Study Area and General Procedures

We collected data on diet composition from parasitic cuckoo and host redstart nestlings using nest boxes in pine forests near Ruokolahti (61°24′N, 28°37′E) in southeastern Finland from May to July, 2012–2016 (see Grim et al. 2009a, 2009b, 2014a and Samaš et al. 2016 for details on logistics and field procedures). This study population shows a high parasitism rate (31%, n = 392 nests).

The quality of parents could differ between naturally parasitized and naturally unparasitized nests: Female cuckoos may target hosts of higher quality (Polačiková et al. 2009) or may disproportionally parasitize host individuals of lower quality due to structural habitat constraints (Grim 2002). In either case, such nonrandom host targeting might confound dietary comparisons if individual host quality covaries with individual host prey selection. Therefore, we randomly removed the cuckoo egg from some parasitized nests and cross-fostered it to another nest. Thus, we created 4 groups of nests: naturally parasitized nests that either remained parasitized (n = 31) or were experimentally unparasitized by us (n = 15), and naturally unparasitized nests that either remained unparasitized (n = 55) or were experimentally parasitized by us (n = 51, using 15 cuckoo eggs from 15 experimentally unparasitized nests and additional 36 cuckoo eggs from nests that were not followed in the present study). To test for nonrandom individual host selection by cuckoo females, we included the predictor of original “parasitism status” (i.e. before cross-fostering) in our statistical models. Because of the randomization and interspersion created via cross-fostering, it was not strictly necessary to control for host pair quality (Hurlbert 1984); however, we conservatively measured host pair quality as provisioning effort (feedings hr⁻¹) and tested its potential covariation with diet composition.

We regularly checked nests throughout the nestling period to video-record host provisioning. Before expected fledging, we measured nestling mass using a digital scale (model AMW-600; American Weigh Scales, Cumming, Georgia, USA) with precision to ±0.1 g. Fledging parameters could be determined for only a subset of video-recorded nests (n = 120 out of 152) because some nestlings and broods did not fledge due to predation (n = 5) or inclement weather (n = 1) or were not followed through to fledging due to logistic reasons (n = 26; i.e. too many synchronous nests in too-distant parts of our large study area, which covered 25 × 7 km; see figure 1 in Samaš et al. 2016). Sample sizes vary slightly (Results) because we failed to record complete data for some nests.

We used 3 methods specifically developed to prevent potential effects of video-recording and observer visitation on nest success. First, cameras were inconspicuously hidden in wooden nest box extensions (see figure 2b in Samaš et al. 2016). Second, we excluded most nest predators mechanically: At the start of the incubation period, we inserted 2 nails into the nest box entrance hole so that redstarts could enter but predators were excluded (see figure 2a in Samaš et al. 2016); this worked because predation rates were almost nil (4%, n = 126 nests). We removed the nails from the nest boxes with cuckoo nestlings shortly before expected fledging; there was no need to do this for the nest boxes with redstart fledglings because host nestlings were small enough to move freely between the nails. Video-recordings confirmed that cuckoo nestlings did not try to fledge before we removed the nails. Third, fledging time was primarily determined from long-term microcamera recordings, which started a day or 2 before the expected fledging date for a particular nest and thus eliminated the necessity of repeated nest checks by a human observer during the period shortly before nestlings fledge when human nest checks may trigger premature fledging (see also Grim 2007b).

We defined fledging success as the proportion of hatchlings that fledged. For redstart broods, each nest represented 1 data point (i.e. the proportion of hatchlings that fledged from each nest). In mixed broods, we scored cuckoo and redstart nestlings separately. Shortly before expected fledging, nest boxes with cuckoos were checked daily and those with redstarts every 1–3 days. For fledging success analyses, the a priori temporal criterion for successful redstart fledging was that nestlings survived to 11 days of age (following Järvinen 1990). For cuckoos, the criterion was 18 days (Grim 2006c), with the day of hatching = day 0 (Li et al. 2016). We managed to follow all cuckoos and all but 1 redstart brood through to fledging; thus, we applied the temporal criterion to only 1 nest. Nests that failed due to predation (n = 5) or inclement weather (n = 1) were excluded in fledging age, mass, and success analyses because we wanted to test only for potential effects of diet type.

Video-recording

Using 2 types of cameras, we video-recorded nestling provisioning from hatching (day 0) to fledging. This whole period covered up to the age of 14 days in redstart nestlings and 20 days in cuckoo nestlings (Grim et al. 2009a, Grim and Samaš 2016). During each day of the nestling period, we video-recorded 15–39 nests with redstart nestlings and 13–37 nests with cuckoo nestlings. Due to this extensive sampling, we were able to identify unusually high (compared with typical dietary studies) numbers of prey items for both host (243–1,003 items per day) and parasitic nestlings (73–827 items per day). Nestlings older than the typical fledging age (i.e. >14 days for redstarts and >20 days for cuckoos) are naturally uncommon (due to fledging), but we managed to video-record 1–11 nests for each day up to the age of 17 days in redstarts and 27 days in cuckoos.

