Experimental manipulation of nocturnal nest cavity temperature in wild blue tits

Advances in the timing of reproduction in temperate species are some of the most well documented biotic responses to increasing global temperatures. However, the magnitude and rate of these advances in timing are not equal across all taxonomic groups. These differences can lead to disruption of interspecific relationships if species respond differently to temperature changes. Understanding the relationship between temperature and phenology is a key step in predicting future population trends for species living in seasonal environments. However, experimentally manipulating temperature in the wild is logistically challenging and has consequently rarely been attempted. In this study we experimentally test whether in-nest temperatures in early spring act as a cue for breeding phenology in a population of wild blue tits (Cyanistes caeruleus). We split nests into three treatments; heated, cooled, and control. In-nest temperature in the heated and cooled boxes was manipulated by an average of ± 0.6 °C from control temperatures using heating devices and ice packs respectively. We assessed the impact of our experimental manipulation on box occupancy and reproductive timing. We found trends towards earlier phenology in heated nest boxes in addition to a higher occupancy rate in cooled boxes, however neither of these trends was found to be statistically significant. Our ability to distinguish statistical signals was hampered by unexpectedly low occupancy rates across all experimental treatments. Based on the results we cannot say if nocturnal in-nest temperature is an important cue for nest box choice or the timing of laying.

boxes was manipulated by an average of ± 0.6 ºC from control temperatures using heating devices 25 and ice packs respectively. We assessed the impact of our experimental manipulation on box 26 occupancy and reproductive timing. We found trends towards earlier phenology in heated nest 27 boxes in addition to a higher occupancy rate in cooled boxes, however neither of these trends was 28 found to be statistically significant. Our ability to distinguish statistical signals was hampered by 29 unexpectedly low occupancy rates across all experimental treatments. Based on the results we 30 cannot say if nocturnal in-nest temperature is an important cue for nest box choice or the timing of can result in energetic demands that cannot be met and a consequent reduction in fitness (Reed,42 Grøtan, et al. 2013; Reed, Jenouvrier, et al. 2013). Therefore understanding how temperature 43 influences breeding phenology is crucial for predicting the impact of climatic change on 44 reproductive success. Despite this importance, the precise role that temperature plays in altering 45 breeding phenology has not yet been determined. 46 47 Matching peak energetic demands to a shifting resource peak is particularly challenging for 48 consumer species, which rely on species in lower trophic levels as their resource. The consumer 49 species must respond to temperature in the same way as the resource in order to maintain 50 synchrony, this is challenging for species with temperature independent development, where 51 embryos have a fixed developmental period. In contrast to species with temperature-dependent 52 development, which respond directly to temperature changes (Perrins 1979), temperature-53 independent species must use predictive cues in order to determine the optimal breeding time (van 54 environmental cues or the specific cues they use to time their phenology, they will shift their timing 58 to differing degrees, which can disrupt synchrony between species (Cushing 1969). For instance, if 59 a prey species advances their phenology at a faster rate than a predator species, due to greater 60 responsiveness to an environmental cue this could result in temporal mismatch. Peak abundance of resources would occur prior to the peak energetic demand, reducing survival and reproductive 62 success of the predator (Reed, Grøtan, et al. 2013). This has been the case for great tits (Parus 63 major) in the Netherlands. Great tits in the Hoge Veluwe population have not advanced their 64 phenology as much as resources despite increases in spring temperatures, whereas peak abundance 65 of their prey species, winter moth caterpillars (Operapthera brumata) has advanced by over one 66 week in two decades (Visser et al. 1998). This has resulted in fitness reductions for these great tits 67 (Reed, Grøtan, et al. 2013). In order to understand how and where these mismatches may occur, it is 68 important to identify the precise cues that individuals use to time their reproductive behaviour. Only 69 through understanding the mechanisms that control reproductive timing can we predict how timing 70 will change into the future and whether temporal mismatch will be likely. 71

