Carry‐over effects on the annual cycle of a migratory seabird: an experimental study

Summary Long‐lived migratory animals must balance the cost of current reproduction with their own condition ahead of a challenging migration and future reproduction. In these species, carry‐over effects, which occur when events in one season affect the outcome of the subsequent season, may be particularly exacerbated. However, how carry‐over effects influence future breeding outcomes and whether (and how) they also affect behaviour during migration and wintering is unclear. Here we investigate carry‐over effects induced by a controlled, bidirectional manipulation of the duration of reproductive effort on the migratory, wintering and subsequent breeding behaviour of a long‐lived migratory seabird, the Manx shearwater Puffinus puffinus. By cross‐fostering chicks of different age between nests, we successfully prolonged or shortened by ∼25% the chick‐rearing period of 42 breeding pairs. We tracked the adults with geolocators over the subsequent year and combined migration route data with at‐sea activity budgets obtained from high‐resolution saltwater‐immersion data. Migratory behaviour was also recorded during non‐experimental years (the year before and/or two years after manipulation) for a subset of birds, allowing comparison between experimental and non‐experimental years within treatment groups. All birds cared for chicks until normal fledging age, resulting in birds with a longer breeding period delaying their departure on migration; however, birds that finished breeding earlier did not start migrating earlier. Increased reproductive effort resulted in less time spent at the wintering grounds, a reduction in time spent resting daily and a delayed start of breeding with lighter eggs and chicks and lower breeding success the following breeding season. Conversely, reduced reproductive effort resulted in more time resting and less time foraging during the winter, but a similar breeding phenology and success compared with control birds the following year, suggesting that ‘positive’ carry‐over effects may also occur but perhaps have a less long‐lasting impact than those incurred from increased reproductive effort. Our results shed light on how carry‐over effects can develop and modify an adult animal's behaviour year‐round and reveal how a complex interaction between current and future reproductive fitness, individual condition and external constraints can influence life‐history decisions.

each of the 6 states the following behaviours using the mean immersion time of the states, and observing temporal patterns: sitting on the water, sustained flight, foraging type 1 (short wet bouts in-between longer dry bouts, perhaps more associated with searching) and foraging type 2 (short dry bouts in-between longer wet bouts) during the day; sitting on the water (mostly wet) and visiting the colony (mostly dry) at night ( Figure S1).
Models were first trained with data from control birds and then applied to data from treatment birds and, when available, from non-experimental years. The most likely sequence of hidden states was calculated and each data point (corresponding to a 10-minute period) was allocated a state.
Unlike (Dean et al. 2012;Freeman et al. 2013) we did not have any other tracking data to validate our classification, therefore we also classified our data in 3 states with simple thresholds (sitting: <3% dry; flight: >97% dry; "foraging": anything in-between), similar methods were used in a number of species (Yamamoto et al. 2008;Lecomte et al. 2010;Catry et al. 2011;Ramirez et al. 2013) including Manx shearwaters (Shoji et al. 2015). All results were comparable, therefore we only present below the results using the HMM classification.