Some nests (10%) were video-recorded during the whole day (24-hr period), but 90% of nests were recorded only between 0800 and 1900 hours (EEST). Each nest was video-recorded 1–18 times (median = 4), always on separate days, and each recording lasted 1.0–4.0 hr, depending on weather conditions, predation, and logistic constraints. In contrast to all previous studies of cuckoo diet, and most studies of avian nestling diet, our recordings covered the whole period from hatching to fledging.

We placed each camcorder at the top of the nest box, where it was attached to a wooden nest box extension (16 cm depth × 15 cm width × 20 cm height; see figure 2b in Samaš et al. 2016). The lens of the camera pointed down into the box, covering both the nest cup and the nest box entrance in its field of view (Grim et al. 2014a). Thus, the recordings allowed us to clearly identify which species of nestling (cuckoo or redstart) got fed in all broods, including mixed broods (i.e. nests in which host and parasitic nestlings shared the nest; Grim et al. 2009a). We used Panasonic HDC-HS80 camcorders (HD quality; Panasonic, Kadoma, Osaka, Japan), which allowed us finer taxonomic identification of prey items (Appendix Table 2,  Supplemental Material Video S1,  Video S2,  Video S3, and  Video S4). Additionally, we used lower-resolution recordings from infrared microcameras (CCD Bird Box Camera 420TVL; SpyCameraCCTV, Bristol, UK) located inside nest boxes; each microcamera was connected to a digital video-recorder (Wireless 2.4 GHz 1 Channel D1 Mini DVR Recorder; SpyCameraCCTV) hidden in an underground box below the nest. Microcameras were primarily used to determine the exact time of fledging of host and parasitic nestlings, as they allowed us to record the brood undisturbed for several days without the need to check the nest in person, which may cause premature fledging (Grim 2007b). However, we also took advantage of these microcamera recordings to determine dietary items, although necessarily at a lower level of taxonomic detail (invertebrate, vertebrate, and plant).

When categorized by the 4 types of nest parasitism status, we video-recorded the following numbers of nests: 31 naturally parasitized, 51 experimentally parasitized, 15 experimentally unparasitized, and 55 naturally unparasitized. We cross-fostered randomly selected cuckoo eggs, and some of the cuckoo nestlings did not successfully evict all host progeny. This was not a result of cross-fostering because mixed broods were similarly frequent among naturally (9 out of 31) and experimentally parasitized nests (9 out of 51; Pearson's χ² = 1.46, P = 0.23), reflecting the natural situation in this cuckoo–host system (Rutila et al. 2002, Grim and Samaš 2016, Samaš et al. 2016, Thomson et al. 2016). This resulted in 64 “solitary” cuckoo broods (nests with 1 cuckoo that successfully evicted all host progeny; Grim et al. 2009a), 70 unparasitized host redstart broods, and 18 “mixed” broods that included 17 cuckoo–redstart broods (nests with 1 cuckoo that did not evict all host progeny and both species were raised jointly; see Grim et al. 2009a) and 1 brood of 2 cuckoo nestlings that both survived to fledging. This last case was a natural case of successful double parasitism; the cuckoos evicted all redstart progeny but neither cuckoo nestling managed to evict the other. Excluding this nest had a negligible effect on quantitative parameter estimates and no effect on our conclusions.

Analyses of Video-recordings

Identification of prey items to a fine level of taxonomical precision (i.e. to genus or species) almost always requires the examination of prey items under a microscope (e.g., Mayer 1971, Trnka 1995, Grim and Honza 1997). Thus, prey items must be removed from studied nestlings via neck-collars (e.g., Trnka 1995, Grim and Honza 1997, 2001). Such methods are invasive, may increase nestling mortality, and are no longer considered ethically acceptable. Therefore, we used less invasive and ethically sound video-recordings, whereby a researcher (Z. Tyller) visually inspected all food items that were visible on our video-recordings of feeding events (Martín-Gálvez et al. 2005, Grim et al. 2014a). This prevented us from detailed taxonomic identification of prey: the majority of feedings (79% out of 20,975, including 8,757 to solitary cuckoos, 1,934 to cuckoos in mixed broods, 1,963 to redstarts in mixed broods, and 8,321 to unparasitized redstarts) contained unidentifiable small insects; identification was not possible because redstart parents moved too fast, prey was too damaged, prey was hidden in the parent's bill, or prey was of typically small body size (see also Grim et al. 2014a). Note that even the determination of prey items collected with invasive neck collars, potentially the most precise method possible, was often unsuccessful, as seen from “indet.” prey items in such dietary studies (e.g., Trnka 1995, Grim and Honza 1997). This does not necessarily represent a disadvantage for any dietary study, including the present one, because various types of insect show similar nutritional and energetic values (more similar to each other than to other diet types such as fruits; Castro et al. 1989).

