Organisms need to time their annual-cycle stages, like breeding and migration, to occur at the right time of the year. Climate change has shifted the timing of annual-cycle stages at different rates, thereby tightening or lifting time constraints of these annual-cycle stages, a rarely studied consequence of climate change. The degree to which these constraints are affected by climate change depends on whether consecutive stages are causally linked (scenario I) or whether the timing of each stage is independent of other stages (scenario II). Under scenario I, a change in timing in one stage has knock-on timing effects on subsequent stages, whereas under scenario II, a shift in the timing of one stage affects the degree of overlap with previous and subsequent stages. To test this, we combined field manipulations, captivity measurements and geolocation data. We advanced and delayed hatching dates in pied flycatchers (Ficedula hypoleuca) and measured how the timing of subsequent stages (male moult and migration) were affected. There was no causal effect of manipulated hatching dates on the onset of moult and departure to Africa. Thus, advancing hatching dates reduced the male moult–breeding overlap with no effect on the moult–migration interval. Interestingly, the wintering location of delayed males was more westwards, suggesting that delaying the termination of breeding carries over to winter location. Because we found no causal linkage of the timing of annual-cycle stages, climate change could shift these stages at different rates, with the risk that the time available for some becomes so short that this will have major fitness consequences.
Organisms living in seasonal environments, such as temperate zones, are generally exposed to predictable changes in environmental conditions that determine the organization of their annual-cycle stages. As some of these stages, such as breeding and migration, are energetically costly, their occurrence needs to be timed correctly to match favourable conditions (Enright, 1970; Gwinner, 1996; Jacobs and Wingfield, 2000; Wingfield, 2008). Annual cycles vary in complexity in a species-specific way: from simple breeding/non-breeding transitions to the inclusion of other energetically demanding stages like migration or moult (Wingfield, 2008). As the total time available for the allocation of these seasonal stages is 1 year, more complex annual cycles will necessarily involve a concomitant increase in time constraints. This is especially true if these different stages are all energetically demanding and need to be temporally segregated (Dietz et al., 2013). Complex annual cycles would thus result in low flexibility as further shortening or overlapping stages would lead to fitness costs (Jacobs and Wingfield, 2000; Wingfield, 2008).
To achieve synchronization between annual-cycle stages and environmental cycles, organisms use predictive cues, with photoperiod and temperature as the most relevant ones (Dawson, 2015; Wingfield, 2008; Winkler et al., 2014). While photoperiod is important as a long-term predictive cue, because it remains unchanged across the years (Gwinner, 1989), temperature provides short-term temporal information for fine-tuning timing within a particular year and allows organisms to match their timing with the year-to-year variation in optimal timing (Dawson, 2005; Schaper et al., 2012; Visser et al., 2004, 2009). The importance of photoperiod and temperature as cues varies across annual-cycle stages. Hence, increasing temperatures due to climate change will cause shifts in the timing of stages that are temperature sensitive, such as reproduction (Knudsen et al., 2011; Parmesan, 2006; Visser and Both, 2005), while other stages may not shift if mainly affected by photoperiod or circannual rhythms (Both and Visser, 2001; Tomotani et al., 2018a; Visser et al., 2004). How this affects the whole annual cycle, however, depends on how stages are interrelated (Crozier et al., 2008; Hera et al., 2013).
One possibility is the existence of a tight link between stages. In this scenario (scenario I: linked), the timing of a given stage depends strongly on the timing of the previous one(s). Such temporal links across stages could be caused, for example, by pleiotropic effects of hormones regulating the transitions between stages (Dawson, 2006; Jacobs and Wingfield, 2000; Williams, 2012). This would mean that even if one stage is sensitive to temperature and the other not, shifts in a given stage would cause similar shifts in the subsequent ones. For instance, in an experiment with captive juvenile Eurasian blackcaps (Sylvia atricapilla), migratory activity was only initiated after the completion of moult; thus, the interval between the two stages was fairly constant among individuals (Pulido and Coppack, 2004). Under this scenario, climate change would not increase or decrease the overlap of stages. However, depending on the magnitude of the shift, stages could be displaced in time and desynchronized from the optimal environmental conditions for that stage.
Alternatively, consecutive stages may not be linked (scenario II: independence). Thus, if the effect of cues varies between stages, climate change will lead to shifts in the timing of annual-cycle stages with different rates, causing an increase or decrease in the degree of overlap between events. This could result in time constraints being tightened or lifted throughout the annual cycle because a shift in one stage would lead to more time in one part of the cycle, but subsequently may result in less time available for other stages. For example, between 1980 and 2000, European pied flycatchers (Ficedula hypoleuca) advanced egg-laying dates more strongly than their arrival dates at breeding grounds, associated with climate change. This has led to a shorter interval between arrival and breeding, constraining how much laying dates could continue to advance with increased climate change (Both and Visser, 2001; Tomotani et al., 2018a).
Finally, it is also possible that both of the above-mentioned scenarios occur in different parts of the cycle (Salis et al., 2017; Tomotani et al., 2016). For example, it has been suggested that pre-breeding stages are more time selected than post-breeding stages, and are, as a result, less plastic (Conklin et al., 2013; Karagicheva et al., 2016). Under that perspective, it would be expected that pre-breeding stages are more likely to fall under scenario II than post-breeding stages. Because these two scenarios have very different consequences in terms of costs of climate change, understanding annual-cycle adaptation to environmental changes requires knowledge on how consecutive stages are temporally linked.
This interrelationship across annual-cycle stages is usually investigated by correlating patterns of timing and duration of annual-cycle stages at the individual (Conklin and Battley, 2012; Echeverry-Galvis and Hau, 2013; Gill et al., 2014; Low et al., 2015) and population levels (Both and Visser, 2001; Finch et al., 2014; Gill et al., 2014; Moyes et al., 2011; Ozgul et al., 2010; Rakhimberdiev et al., 2018). Such studies are necessary to describe when in the annual cycle individuals may be shifted or constrained by, or otherwise suffer from carry-over effects. However, to fully understand the temporal links across stages, it is necessary to experimentally test for causality of these interrelations, thus controlling for potential confounding effects. A key way of doing this is to manipulate the timing of a given stage and look for cascading effects on the timing of consecutive stages.
A number of studies have directly or indirectly tested the link between consecutive stages, particularly by looking at the trade-off between moult and breeding (see Tomotani et al., 2018b, for a review). In many cases, the purpose of these studies was to understand how costly breeding would cause a delay in the onset of moult, for example via food availability manipulation (Siikamäki, 1998) and experimental delays in the timing of breeding (Hemborg, 1998; Nilsson and Svensson, 1996; Siikamäki, 1998), where delays lead to a higher moult–breeding overlap. Hemborg (1998) carried out a comprehensive study of the relationship between the timing of breeding and the timing of moult. By manipulating hatching dates, he showed a different control for the moult onset of males and females: the manipulations did not affect female moult (consistent with scenario II), while it had an effect for males (consistent with scenario I). One limitation of this study is the use of moult scores instead of actual dates when moult started. This makes it more difficult to assess whether individuals really started to moult at a certain chick age or calendar day (true scenario I versus II).