72
Attempts to identify the cues that drive reproductive phenology consist of several approaches. The 73 most commonly used approach is to conduct regression based analyses on observational data (van 74 what it would be in the wild (Lambrechts et al. 1999). night by ± 0.75 ºC from two to four weeks prior to the median clutch initiation date of each year; 130 however, no difference in phenology was detected between treatments. Here we repeat and extend 131 this study using a population of wild blue tits. We heated and cooled 87 (N = 29 per treatment) nest 132 boxes by an average of ± 0.6 ºC from 6 pm to midnight from 17 days before the first egg was laid, 133 and testing whether this manipulation influences nest box choice, commencement of nest building 134 and clutch initiation date. We hypothesise that female blue tits will preferentially choose to nest in 135 warmer nest boxes in order to minimise their nocturnal energy expenditure (Ardia et al. 2009; 136 Vedder 2012; Dhondt & Eyckerman 1979). Furthermore, we predict that the temperature of the nest 137 cavity will also act as a cue for clutch initiation date (Dhondt & Eyckerman 1979;O'Connor 1978), 138 therefore the warmer nests will also experience the earliest clutch initiation dates and the coolest 139 boxes, the latest clutch initiation date. diameter too small to allow larger species, such as great tits, to enter the boxes. By using these 148 boxes rather than those with larger entrance holes, which are also present in the study area, we were 149 able to exclude great tits from our experimental manipulations. We use 87 of these blue tit nest 150 boxes as the focus of our study (Figure 1). These 87 were chosen because they are located close 151 together in a topographically similar area of the woodland, consequently limiting habitat 152 differences. Proximity of the boxes was essential for this experiment to ensure all experimental 153 treatments could be set running within a two hour window each evening. these ambient temperature iButtons were distributed in a grid system across Wytham woods with positions chosen to reflect the density of nest boxes. QGIS (Team 2016) was used to match each 163 experimental nest box to a nearest ambient temperature iButton, therefore allowing the calculation 164 of local ambient temperature for each box. The measure of local ambient temperature used in this 165 study is mean temperature for the experimental period, from 1 st April to 6 th May or until the first egg 166 is laid, whichever is first.

177
Of the 87 experimental boxes, 29 were assigned to each of the three treatment types: heated, cooled, 178 and control. Boxes were assigned to treatments using a stratified sampling technique, taking 179 account of historical box popularity. Historical popularity was determined by calculating the 180 proportion of years in which each box has been occupied since its placement in the woodland. 181 Boxes were then grouped into bands of popularity in increments of 10 %, within each group 182 treatments were assigned randomly using R (Team 2008). This assignment method was used to 183 prevent a clustering of treatments by historical popularity as this metric showed a significant 184 negative correlation with mean clutch initiation date of the nest box (EST = -3.18, SE = 1.46, P = 185 0.03), with the historically most popular boxes having the earliest clutch initiation dates. This 186 sampling approach also created an even likelihood of box occupancy across treatments, therefore 187 allowing us to explicitly test whether our experimental manipulation altered the popularity of 188 individual boxes. We tested whether other spatial components, such as altitude, influenced the mean 189 clutch initiation date of each box, but this was not statistically significant (EST = 0.02, SE = 0.03, P 190 = 0.52). Consequently, we did not take account of altitude when assigning treatments. This 191 assignment of treatments does mean individuals can move between different treatments if they do 192 not roost in the same nest box every night, however, it also avoids temporal or spatial clustering of 193 treatments and is essential for the exploration of in-nest temperature influence on occupancy. The heating device comprised (see Figure 2 for details of device construction) of a three-volt 202 filament bulb secured in a mount within a metal topped polystyrene disc (approximately 2.5-3.5 cm 203 thick), which was placed on top of a plastic ring with a 5 cm gap at the front (also 2.5 cm in height). 204 The bulb was powered by two rechargeable batteries, size C, located beneath the polystyrene disc 205 within the plastic ring (see Figure 2). The bulb was covered from above by a 5 cm diameter metal 206 disc, to hide the light produced. Heating was achieved by the bulb heating the metal disc and 207 convectional transfer of the heat up into the nest cavity. These heated boxes also contained four, 208 non-frozen, Thermafreeze ice packs (Thermafreeze EU, black ice packs) in the same configuration 209 as the cooling device (see below) to ensure that all experimental boxes had the same outward 210 appearance. 211 The cooling device comprised two frozen Thermafreeze ice packs secured at the back of each nest 217 box and two on the nest box door. Ice packs at the back of the box were secured using the same 218 metal topped polystyrene disc as used in the heating device, with another polystyrene disc below 219 (around 2.5 cm thick) to achieve the same thickness as the heating device but without the plastic 220 ring and batteries. Ice packs on the nest box door were secured using rubber bands. All ice packs 221 had a black outer side to camouflage them from the birds once the nest box was closed. 222

223
The control treatments comprised the same construction as the cooled treatment but with non-frozen 224 ice packs. Therefore, all boxes had the same appearance, a raised metallic box floor of 225 approximately 5 cm above the true base, with ice packs mounted front and back. 226

227
Both the batteries and frozen ice packs had an effective life of around six hours and therefore were 228 replaced each evening around 2.5 to 1.5 hours prior to sunset so that the treatments were running 229 during the night. This period of the day was chosen as female blue tits usually roost overnight 230 within their nest cavity prior to nest building and therefore should be present within the box to 231 experience the temperature manipulation (Kluijver 1950;Perrins 1965a;Kidd et al. 2015). 232 Temperatures were altered on average by ± 0.6 ºC from the control boxes from sunset (around 6 233 pm) to midnight. Batteries and ice packs were replaced each evening; control boxes were also 234 visited to achieve a uniform disturbance level across boxes. Some damage to equipment did occur 235 (pecks to polystyrene inserts) and equipment was either repaired or replaced to ensure the 236 treatments could continue. Experimental treatments could not be changed on two nights during the