The snout-vent length (SVL) of lizards was estimated by a professional herpetologist (M. Veselý) using the known length of host redstart bills as a reference. The mass of lizards was then estimated based on the known allometric relationship between SVL and mass (Meiri 2010).

In the analyses of fledging mass, age, and success, we present results partitioned into 4 categories of nestling diet. “Plant” included nests where whortleberries (Vaccinium myrtillus) and lingonberries (Vaccinium vitis-idaea) were detected in addition to typical invertebrate diet items. “Vertebrate” included nests where viviparous lizards (Zootoca vivipara) were detected in addition to typical invertebrate diet items. Nests where nestlings were fed both vertebrates and plants in addition to invertebrates were classified as “Both.” We did not detect any other plant or vertebrate species in nestling diets. All of the other video-recorded nests were assigned to the category “Invertebrate.” We managed to collect data on dietary items (Table 1) for 83 cuckoo nestlings (64 solitary plus 19 nestlings in 18 mixed broods) and 86 redstart broods (70 unparasitized plus 16 in mixed broods; i.e. we failed to record any dietary information for redstarts in 1 of the mixed broods).

TABLE 1.

Prey abundance (n = number of food items), dominance (D, %), and frequency (F, %) of food items (see Appendix Table 2 for definition of these terms), and number of nests where each type of prey was fed by Common Redstarts to parasitic Common Cuckoo nestlings (n = 4,998 items from 10,691 feedings of 83 nestlings in 82 nests) and to their own nestlings (n = 4,413 items from 10,284 feedings of 86 broods in 86 nests; data from 1 out of 87 video-recorded broods were missing). Numbers of items are lower than numbers of feedings because dietary composition of many feedings could not be determined. Note that at some nests both plants and vertebrates were consumed, and therefore total sample sizes (nest, above) are smaller than the simple summary of number of nests in the table would suggest. Invertebrates were fed to nestlings in all nests. See Appendix Table 2 for a more detailed taxonomic determination of invertebrate diet items and  Supplemental Material Video S1,  Video S2,  Video S3, and  Video S4 for examples of feedings.

Sampling: Effort and Limitations

Despite massive sampling effort, and similarly to any other dietary study, we cannot exclude the possibility that we did not detect vertebrate or plant dietary components at nests included in the “Invertebrate” category. However, such misclassification would only introduce noise into our analyses and decrease the power of our tests to detect statistically significant differences among groups. We also stress that our temporal coverage of nests from hatching to fledging is much longer than that of any typical avian dietary study. This is because it is not possible to apply neck collars to freshly hatched nestlings or to nestlings shortly before fledging (see any of the dietary studies that we have cited above), but cameras can be used to cover the complete nestling period (this study). Also, the majority of dietary studies have not sampled particular nests repeatedly (as we have done), including statistically controlling for such repeated sampling (see Statistical Analyses, below). Crucially, the focal prey types of interest in our study (i.e. fruits and lizards) were easier to identify than the more common food items (invertebrates) due to their uniquely large size (lizards) and color (fruits;  Supplemental Material Video S1,  Video S2,  Video S3, and  Video S4). Fruits were further unambiguously confirmed by the color of feces of nestlings from the same nests, even after fledging (T. Grim, Z. Tyller, and P. Samaš personal observations). Thus, any biases in the present study will be smaller than those in typical avian dietary studies.

Sampling in the present study can be considered sufficiently representative: Data were collected over 5 breeding seasons (compared to 1–3 seasons used in previous studies), and numbers of sampled nestlings and numbers of samples (∼20,000 feedings) are much larger than those in previous studies (typically several dozens of feedings). On the other hand, the total number of successfully identified prey items is smaller. This follows from host feeding strategies and logistic constraints. First, unlike some passerines that bring a single large food item per feeding (e.g., tits; Naef-Daenzer and Keller 1999, Barba et al. 2009), redstarts typically bring multiple small items per feeding (similarly to, e.g., reed-warblers; Grim and Honza 1996, Davies 2015). Second, these multiple food items are largely hidden inside the bill cavity, preventing their identification without invasive methods such as neck collars (Grim and Honza 1996).

Finally, our samples were not well balanced between typical purely invertebrate-fed nestlings and broods and those fed additionally with plants or vertebrates, both for cuckoos (69 vs. 11 nestlings) and for redstarts (78 vs. 9 broods). Note that the unbalanced samples resulted from biological reality, that is, low numerical dominance of alternative food types, and not from poor sampling effort (which was massive by any standards).