Natural differences in lay date prior to the experiment and individual consistency in phenotypic quality
In the year of manipulation, natural differences in laying date existed between groups. On average during the year of manipulation, control birds laid 3.3 ± 0.7 days before the colony median lay date, "shorter effort" birds laid 7.9 ± 1.7 days after the colony median, and "longer effort" birds laid 6.4 ± 1.4 days before the colony median. The laying dates in the manipulation year ranged from 14 days before to 20 days after the colony median, close to the actual range of laying dates at this colony. Birds in the "lower effort" treatment group laid later than control birds (LMM: n control =12, n lower effort =12; parameter estimate = 11.3 ± 1.9, R 2 = 0.63, χ 1 2 = 23.4, P < 0.001) while birds in the "higher effort" treatment group laid earlier (despite a lack of significance the effect size is relatively important) (LMM: n higher effort =14; parameter estimate = -3.2 ± 1.6, R 2 = 0.19, χ 1 2 = 3.5, P = 0.060).
Attributing the shifts in lay date and egg mass we observed in the "higher effort" pairs following manipulation to an effect of the manipulation requires the assumption that individuals are consistent in their quality and hence in their breeding variables such as laying date and egg mass (otherwise one could argue that the early laying birds chosen for the "higher effort" group were having a good year and simply returned to their average state the next year, regardless of the manipulation). Individual consistency in laying date and egg mass has been shown by previous studies of Manx shearwaters (Brooke 1978, Brooke 1990), but we did our own analyses to check whether this was the case in three different ways.
First, the pattern in laying date observed in the experimental year was similar the previous year: "lower effort" birds laid significantly later than controls (LMM: n control =9, n lower effort =9; parameter estimate = 5.9 ± 2.4, R 2 = 0.36, χ 1 2 = 5.5, P = 0.019) while there was no significant difference (but an effect size similar to the experimental year) between control and "higher effort" birds (LMM: n higher effort =9; parameter estimate = 1.3 ± 2.5, R 2 = 0.19, χ 1 2 = 0.3, P = 0.579), which indicate a lack of major changes in lay date between the experimental year and previous years.
Second, we tried to match pairs from the "higher effort" group to un-manipulated pairs which laid at the same time during the experimental year, to test whether the shift in lay date and egg mass observed in our study after manipulation was not due to these early laying birds returning to their normal average state. We could only match 5 "higher effort" pairs to un-manipulated pairs. The 5 "higher effort" pairs laid 7.8 ± 2.0 days later in the year after manipulation than their matched un-manipulated pairs (paired T-test, t= -3.30, df = 4, P = 0.029).
Similarly, while the 5 "longer effort" pairs laid eggs 1.4 ± 1.0 g lighter than they did prior to manipulation, the un-manipulated pairs laid eggs 0.4 ± 1.0 g heavier, but the difference was not statistically significant (paired Ttest, t= 1.09, df = 4, P = 0.306). While the sample sizes are small, these results suggest that the delay in lay date and decrease in egg mass we observed after manipulation in the "longer effort" pairs was indeed due to the manipulation, and not because the birds were in an "abnormal" state in the experimental year and returned to their normal state after a good year.
Finally, we used 44 un-manipulated pairs for which we had data on timing of laying and egg mass for 2 breeding seasons during the 2011-2014 period (this period covers all the data included in our study). On average, the difference in laying date between two years was 0.16 ± 0.81 days, not statistically significant (paired T-test, t=-0.2, df = 43, P = 0.845). The difference in egg mass was 0.23 ± 0.49 g, also not statistically significant (paired T-test, t=-0.5, df = 36, P = 0.645). These results held when we looked at pairs for which we knew lay date for 2 consecutive years and breeding success for at least the first of these 2 years (n=29). Taken together, these results agree with previous studies and show that un-manipulated Manx shearwaters are consistent year to year in their laying date and egg mass. Therefore, we are confident that the results we observed in our study are due to the effect of the manipulation. Chick mass and fledging success are known to be strongly associated with laying date and egg size in shearwater species (Perrins 1966;Perrins, Harris & Britton 1973;Ramos et al. 2003), so the consistency in laying date and egg mass is likely to lead to consistency in the other variables.

Sample sizes
For several reasons sample sizes vary between analyses. To avoid confusion, here are some explanations about why numbers vary so much. Please note that all sample sizes are given for every analysis, in the main text or in Table 1.
For laying dates and breeding variables in general, all nests were rigorously monitored from hatching to fledgling, in the 2 experimental years (i.e. 62 nests in total). While many of the nests were monitored from the start of the season and also in previous and following years, some nests were only monitored from the incubation stage, i.e. we did not know the laying date. It would have been possible to make an approximation by substracting 51 days to hatching date (the average duration of incubation in this species), but we preferred only to include dates we knew with certainty. Sample sizes in the years before and after manipulation also tend to be smaller, because some birds did not return, or changed burrow and were not detected.
As far as the sample sizes during the winter are concerned, the decreasing sample sizes as winter passes are due to more and more devices failing to record data. As a result, we have a large number of autumn migration tracks (all devices were working at the end of the breeding season when we deployed them), but we lost a few devices during the wintering period, and a few more during the spring migration. Similarly, an unfortunate rate of failure of devices in the years prior to the experiment led to smaller sample sizes for the within-group longitudinal comparisons in wintering at-sea behaviour. foraging and (d) flying), for the breeding season and the overwintering period, for each group ("lower effort" treatment in white and black stripy pattern, control in light grey and "higher effort" treatment in dark grey).
Supplementary Tables   Table S1. Number of complete tracks collected during each period of the annual cycle (excluding pre-laying and incubation), for experimental and non-experimental years. Numbers within brackets represent the number of incomplete tracks (due to device failure).