Here, we tested whether experimental shifts in the termination of breeding, as a key annual-cycle stage, causally affect the timing of subsequent stages. We experimentally shifted hatching dates as a proxy for the termination of reproduction and measured the timing of post-breeding moult and migration. Previous studies have suggested a difference in the control of the timing of moult in females and males, which leads to a sexual conflict as males and females unequally prioritize their current reproductive success (Hemborg, 1998, 1999; Hemborg and Merila, 1998; Tomotani et al., 2018b). We were mostly interested in how adult males were affected; because they supposedly invest less in their current reproduction, males present a larger moult–breeding overlap than females. Moreover, males have a higher fidelity to the breeding area (Both et al., 2016), making them more suitable for evaluating the effects of our manipulations. Thus, after the chick-rearing phase, we monitored the males' post-breeding moult by temporally keeping them in captivity, their moult control via analyses of the hormone prolactin and, after release, the timing of their autumn migration and wintering site selection using light-level geolocators.
The timing of moult and departure could (scenario I) depend on a previous stage (linked). Such a previous stage could be: (Ia) termination of breeding, the manipulated stage; thus, shifting the termination of breeding also shifts the onset of moult and migration; or (Ib) spring arrival or onset of breeding, which are non-manipulated stages. Alternatively, (scenario II) timing does not depend on a previous stage (independence). It could, for example, simply depend on photoperiod or other cue(s), but not on the timing of breeding. Notice that in both scenarios Ib and II, the timing of moult and migration departure will not be related to the manipulation, but, while between-year variation in the timing of moult and migration will certainly exist for scenario Ib, this will not necessarily be the case for scenario II (i.e. if only related to photoperiod). Thus, we cannot completely distinguish between Ib and II as we would need an experimental period of more than 2 years to estimate year to year variation, but we can provide contrasting predictions for scenario Ia versus Ib and II.
MATERIALS AND METHODS
Study area and study species
European pied flycatchers, Ficedula hypoleuca ([Pallas], 1764), are small long-distance migratory birds that reproduce in Europe and Western Asia and winter in West Africa (Ivory Coast in the case of this study population). Because of their acceptance of artificial nest boxes and low nest desertion rates, it is possible to conduct field experiments and precisely monitor their breeding success. Voucher material of this study population was deposited in the ornithology collection of the Naturalis Biodiversity Center (Leiden, The Netherlands) under the inventory numbers RMNH.5092348 and RMNH.5092349.
Fieldwork was conducted during 2013 and 2014 in the Hoge Veluwe National Park (The Netherlands; 05°51′E, 52°02′N). Forested areas in the park are dominated by pedunculate oaks (Quercus robur), the exotic northern red oaks (Quercus rubra), Scots pines (Pinus sylvestris), larches (Larix sp.) and birches (Betula sp.). We provide around 400 nest boxes year-round in an area of 171 ha., which are occupied in spring by cavity-nesting passerines such as pied flycatchers, great tits (Parus major), Eurasian blue tits (Cyanistes caeruleus), nuthatches (Sitta europaea) and coal tits (Periparus ater). We collect long-term field data on arrival dates of males, nest building of females, female egg-laying dates, chick-hatching dates, brood success, and adult and chick basic biometrics.
In order to place our results into an ecological context, we note that 2013 was a cold spring (April–June average temperature of 11.25°C versus 12.65°C for 1995–2015) with a low density of caterpillars (average biomass of 2.4 g m−2 day−1 versus 14.18 g m−2 day−1 for 1995–2015) and peak in caterpillar density on 30 May. The average flycatcher laying date was 9 May and the average number of fledglings was 5.78 (excluding manipulated broods), above the average of 5.6 for 1995–2015. In contrast, 2014 was a warm spring (13.21°C), also with a low caterpillar density (biomass of 0.67 g m−2 day−1) but with a peak on 3 May. The average flycatcher laying date was 6 May and the average number of fledglings was 4.79 (excluding manipulated broods), i.e. below average.
Foster pairs used to incubate experimental eggs (see below) consisted of naturally nesting great tits and blue tits found in the nest boxes of the Hoge Veluwe or in Heikamp, The Netherlands, a nest box population adjacent to the Hoge Veluwe.
Our experiment involved advancing and delaying the hatching date in two sets of breeding pairs, using a third set as a control. To test for causal effects of shifting reproduction on the subsequent stages, we did not just use the natural variation in hatching dates. Instead, we incubated first-laid eggs before the natural incubation time and delayed incubation of later laid eggs after the natural incubation time. This resulted in an advancement and delay of hatching before and after the natural range of hatching dates.
All nest boxes in the study area were checked twice per week until we found pied flycatcher nest material. We then inspected the nests daily to determine the date the first egg was laid. When we found three or more nests with first eggs laid on the same day, we randomly assigned them to a treatment (or randomly excluded them from the experiment, in cases where we had sufficient experimental groups or there were not enough nests to create additional groups of three nests).
In order to create differences in chick-hatching dates, our treatments consisted of a combination of manipulating the start of the egg incubation date, and thereby hatching date, and a split–brood design (Fig. 1). We took advantage of the fact that these birds lay a single egg per day and usually start incubating their eggs only when the last one is laid.
One day after finding the first egg, when the three females had laid their second egg, we collected all six eggs, replacing them with plastic dummy eggs to prevent desertion. This clutch of six eggs (the typical clutch size for a pied flycatcher) composed of two eggs from each of the three nests was then placed under an already incubating great tit or blue tit (the ‘foster incubator’) – therefore, the initiation date of the incubation of this clutch was 4 days earlier than normal. We continued collecting the remaining eggs daily until the sixth egg was laid, always replacing them with dummy eggs. When the sixth egg was laid, we considered the clutch to be complete. In cases where a female laid a seventh egg, the egg was removed from the experiment. All these newly collected eggs (laying order three to six) were stored in a field shed at ambient temperature and turned twice per day to prevent damage to their viability (Visser and Lessells, 2001).
Once the female assigned as the ‘control’ laid her sixth egg, we swapped all plastic dummy eggs with a clutch created from eggs with laying order four and five. Because we gave the female six eggs on the same day that she laid her sixth egg, we did not modify the time when she started incubating or the hatching date of the eggs (Fig. 1). Thus, the birds hatched their chicks at the expected time. Eventual seventh eggs were removed as explained above.
As soon as the female assigned as ‘advanced’ started incubating her dummy eggs continuously (i.e. when we found the female at least three times in a row incubating the dummy eggs within an interval of a few hours), which happened between 1 and 3 days after the last egg was laid, we replaced the dummy eggs with the six pre-incubated eggs (previously transferred to a foster incubator as described before). As mentioned above, because most pied flycatcher females typically start incubating their clutch once completed (six eggs), we advanced the hatching of these eggs by 4 days (Fig. 1). Thus, these birds hatched their chicks 4 days earlier than expected.
