Study species and sampling
We used 125 Zebra Finches (66 females; 57 males), originating from 73 broods, reared from stock that were part of a long-term experiment, in which natal brood size and energy expenditure required for foraging were manipulated (De Coster et al. 2011; Koetsier and Verhulst 2011). With respect to the foraging cost manipulation, we only used control birds that had easy access to food. Parents were paired randomly and housed in pairs during breeding; hence, paternity was known with certainty. The inbreeding level is low in our Groningen Zebra Finch population (Forstmeier et al. 2007). Four days after the first chick of a brood hatched, we conducted a brood size manipulation, in which brood size was standardized to either 2 or 6 young (both within the natural range). Our aim was to cross-foster all individuals in this procedure, but due to logistic constraints, we cross-fostered 75 % of the chicks (N = 94).
We measured telomere lengths in DNA from red blood cells. Blood was collected from the brachial vein into heparinised capillaries. Samples were suspended in 2 % EDTA buffer, and within 2 days the red blood cells were spun down, and the pellet was stored in glycerol buffer at −80 °C after snap freezing. We used blood samples collected in 2006–2010. Storage time prior to analysis (0–6 years) did not affect telomere length (F
6,152 = 0.66, p = 0.68). The samples were analysed, divided over seven gels, and timing (batch) of analysis did not affect telomere measurements (F
6,152 = 1.48, p = 0.19). Hence, storage time and timing of analysis were not included in the analyses.
On average, individuals were 132 days old (SE = 11.6, range = 9–636) when a blood sample was collected. For 18 individuals, we analysed two to three samples, as these individuals were sampled multiple times in life, resulting in a total of 158 telomere length estimates. On average, the samples after the first sample were taken at an age of 940 days (SE = 52.5, range = 609–1572). To simplify the models for estimating heritability, and because we had repeated measurements of only a subset of all individuals (18 out of 125 individuals), we used the average telomere length as a Terminal Restriction Fragment (TRF) estimate per individual to calculate heritability. In order to estimate variance components that might influence phenotypic similarity, we used the complete data set, including repeated measurements. Because telomere length generally declines with age, we controlled for age in all analyses.
TRF assay
The TRF assay was conducted following Salomons et al. (2009). In summary, 5 μl of red blood cells were suspended in an agarose solution to form an agarose plug (0.8 %; following the manufacturer’s protocol, CHEF Mammalian Genomic DNA Plug kit, Bio-Rad Laboratories, Inc., USA). The cells in half a plug were digested overnight at 50 °C with Proteinase K. DNA was then digested overnight at 37 °C using a mixture of three restrictions enzymes, HindIII (60 U), HinfI (30 U) and MspI (60 U), in NEB2 buffer (New England Biolabs, Inc., Beverly MS, USA).
The restricted DNA and the size standards (Molecular Weight Marker XV, Roche and 1 kb DNA ladder, New England Biolabs) were electrophoresed through a 0.8 % agarose gel by pulsed field gel electrophoresis at 14 °C for 24 h (3.5 V/cm, initial switch time 0.5 s, final switch time 7.0 s). Gels were dried with a gel dryer (Bio-Rad, model 538) and hybridized overnight with 32P-labelled oligo (5-CCCTAA-3)4, which labelled the single-stranded overhang of the telomeres. Since the DNA was not denatured as in Southern blot techniques, no 32P-labelled oligo marked interstitial repeats. The radioactive signal of the marker was detected by a phosphor screen (PerkinElmer Inc., USA), and analysed using a phosphor imager (Cyclone TM Storage Phosphor System, PerkinElmer).
Telomere length varies among cells and chromosomes (Lansdorp et al. 1996); and hence, the TRF assay results in a smear, instead of a clear band. The distribution of telomere lengths was calculated based on densitometry (Haussmann and Mauck 2008) in the open-source software IMAGEJ v. 1.38x (Salomons et al. 2009). The average labelled telomere length per lane was calculated as: Σ (OD
i
× L
i
)/Σ (ODi), where OD
i
is the optical density output at position i, and L
i
is the length of the DNA (bp) at position i. OD is corrected for the background by subtracting the average grey value of non-DNA containing gel in IMAGEJ. Our lower limit was 2.3 kb, which falls within the smallest band of the 1 kb DNA ladder, which is 1 kb, and our upper limit was an extrapolated value of 80 kb based on the Molecular Weight Marker XV, which has a range of 2.4–48.5 kb, because telomere lengths of the Zebra Finches exceeded the Molecular Weight Marker XV. Note, however, that the extrapolation comprised < 1.5 cm on the gel (±7 % of the total length used), and that there was a strong correlation between calculations of telomere length based on the Molecular Weight Marker XV (up to 48.5 kb) and the same samples quantified with the extrapolated marker (up to 80 kb) (r = 0.82). Based on the repeated measures of 18 individuals, individual variation in TRF assays was 78 % of the total variance in telomere lengths. Since repeat abilities of our TRF assays are high (Jeanclos et al. 2000; Haussmann and Mauck 2008; Salomons et al. 2009) and the analysis is time consuming, all samples were run once.