As in all previous studies (Introduction), several items (e.g., a caterpillar, a spider, and an unidentified insect) could be identified in a single feeding. We simply analyzed all items from all feedings and present the data pooled per whole study population (Table 1, Appendix Table 2). This approach parallels all previous studies and makes our results comparable with previous work.

Statistical Analyses

First, we tested whether diet composition was specific to host pairs. Using repeated samples per pair (1 sample = a single 1-hr video-recording per nest per day, i.e. there was no more than 1 sample from the same nest and day), we calculated the repeatability of diet composition per host (parent) pair. Nests that provided only a single sample, e.g., due to early predation, were naturally excluded from repeatability analyses. Because unusual food items were not provided during the first week of nestling development (Results), we estimated repeatability only for the later part of the nestling period, when both usual and unusual diet items were provided. However, including data from the first week of nestling development led to virtually the same parameter estimates (results not shown). We scored the data as a binary variable: unusual food items (i.e. fruits and lizards pooled) present vs. absent. Further, we calculated repeatability separately for “fruits” and “lizards.” Repeatability estimates and their 95% CIs were calculated using a generalized linear mixed model with a binary response (logit link), using the formula r = VB / (VB + VE + π²/3), where VB denotes between-individual variance, VE is the residual variance (always fixed to 0 for binary response variables), and π²/3 is the inherent distribution-specific variance (for details see Nakagawa and Schielzeth 2010, Samaš et al. 2011, Grim et al. 2014b).

Next, we calculated 8 linear mixed models separately for each combination of nestling species (redstart, cuckoo) and fledging (mass, age, success) or provisioning parameter (parental provisioning effort). We used generalized linear mixed models with identity links when analyzing the continuous response variables of nestling fledging mass (g), fledging age (days), and provisioning effort (feedings hr⁻¹). We employed a generalized linear mixed model with a logit link when analyzing the binomial response variable of fledging success (fledged vs. died in the nest).

All models contained “diet type” (categorical variable, with the categories “Invertebrate,” “Vertebrate,” “Plant,” and “Both”) as a fixed predictor. In the fledging success analysis, the predictor “diet type” had just 2 categories (“Invertebrate” and “Noninvertebrate”). We pooled the “Vertebrate,” “Plant,” and “Both” categories into a single “Noninvertebrate” category for this particular analysis because all nestlings fed with lizards successfully fledged (thus, there was no variation in the data and the model failed to converge). We summarized the fit of models using marginal (R²_m) and conditional (R²_c) R-squared accompanied by Akaike's Information Criterion corrected for small sample sizes (AIC_c), as recommended by Nakagawa and Schielzeth (2013) and Lefcheck (2015).

In analyses with a fledging parameter as the response variable (fledging mass, age, success), we controlled for temporal between-year variation by including the variable “year” (5 levels, 2012–2016) as a random effect because we did not have any specific year-based predictions (Bolker 2015). Each redstart nest had several nestlings and thus we additionally controlled for within-nest variability by including the random effect redstart “brood ID.” This was applied only to models with the response variables “fledging mass” and “fledging age” because for “fledging success” all predictors and the response were quantified at the brood level (fledging success was defined as the proportion of hatchlings that fledged per whole brood). In analyses with provisioning effort as the response variable, we controlled for variation in both nestling age and brood ID by employing a random slope model (i.e. nestling age was statistically nested within brood ID, following the guidelines of Bolker 2015). This model took into account the fact that provisioning rate in relation to nestling age differed across broods. Further, nestling growth performance and fledging success may vary because of cohabitation (Grim et al. 2009a) and individual host quality (Polačiková et al. 2009). Therefore, we additionally tested for the potential confounding effects of “brood type” (categorical variable: mixed or not) and “parasitism status” (categorical variable: nest originally parasitized by a cuckoo female or not).

The timing of breeding within a season in relation to the occurrence of an unusual diet was controlled for by the binary variable “fruit availability” (yes or no) in each particular season. The onset of ripe berries typically comes in the middle of the redstart breeding season, so each year we had active nests when fruits were both unavailable and available. The year-specific date of whortleberry ripening was obtained from Metla ( http://www.metla.fi/cgi-bin/feno/kuvasarja/mu/10.5/30.8/3/en). The occurrence of young lizards has not been monitored in southeastern Finland, and thus we did not have any specific predictor for the availability of vertebrates; however, anecdotal data suggest that lizards, including young ones, are available throughout the whole redstart breeding season (J. Rutila personal communication).

