Methods

The Seychelles warbler model system

The Seychelles warbler is a small insectivorous passerine currently distributed across five islands in the Seychelles. The population on Cousin Island (29 ha; 4°20’ S, 55°40’ E) – containing ca. 320 individuals – has been extensively monitored since 1986 (Komdeur, 1992; Hammers et al. , 2015). Since 1997, nearly all individuals (>96%) have been ringed with a unique combination of a British Trust for Ornithology (BTO) metal ring and three colour rings for identification (Raj Pant et al. , 2020) (Richardson et al. , 2001). Individuals were usually first caught as nestlings, or as dependent juveniles (<8 months old) with mist nets (see Kingmaet al. , 2016 for details). Juveniles were aged as fledglings (1–3 months), old fledglings (3–5 months) or sub-adults (5–12 months) based on behaviour and eye colour (Komdeur, 1992). Since the resighting probability of individuals during the major breeding season is close to one – 0.98 for individuals ≥2 years-old (Brouwer et al. , 2006) – and dispersal from the island is virtually absent (Komdeur et al. , 2004), individuals that were not observed during the major breeding season were assumed dead. First year survival is 0.61 ± 0.09 SE, increasing to 0.84 ± 0.04 SE annual survival in adults (Brouweret al. , 2006). For individuals reaching fledgling age, the mean life expectancy is 5.5 years (Komdeur, 1991), and the maximum recorded lifespan is 19 years (Hammers and Brouwer, 2017).
The population is structured into ca 115 clearly defined territories (Kingma et al. , 2016). The availability of the warbler’s invertebrate prey (Komdeur, 1992) varies considerably due to the interacting effects of defoliating salt spray (along coastal territories), tree species abundance, elevation and rainfall (Van de Crommenacker et al. , 2011). Territories are defended year-round by a single dominant breeding pair, but ca 40% of territories include an additional 1–5 sexually mature subordinates, often past offspring of the same dominant pair (Richardson, Burke and Komdeur, 2002).
The majority of breeding activity (94% of territories) occurs from June to August, but a minor breeding season also occurs from January to March (Komdeur and Daan, 2005). Breeding attempts usually consist of one-egg clutches (Komdeur, 1994). Only females incubate while both sexes provision chicks and fledglings for ca three months post-fledging. Around one third of subordinates also provide alloparental care to group offspring, hereafter ‘helpers’ (Komdeur, 1994; Hammers et al. , 2019). About 44% of female helpers are also co-breeders (Richardson et al. , 2001; Raj Pant et al. , 2019). The offspring of cobreeders are jointly cared for by the subordinate female and dominant pair (Richardson et al. , 2001; Bebbington et al. , 2018). The frequency of extra-pair paternity in the population is high (ca 41%; Raj Pant et al. , 2019) and such paternity is nearly always gained by dominant males from other territories (Richardson et al. , 2001), but males only provide parental care in their own territory.

Data collection

Our study uses data collected from 1995 to 2015. Each year (June–September), during the major breeding season, each territory was visited at least every two weeks to determine the identity and status of group individuals. During visits, the dominant female was followed for at least 15 minutes to assess breeding activity (Richardson, Burke and Komdeur, 2007). Territories with an active nest were visited every 3–4 days until the nestling(s) have fledged or the breeding attempt failed. Observations of incubating and/or provisioning were used to estimate hatching/fledging dates, and to determine whether any subordinates present in the territory were helpers (Richardson, et al ., 2002; van Boheemen et al. , 2019). For each territory, the availability of food was calculated (following Komdeur, 1992). Briefly, the number of insects (on the undersides of leaves) was multiplied by the percentage cover of broad-leaf vegetation within territories. This number was then divided by the number of adult territory occupants to give food availability per individual (Brouwer et al. , 2006).
During each major breeding season, as much of the adult population as possible (normally around 30%) was caught and re-sampled: ca 25 ul of blood was taken from the brachial vein and stored in 100% ethanol (Richardson et al. , 2001). DNA extracted from the blood samples (following Richardson et al. , 2001) was used to confirm sex and assign parentage using MasterBayes 2.52 based on genotypes derived from 30 microsatellite loci (for details see Sparks et al. , 2020). The presence of haemosporidian infection (Haemoproteus nucleocondensus ; hereafter referred to as malaria) - the only known parasite in the Seychelles warbler (Hutchings, 2009) - was screened for following Hellgren et al . (2004). In the Seychelles warbler, nearly all individuals become infected with malaria in their first year (Hammers et al. , 2016). Thus, in our study (of individuals >1 year of age), birds in which we did not detect malaria may be in the latent infection stage – where parasites persist in organs - or have cleared infection, while infected individuals are either in the later chronic stage of infection, or in a subsequent relapse or reinfection. Infection with malaria does not appear to have an impact on annual survival in the Seychelles warbler (Hammers et al. , 2016) but has been linked to telomere attrition in the great reed warbler (Acrocephalus arundinaceus : Asghar et al. , 2015).
Relative Telomere Length (RTL; the concentration of amplified telomeric DNA relative to that amplified at GAPDH – a single copy gene) had previously been measured using qPCR as part of another study (Spurgin et al. , 2018). Since avian erythrocytes are nucleated and vastly outnumber other blood cell types, blood RTL is effectively a measure of erythrocyte RTL (Stier et al. , 2015). Individuals with two or more RTL measurements were used in the current study, with the difference between consecutive pairs of RTL measurements (ΔRTL) as the response variable. We excluded RTL measurements from young individuals (< 1 year), as previous work in this population has shown that within-individual rate of attrition per annum is an order of magnitude greater in the first year compared to adult life (Spurgin et al. , 2018). For consistency, we focused on RTL measurements from catches only within the major breeding season, since inter-seasonal ∆RTL could reflect seasonal effects on RTL. Individuals are caught opportunistically, meaning that the follow-up period between RTL measurements (hereafter ∆RTL period) ranged from one year (i.e. consecutive seasons) to 9 years (Fig. S1). ∆RTL did not exhibit a significant relationship with the ∆RTL period. The final dataset comprised 359 ΔRTL measures from 227 adults.
Reproductive effort was measured as the number of offspring raised by an individual in the ∆RTL period; specifically, offspring that had hatched after time 1 and had reached independence (3 months old) before time 2. Social offspring – those for which a dominant breeder provides parental care – are determined from behavioral observations during nest attempts. Offspring are genotyped to identify genetic parentage. This is an underestimation of total offspring produced, since we excluded offspring for which parents could not be assigned (ca 15% of offspring ) and some offspring are likely to have died before being sampled (Edwards, Burke and Dugdale, 2017). We used the number of social offspring as our estimate of reproductive output in males (since males do not care for offspring sired in other territories). Females (dominants or co-breeder) always contribute to the care of any offspring in the nest; (Richardson, Burke and Komdeur, 2002). Thus, female reproductive output was the number of co-bred offspring. For both sexes, the majority of individuals had 0–2 offspring within each ∆RTL period (Fig. S1). Offspring number was positively correlated with the ∆RTL period (Pearson’s r = 0.69, df = 357, P < 0.001, Fig. 1S), meaning that individuals typically produced one offspring every two years.
During the ∆RTL period, we averaged food availability (insect abundance per individual per field season) across field seasons. Reproductive effort - in terms of time spent incubating and provisioning - of dominant breeders is reduced by the presence of helpers, including co-breeders (Hammers et al. , 2019; van Boheemen et al. , 2019). Reduced telomere attrition in dominant females has been associated with the presence of helpers in a previous study (Hammerset al. , 2019). Therefore, we determined whether nest helpers (including co-breeders) were present in the territories of individuals that produced offspring.