The females assigned as ‘delayed’ only received a clutch of six eggs (with laying order three and six), replacing the dummy eggs, 5 days after the female laid the sixth egg (on the hypothetical ‘egg day 11’). This means that the starting point of incubation of these eggs was 4 days later than normal (Fig. 1). Thus, delayed birds hatched their chicks 4 days later than expected.
The procedure was repeated throughout the season as we tried to span the whole range of egg-laying dates as long as we found three females that laid their first egg on the same day. In a few cases (three nests in 2014), we allowed 1 day difference between nests to be able to create enough groups. In total, we managed to successfully manipulate 117 broods: 57 in 2013 (19 experimental groups) and 60 in 2014 (20 experimental groups). In five broods in 2013 and five in 2014, females laid five instead of six eggs; clutches of experimental groups involving these broods were completed by adding one of the occasional seventh eggs collected throughout the experiment (see above) – note that this procedure did not alter the timing of treatment for the adult birds, just the composition of egg origins in the brood. As a result of desertions at incubation or chick stage, not all manipulations were successful. Whenever possible (four nests in 2013 and two in 2014), deserted manipulated clutches were moved to new pied flycatcher nests with a maximum 1 day of laying date difference from the original experimental nest – this procedure also did not affect the timing of treatment for the adult birds. Thus, the real starting number of manipulated broods was 111 in total: 55 (18 advanced, 18 control and 19 delayed) in 2013 and 56 (19 advanced, 18 control, 19 delayed) in 2014. It is important to note that 8 days is a substantial difference in chick-hatching dates for our birds. Average standard deviation of chick-hatching dates per year of the entire population is just 5.2 days (mean for all years 1964–2012).
The number of hatched chicks did not differ across treatments (n=111, χ2=0.23, P=0.89, advanced 1.68±0.07, control 1.72±0.07, delayed 1.71±0.07), but we did not allow differences in brood size of more than one chick as this would affect the parental effort. Therefore, we checked all nests with 2 day old chicks and, in the case of nests with fewer than five hatched chicks (seven nests in 2013, four in 2014), we added extra chicks with a maximum 1 day difference from nearby woodlands (e.g. other field sites like Westerheide, 5°50′E, 52°00′N, The Netherlands), raising the brood size to five chicks. Again, this procedure did not alter the timing of treatments, only the composition of sibling chicks in a nest. Treatments did not differ in the number of 2 day old chicks (either before or after the brood size correction).
Later during the experiment, we lost three nests (2013) through natural causes (death of chicks, adult desertion at a later breeding stage or predation events) and not all males could be caught at either chick day 7 (three males in 2013, three in 2014) or chick day 12 (seven in 2013 and three in 2014). Therefore, some variation existed in sample size across the different analyses and this is reflected in the degrees of freedom for each analysis.
Although only females incubate in this species, our treatment was effective in manipulating both sexes, as males and females are involved in chick rearing and thus the termination of breeding was modified for both sexes. Moreover, because the chick-hatching dates was literally moved in time, our manipulations were also useful to test effects of hatching date on components of individual chick timing (Tomotani et al., 2016) and fitness.
The arrival date of individual males was assessed by daily identifying newly arriving males in the study area from early April until early May, using individual plumage characteristics and colour ring combinations (Both et al., 2016; Tomotani et al., 2017; Visser et al., 2015). Once a pied flycatcher pair settled in a nest box and started building a nest, these were checked frequently to assess the exact date of egg laying, clutch size, the exact date of chick hatching and the number of hatched chicks.
After the manipulations took place, the breeding adults were caught when chicks were 7 days old and then when they were 12 days old and their primary moult (henceforth ‘moult’) was scored in both cases. Moreover, in order to obtain a more precise moult data, males were not released after being caught on chick day 12 but taken to the Netherlands Institute of Ecology (NIOO-KNAW, Wageningen; see ‘Housing conditions’, below). Females were left in the field to continue to care for the chicks. In 2014, a subset of the chicks was also taken to the institute for a complementary study (Tomotani et al., 2016) and females from those nests were also taken to the institute.
When in captivity, birds were housed in individual cages (90 cm×50 cm×40 cm) and provided with food (mealworms, crickets and flies) and water ad libitum. Conditions of light and temperature were kept as close to the natural situation as possible, with all windows open, allowing birds to be exposed to natural illumination. This was supplemented by two fluorescent lamps connected to a timer that switched on 30 min after sunrise and off 30 min before sunset. Therefore, birds were exposed to the natural photoperiodic variation with natural duration of twilight periods. After 14 days in captivity, they were returned to the Hoge Veluwe and released close to their breeding nest box.
Timing of moult
In 2013, adults were further checked for moult after 7 and 14 days of arrival in the laboratory (chick age 19 and 26 days). In 2014, however, moult was scored every day in most of the birds so the date the first feather was dropped could be pinpointed in most occasions. Moult score was based on which primary feather(s) was being moulted and how much it had grown (following Dawson and Newton, 2004, but discriminating between old and dropped feathers). The score was then converted to the mass of newly grown feathers, which was assumed to increase linearly through time (Dawson and Newton, 2004). Moult onset was then calculated by regressing a line through the different measurements of grown feather mass on different days. We used the method described in Dawson and Newton (2004) for calculating moult onset based on the feather mass grown and the R ‘moult’ package (Erni et al., 2013), with the formula developed by Underhill and Zucchini (1988) to estimate the mass of growing feathers for each individual bird. Because we also had the dates that the birds lost their first primaries for most birds in 2014, we could compare data from the calculated and observed onsets. The calculated onsets differed from the observed first moult days by an average of 0.4 days (maximum 2.3 days); therefore, we are confident that our calculated moult onsets are accurate. In cases where we only had a measurement of new feather mass for one day (for example, a bird that was only observed moulting on the last time it was caught in the field), we first calculated the mean value of moult speed for the year (based on the slopes of the linear regression for all other individuals) and then used those calculated mean values of moult duration as a slope in the linear regression together with feather mass to calculate the onset of moult for that individual. This assumes that there is little variation in the speed of moult among individuals, which is not necessarily true as late-moulting birds moult faster (Dawson, 2004). However, we assume this was a minor issue for our calculations, as (a) the number of birds affected was fairly small (two out of 49 moult data in 2013 and three out of 54 in 2014), and (b) those individuals affected did not have consistently very early or very late laying dates.