If fruit was available only later in the season, and if late nests produced fledglings of lower mass overall, then there would be a potential problem: the effect of “fruit availability” on fledging parameters would be confounded with seasonal effects. An alternative way to control for the potential effects of fruit availability would be to include only nests that had nestlings in the period when fruits were available. However, such data trimming would have caused us multiple problems. First, we would have had to exclude ∼70% of our data points, leading to poor statistical power of our tests. Second, excluding data from the early part of the season, when fruits were not available, would have automatically excluded some of our (already limited) sample of nests where lizards were fed to nestlings (lizards were available the whole season, independently of fruits), leading to even poorer statistical power. These limitations, of course, do not solve the potential issue of “fruit availability” being confounded with seasonal effects. Therefore, we ran separate alternative analyses in which we directly tested whether seasonal effects could explain the fledging patterns. We removed the predictor “fruit availability” and instead included a new predictor, “hatching date” (the rest of the models remained identical to models reported in Appendix Tables 3 and 4). Hatching date was centered on the mean within each year to exclude any confounding effects of between-year variation (following Grim et al. 2011). We ran these models for fledging mass, age, and success. Hatching date was not statistically significant in either full or final models, the statistical significance of the other predictors did not change qualitatively (i.e. no nonsignificant predictors became significant, and vice versa), and parameter estimates remained similar to those in our original analyses (i.e. analyses with “fruit availability”). Although the conclusions remained the same when we used “fruit availability” and “hatching date,” we preferred using the former predictor because it directly reflected biological reality (i.e. the availability of the alternative plant diet item in each specific season), whereas the latter predictor merely assumed in an unspecific way that conditions were poorer at the end of the season. However, this assumption was not supported by our data: “weather” (see below) was not correlated with “hatching date” for cuckoos (r = 0.02, P = 0.88, n = 74) or redstarts (r = –0.19, P = 0.11, n = 72).

Weather, including precipitation and temperature, may affect food supply and nestling growth (Pérez et al. 2016). Weather data were acquired from the Finnish Meteorological Institute ( https://en.ilmatieteenlaitos.fi/open-data). We tested potential weather effects with 2 predictors, total daily rainfall (a continuous variable measured in mm per day) and average daily temperature (a continuous variable measured in °C). In the analyses of fledging parameters, we averaged rainfall and temperature for the whole nestling period for each particular brood (i.e. from the date of hatching till the date of fledging). We detected high collinearity between the predictors of rainfall and temperature (variance inflation factor: VIF > 7; Zuur et al. 2010); the variables were strongly negatively correlated (Pearson's r = −0.87, P < 0.001). Thus, we used principal component analysis (PCA) to create a new single variable, “weather,” representing both predictors. The first principal component, PC1, explained 94% of the total variance and was positively correlated with rainfall (Pearson's r = 0.65, P < 0.001) and negatively correlated with temperature (Pearson's r = –0.51, P < 0.001). In the analyses of provisioning effort, we controlled for weather on the particular day of video-recording of provisioning behavior. There was no collinearity between the 2 predictors in this case (i.e. rainfall and temperature at the day level); therefore, we used both as separate predictors in the provisioning effort analyses.

In the analysis of fledging mass, we included an additional predictor of nestling age (continuous variable, measured in days) when fledging mass was measured (“age weighed”; this was always after nestlings reached the asymptotic phase and shortly before they fledged). For analyses of redstart data, we additionally included another potential predictor, “brood size” (as a continuous variable). Brood size was not included in parasitic nestling models because brood size did not vary for solitary cuckoos (brood size always = 1), whereas it was always larger than 1 for mixed broods; therefore, there would be multicollinearity between “brood size” and “brood type.” In the analyses of provisioning effort we included the additional predictors “daytime” (continuous; in hours) and its quadratic term “daytime²” (because provisioning effort can be expected to be nonlinearly related to the hour of the day, with increased feeding in the morning and evening) and the “date” of video-recording (because food availability may theoretically decline late in the season leading to lowered provisioning effort).

We present results from both the full (as recommended by Forstmeier and Schielzeth 2011) and final reduced models (as recommended by Grafen and Hails 2002). We used backward elimination of nonsignificant terms. We kept the main predictor of interest, i.e. “diet type,” in the models until the final step, regardless of its significance (following Grafen and Hails 2002). Additionally, in another series of analyses, we added each previously removed covariate (separately, one by one) and confirmed that it was nonsignificant in all the final models. In models with identity links, we checked the assumptions of normality of residual errors, linearity of effect, and homogeneity of variances by visual inspection (Ieno and Zuur 2015), and found them satisfactory.