263
A power analysis was conducted prior to starting the experiment to determine whether differences 264 were likely to be detected between treatments given the number of nest boxes available. Based on 265 the average historical occupancy of the study boxes (67 %) we would expect an occupancy rate of 266 19 boxes per treatment. With this number the power analysis showed that we could expect a 75 % 267 chance of detecting a mid to strong signal (an effect size of 0.4 or higher). We would expect a mid 268 to high effect size based on previous work that has found a strong correlation between temperature 269 and clutch initiation date (Perrins 1970;Perrins 1973). Furthermore, analyses of the relationship study 12 % of individuals (8 % of breeding attempts) returned to the same box they had previously 283 bred in, suggesting that for the majority of individuals previous location is not the only driver of 284 nest box choice. Therefore we used one way ANOVAs and a Welch's T-Test to explore the 285 difference in historical popularity, altitude and ambient temperature between occupied and 286 unoccupied boxes. A Welch's T-Test was used for local ambient temperature due to 287 heteroscedasticity of temperature data. The whole dataset was used for these analyses as there was 288 not enough data to split by treatment. 289

290
Analyses of the influence of experimental treatments on breeding phenology were performed using 291 one way ANOVAs to determine if there are differences between phenology measures in each 292 treatment group. The phenology measures considered here were the date of nest building (the date 293 on which moss was first observed in the nest box), the clutch initiation date (date when the first egg 294 is laid), and date of clutch completion (date when the last egg in the clutch is laid). Regression 295 analyses were also performed to assess the influence of environmental variables on clutch initiation 296 date. Environmental variables tested were altitude and local ambient mean temperature in a single 297 linear regression to assess the combined influence of these variables on variance in phenology. 298

Ethics statement 299
Work was subject to review by the Department of Zoology ethical committee, University of 300 Oxford. All work adhered to UK standard requirements and was carried out under Natural England 301 licence 20114732. This experimental methodology was discussed with, and approved by the 302

Departmental Animal Welfare Ethical Review Body (AWERB). Field work took place in Wytham 303
Woods (lat. 51°46'N, long. 1°20'W), private land that belongs to the University of Oxford; for 304 permission contact the Conservator, Nigel Fisher. No endangered or protected species were 305 involved in the study. A breakdown of nest building, abandonment and chick mortality by treatment 306 can be found in supporting information Table S1, however these rates did not differ from those in 307 non-experimental boxes. 308

309
The influence of experimental treatment on occupancy 310 Occupancy in our experimental boxes was 16 %, which was a decline of 73 % from the preceding 311 year (2014). Adjacent non-experimental boxes had a blue tit occupancy of 14 % (Table 1) boxes had higher historical popularity than unoccupied boxes (F(1,82) = 6.15, P = 0.02, see Figure  315 3), and this pattern was consistent across treatments. It should be noted that three of the occupied experimental boxes (N = 17) were occupied by marsh tits (Poecile palustris). As these are a 317 different species to our study species, this could influence some of our results. Therefore we 318 removed these nests from our analyses, leaving a sample size of N = 14. 319