To assess the internal state of the animals, in 2013 we also performed an additional blood sampling for hormones. Prolactin is a hormone that could be involved in the transition between moult and breeding (Dawson, 2006), mediating the stress responses during breeding and having a role in the trade-off between parental effort and self-maintenance (Angelier and Chastel, 2009). Prolactin levels decline towards the end of the season, being lower when breeding is over, and a decline in prolactin is expected to trigger moult onset (Dawson, 2006). We expected that, if animals adjusted their moult in relation to the termination of breeding, we would also observe a shift in their prolactin levels, as we assumed that prolactin would be dependent on chick age (older chicks, lower values). Because birds were sampled at the same chick age, we would thus expect that advanced, control and delayed birds from the same group (same laying date) would have the same prolactin level and that it would decline over time equally for all treatments (similar to scenario Ib). In 2013, all males taken to the laboratory were blood sampled on the day they were captured (chicks of 12 days old). We collected 100 µl of blood from the jugular vein of each individual as soon as the bird arrived at the laboratory, to prevent any effect of captivity stress on prolactin levels (Angelier and Chastel, 2009). Samples were immediately centrifuged and the plasma was collected and stored at −20°C. Samples were then shipped on dry ice to the Roslin Institute and levels of plasma immunoreactive prolactin were determined using a radioimmunoassay for recombinant-derived European starling (Sturnus vulgaris) prolactin described by Bentley et al. (1997). All samples were measured in duplicate in a single assay. The sensitivity of the assay was 0.3 ng ml−1, the intra-assay coefficient of variation was 7.6%, and serial dilution of individual samples ran parallel along the standard curve within the dilution range assayed.
Timing of migration
Before being released back into the field, a subsample of the males (38 out of 42 in 2013 and 30 out of 50 in 2014, including individuals from all treatments across the whole range of laying dates), was deployed with 0.5 g light-level geolocators (Intigeo-W50, Migrate Technology Ltd, Cambridge, UK) using a leg-loop harness build with an 0.7 mm elastic nylon cord and adjusted to each bird with a knot in the posterior mounting tube, between the device's mounting loops.
Geolocators were recovered in the subsequent years (12 from the 2013 cohort: 10 in 2014, two in 2015; seven from the 2014 cohort: six in 2015, one in 2016) when the birds returned to breed. Data were processed similarly to Åkesson et al. (2012), but because of the imprecision of latitude data, we only used the information collected for longitude (inferred from local solar noon/midnight). Twilight transitions were determined using TransEdit (British Antarctic Survey, Cambridge, UK) with a single threshold value of 5, minimum daylight period of 1 h and minimum night period of 4 h. Positions were obtained using the software BirdTracker (British Antarctic Survey), which gave us two positions per day (noon and midnight). Data were then visually inspected to detect large changes from the study area longitude, indicating a departure from breeding grounds, as pied flycatchers move to the west, following the African coast during the autumn migration (Ouwehand and Both, 2016; Ouwehand et al., 2016).
Effects of logger clock drift on longitude were tested as described in Ouwehand et al. (2016). There was no noticeable clock drift effect (always <1 min) and treatments did not significantly differ in terms of clock drift. Regardless, we decided to use the average longitude of the first winter month (December) to extract the wintering longitude as the drift supposedly gets larger the longer the period after deployment.
To assess the effects of our experiment on fitness, we measured breeding success and adult and chick condition (body mass) at the end of breeding. As a second metric of male condition, we also measured the quality of the feathers moulted after the experiment and collected in the subsequent year because a constrained moult is expected to lead to poor feather quality (Echeverry-Galvis and Hau, 2013; Vágási et al., 2012). Finally, we also obtained data on parental care in terms of the number of visits to the nest by each parent.
To measure breeding success, nests were regularly checked until all chicks fledged to obtain information on the number of fledged chicks and overall brood success/failure.
To assess adult and chick condition, we took biometric measurements in the field of males, females and chicks. When chicks were 7 days old, both adults were caught and we took measurements of tarsus length and mass. Later, when chicks were 12 days old, males and females were caught again and weighed for a second time. Adults and chicks were also blood sampled for paternity analysis (Tomotani et al., 2017).
To assess feather quality, in the years following the experiment (2014 and 2015), we sampled the fourth innermost primary of males that were in the experiment in the previous year, i.e. the primary feather grown immediately after the experiment. We estimated the bending stiffness of feathers (mN mm−1) by testing the force necessary to bend the feather when applying a load at 32 mm from the base of the rachis. We moved the load down at a speed of 4 mm min−1 and took the first 60 (mass–time) points immediately after contact of the load with the feather (1 point per second). Bending stiffness was estimated for each feather based on the slope of the relationship between feather mass and time (de la Hera et al., 2010).
To measure parental care, we fitted transponders with a unique individual code to each of the adults when their chicks were 7 days old and installed transponder readers at the entrance of each nestbox to continuously measure the visits of each parent. To avoid disturbing the nest too much in 1 day, readers were only installed when chicks were 8 days old and then removed when chicks were 12 days old and parents were caught for the second time. Thus, we only used transponder data obtained when chicks were 9, 10 and 11 days old. We collected data for most nests in 2013 (43 nests – because of a failure in some equipment and nest desertions at an early stage) and for a subsample of the nests in 2014 (6 nests).
The readers detect the presence of the transponder in the vicinity of the nest box every second. This means that if a bird is perched in front of the nest box, the equipment will detect multiple records of this individual, even though it is not visiting the nest box multiple times. In order to filter out these pseudo-visits, in 2015 we recorded transponder data together with video for another experiment (Tomotani et al., 2018b) and used the video data to discriminate which intervals actually reflected real visitations to the nest. We then used this information to clean up the data from 2013 and 2014. We are aware that year-to-year differences may exist, but reader configuration was kept constant across the 3 years, meaning that those differences should be relatively small.
In order to assess longer-term effects of our manipulations (as reversible state effects; Senner et al., 2015), we looked at the probability of finding the manipulated birds the following years as well as the breeding dates of males and females in the following year.
All analyses were performed in R version 3.2.1 (http://www.R-project.org/). Most of the females were not taken to the laboratory (eight females taken to the lab) and were not moulting on the last date they were captured (chicks with 12 days old); we also did not deploy geolocators on females. Thus, we only present results relative to males here.
We found a substantial difference in the spread of dates between the two years: mean±s.d. egg-laying April date in 2013: 38.63±4.81; in 2014: 36.41±8.36. Thus, in all analyses that involved more than one year, we centred the dates (expected or manipulated hatching date) on the mean date instead of using the raw dates as fixed effects to avoid mixing up within-year effects caused by our treatments and the natural variation that occurred between years. This procedure was based on all data collected for our population, not only experimental nests. Although some individuals were present in both years of data collection, they were too few (six birds) to allow the use of ‘individual’ as a random effect in a mixed effect model approach. Therefore, if birds were present in the experiment twice, we had to randomly exclude one of the observations from the dataset.
In most cases, we used two separate models: (1) a model including the expected hatching date (henceforth ‘expected hatching date’, i.e. the egg-laying date+18 days of hypothetical clutch completion and incubation), which represents the non-manipulated value for seasonal changes, and (2) a model including the real hatching date (henceforth ‘manipulated hatching date’), the actual date of hatching after the manipulations took place. Results arising from these models are complementary and it is important that they are looked at in conjunction to discriminate between the described scenarios.
Scenario Ia: linked (termination of breeding)
If the timing of moult and/or departure is causally affected by the termination of breeding, the moult and/or departure dates will differ between treatments when analysed in relation to the expected hatching date. Thus, when data are plotted in relation to the expected hatching date there will be three parallel lines, i.e. a significant difference of moult and/or departure date between treatments. In other words, for a given date, the line of the advanced broods it will fall earlier than (below) that for the control, and the line of the delayed broods will fall later than (above) that for the control (Fig. 2A). Whereas, when data are plotted in relation to manipulated hatching date, the points will fall on the same line, because the lines are simply moved along the x-axis (Fig. 2B).
