The decline of the Turtle Dove: Dietary associations with body condition and competition with other columbids analysed using high‐throughput sequencing

Dietary changes linked to the availability of anthropogenic food resources can have complex implications for species and ecosystems, especially when species are in decline. Here, we use recently developed primers targeting the ITS2 region of plants to characterize diet from faecal samples of four UK columbids, with particular focus on the European turtle dove (Streptopelia turtur), a rapidly declining obligate granivore. We examine dietary overlap between species (potential competition), associations with body condition in turtle doves and spatiotemporal variation in diet. We identified 143 taxonomic units, of which we classified 55% to species, another 34% to genus and the remaining 11% to family. We found significant dietary overlap between all columbid species, with the highest between turtle doves and stock doves (Columba oenas), then between turtle doves and woodpigeons (Columba palumbus). The lowest overlap was between woodpigeons and collared doves (Streptopelia decaocto). We show considerable change in columbid diets compared to previous studies, probably reflecting opportunistic foraging behaviour by columbids within a highly anthropogenically modified landscape, although our data for nonturtle doves should be considered preliminary. Nestling turtle doves in better condition had a higher dietary proportion of taxonomic units from natural arable plant species and a lower proportion of taxonomic units from anthropogenic food resources such as garden bird seed mixes and brassicas. This suggests that breeding ground conservation strategies for turtle doves should include provision of anthropogenic seeds for adults early in the breeding season, coupled with habitat rich in accessible seeds from arable plants once chicks have hatched.

