Characteristics determining host suitability for a generalist parasite

Host quality is critical for parasites. The common cuckoo Cuculus canorus is a generalist avian brood parasite, but individual females show strong preference for a specific host species. Here, we use three extensive datasets to investigate different host characteristics determining cuckoo host selection at the species level: (i) 1871 population-specific parasitism rates collected across Europe; (ii) 14 K cases of parasitism in the United Kingdom; and (iii) 16 K cases of parasitism in Germany, with data collected during the period 1735–2013. We find highly consistent effects of the different host species traits across our three datasets: the cuckoo prefers passerine host species of intermediate size that breed in grass- or shrubland and that feed their nestlings with insects, and avoids species that nest in cavities. Based on these results, we construct a novel host suitability index for all passerine species breeding in Europe, and show that host species known to have a corresponding cuckoo host race (gens) rank among the most suitable hosts in Europe. The distribution of our suitability index shows that host species cannot be classified as suitable or not but rather range within a continuum of suitability.

insects, although not significantly so in the UK (Fig. 1c, Tables 1 and 2); (iv) species with either larger or smaller body size are used less than species with intermediate sizes (Fig. 1d, Tables 1 and 2); and (v) species with smaller population sizes have fewer parasitism events than species with larger populations both in the UK and Germany (Table 1). However, from the current analyses we are not able to tell if there is a deviation from what would be expected if hosts are being used at random as expected from population size. Nest height, nest depth and overlap in breeding period do not affect parasitism in any of the three datasets (Tables 1 and 2). We also note that results are qualitatively and quantitatively very similar if we include all populations regardless of number of nests or if we exclude all populations with less than ten nests from our analysis.
The distribution of our host suitability index calculated for all European passerine species does not show a clear bimodal separation between suitable and unsuitable host species, but rather a continuum from low to high suitability (electronic supplementary material, Table S1 and Fig. 2). However, all species with a recognized corresponding cuckoo gens are ranked towards the high suitability end of the index (Fig. 2). Furthermore, the host suitability index, which is based on the model of parasitism rates across Europe, shows a strong correlation with the number of parasitism events in both Germany and the UK (Fig. 3a,b).

Discussion
Species vary in their quality as hosts for parasites, as manifested through host-specific variation in parasite reproductive success 3,67 . Such variation in parasite success is also evident in avian brood-parasite systems 68 and brood parasites should selectively target hosts that maximize the probability of successful fledging of the parasitic chick. Since there is pronounced variation in utilization among potential hosts, selection by cuckoos is clearly not random 25,26,28 . We find highly similar effects of the different ecological host traits on cuckoo parasitism in our three independent datasets. According to our results, the cuckoo prefers host species of intermediate size that feed their nestlings with insects, and tends to avoid species that nests in cavities or breed in forest or rocky areas, but we find little effect of nest height, nest depth and breeding overlap. Importantly, there is no single variable explaining host use by cuckoos, but rather a combination of variables that together influence parasitism rates.
Host body size is clearly important for host selection by cuckoos. Intermediate-sized passerines are in general parasitized at higher rates than smaller or larger species. Use of the smallest passerines may be hampered by inefficient incubation of the parasitic egg and inadequate provisioning of the parasitic chick. The largest passerines may be avoided for the same reasons, and in addition large host nests, eggs and chicks may render it difficult for the cuckoo chick to evict potential competitors. However, nest cup depth was not an important predictor of cuckoo parasitism in our analyses, despite a deeper or steeper-sided nest tending to render eviction more difficult 62,69 . Experiments have shown that cuckoo chicks growing up together with host chicks suffer significantly lower probability of survival than when raised alone [69][70][71] , but see 72 . It is possible that the combined effect of larger eggs and nest steep-sidedness would render the largest passerines unsuitable as cuckoo hosts.
The cuckoo chick is dependent on invertebrate food, and seed eating species have therefore been considered unsuitable as cuckoo hosts e.g. 10 . Nevertheless, species like greenfinch Carduelis chloris and linnet C. cannabina rank among the 10 most commonly used hosts in UK, and there was no significant effect of nestling food on parasitism in UK (Table 1). Greenfinches have been able to raise cuckoo chicks 73 , but this observation alone is not sufficient to conclude that they are high quality hosts, as we do not know the condition of the fledgling cuckoos and hence the likelihood of recruitment of cuckoos raised by greenfinches. Furthermore, none out of 20 cuckoo eggs in linnet nests recorded in the BTO Nest Record Scheme resulted in successfully fledged cuckoo chicks 25 . In contrast to UK, our data disclose that few cuckoo eggs have been found in this abundant species in Germany. Hence, the most plausible explanation for the relatively high use of seed eaters in UK is "mislaid" eggs by cuckoos belonging to other tribes. The dunnock Prunella modularis, one of the favourite hosts in UK but less  Species nesting in cavities often have small entrance holes and deep nests 74 . The small entrance hole poses great problems for the female cuckoo attempting to successfully place her egg into the nest cup 74 and, even if she succeeds, her chick may grow too big to escape and become trapped inside. Chicks may also struggle to evict competing eggs and nestlings. Cavity nesters are therefore regarded as unsuitable hosts 35,71,75 , a prediction confirmed by our results.
