of daily feeding trips females made using a generalised linear mixed model GLMM fitted with a Poisson distribution. Population, female age, treatment, their interactions and nestling age were added as fixed factors and brood size at each nestling age as a covariate. We also added relative clutch initiation date and brood size as covariates and nest identity as random factor. We used GLMMs to compare nestlings’ body mass with population, female age and their interaction as fixed factors. Female identity was added as a random factor.
GLMs were run using the GENMOD procedure and GLMMs the GLIMMIX procedure in SAS University Edition (SAS Institute Inc., Cary, NC, USA). Distributions for the different models were chosen depending of the type of data. Poisson distribution was used for counting data (nestling provisioning, breeding season length, clutch size, brood size), binomial distribution was used for binary data (0/1: hatching and fledging success), gamma distribution was used for continuous data when all data were superior to 0 (corticosterone levels, stress response, nestlings’ body mass) and Gaussian distribution was used for continuous data that included positive and negative values, after checking for normality of the residuals (negative feedback). Post‐hoc comparisons were performed using Tukey‐Kramer multiple comparison adjustment to obtain corrected p‐values. Probability levels < 0.05 were considered significant. Data are presented as mean ± SE. Figures 1, 2 and 4 were made in R 3.6.2 (R Core Team 2019, Vienna, Austria) and Fig. 3 in sigmaplot 14 (Systat Software, Inc, SanJose, CA, USA).
Source: Ecology - nature.com