Scenario Ib: linked (other stage)
If the variation of moult and/or departure is correlated to the natural variation of another stage (for example, the egg-laying date of the female) without a causal relationship to the termination of breeding, the moult and departure dates will not differ between treatments when analysed in relation to the expected hatching date. Thus, when plotted in relation to the expected hatching date, the three lines will fall on top of each other, with no significant treatment effect. This means that, for a given hatching date, there will be no difference in the timing of moult or departure between advanced, control or delayed treatments (Fig. 2C). But when plotted against the manipulated hatching date, the treatments will be seen as separate parallel lines simply because the lines move along the x-axis (Fig. 2D).
Scenario II: independence
Finally, it is also possible to have a complete lack of effect of the previous stages on the timing of moult and/or migration, meaning that the timing of moult and departure is solely regulated by photoperiod. This will be seen as the same onset of moult and/or departure for all individuals, regardless of when they bred (Fig. 2E,F).
Timing of breeding, moult and migration
To be sure that there were no differences between the birds due to chance prior to the experimental manipulations, we used multiple linear regressions to test whether the pre-manipulation male arrival and female egg-laying dates differed in relation to treatment.
We then tested the effects of our manipulations on hatching date, prolactin concentration, autumn migration and wintering longitude, also using multiple linear regressions. Except for hatching date, all tests were carried out with the two above-mentioned models (in relation to the centred expected or manipulated hatching dates). Because date effects are not necessarily linear, in all cases we also tested for the quadratic effect of (expected/manipulated) hatching date. The analysis also included ‘year’ as a covariate if we had 2 years of data (all traits except prolactin concentration). We also tested in separate analyses whether departure date was explained by the onset of moult or the previous un-manipulated arrival date of the male to the breeding grounds.
Out of 102 males identified in the 2 years (from 111 experimental nests), we could reliably calculate the individual onset of moult for 80. The remaining birds were either not captured on chick day 12 (N=10) or not moulting by the time they had to be released (N=12), as we wanted to guarantee the same chance of return for all experimental birds. Therefore, the difference in the onset of moult in males from advanced, control and delayed groups could not be assessed using a simple mean group comparison. Instead, we analysed moult onset using Cox proportional hazard models (Cox, 1972), as this model allows the inclusion of ‘censored’ data and was much more reliable for comparing the onset of moult in our different groups. Moult onset was, therefore, our ‘time to event’ variable and the last day of observation of ‘not moulting’ birds was included in the analysis as ‘censored’. As in previous cases, to test the effect of treatment, we used two separate models correcting for either the (linear/quadratic) centred expected or manipulated hatching date, with year as a covariate.
The amount of moult–breeding overlap is a consequence of how moult is scheduled in relation to the termination of breeding. An analysis using this trait has the disadvantage of only allowing the inclusion of non-censored data, but it allows the use of multiple regressions instead of hazard models, facilitating visualization of the results. We thus analysed the moult–breeding overlap (excluding non-moulting individuals) as a complementary way of assessing the potential effects of our treatments. Moult–breeding overlap was calculated as the difference in days between the day of moult onset and the day when chicks were/would be 15 days old (the fledging date). Negative values represent an overlap, while positive values indicate an interval between fledge dates and onset of moult. This was tested in a similar way to the previous multiple regressions, correcting for either the centred expected or manipulated hatching date as linear and quadratic terms, also including year as a covariate.
We used generalized linear models in R (logit-link and Binomial error-distribution) to test whether the probability of brood failure was explained by treatment and a linear regression to test for the effect of treatment on the number of fledglings. In both cases, we tested in two separate models the effect of treatment, correcting for either the centred expected or the centred manipulated hatching date (both as linear and quadratic terms), also including year and starting number of chicks as a covariate.
For adult and chick body mass, we used mixed effect models with the brood into which the chick hatched (ontogenetic effect) and original brood (genetic effect) as random effects and expected or manipulated hatching date as the fixed effect. Year was also included as a fixed effect for adult and chick mass, and tarsus length was used as a fixed effect as a correction factor for structural size. We also included the three-way interaction between year, treatment and hatching date. Because not all chicks could be assigned to a genetic mother, we could not include all of them in this analysis.
To test for treatment effects on the number of visits, we also used mixed effect models and included treatment, year, sex, chick age (9, 10 or 11 days old) and their interaction as fixed effects. Because males and females provided food to the same nest, we also fitted individual and nest as random effects.
Finally, feather quality (bending/stiffness) was analysed using multiple regression, with treatment and year as explanatory variables.
To test for possible differences caused by making clutches of eggs based on the laying order (first and second eggs make up the advanced group, third and sixth egg make up the delayed group, and fourth and fifth egg make up the control group), we designed a control experiment in which we created clutches containing only eggs from those laying orders, but without manipulating the chick-hatching dates. This experiment was conducted in 2013 in a separate breeding population in the forests of Buunderkamp (The Netherlands; 5°45′E, 52°01′N). Triplets of nests (N=8 triplets, 24 nests but one control nest was lost as a result of the death/desertion of the female) with the same egg-laying dates were used to form a group and one of the treatments was randomly assigned to each of the nests: clutch containing only egg 1 and 2, clutch containing only egg 3 and 6, or clutch containing only egg 4 and 5. On the day the females laid the sixth egg, eggs were swapped among the nests and treatments were created.
We used multiple regressions to test for effects of treatment on chick-hatching dates and hatching success (with egg-laying or chick-hatching dates as covariates) and mixed effect models with brood as a random effect to test for effects of treatment on chick mass (with egg-laying or chick-hatching dates as covariates as both linear and quadratic terms).
There was no effect of treatment on hatching date (F2,20=0.27, P=0.77, advanced 56.75±0.66, control 57.43±0.71, delayed 57.25±0.66), hatching success (F2,20=1.93, P=0.17, advanced 5.88±0.22, control 5.57±0.24, delayed 5.25±0.22) or chick mass (F2,19.99=0.57, P=0.58, advanced 13.64±0.32, control 13.69±0.35, delayed 14.09±0.33). This indicates that any effect of our manipulations on timing was solely caused by the shifts in timing of egg hatching and not due to possible differences in egg quality correlated to the egg-laying order.
