Biology of the Thorn-tailed Rayadito and the study populations
The Thorn-tailed Rayadito (Passeriformes: Furnariidae) is an insectivorous endemic species of the South American temperate forest [36]. Thorn-tailed Rayadito are small (~11 g) and usually lay one clutch per breeding season during the austral spring [37]. Thorn- tailed Rayaditos can live at least nine years (nestlings marked and recaptured nine years later) and their mean the lifespan is 4.8 yr [38]. The nest construction period lasts 6–15 days, and the incubation period is 9–15 days. Eggs are laid on alternate days and the Thorn-tailed Rayadito postpones incubation until after the clutch is complete [37]. As Thorn-tailed Rayaditos are secondary cavity nesters they will adopt artificial boxes, which we monitored, in two populations in Chile over nine reproductive seasons (2008-2017). We monitored 170 nest boxes in Fray Jorge National Park (30°38 S, 71°40 W), the northernmost population (lowest latitude) of the species’ distribution. At this site there is a relic forest composed mainly of Olivillo (Aextoxicon punctatum), occurring in patches at the top of the coastal mountain range where fog-induced microclimatic conditions allow the forest to exist in this otherwise semiarid region [39]. In Puerto Williams, on Navarino Island (55°40 S, 67°40 W), which represents the southernmost (and thus highest latitude) limit of the species’ distribution, we monitored 150 nest boxes. At this site, the vegetation is characterized by deciduous Magellanic forest, whose characteristic species are Lenga Beech (Nothofagus pumilio) and Ñirre Beech (Nothofagus antarctica).
Monitoring, capture procedures and blood sampling
Hormone levels were sampled in the high latitude population in 2008, 2010 and 2011 and in the low latitude population in 2010 and 2011 [34]. Subsequently, we continuously monitored the populations until 2017. To check for nest box occupation, nest boxes were monitored on a weekly basis. When occupied, the frequency of monitoring was increased in order to detect laying dates (date of first egg) and hatching dates. Adults were captured in their nests between 08:00 and 12:00 with a walk-in trap located in the nest-box entrance hole [37]. For hormone analysis, blood samples (ca. 50 μL) were obtained by puncturing the brachial vein with a sterile needle and collecting blood into heparinized micro-hematocrit capillary tubes. In our study, samples were collected within 3 min of capture [34]. Samples were stored on ice until the end of the sampling period (maximum of 5 h) and were then centrifuged for 5 min at 8000 rpm to separate the plasma from red blood cells. The plasma was aspirated with a Hamilton syringe and stored (at -20 °C) until assayed for total CORT content (University of California, Davis). Tarsus length and the weight of each bird were measured. Nestlings and adults were banded with individual metal bands (National Band and Tag Co., Newport, Kentucky, USA and Split Metal Bird Rings, Porzana Ltd, UK) or with a numbered band provided by the Servicio Agrícola y Ganadero (SAG), Chile. This procedure was carried out with the permission of SAG, and the Corporación Nacional Forestal (CONAF), Chile.
CORT levels (i.e., concentrations) in the plasma were determined using direct radioimmunoassays. To determine the efficiency of hormone extraction from the plasma, 20 µL of 2000 cpm of tritiated Cort was added to all samples and incubated overnight. Hormones were extracted from the plasma using freshly re-distilled dichloromethane. The aspirated dichloromethane phase was evaporated using a stream of nitrogen at 45 °C. Samples were then reconstituted in phosphate-buffered saline with gelatin. All samples were run in duplicate, and intra-assay variation for Cort ranged from 11.2% to 14.7%, while inter-assay variation was 12.73%. The plasma volumes of the samples varied from 5 to 15 μL.