Habitat showed a strong relationship with the number of parasitism events in all three datasets. Species breeding in shrubland and grassland were preferred by the cuckoo whereas species breeding in forests and rocky habitats were largely avoided. Wetland species were utilized relatively frequently in Germany, but not in the two other datasets. The cuckoo is dependent on high vantage points from where it can search for available host nests 58 , which may render species breeding in rocky areas unsuitable in most cases. Moreover, most of the species in UK and Germany breeding in rocky areas have small population sizes/densities, potentially making it more difficult for cuckoos to maintain a viable population. Many potential host species breeding in forest habitat are cavity breeders (like tits, Paridae) and of larger size (like thrushes, Turdidae), which may explain the relatively less use of forest breeding species than those breeding in other habitats. Wetland breeding species are apparently more used in Germany than in UK, which seems to be due to utilization of great reed warblers Acrocephalus arundinaceus in Germany, a species that is absent from UK.
Cuckoos are dependent on the ability to synchronise the timing of their breeding with that of their hosts. However, our analyses did not provide a significant effect of overlap in breeding season between the potential host species and the cuckoo. The reason for this could simply be that most passerine species overlap in duration of breeding season, and those with a small overlap in breeding season are generally large species or cavity breeders and therefore not suitable anyway.
Nest height above ground was also not a significant predictor of cuckoo parasitism in our analyses. This is contrary to the findings of Martín-Vivaldi et al. 76 , who suggested that cuckoos have difficulties finding host nests on the ground. They found lower egg rejection in ground-nesting passerines and hence concluded that ground-nesters are rarely used by cuckoos. Several cuckoo gentes, however, utilize hosts breeding on the ground. In the UK for instance, ground-nesting meadow pipits Anthus pratensis are among the most common hosts 41 . Moksnes and Røskaft 9 mention Anthus, yellow wagtail, white wagtail, blue and Emberiza cuckoo egg morphs being found in ground-nesters. Previous within-species analyses are also in line with our findings; nest height was not a predictor of parasitism in marsh Acrocephalus palustris and reed warblers A. scirpaceus 24,77 . There is considerable interspecific variation in the ability of hosts to recognize and reject cuckoo eggs e.g. 35,37 , which may influence our estimation of parasitism rates. In host species with well-developed egg rejection abilities, a poor mimic may be removed before its presence could be detected in the nest. This plausible scenario may lead to an underestimation of relative parasitism in such species compared to hosts that are poor rejecters. The data available in the present study, except those eggs stored at museums, do not allow us to assess egg mimicry since there is no description of egg appearance of either host or parasite in most sources. Variation in egg rejection among species may therefore blur the apparent suitability of different species over time. This could potentially make it harder for us to detect the factors important for host suitability, but is unlikely to contribute to false positive effects in our analyses. While some suitable hosts may not have been identified and hence misclassified in our suitability index, there is no reason to suspect an equivalent bias towards detection of parasitism in unsuitable hosts, because rejection behaviour is likely to have been selected for due to historic parasitism.