The Animal Experimental Committee of the Royal Netherlands Academy of Sciences (KNAW) approved all experimental procedures, following the Dutch law regulating animal experimentation (protocol NIOO 13.04 and amendment NIOO 14.04). Eggs of foster parents were removed from foster nests prior to incubation and discarded (under European law, eggs are not considered vertebrates) and foster parents (great tits and blue tits) had their nests removed after the experiment. In some cases, those birds were re-sighted in the study site starting a new brood. Occasional seventh eggs of pied flycatchers were removed and, if not used to complete clutches, were sampled for DNA to estimate extra-pair paternity in a separate study (Tomotani et al., 2017). Experimental manipulations did not affect the fledging probability of the pied flycatcher chicks: nine out of 97 experimental nests (over both years) failed at some point during the experiment, which is within the natural range of brood success for our area. Out of those nine nests, three failed immediately after hatching before the adults were caught, five failed between the first and the second capture, and only one failed after the male was taken to the laboratory. The number of fledged chicks was not different between manipulated and un-manipulated nests. When kept in the laboratory, birds initially decreased their body mass in the first 3 days and then returned to their normal mass, sometimes becoming heavier prior to release. It was not possible to compare the return rates of the manipulated and un-manipulated birds in our population during those 2 years because the majority of the un-manipulated birds were the late breeders. In comparison to the average survival across years, the return rate of the manipulated males was lower than the average survival probability for our population (21% versus 47%). However, our recovery success of birds deployed with geolocators in 2013 was similar to that reported for a nearby population (32% versus 34%; Ouwehand and Both, 2017). Therefore, while it is possible that our manipulations affected the survival of the birds, year-to-year differences in survival may also have played a role (with a naturally lower survival in 2014).
Full statistical results are presented in Tables S1–S5. Dates are given in April days unless otherwise specified.
Effect of the manipulation
There was no difference in arrival date or egg-laying date of birds in the different treatments prior to the experiment. Egg-hatching dates were 3.5 days earlier in the advanced group and 3.5 days later in the delayed group relative to the control, showing that our experiment had the intended effect (F1,98=268.01, P<0.01; Fig. 3A,B). Therefore, we are confident that our experimental manipulation of hatching dates was successful.
Timing of moult
There was no effect of treatment on the onset of moult, independent of using the expected or the manipulated hatching date (treatment term: χ2=2.25, P=0.32; Fig. 3C,D). There was also no linear or non-linear effect of (expected or manipulated) hatching date on the onset of moult. These results are in accordance with scenario II (independence) and suggest that all birds had a similar moult onset that did not depend on the termination of breeding (Fig. 3C,D). In addition, there was a significant year effect (average onset of moult: 16 June in 2013 and 11 June in 2014; χ2=16.13, P<0.01).
In line with this lack of adjustment of moult in relation to breeding, birds in the delayed treatment had a larger moult–breeding overlap than those in the advanced and control groups when controlling for the expected hatching date (F2,69=4.00, P=0.02, advanced 4.38±1.61 days, control 3.14±1.61 days, delayed −0.71±1.45 days; Fig. 3E). There was also a significant quadratic effect of the expected hatching date (F2,69=7.88, P=0.01, estimate=−0.07±0.03; Fig. 3E), i.e. the weak effect of hatching date was only found in very late birds. However, when we compared this quadratic trend with two linear regressions (post hoc broken-stick analysis), we found that two regressions had a better fit than the quadratic term (adjusted R²=0.47 for quadratic term, adjusted R²=0.5 for two regressions with a break point at centred expected hatching date of 5). Dividing the dataset before and after centred day 5 and testing the two regression separately, we found that the first regression was significant (F1,60=29.8, P<0.01, slope=−1.04±0.19; Fig. 3E), but the second one, which included the late breeders, with a much lower sample size, was borderline non-significant (F1,5=0.8, P=0.05; Fig. 3E). Finally, the relationship between moult–breeding overlap and manipulated hatching date was significant and very close to 1 (F1,72=50.4, P<0.01, slope=−1.06±0.15; Fig. 3F), so for each day that hatching was delayed, overlap increased by 1 day. This supports the moult onset analysis and suggests that onset of moult is similar for all birds, being independent from the termination of breeding.
The prolactin concentration measured in males when chicks were 12 days old was significantly higher in advanced birds and similar in control and delayed birds in the model controlling for the expected hatching date (treatment term: F2,31=4.43, P=0.02, advanced 7.94±2.47, control 7.15±2.47, delayed 7.16±2.41; expected hatching date term: F1,31=5.42, P=0.03, slope=−0.15±0.07; Fig. 4A), with the treatment effect disappearing in the model that uses the manipulated hatching date (manipulated hatching date term: F1,33=11.62, P<0.01, slope=−0.12±0.04; Fig. 4B). This is in agreement with scenario Ia (linked – termination of breeding) and suggests a causal effect of hatching date on prolactin concentration. Thus, advanced birds had the same prolactin levels as individuals that had a naturally earlier breeding time and were ‘moved’ to an earlier point in the curve of natural seasonal decline in prolactin. This also means that chick age did not affect prolactin levels.
Timing of migration and wintering longitude
Males departed from the breeding grounds between 24 July and 18 August, which was on average 50.5 days after fledging of their chicks, and 33.5 days after release from the lab. We retrieved 32% (7 out of 22) of the advanced males with geolocators, 36% (8 out of 22) of the controls and 17% of the delayed males (4 out of 24) (but these differences are not significant, χ2=2.58, P=0.28).
The date of departure from The Netherlands to Africa was positively correlated with the expected hatching date (F1,17=43.16, P<0.01, slope=1.19±0.18; Fig. 3G), with no indication of an additive treatment effect (Fig. 3G), supporting scenario Ib (linked – other stage). Thus, birds that naturally bred earlier also departed earlier. However, according to our prediction (Fig. 2), we would also expect to find a significant treatment effect, which was not the case (F2,15=1.5, P=0.27; Fig. 3H). One possibility for this result is that we have a clear segregation of treatments across the manipulated egg-hatching date axis as this is the essence of our manipulations (i.e. early hatched were primarily advanced and late hatched were primarily delayed). This makes the testing of a treatment effect difficult as the model will fit a line through three discrete groups of points which are at different positions on the x-axis. To correct for this, as a post hoc test, we performed a second analysis constraining the slope of each treatment to the same general slope of the expected hatching date term (obtained in the previous analysis with treatment corrected for expected hatching date: 1.19). With this constrained slope, this second analysis pointed to a significant treatment effect (F1,16=3.95, P=0.04, estimate in the presence of the slope: advanced 138.61±1.70, control 134.93±1.59, delayed 130.78±2.24), which thus does support scenario Ib (linked – other stage).
Departure date was not correlated with individual moult onset (F1,15=2.64, P=0.13, slope=0.39±0.24, excluding non-moulting birds) or to the arrival date to the breeding grounds earlier in the same breeding season (F1,12=0.28, P=0.61, slope=0.27±0.51).
Wintering longitude was significantly affected by treatment (F2,16=3.96, P=0.04, advanced −7.40±0.45, control −7.00±0.42, delayed −9.01±0.59; Fig. 5), with delayed birds wintering ca. 200 km more westerly than advanced and control birds. We did not find year effects on wintering longitude; however, we found a curious additional effect of natural hatching date, but in the opposite direction from that of our treatments, with naturally early birds wintering more westerly than late ones (natural hatching date: F1,15=5.13, P=0.04, slope=0.11±0.05).
When breeding success was analysed, there was no significant treatment effect on the probability of brood failure or on the number of 12 day old chicks. The final number of chicks was obviously dependent on the number of chicks hatched (5 or 6), which was not related to treatment (F1,100=11.50, P<0.01, slope=0.82±0.24).