We were specifically interested in differences between the potential effects of the following diet combinations: invertebrate vs. vertebrate, invertebrate vs. plant, and invertebrate vs. both. We did not have any specific predictions about differences between plant vs. vertebrate diet effects: both effects should be detrimental, but we did not find any information in the published literature to allow us to predict the absolute or relative magnitude of the 2 effects. Thus, there were no grounds to predict whether the effect of fruits would be larger or smaller than that of vertebrates. Therefore, we used post hoc Dunnett's tests with invertebrate diet (typical for both redstart and cuckoo nestlings; Glutz von Blotzheim et al. 2001) set as a reference level. The results of the post hoc tests remained the same when applied to both full and final models (Appendix Tables 3–5). For simplicity, we present results based on final models (Figure 1). All analyses were performed in R 3.3.2 (R Core Team 2016). All estimates are means ± SE.

FIGURE 1.

(A) Fledging mass and (B) age of Common Cuckoo nestlings provisioned by Common Redstart hosts with either a typical diet of insects (Invertebrate, n = 62 nestlings), or a diet containing additionally either young lizards (Vertebrate, n = 2), fruits (Plant, n = 7), or both lizards and fruits (Both, n = 2). (C) Fledging mass and (D) age of Common Redstart broods (brood averages, i.e. without pseudoreplication) provisioned by their parents with either a typical diet of insects (Invertebrate, n = 60 broods) or a diet that also (i.e. additionally to invertebrates) contained either young lizards (Vertebrate, n = 5), fruits (Plant, n = 3), or both lizards and fruits (Both, n = 1). Total sample sizes are smaller than those for diet composition (Table 1) due to predation, weather effects, and logistic reasons. Numbers within bars are P-values from post hoc Dunnett's tests comparing Invertebrate (reference category, gray bars) with either Vertebrate, Plant, or Both (black bars). Values are least square means + SE from final models controlling for other significant predictors where appropriate (see Appendix Table 3).

Alternative Explanations

Theoretically, poorer fledging parameters of cuckoos fed with unusual dietary items could be a by-product of individual host pair selection by cuckoos, that is pair quality (Grim 2002, Polačiková et al. 2009). We excluded this potential confounding factor experimentally by cross-fostering randomly selected cuckoo eggs between naturally parasitized and unparasitized nests (see Methods). We also explicitly statistically tested this potential confounding effect (Appendix Tables 3–5). None of the various fitness-related parameters covaried with host parasitism status for cuckoo nestlings. To our knowledge, cross-fostering was done in only one previous dietary study (Martín-Gálvez et al. 2005), which highlights that the results of the present study should be robust.

However, randomization via cross-fostering does not automatically ensure interspersion of treatments (Hurlbert 1984). Despite our efforts, we could still have inadvertently assigned cuckoos to host pairs that showed overall poorer provisioning capacity, because of which they may have fed nestlings with unusual diet items, although we find it extremely improbable that such a coincidence would have arisen in 5 separate breeding seasons. In this scenario, hosts of lower quality would have collected suboptimal prey to feed to nestlings. Clearly, this scenario is unlikely because the redstart nestlings fed with fruits and lizards did not show poorer fledging parameters (Figure 1). Indeed, we did not detect any differences in provisioning capacity (i.e. feeding frequency) between hosts feeding cuckoos with the different types of diet (Appendix Table 5).

Importantly, host vs. parasite fledging success was not correlated within mixed broods. These analyses clearly reject the alternative “poor parent” scenario and further support our prediction (Introduction) and results that the presence of fruits and lizards in the diet is causally responsible for the poorer fledging parameters of cuckoo nestlings.

Finally, poor foraging conditions could also have caused the observed patterns. Poor conditions could manifest themselves either spatially (at particular nests located in poor habitats) or temporally (in particular years or particular parts of the breeding season with inclement weather).

A spatial confounding effect is highly unlikely for 3 reasons. First, our study sites were highly homogeneous because they were all located in industrial forests of similar age (Samaš et al. 2016). Consequently, the structure of the vegetation layers, and prey availability, was highly uniform. Second, we randomly cross-fostered cuckoo eggs among available synchronous redstart nests, eliminating any potential confounding effect of spatially specific poor conditions: Nests with host vs. parasite nestlings were interspersed in space (sensu Hurlbert 1984). Third, no matter how “poor conditions” may have been manifested, we established quantitatively and statistically that nestlings who were fed with unusual food items were fed at similar frequencies as nestlings fed with “standard” invertebrates.