Our host suitability index based on population-specific parasitism rates correlated well with the number of parasitism events both in Germany and UK. In both countries, we also observed that few species were used more than would be expected by their suitability. On the other hand, quite a few species are being used less than predicted purely by the host suitability index. Although one should always be careful in the interpretation of variables based on estimates from statistical analyses, this bias suggests that the factors we have investigated may together act to modulate the suitability of species as cuckoo hosts. There may also be other limitations that we have not been able to detect with the current dataset, however, such as local variation in population sizes and rejection ability. Despite these possible caveats, the strong correlation between the host suitability index and the number of parasitism events suggest that it is useful as a species level index of host suitability among European passerines. When we then look at the distribution of this index among species (Fig. 2), it becomes clear that it would be too simplistic to regard species being either suitable or not as hosts for the cuckoo, but rather that the various species show various degrees of suitability. The host species with a corresponding cuckoo gens (classified based on egg mimicry) are all, as expected, placed among the most suitable hosts. According to the index there are a fair number of additional species that appear to be suitable for parasitism, but apparently without any gens attached to them. There are several possible explanations for this pattern. Firstly, some of these species only have very small population sizes in Europe, rendering them unsuitable as cuckoo hosts here but not necessarily in areas where they are more abundant. Little buntings Emberiza pusilla and Blyth's reed warblers Acrocephalus dumetorum, for instance, are regularly parasitized in parts of Russia by cuckoos laying mimetic eggs 78,79 . Secondly, host use in some areas of Europe is poorly known, especially the southern and eastern parts. Hence, gentes that are still unknown to us may exist, such as cuckoos targeting those Sylvia warblers with a southern distribution (e.g. 80,81 ). Thirdly, some cuckoo gentes, e.g. the dunnock gens, do not mimic the eggs of their hosts 10 . Hence, a classification based on egg appearance alone would result in missing some of the existing cuckoo gentes (e.g. 82 ). Finally, as stated above, even though we have included many factors of importance for cuckoo host selection in constructing the host suitability index, there may still be others.
In many systems with generalist parasites, like ecto-or endo-parasites, the parasite is limited by dispersal between species. This is clearly not the case for the cuckoo and most other brood parasites. The bitterling Rohdeus sericeus is a parasitic fish that shares many of the same attributes as avian brood parasites and a similar pattern emerges for their host use. Investigations of four of their potential mussel hosts (Anodonta anatine, A. cygnea, Unio pictorum and U. tumidus) reveal differential suitability of the different host species, with the most suitable host offering twice as high survival for embryos as compared to the least suitable species of the four, and the two other species offer intermediate survival probabilities 3 . Furthermore, the bitterling prefers the four different hosts in the exact same order as their suitability 4,83 . The brown-headed cowbird Molothrus ater is, like the common cuckoo, a generalist brood parasite. In a study of nests of 34 potential host species, 18 were parasitized by cowbirds and a large range of parasitism frequencies were observed 84 . This suggests that even though a range of hosts can be used by generalist brood parasites, they are used in different frequencies according to factors that affect their suitability. In general, the nature of suitability indices will depend on parasite requirements. In the present analyses, we have selected host characteristics that have been hypothesized to explain variation in host use by the common cuckoo. For other parasites, there may be additional host traits that could be of importance in this sense (e.g. intraspecific variation in size and morphology and interspecific variation in coloniality 85 ), and obviously the suitability index would also depend on the level of host specialization of the parasite.
In this study, relying on three novel large datasets, we have disclosed characteristics of potential hosts that may be important for cuckoo host selection at the species level. Host body size, nest structure, habitat and food type for the chicks are all important predictors of cuckoo parasitism, either independently or in combination. The same set of predictors explained variation in host use both in Germany and the UK, even though the actual species used varied somewhat between the two countries. Our findings offer a basis for more thorough analyses of temporal and spatial variation in cuckoo host use. We have shown that the relative importance of a suite of host characteristics on parasite utilization can be modelled statistically by using data from a subset of hosts in specific geographical areas. The outputs from such exercises can then be used to construct host suitability indices on a larger geographical scale for a larger set of species with unknown status as hosts, but where data on life history traits can be retrieved. Our results also demonstrate that potential cuckoo hosts should not distinctly be considered suitable or unsuitable, but rather be placed on a suitability continuum, with the majority of species located towards the more suitable end. We may predict similar patterns in other generalist parasites: many host traits going into their suitability index and similar distributions for the suitability of potential host species. More generally, such suitability indices may be valuable for predicting the potential for host use (current and future) in a whole range of host-parasite systems. This may be increasingly important for understanding species interactions in a world where both parasites and their potential hosts may have to shift their ranges due to climate change or human induced alterations of landscapes. Ethics. The data used in this study were entirely retrieved from the literature, museum collection, databases, etc.

Material and Methods
Population-specific parasitism rates and cases of parasitism (cuckoo egg or chick) were obtained through various literature search, resulting in data ranging from the period 1735-2013 with the majority of cases from 1850 onwards, originating from more than 7,000 publications meticulously browsed by BGS (ISI Web of Science and Biodiversity Heritage Library, Google Scholar and the Natural History Museum library in Tring, UK, communication with British and German ornithologists, ringing and nest record schemes, museum egg collections and unpublished notes or reports stored in libraries and museums).
Firstly, we investigated 1871 population-specific parasitism rates from 139 passerine species, collected across Europe (https://doi.org/10.5061/dryad.9r0n681). We only included parasitism rates based on a minimum of five nests (including parasitized and non-parasitized nests). Although we find five nests to be an appropriate cut-off for the number of nests needed to qualify as a population in these analyses, the number is not based on previous knowledge. We have therefore also undertaken the analyses with (1) all data included regardless of sample size, and (2) populations with ten or more nests included.