There were no treatment effects on the fitness components measured for the adult birds. We found a significant three-way interaction between treatment, sex and year on adult mass, measured when their chicks reached 12 days old (F3,80.42=3.61, P=0.02). However, when the dataset was divided and 2013 and 2014 data were analysed separately, there was no significant treatment difference. The small sample size may have produced a misleading pattern when a three-way interaction was fitted; therefore, for the present work, we should assume that there was no relationship between treatment and adult mass. The only significant difference was between the sexes in 2013, with males being lighter than females (F1,44.83=9.21, P<0.01). There was also no difference between treatments in the quality of the male feathers (feather bending/stiffness), produced after our manipulations and collected in the following year. Finally, there was no effect of treatment on the number of visits to the nest per day. From all traits tested, only sex had a significant effect, with females feeding significantly more than males (F1,71.26=13.62, P<0.01). Effects of individual birds (as random effects) were also significant (P<0.01) and explained 75.66% of the variance, because, as expected, some individuals visit the nest much more often than others.
There were also no treatment effects on the fitness components measured for the chicks. Chick mass was not causally affected by treatment, as is evident in the non-significant treatment effect in the analysis including the egg-laying date. There was a three-way interaction between treatment, hatching date and year (F2,151.41=6.08, P<0.01) but, again, when the dataset was divided and 2013 data were analysed separately from 2014 data, both treatment and hatching date–laying date effect disappeared. This means that there was no effect of treatment, laying date or hatching date on chick mass.
Carry-over effects (reversible state effects)
Treatments did not differ in the probability of finding the manipulated birds the next year (χ2=0.83, P=0.66) or any of following years (up to 2016, χ2=1.04, P=0.59). The breeding date of males and females in the following year was also not affected by treatment (F1,26=0.39, P=0.68), only by their un-manipulated laying date in the previous year (estimate=0.58±0.27, F1,40.98=4.56, P=0.04) and the year in which the experiment was conducted (2013 11.66±10.15, 2014 17.13±9.61; F1,39.72=8.73, P=0.02).
Our results show that in male pied flycatchers, the timing of moult is independent of the termination of breeding (scenario II: independence). Moult, however, is also not completely dependent on photoperiod as moult onset varies from year to year (Fig. 6). Alternatively, moult could dependent on a so-called ‘remote timer’ (Helm and Gwinner, 2005). The timing of autumn migration, in contrast, depends on a previous stage, but not on the termination of breeding (scenario Ib: linked – other stage). Therefore, male pied flycatchers do not compensate for the earlier or later hatching of their chicks by changing the onset of their moult or their departure date. Instead, if hatching date is shifted to an earlier or later date, it affects the degree of overlap between breeding and moult. Although we did not detect short-term effects on fitness caused by our treatments, delaying the chick-hatching dates caused adult males to winter at a different location (Fig. 5), suggesting carry-over effects of moult–breeding overlap on wintering behaviour. However, these effects disappear in the following year consistent with a ‘reversible state effect’ (Senner et al., 2015).
While our results suggest that the timing of moult is relatively inflexible for most individuals, other studies have suggested that timing of moult in males is related to breeding. For example, Dawson (2006) showed a relationship in starlings between moult onset and gonadal regression. Because we did not collect data on testicular regression, it is not possible to infer whether moult in our birds was or was not correlated with gonadal regression. This aspect must remain speculative, but we do not discard it as a possibility. The interpretation of whether breeding affects moult thus may depend on how ‘termination of breeding’ is defined. For our purposes in the present study, when assessing overlap of annual-cycle stages it seems to be more meaningful to assume the end of chick care as the termination of breeding, especially in bird species where both parents contribute to chick care, as in pied flycatchers.
Hemborg (1998) also conducted an egg manipulation experiment with pied flycatchers and showed that advanced and delayed males did not differ in their post-breeding moult stage, which is the opposite of what we found. Because our experimental design is not directly comparable to his (i.e. his design depended on broods that lay eggs a number of days apart, while the present experiment used broods with the same laying dates), it is difficult to determine what exactly caused these differences. Pied flycatchers from Sweden breed later than birds from The Netherlands and are supposedly more time constrained to moult (Hemborg had up to 80% of the males and 50% of the females moulting at chick day 13, while we had at most 24% and 6%, respectively, at chick day 12); thus, they may require some different strategies to be able to migrate in time (Hemborg et al., 1998, 2001). An inflexible timing of post-breeding moult was also found in other long-distance migrants such as the bluethroat, Luscinia svecica (Lindström et al., 1994). It is, thus, possible that flexibility in the timing of post-breeding moult varies among species and populations (Helm and Gwinner, 2006; Jenni and Winkler, 1994; Newton, 2009).
In our experiment, an independent timing of moult is supported by the prolactin results. There is good evidence for the role of prolactin in the regulation of moult (Dawson, 2006; Dawson and Sharp, 2010; Payne et al., 1972; Williams, 2012). For instance, administration of prolactin leads to moult inhibition in some species (Payne et al., 1972), while a decline in prolactin (but not the hormone level per se) could act as the moult trigger (Dawson, 2006). We observed higher levels of prolactin in advanced birds in comparison to control and delayed animals (Fig. 4), together with a smaller amount of moult–breeding overlap in birds from the advanced group (Fig. 3). We would expect equally high levels of prolactin at the same chick age if the moult control depended on the termination of breeding. Thus, the prolactin results also support scenario II (independence) for moult onset.
The lack of effect of the manipulations on the timing of moult, and moult occurring at a certain calendar day for most birds, suggests either that moult is independent of hatching or that the decision to start moulting is made at an earlier, un-manipulated, stage in the season than hatching date. While our experiment is not able to test for causality of other stages, it is possible to test for these two possibilities using the natural variation in moult onset. For a few individuals, we had data on moult and laying dates in both 2013 and 2014. We used these data to compare for each individual how the difference in laying dates between 2013 and 2014 was related to the 2013–2014 difference in moult onset. These results are present in Fig. 6. If moult onset is related to a previous stage, such as laying date, points would fall on the hypothetical line a, while a complete independence would make the points fall on the hypothetical line b. We found that moult onset did not depend on a previous stage (the slope was not significantly different from 0, F1,8=2.92, P=0.13; but was significantly different from 1, F1,8=21.2, P<0.01), not falling on line a. But it is also interesting to notice that it also did not fall on line b, so it does not happen on exactly the same day in different years (intercept was significantly different from 0, estimate=−7.38, P<0.01). These results suggest that the timing of moult is not solely regulated by photoperiod and may also be related to an environmental variable that that varies from year to year. Alternatively, flycatchers could use a ‘remote timer’, with the timing of moult set during spring migration. Because there is year-to-year variation in photoperiodic exposure due to a year-to-year variation in the timing of migration, there would also be variation in the timing of moult (Helm and Gwinner, 2005; Sharp, 2005).