Statistical analyses
We compiled a pedigree using the Groningen Zebra Finch database. Data on ancestry were available for four generations of birds, with the earliest records dating back to 2004. We used pedigree data pruned back to the 125 phenotyped individuals, plus 143 unphenotyped individuals linking the phenotyped birds. The pedigree contained 112 individuals in a full sibling relationship, 45 maternal half-siblings and 62 paternal half-siblings. Half-sibling comparison facilitated attempts to separate genetic and environmental components. We over-represented paternal half-siblings in our data collection, because females lay the eggs and may thereby potentially exert a greater environmental influence on offspring telomere length, and we were primarily interested in the genetic component of the variance. For further details of the pedigree, see Table S1 (Supplementary material).
We calculated the heritability of telomere length with an ‘animal’ model (Kruuk 2004), using a Bayesian approach (Hadfield 2010), estimating the posterior mode and 95 % credible intervals (95 % Cred. Int.) for fixed effects, variance components and heritability. In short, an ‘animal’ model uses a pedigree to calculate the proportion of the phenotypic variance that is due to additive genetic effects, by comparing the covariance due to additive genetic effects in a phenotype between relatives. We calculated heritability using the package MCMCglmm (2.15) in R 2.14.1 (Hadfield 2010; R Development Core Team 2011) with 10,000,000 iterations, a burn-in of 2500,000 and a thinning interval of 5000. Autocorrelation between sampled iterations was < 0.08. We used default priors for fixed effects, parameter expanded priors for the random variance structure (variance = 1, degree of belief = 1, prior mean = 0, prior covariance matrix = 500), and non-informative inverse-Wishart priors for the residual variance structure (variance = 1, degree of belief = 0.002). We applied several different prior distributions to confirm that our estimate of additive genetic variance was robust to prior specification.
Telomere length was normally distributed (Shapiro–Wilk W = 0.989, p = 0.25, N = 158). Exploratory analyses indicated that including sex and the logarithm of age (we log-transformed age because telomeres shorten faster early in life) as fixed effects improved our model fit. For individuals with an average telomere length of multiple TRF estimates, we used the logarithm of the average age at sampling. Because the data set did not have sufficient power to discriminate between the random variance components explaining environmental (birth nest and permanent environment, meaning an individual’s own common environment) and parental effects, we used a naïve model including only the pedigree component as a random effect. We therefore estimated the variance components as:
$$V_{\text{P}} = V_{\text{A}} + V_{\text{R}}$$
(1)
where V
P
is the phenotypic variance, accounting for the fixed effects of sex and the logarithm of the age at which an individual was sampled; V
A is the additive genetic variance and V
R the residual variance. We then calculated heritability as:
$$h^{2} = \frac{{V_{\text{A}} }}{{V_{\text{P}} }}$$
(2)
We compared h
2 from the ‘animal’ model with the heritability estimate based on the intraclass correlations for sibships (Falconer and Mackay 1996), to evaluate robustness of our findings. The individual least square mean estimate of telomere length was used from a model including the logarithm of age. For the full sibling comparison, we used a general linear mixed model (GLMM) with family identity as a random effect, where a family is defined as a set of full siblings. Second, we made a half-sibling comparison in a subset of the data containing half-siblings, by building a GLMM with either mother identity or father identity as a random effect. Here we used a mean value of telomere length for each birth nest, to prevent our estimate from being biased by pseudo-replication by full siblings. This approach allowed us to compare similarity between maternal- and paternal half-siblings, testing for a trans-generational effect from mother or father. Sample sizes for all sibling relationships can be found in Table 4. Some of the individuals in the full sibling comparison were also included in the maternal (N = 14) and paternal (N = 31) half-sibling comparison. Following Falconer and Mackay (1996) equation 9.8, heritability in the full sibling data set was calculated by multiplying the correlation between full siblings by two:
$$t_{\text{FS}} = \frac{{\frac{1}{2}V_{\text{A}} + \frac{1}{4}V_{\text{D}} }}{{V_{\text{P}} }}$$
(3)
where t
FS is the correlation between full siblings and V
D the dominance variance. In general V
D is relatively small compared to V
A, and hence could be ignored (Falconer and Mackay 1996). This was confirmed by the similarity of our heritability estimates of full siblings and half-siblings (see “Results”). Heritability using the half-sibling data set was calculated by multiplying the correlation between half-siblings by 4, according Falconer and Mackay (1996) (equation 9.6)
$$t_{\text{HS}} = \frac{1}{4}\frac{{V_{\text{A}} }}{{V_{\text{P}} }}$$
(4)
where t
HS is the correlation between half-siblings.
As we could not include random effects in the ‘animal’ model, nor separate variance components in the intraclass correlations, we estimated variance components that may influence phenotypic similarity in telomere length with a GLMM. We ran a GLMM with sex and the logarithm of age as fixed effects, where our response was telomere length at a given age, rather than one averaged value, and included individual as random effect in the model. In Table 2, we describe which environmental and genetic effects are embedded in each variance component. We defined phenotypic variance components as significantly different from zero if their confidence intervals (CI) did not overlap with zero.
Table 2 Separation of phenotypic variance components into genetic, environmental and parental effects
Ethics statement
The brood size manipulations and long term foraging experiment with Zebra Finches, including blood sampling, have been approved by the animal welfare ethics committee of the University of Groningen (according to Dutch law), under license number 5150.