Statistical analysis

Using RStudio (v1.2.5033, Rstudio Team 2020) we tested whether food availability, reproductive output, helper presence and malaria status predicted ∆RTL, with the prediction that high food availability, low reproductive output, helper presence and no malaria infection - or a combination of these factors - would result in telomere lengthening. We adjusted ∆RTL following Verhulst et al. (2013a); this method subtracts the mean difference between successive samples expected from the regression-to-mean effect, estimated by the correlation between successive samples. In our dataset, this correlation was very weak (Peason’s r = 0.06, df = 357 P = 0.22), as expected given the low within-individual consistency of RTL in this system (Spurgin et al. , 2018). This results in an adjusted ∆RTL (hereafter ∆RTL) which is equivalent to RTL at time 2; positive values indicate longer RTL and negative values indicate shorter RTL, relative to the population mean RTL. As expected, ∆RTL was strongly correlated with unadjusted ∆RTL (Pearson’s r = 0.71, df = 357,P < 0.001), meaning that individuals with lengthened or shortened RTL (relative to their initial RTL) tended to have more positive or more negative ∆RTL, respectively.
The association between factors and ∆RTL was tested using Linear Mixed Models (package lme4 v1.1-25; Bates et al. , 2015). We deliberately focused on a restricted set of fixed effects – chosen a priori based on logic and evidence of influencing telomere dynamics – to avoid data dredging, which could generate false-positive associations. Chosen fixed effects included mean food availability (continuous), number of offspring (continuous), malaria status (infected or uninfected), helper presence (yes or no). We also included age at time 1 (continuous), ∆RTL period and logical two-way interactions; for example, the effect of offspring production on telomere maintenance may depend on food availability and/or helper presence. Effects were likely to differ between the sexes due to differing investments in reproduction, so to investigate sex-specific differences, while also avoiding the need for complex three-way interactions, separate models were created for males and females. We included the sample year of the first RTL measurement as a random factor. Variation in RTL between qPCR plates (Sparks et al. , 2020) could contribute to variation in ∆RTL, since RTL measurements from longitudinal samples were run on separate plates. Therefore, the plate identities of both RTL measurements per ∆RTL were included as random factors. Since individuals with three or more RTL measurements had multiple measures of ∆RTL, individual identity was also included as a random factor. Full models were reported after removing non-significant interactions. Offspring number, food availability and ∆RTL period were log10 transformed (for normality) and mean centered to remove collinearity between their main effects and interaction (Schielzeth, 2010). Since most individuals had 0–2 offspring, offspring number was reduced to a categorical variable (zero, one or ≥2) for graphical interpretation of interactions.
To test whether ∆RTL influenced subsequent survival, we performed a Cox proportional hazards regression analysis (package survivalv3.2-7; Therneau, 2014). The response variable was the number of years an individual lived beyond the sampling date of its last RTL measurement. We included 16 individuals that were still alive in 2020 as right-censored data points. 13 individuals translocated to other islands post-sampling were excluded, leaving 214 individuals. Predictor variables were ∆RTL, sex and age at last RTL measurement (since older individuals are expected to have shorter remaining lifespans). Where multiple measures of ∆RTL were available per individual, we used only the last measurement, meaning that each ∆RTL value represents the last known change in telomere length before the individual died. We were interested in whether predicted hazard ratios exhibited a proportional change across the range of ∆RTL values, or whether the association was nonlinear. Therefore, we modelled ∆RTL as both a linear and quadratic function. In this model, positive hazard coefficients would indicate a decreased probability of survival with increasing values of the predictor variable. Hazard ratios represent the effect size of predictors; for example, a hazard ratio of 2 indicates that the risk of death is twice as high for the corresponding change in a predictor variable.