In 2008, baseline Cort was determined from 29 samples in the high latitude populations. In 2010, baseline Cort was determined from 15 samples in the low latitude population and 8 from the high latitude population. In 2011, baseline Cort was determined from 23 samples in the low latitude population and 17 from the high latitude population. In total, 92 blood samples were analyzed, 38 from the low latitude population and 54 from the high latitude population. There were some instances where by chance we obtained samples from the same individual in different breeding seasons (2008, 2010 and 2011). In the present study these duplications have been excluded (five samples collected from the low latitude population and ten samples from the high latitude population). Summarizing, we included 34 bird encounter histories (with accompanying covariate: baseline Cort) (see below) from the low latitude population and 44 bird encounter histories (with its covariate: baseline Cort) from the high latitude population.
Capture-Mark-Recapture information and analytical procedure
As we had previously observed an absence of differences in baseline Cort levels between males and females [34] together with an absence of difference in survival probability between the sexes [40] we didn’t separate capture history by sex (we didn’t create two groups) (e.g., [32]). In order to maintain the same number of capture events (i.e., eight reproductive seasons), in the high latitude population we used encounter history form 2008 to 2015 and in the low latitude population from 2010 to 2017. For example, an encounter history of “10000100” indicates that the individual was captured and marked in the first reproductive season, not captured from the second to the fifth reproductive season, captured in the sixth reproductive season and not captured (may be alive or maybe not) in the last two reproductive seasons. In the low latitude population the number of capture-recapture occasions per individual ranged from 1 to 6 (1 = 16, 2 = 7, 3 = 6, 4 = 3, 5 = 1 and 6 = 1), resulting in a total of 71 capture-recaptures of 34 individuals. In the high latitude population the number of capture-recapture occasions per individual ranged from 1 to 8 (1 = 22, 2 = 11, 3 = 5, 4 = 2, 5 = 2, 6 = 1 and 8 = 1), resulting in a total of 91 capture-recaptures of 44 individuals.
To examine the effect of baseline Cort on the probability of survival we applied the Cormack-Jolly-Seber (CJS) model [41] as implemented in MARK software, release 5.1 [42,43]. We used captures to calculate the return rate of birds, which depends on their probability of 1) surviving and coming back to the sampling site (Φ, the apparent survival probability) and 2) being encountered (p, the encounter probability). We generated different models with restrictions on the parameters [33,44]. For example, apparent survival Φ (or the rate of recapture p) can be restricted in such a way that it can held constant throughout sampling periods – denoted as (•), or varied between sampling intervals (1-year interval) - denoted as (t).
We first tested whether our data fitted the full time-dependent CJS model (return rate = Φ(t)p(t)) using the median ĉ estimator provided by MARK to estimate the overdispersion of our data. The ĉ value is used in MARK to adjust the AICc through quasi-likelihood, resulting in a QAICc, whenever ĉ departs from 1.0. This variance inflation adjustment allows the use of data sets that depart from the assumptions of binomial distribution [45]. We based our goodness-of-fit on the CJS model because as yet there are no tractable goodness-of-fit methods available for models with an individual covariate. The approach generally recommended is to perform goodness-of-fit tests based on the more general CJS model and to use the same ĉ value for the models that contain covariates [43]. As the median overdispersion factor (ĉ) was always inferior to 3, we used CJS models for further analyses. However, when the median ĉ was superior to 1, we multiplied the variance-covariance matrix by the median ĉ to control for the overdispersion of our data. For both populations, the model with the lowest 2nd order Aikake information criteria (AICc or QAICc when the matrix was multiplied by the median ĉ) was selected as the basic model. For both populations, the null model (return rate = Φ(•)p(•)) (see Results) was the basic model for the subsequent analysis (the lowest AICc or QAICc).
We analyzed the effect of baseline Cort on inter-annual survival by including baseline Cort as a covariate in the basic models and selecting the model with the lowest AICc (or QAICc). Then, we used likelihood ratio tests to determine the significance of the covariates. Some models incorporated Cort as a linear function of survival (denoted: Cort); others were tested for a curvilinear survival function using the square of Cort as a covariate (denoted: Cort-sq). MARK automatically standardized all covariates by subtracting the mean from each and dividing by the standard deviation.