In contrast to the breeding/moult transition, information on the transition to the migratory phenotype is more sparse in the literature. Some studies have demonstrated a possible hormonal control mediated by testosterone and corticosterone in the transition between spring migration and breeding (Ramenofsky and Wingfield, 2006; Tonra et al., 2013) and also in stopover decisions (Eikenaar et al., 2017). Autumn migration, however, may still have a different control (Agatsuma and Ramenofsky, 2006; O'Reilly And Wingfield, 1995). Our data suggest that departure time from the breeding grounds is related to the onset but not the termination of breeding (Fig. 3). While we are unable to assume causal relationships because of the lack of experimental manipulation of these other stages, it is still interesting to speculate. We found a positive correlation between the arrival of the males and the egg-laying date of their females (used as a proxy for the expected hatching date, prior to experimental manipulation, estimate=0.52±0.16, F1,69=10.46, P<0.01) and also between the egg-laying dates of the females and the date of the males' departure back to Africa (estimate=8.09±2.47, F1,14=10.71, P<0.01; see also Ouwehand and Both, 2017, for a similar trend). This suggests that male departure date is set relatively early in the season and may even be related to the timing of spring arrival, although our data do not show a significant correlation between male spring arrival and autumn departure dates (but our sample sizes are relatively small). This suggests that male pied flycatchers cannot or will not delay their departure to accommodate a late chick hatching or delay their moult. A delay in chick hatching or moult would increase the time pressure for chick rearing and growth and adult feather replacement, with potential fitness costs (Dawson, 2004; Tomotani and Muijres, 2019; Tomotani et al., 2016; Vágási et al., 2012).
It is striking that flycatchers depart soon after the breeding season and that males relatively frequently bear the costly overlap between chick feeding and moult, suggesting a great benefit to departing early. Interestingly, the ongoing advance in mean laying date in our breeding population since 1980 (Tomotani et al., 2018a; Visser and Gienapp, 2019; Visser et al., 2015) could have led to some relaxation in moult–breeding overlap, because moult was found to be unrelated to laying date within years. However, there was a difference between years in the start of moult, being earlier in the warmer 2014 than in the colder 2013, which suggests that this annual-cycle stage may also vary over the years. Indeed, in the last 10 years, male moult onset has advanced faster than breeding (Tomotani et al., 2018a). Autumn departure has also been shown to advance in single-brooded species migrating through Switzerland (Jenni and Kery, 2003), although not clearly in citizen science observations in Britain (Newson et al., 2016). It is thus not clear yet how a change in the timing of one annual-cycle stage over time impacts on later stages, but for pied flycatchers, it seemed that any delays during autumn migration are compensated during winter (Ouwehand and Both, 2017).
It is interesting that we did not find short-term fitness costs of moult–breeding overlap as there was no effect of treatment on any component measured. Williams (2012) suggested that there is a lack of empirical evidence in the literature supporting the energetic costs of overlapping moult and breeding (but see Tomotani and Muijres, 2019; Tomotani et al., 2018b). However, such costs could appear at a later stage in the annual cycle (Williams, 2012). Our data support this idea, as birds in different treatments also differed in their wintering location, with delayed birds wintering further west than advanced and control birds (Fig. 5). Interestingly, the differences in wintering longitude that we observed resemble the population difference of pied flycatchers seen across different European breeding populations (Ouwehand et al., 2016), in which naturally late breeding populations wintered further west than naturally earlier populations. A similar effect was, however, not observed in birds that naturally breed earlier or later either in our population or in another Dutch flycatcher population deployed with geolocators (Ouwehand and Both, 2017). These observations support a causal relationship between moult–breeding overlap and wintering behaviour. We can speculate: (i) that such a pattern appears as a result of a change in body condition of birds in the delayed treatment that made individuals less competitive to secure a winter territory and/or made them less loyal to their former wintering site – although this could also lead to a later autumn departure, for which we found no evidence; or (ii) the possibility that birds that we recovered were the ones that changed behaviour and wintered in a different location, while the others did not return to the breeding grounds and thus are not represented in our dataset. Regardless, we suggest that delaying breeding may have carry-over effects on the wintering distribution of pied flycatchers.
Finally, the lack of causal relationship between the termination of breeding, moult and departure means that these three life history stages may be differently affected by climate change. When climate change advances the timing of breeding, but the timing of moult does not shift at the same rate, this would either increase the time available between breeding and moult (thus reducing the probability of a moult–breeding overlap) or increase the amount of overlap. However, this overlap will increase when moult advances faster than egg-laying dates (Dawson, 1998; Tomotani et al., 2018a; Visser et al., 2009). Even if some time is gained between moult and breeding, there will be time loss in another portion of the cycle, as has already been seen in the time between arrival and breeding in the Hoge Veluwe population of pied flycatchers (Both and Visser, 2001; Tomotani et al., 2018a). In summary, our results suggest that while the advancement of breeding caused by climate change will increase the time pressure for arrival dates, it may relax the pressure for summer moult and autumn migration, and this way may lead to a mismatch of annual-cycle stages.
We are grateful to the board of the National Park ‘De Hoge Veluwe’ for permission to conduct our research on their property; and Geldersch Landschap en Kasteelen and Staatsbosbeheer for permission to work on Heikamp and Buunderkamp. Henri Bouwmeester, Ezra Caglar, Susanne Maas, Evert Boesenkool, Piet de Goede and Peter de Vries gave us assistance with fieldwork. Kate Lessels helped with the control experiment in Buunderkamp. Raymond Klaassen and Janne Ouwehand assisted with geolocator deployment and Raymond Klaassen gave advice on data analysis. Samuel Caro assisted with hormone sampling. Barbara Helm, Kees van Oers and Suvi Ruuskanen advised on how to keep pied flycatchers in captivity. Marylou Aaldering, Ruben de Wit, Franca Kropman, Coretta Jongeling and Anouk de Plaa took good care of our birds. Christa Mateman and Ezra Caglar genotyped and assigned the paternity of the chicks, and Louis Vernooij managed the data. Kate Lessells, Philipp Gienapp and Callum Lawson gave advice on the statistical analysis. Finally, we are grateful to the referees who carefully read our manuscript and provided constructive comments to improve it.
Conceptualization: B.M.T., I.d.l.H., C.B., M.E.V.; Methodology: B.M.T., I.d.l.H., C.Y.M.J.G.L., C.B., M.E.V.; Formal analysis: B.M.T., M.E.V.; Investigation: B.M.T., I.d.l.H., C.Y.M.J.G.L., B.v.L., S.L.M.; Resources: S.L.M., M.E.V.; Data curation: B.M.T., S.L.M.; Writing - original draft: B.M.T., M.E.V.; Writing - review & editing: B.M.T., I.d.l.H., C.Y.M.J.G.L., B.v.L., S.L.M., C.B., M.E.V.; Supervision: M.E.V.; Project administration: B.M.T., M.E.V.; Funding acquisition: B.M.T., M.E.V.
B.M.T. was supported by a doctoral grant from CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico, Brazil; proc. no. 237790/2012-2) and S.L.M. by the Roslin Institute strategic grant funding from the Biotechnology and Biological Sciences Research Council (BB/P013759/1).
The authors declare no competing or financial interests.