A temporal confounding effect is also highly unlikely, for 2 reasons. First, parasitized and unparasitized nests were interspersed in time (sensu Hurlbert 1984) across each of the 5 breeding seasons that we sampled; thus, nests with both cuckoo and redstart nestlings were synchronous and therefore experienced similar weather conditions. This is inevitable in mixed broods in which both parasitic and host nestlings experience identical conditions. Second, we explicitly quantified and statistically tested for potential confounding weather effects (rain, temperature); as expected due to the previous point (synchrony), the patterns of nestling growth and success that we detected were not explained by weather (cf. Pérez et al. 2016). Analyses based on the alternative predictor of “hatching date” did not change our conclusions. There was no decline in provisioning rates late in the season for redstarts and there was an increase for cuckoos (Appendix Table 5), and cuckoos survived better (not worse) overall when fruits were generally available (Appendix Table 4). These patterns are mostly the opposite of what would be expected under the assumption of confounding seasonal effects, and therefore provide strong support for our conclusions (causal effect of unusual dietary items).

Thus, although prey availability may vary in space and both within and between breeding seasons, such confounding effects were excluded in the present study via experimental design and statistical control (see full models including all these covariates; Appendix Tables 3–5). Previous studies of cuckoo diet rarely adopted an experimental approach (Martín-Gálvez et al. 2005) and never tested for potential confounding effects (e.g., Trnka 1995, Grim and Honza 1997, 2001). We believe that future studies would benefit from employing these improvements because they allow the detection of correlates and causes of parasite and host nestling fitness.

In summary, we attempted to control for many potential confounding factors (Appendix Tables 3 and 4). Based on these analyses, we suggest that the decreased mass and delayed fledging of cuckoo nestlings fed with some unusual diet items were not by-products of the timing of nesting within the breeding season, between-year variation, parasitism status (with cuckoo females targeting hosts that did not produce nestlings with better fledging parameters), or other confounding factors whose potential effects we statistically excluded (final models) or controlled for (full models). Fledging mass patterns were not confounded by the nestling age at which the fledging mass was measured; this is not surprising because we measured all nestlings shortly before fledging and always after they reached the growth asymptote.

Potential Fitness Consequences

Lower mass at fledging and delayed fledging suggest a dietary cost of some noninvertebrate diets (Figure 1), which may affect postfledging survival and recruitment and thus a brood parasite's fitness (Ladin et al. 2015). For example, Naef-Daenzer et al. (2001) found that Great Tit fledglings that died soon (<20 days) after fledging had ∼5% lower fledging mass than fledglings that survived. In our study, cuckoo nestlings that were fed plants, vertebrates, or both had lower fledging masses (by 20%, 20%, and 27%, respectively) than purely invertebrate-fed cuckoos. Fledging age was prolonged by 19% in cuckoos fed both fruits and vertebrates, but did not differ between the other diet categories.

These substantial values strongly suggest that the unusual fruit and lizard diet (the latter only in combination with the former) may indeed translate into decreased parasite fitness during the postfledging period. This view is further supported by quantitatively similar fledging mass decreases between plant- and vertebrate-fed solitary cuckoos and those sharing the nest with host redstart nestlings in a previous study in the same study site (Grim et al. 2009a): Cuckoos in mixed broods had 26% lower mass than solitary cuckoos (i.e. cuckoos raised alone) and their fledging age was prolonged by 15%. Cuckoos in mixed broods suffered drastic prefledging mortality, with fledging success decreased to 44% of that of solitary cuckoos (Grim et al. 2009a; see also Grim and Samaš 2016). In contrast, in the present study, both plant- and vertebrate-fed cuckoos all fledged (predation and weather effects excluded). However, poor body condition at fledging of cuckoos with unusual diets suggests that mortality may have taken its toll shortly after fledging. Indeed, preliminary data on postfledging dispersal and survival show that cuckoo nestlings from mixed broods always die within several days after fledging (Grim and Rutila in press).

Host nestlings fed lizards or fruits did not show statistically significantly lower fledging success, mass, or delayed fledging. This was despite the fact that redstart nestlings consumed such items at relative ages similar to those of cuckoos, namely in the last two thirds of the nestling period. This finding is roughly parallel to that of Yang et al. (2013), who reported that Verditer Flycatcher nestlings were able to digest insects with hard exoskeletons, whereas cuckoo nestlings fed with the same prey items grew poorly and even died long before potential fledging. The same applies to plant-provisioning hosts of Brown-headed Cowbirds (Rothstein 1976).

There was partial prefledging mortality in redstart host broods fed with invertebrates and plants, but not vertebrates. However, these differences were not statistically significant. The potential causes of this mortality in our redstart population include inclement weather conditions (e.g., O'Connor and Morgan 1982), partial predation (e.g., Thompson et al. 1999), hatching asynchrony (e.g., Slagsvold and Wiebe 2007), and a beak deformity that was observed in one of the dead nestlings.