The European turtle dove (hereon referred to as turtle dove) is the UK's and one of Europe's fastest declining breeding bird species (Hayhow et al., 2017;PECBMS 2015). It is classified as a farmland specialist in the UK, although elsewhere it is also associated with open woodlands and forest borders (e.g., Bakaloudis, Vlachos, Chatzinikos, Bontzorlos, & Papakosta, 2009;Dias et al., 2013). Turtle doves and stock doves feed only on seeds (Browne & Aebischer, 2003;Murton, Westwood, & Isaacson, 1964), whereas other columbids will also take leaves and other plant matter (Murton et al., 1964;Wilson, Morris, Arroyo, Clark, & Bradbury, 1999). Previous microscopic analysis of faecal samples has shown that the diet of the turtle dove changed from mainly noncultivated (natural) arable plants in the 1960s (Murton et al., 1964) to mainly cultivated food resources (mostly wheat [Triticum aestivum] and oilseed rape [Brassica napus]) in the 1990s (Browne & Aebischer, 2003). The turtle dove diet switch occurred concurrently with decreases in the abundance of many natural arable plants (Storkey, Meyer, Still, & Leuschner, 2012), along with a decrease in reproductive effort and a rapid population decline (Browne & Aebischer, 2004). It is postulated that this dietary switch may be associated with a reduction in food availability during key periods of the breeding season when seeding natural arable plants have become scarce as a result of agricultural change (Browne & Aebischer, 2004). For example, increases in autumn-sown crops, with associated fertilizer and herbicide applications and a consequent reduction in the area of overwinter fallow, have adversely affected populations of natural arable plants that persist overwinter in fallow land or germinate after spring tillage, thus reducing the availability of accessible seed for breeding birds (Smart, Firbank, Bunce, & Watkins, 2000). There is also uncertainty about the dietary quality for turtle doves of the anthropogenic foods that have largely replaced natural arable plant seeds (Pruitt, Hewitt, Silvy, & Benn, 2008).
Recent developments in genetic analysis of diet have led to the possibility of using molecular barcodes amplified from faecal DNA and analysed using high-throughput sequencing (HTS), a method with higher resolution and improved accuracy when compared to traditional microscopic methods (Ando et al., 2013;Galimberti et al., 2016). Standard barcode analyses of plant species use parts of the rbcL and matK genes, which can provide species-level discrimination of 75% when combined (de Vere et al., 2012). However, limitations on amplicon length in HTS (current maximum of 2 × 300 base pair reads on Illumina Miseq;Illumina 2016), as well as the need to design primers that will amplify shorter barcodes to detect degraded DNA in faecal samples (Ando et al., 2013;King, Read, Traugott, & Symondson, 2008;Pompanon et al., 2012), have meant in practice that these gene regions provide limited discriminatory powers for analysis of faecal samples from herbivores (Pompanon et al., 2012).
The ITS2 nuclear gene has been proposed as a target for the design of short-length barcodes suitable for dietary analysis (Bradley et al., 2007) with a high species-level discrimination for identifying medicinal plants (92.7%; Chen et al., 2010) and herbivorous insect gut contents (61.6% for the Zingiberales order; García-Robledo, Erickson, Staines, Erwin, & Kress, 2013), suggesting ITS2 may have higher resolution than more widely used short-length barcodes (Hollingsworth, Graham, & Little, 2011). A major criticism of ITS2 is the lack of reference sequences available for this region (Hollingsworth et al., 2011); however, the latest update to the ITS2 database has doubled the number of reference sequences available to 711,172, of which 208,822 belong to the Chloroplastida (Ankenbrand, Keller, Wolf, Schultz, & Förster, 2015 Gann et al., 2018). This is considerably higher than either trnL or rbcL short-amplicon primers (which identify 34% and 42% of plant sequences, respectively, to genus level; Pompanon et al., 2012) and avoids the need to use multiple gene targets to maximize identification. In practice, in vitro tests of 202 UK and tropical plant species showed that 99% were amplified by the Moorhouse-Gann et al.
Here, our aim was to apply HTS to identify dietary components from columbid faecal samples and test three hypotheses: 1. Turtle dove diet currently shows strong overlap with that of other UK columbids, suggesting competition for limited food resources.
2. Anthropogenic food resources, such as cultivated crops and artificially provided food for songbirds at bird tables, are associated with poorer condition in both adult and nestling turtle doves.
3. Turtle dove diet shows both inter-and intra-annual variation, with anthropogenic food resources more important early in the turtle dove breeding season. Adult columbids were caught using whoosh and mist nets (Redfern & Clark, 2001) at temporarily baited sites in areas either where birds had previously been seen feeding, or where farmers provided grain, during May, June and July 2011-2014. Thus, we expected a small amount of mixed seed to be present in faecal samples of adult columbids if they were regularly using baited sites. When caught, birds were weighed and maximum wing chord measured (Redfern & Clark, 2001). Adult turtle doves were fitted with tail-mounted Pip3 radio-tags (Biotrack, Dorset, UK) weighing 1.7 g (<1.5% of body mass), to help in locating nests. All adults were caught prior to them having chicks in the nest, ensuring we were identifying components of adult diet, rather than seeds collected for regurgitation to nestlings. As well as adult turtle doves (n = 26), we also collected faecal samples from adult collared doves (n = 6) and stock doves (n = 12).

| Sites and field collection
Faecal samples were collected either directly from the bird or from the inside of clean bird bags within which the birds were temporarily held after capture. All faecal samples were frozen at −20°C as soon as possible after collection (1-8 hrs) until subsequent analysis.
Nests were located by monitoring the movements of radiotagged turtle doves and by cold-searching suitable habitat for all columbid species. Nests were checked every 2 days, and when nestlings were seven (turtle dove n = 66 and collared dove n = 5) or 10-14 days old (stock dove n = 3 and woodpigeon n = 22), they were ringed, weighed and faecal samples collected. Different sampling ages were due to different nestling growth rates between species (Robertson, 1988), precluding the sampling of turtle doves later than 7 days old when they were capable of leaving the nest prematurely.
At this age, nestlings are fed seeds and not crop milk (confirmed by F I G U R E 1 Locations of study sites from where faecal samples were collected. Sites where only nonturtle dove faecal samples were collected are shown as black dots, although turtle doves were also present at these sites; red dots denote sites from which turtle dove faecal samples were collected in addition to those of other columbids. Further site and faecal sample collection details are provided in Appendix 1. Contains Ordnance Survey data © Crown copyright and database right 2017 [Colour figure can be viewed at wileyonlinelibrary.com] examining the crop contents of three nestlings found dead under their nests at 3-5 days old; J. Dunn, personal observation). Multiple faecal samples from nestmates were processed separately and data subsequently pooled for statistical analyses. Faecal samples from nestlings were collected between June and September, 2011-2014.