Secondly, we investigated 16,515 cases of parasitism from 100 passerine species in Germany and 14,507 cases of parasitism from 78 passerine species in UK (https://doi.org/10.5061/dryad.9r0n681). One potential bias using these data is that parasitism rates are generally overestimated because populations that are likely to be parasitized are also more likely to be investigated for parasitism. However, our main question does not relate to actual parasitism rates, but rather to how host life-history traits affect relative parasitism rates, and we have no reason to believe that parasitism rates are overestimated relatively more for species with specific ecological characteristics.
The dataset on population-specific parasitism rates contains 2696 cases from UK and 2660 cases from Germany that are also included in the "cases of parasitism" datasets from UK and Germany. On the other hand, the dataset on population-specific parasitism rates includes additional cases from UK and Germany, where parasitism rates were reported to be zero (these are of course not included in the "cases of parasitism" datasets).
We selected the following variables as predictors of variation in parasitism between species: (1) Nest cup depth: Inner height of nest cup from bottom to rim (cm) 86 ; ( Using the dataset on population-specific parasitism rates in Europe, we ran a binomial generalised linear mixed effects model with counts of parasitized and unparasitised nests as response variable, using the glmer function in the lme4 package 88 in R 89 . In this mixed model, we included species as a random effect. We have chosen not to include any phylogenetic effects in our analysis, because we assume that cuckoo host preferences in Europe were established after most passerine species evolved, and, therefore, do not expect cuckoo parasitism rates to be affected by the phylogeny of the species, but only by their actual trait values. Closely related species may have similar parasitism rates, but we believe that is due to similarities in their ecology rather in their phylogenetic history. We analysed predictor variables 1-7 listed above and additionally included the square of bird size to allow for a non-linear relationship, as we expected that species could potentially be both too small and too large to be suitable as hosts for the cuckoo. Generally, fixed factors were not markedly correlated, but nest height and habitat type grassland showed a correlation of 0.38, while body size and domed nests showed a correlation of 0.31. All other correlations had absolute values below 0.3. Data from the UK and Germany were analysed separately. In each country we ran a hurdle regression model using the pscl package 90,91 in R. Hurdle models are well suited to handle datasets with excess zeros, such as ours, since we have many records of host species with no parasitism. In Hurdle models, two different components are estimated: (i) a truncated count component and (ii) a hurdle component. The latter component estimates the zero vs. larger counts as a binomial process, while the former component excludes the zeros and models all the positive counts of parasitism. In our case a negative binomial distribution fitted the first component best. All predictor variables, 1-7 listed above, were part of both the binomial and the negative binomial components of the model and the model structure was the same in both cases.
In these hurdle models for UK and Germany, each passerine species breeding in the given country was included as a data point in our model. In each case, our response variable was the number of parasitized nests recorded. In these datasets we do not know the number of non-parasitized nests in the populations where these parasitism events were recorded. Therefore, we also included population size of each species in the given country (variable 8 listed above) to account for the uneven availability of the different species. All continuous predictor variables were log-transformed except overlap in breeding period. To create figures illustrating the significant effects, new models were run without non-significant predictor variables. To determine the model predictions from the hurdle models for UK and Germany, we used the predict function in R. We used the response predictions and kept all other variables constant whilst varying the focus variable within its observed range. For graphical purposes, these estimates were then scaled so that the highest estimate was the same as the highest estimate for predicted parasitism rates to allow comparison of patterns between the otherwise incomparable model estimates (rates vs numbers).
To determine the model predictions for the parasitism rates across Europe, we used the sim function in the arm package 92 in R to simulate the posterior distributions of the parameters in our fitted model. From these distributions, average effect sizes and credibility intervals for different parameter combinations were calculated and used to make the figures. We chose the parasitism rates as reference values in our figures because these have an intuitive biological meaning while the predictions from the hurdle models provide a less obvious meaning biologically.
Next, we used the model predictions from the binomial model of parasitism rates as a species-specific index and a measure of suitability of each species as a cuckoo host. This index was scaled so that the most suitable species had an index of 1 and the least suitable species have an index of 0. Furthermore, we report the distribution of this host suitability index calculated for all European passerines (electronic supplementary material, Table S1). By separating between species where a corresponding cuckoo gens has been described based on egg characters 8-10,23,24 , we disclose how well our index relates to the number of parasitism cases found in the independent datasets for Germany and UK, and hence validates that these numbers do reflect parasitism rates.