Does in-nest temperature influence nest box choice?
334 The number of nest boxes occupied during our experimental period varied between treatments 335 (Table 1). Cooled boxes had the highest proportion (24 %) of boxes occupied, followed by control 336 boxes (17 %) and heated boxes (7 %) have the lowest levels of occupancy. These differences were 337 not statistically significant (χ 2 = 3.97, DF = 2, P = 0.14). We explored additional factors that may 338 explain variation in occupancy. Neither altitude nor local mean ambient temperature differed Cooled Does in-nest temperature influence breeding phenology? 349 For these analyses N = 13 (two heated boxes, five control boxes and six cooled boxes), one nest 350 was removed from analyses due to lack of a clutch initiation date as laying occurred after the 351 conclusion of experimental treatments. On average, birds nesting in heated boxes started nest 352 building, laying and completed their clutches earlier than birds nesting in the control or cooled 353 boxes (Figure 4), however none of these differences were statistically significant (nest building 354  (Table 2). Secondly, it was also clear that the parent blue tits found the disturbance 386 and/or altering of the box appearance a deterrent from nesting in these boxes. Evidence that the 387 birds were disturbed by the presence of the equipment was the pecking of polystyrene mounts in 388 many of the boxes. This demonstrates the birds were not deterred from entering the box but also appeared to want to remove the heating and cooling devices. Evidence that birds entered the 390 experimental boxes was also found in the form of signs of roost (faeces and feathers) found in 391 several of the experimental boxes. However this use of boxes did not translate into nest building. 392 Smaller nest cavities have been linked to higher nestling mortality (Mertens 1977) and could 393 therefore deter parent birds. Additionally, the blue tit occupancy of the experimental nest boxes 394 declined by a larger proportion than the non-experimental boxes (73 % and 48 % respectively, see 395 Table 2). However, this cannot definitively be attributed to our experimental treatment because 396 declines in both blue tit and great tit numbers were seen throughout the whole of Wytham Woods 397 from 2014 to 2015. The lack of increased blue tit occupancy in the surrounding non-experimental 398 boxes, suggests that birds deterred from the experimental boxes did not relocate to nearby unaltered 399 boxes. However, disturbance did not seem to influence marsh tits to the same degree, their numbers 400 remained at a similar level of occupancy to historical records, suggesting they were less deterred by 401 the experimental set up than blue tits. Further refinements of techniques to minimise the amount of 402 equipment required to heat and cool boxes would improve these experimental techniques. Caution 403 should be employed for future blue tit nest box manipulations. 404 The impact of low sample size on statistical analyses 405 The sample size achieved in this study was exceptionally low, this heavily shaped our ability to 406 draw statistically supported conclusions from this experimental work. With a sample size of 13 407 across all three treatments (two heated, five control, and six cooled), very little could be ascertained 408 statistically from the resulting dataset. As analyses were planned prior to running of the experiment, 409 they were duly conducted. However, even if statistically significant results are found, we cannot 410 confidently reject a null hypothesis. The inability to determine statistical results stems from 411 fundamental issues with multiple testing and p-values. 412

413
The rise of inappropriate and misleading statistical analyses is a growing concern across many 414 scientific disciplines, in particular psychology, medicine, ecology, and evolution ( variables to a focal explanatory variable (treatment), which were all chosen a priori. As a result, we 430 move away from discussion of statistical significance in this study, as it cannot be relied upon in 431 this instance (Mcshane et al. 2017). Instead we focus on the effect size of any trends found. 432

Trends found
433 Trends shown in this analysis are that historical occupancy levels (popularity) also appear to predict 434 occupancy during the experimental period. Boxes occupied during our experiment were occupied 435 for, on average, 15.6 % more of the historical time period than the unoccupied boxes. It seems 436 logical that the historically most popular boxes are also those occupied most readily during the 437 experimental period. Neither altitude nor local mean temperature appear to explain popularity, 438 either historically or during this experiment. Further analyses of different box positions would be 439 required, manipulating specific aspects of the box environment, for example, aspect or oak density, 440 to determine the drivers of occupancy/popularity of nesting sites. Although cooled boxes had the 441 highest number of occupied boxes, this was statistically indistinguishable from control or heated. 442 The difference between treatments was a maximum of four boxes. Past studies have found a slight 443 preference for cooler nest boxes (Dhondt & Eyckerman 1979), however, this could also be a chance 444 difference, substantially more analyses would be required to draw definitive conclusions. 445

446
No influence was found of in-nest temperature on breeding phenology. There was some overlap 447 between the cooled and control groups for all measures of phenology, however, mean differences 448 had effect sizes of several days. On average heated boxes initiated nest building and egg laying 449 approximately three to four days prior to control boxes and six to four days prior to cooled boxes. 450 The magnitude of these effect sizes from < 1ºC change in in-nest temperature, predominantly driven 451 by the heated treatment, suggest we cannot rule out an influence on nest box temperature on 452 breeding phenology even if it could not be statistically supported in this study. No influence was 453 found of environmental variables on clutch initiation dates. As ambient temperature typically 454 correlates strongly with clutch initiation dates across our long-term data, we would expect that in 455 this case the sample size was too low to distinguish a signal or the influence of ambient temperature 456 was dampened by the experimental manipulation. However, from the current data, we cannot 457 discriminate between the two possibilities. We were unable to find statistical support for the 458 relationships observed by Dhondt  This study would not have been possible without the hard work of Tony Price and John Hogg, who 466 assisted in the design and manufactured the heating and cooling devices for this experimental work. 467 We would like to extend our gratitude to both Tony and John for all of the time they put in to this 468 project. We are also grateful to all of the Wytham fieldworkers who collected population census 469 data on the Wytham great tits. In particular, we would like to give a special mention to Keith 470 McMahon, Stephen Lang, Koosje Lamers and Nico Alioravainen, who assisted with the changing 471 of batteries and ice packs for this experiment come rain or shine. This work was supported by 472 NERC grant NE/K006274/1 to Ben Sheldon. 473