Unusual Prey Items

Fruits have previously been reported in the redstart nestling diet (Glutz von Blotzheim et al. 2001) but only once in the diet fed by other hosts to cuckoos (Martín-Gálvez et al. 2005). Thus, one could expect the effect of unusual plant items to be larger for cuckoos than for host nestlings, that is, a large effect of novel prey to which they are not adapted on parasitic nestlings vs. little effect of the same prey on host nestlings who are at least partly adapted. Indeed, as predicted, cuckoo nestlings fed plants showed lower fledging masses (but not delayed fledging), but there was no such effect for host nestlings.

Lizards were by far the largest prey brought to nestlings by hosts, with an estimated body mass of several grams. In contrast, typical insects brought by hosts to cuckoos were several orders of magnitude smaller, with body masses at the scale of dozens of milligrams (Grim and Honza 1997). For example, the body mass of a single lizard (0.5–3.0 g) is equal to 13 to 75 caterpillars (1 fully grown caterpillar weighs 40 mg; Naef-Daenzer and Keller 1999). This comparison refers to caterpillars that are at their maximum size; redstarts, of course, also fed smaller, not fully grown caterpillars to nestlings. Thus, comparing simple number dominance of prey items (Table 1) massively underestimates the contribution of lizards to the diet. This is even more pronounced because caterpillars were by far the largest insect items that redstarts brought to their own or parasitic nestlings. Therefore, the mass and, by implication, the energy content of lizards are ∼2–3 orders of magnitude larger than their numerical dominance (Table 1) would suggest. In light of these differences, it can be reasonably expected that feeding lizards to cuckoo nestlings will affect them.

However, lizards might be too rare as food items to cause detrimental effects by themselves (Table 1). Indeed, the significant combined effects of plants and vertebrates in the diet, coupled with the nonsignificant effects of lizards alone, in the analyses of both fledging mass (Figure 1A) and fledging age (Figure 1B) suggest that lizards only exacerbate the negative effects of plants (which also show statistically significant negative covariation with cuckoo fledging mass when fed without lizards). Still, we note that the effect of lizards on fledging mass (but not on fledging age) is in the predicted direction and the lack of statistical significance may stem from the low number of nestlings that were fed only lizards (i.e. not also plants) in conjunction with invertebrates. In general, large prey items, such as lizards that are several centimeters long, may be problematic for small altricial nestlings to swallow (Turtumøygard and Slagsvold 2010) and may impair growth (Slagsvold and Wiebe 2007). Yet, observations of feeding events involving lizards did not suggest that either host ( Supplemental Material Video S2) or parasitic nestlings ( Supplemental Material Video S1) had any problem swallowing these food items. Therefore, we hypothesize that the decreased growth and delayed fledging of parasitic nestlings fed lizards in combination with plants might have been caused by variation in digestive efficiency between host and parasitic nestlings. Digestive efficiency has not been studied in the cuckoo and represents a potentially fruitful avenue for future research (see also Soler et al. 2014).

A literature review suggests that the range of the cuckoo nestling diet may be much broader than previously thought (T. Grim and P. Procházka personal observations). For example, anecdotal records report relatively large vertebrates in the diets of host Great Reed-Warblers, namely fish (common carp [Cyprinus carpio]; Mayer 1971, Mikulica et al. 2017) and amphibians (the European tree frog [Hyla arborea]; Trnka 1995, Mikulica et al. 2017). Our work further widens the range of these unconventional vertebrate dietary components.

Soler (2008) suggested that feeding parasitic nestlings with unusual prey items represents a subtle form of nestling discrimination. However, in this study, the numbers of unusual prey items fed to cuckoo and redstart nestlings were similar. Generally, we doubt that hosts are able evolve specific cuckoo nestling feeding habits as a form of nestling discrimination. First, nestling discrimination is typically represented by dramatically different host behavior toward alien offspring, in the form of starvation, desertion, pecking, or outright nestling removal (reviewed by Grim 2006a, 2011). Second, changing diet composition would provide a very inefficient antinestling defense (Yang et al. 2013). This is because shifting dietary habits would not ameliorate the majority of parasitism-related costs. First, eviction of the host's own progeny by the cuckoo hatchling typically occurs before the death of the cuckoo nestling due to an unsuitable diet (Grim et al. 2009b, 2011). Second, shifting dietary habits would be costly to hosts because it would affect most of the host's biology unrelated to parasitism (Yang et al. 2013). This would be especially so in the case of lizards, which show low availability simply because their body size is much larger than that of insect or spider prey. Lizards also surely require more effort to catch, handle, and transfer than typical invertebrate food items (“loading effect” sensu Carlson and Moreno 1982). Instead, we suggest that host diet selection may represent a general life history trait (sensu Grim et al. 2011) that either facilitates (Brooke and Davies 1989) or prevents (Yang et al. 2013) coevolution between parasites and (potential) hosts, even though it most likely did not evolve as a specific antiparasite defense.