| Construction of a DNA barcode reference library
Seeds were collected in the field from 24 plant species, supplemented by seeds from nine species known to be commonly present within commercial seed mixes (Appendix 2). We downloaded sequences from an additional 19 species from GenBank to ensure that all species previously recorded in turtle dove diet (Browne & Aebischer, 2003;Murton et al., 1964), as well as other plant species commonly found at our field sites, were included in the barcode library (Appendix 2; Moorhouse-Gann et al., 2018). We extracted DNA from all species using a standard salting-out protocol (Randall, Sornay, Dewitte, & Murray, 2015) and confirmed in vitro that our

| Identification of plant species
Our Illumina run resulted in 12,592,989 paired-end reads, which were filtered for quality using Trimmomatic v0.32 (Bolger, Lohse, & Usadel, 2014) with a minimum quality score of 20 over a sliding window of 4 bp, retaining sequences with a minimum length of 135 bp resulting in 10,138,058 sequences. These were aligned using FLASH (Magoč & Salzberg, 2011), resulting in 9,921,248 aligned sequences.
These were demultiplexed into faecal sample-specific files using the MID tag sequence with the "trim_seqs" command in Mothur (Schloss et al., 2009), which also removes the MID and primer sequences from the reads. After eliminating reads without an exact match to primer sequences and MID tags, 6,105,478 sequences remained (mean ± SE for samples: 42,917 ± 2,871; for negatives and unused tag combinations: 1,930 ± 382). We then used the "derep_fulllength" and "uchime2_denovo" commands in the USEARCH software v9.2.64 (Edgar, 2010) to remove any sequences with fewer than 10 copies within a faecal sample and any potential chimeric sequences, resulting in 12,608 unique sequences. Analysis of species discrimination at the ITS2 region (Moorhouse-Gann et al., 2018) suggests this region to be unsuitable for an approach of clustering similar sequences into molecular operational taxonomic units (MOTUs) due to the loss of ability to distinguish between species prior to the grouping of multiple polymorphisms within some plant species.
Therefore, we adopted a closest matching sequence approach to identify species within our samples (e.g., Hawkins et al., 2015).
We took a sequence read-number approach to deal with any background contamination. First, we examined sequences found only in samples with unused MID combinations (n = 20) as these could only be attributed to background contaminants or "tag jumping" (Kircher, Sawyer, & Meyer, 2012;Schnell, Bohmann, & Gilbert, 2015). The highest number of reads for any of these sequences was 139, so we re-ran our initial dereplication step (using "derep_fulllength" in USEARCH) with this new sequence read threshold. This resulted in 1,192 unique sequences, which we then assigned to taxonomic unit using the BLAST algorithm (Altschul et al., 1997) to search GenBank, combined with new sequences from our barcode library (GenBank Accession nos KT948614-KT948638). If a sequence had the smallest e-value matching only one species on GenBank, with >99% sequence identity, we assigned the sequence to that species (Hawkins et al., 2015). If the sequence matched more than one species from the same genus, tribe or family, we assigned the sequence to the lowest common taxonomic unit up to the family level. Any sequence with <90% match to the closest matching species on GenBank, or for which BLAST returned no significant match (n = 80), was discarded, as was any sequence for which the closest match included a bacterium or fungus (n = 64). Next, to deal with any specific contaminants within our samples, we examined each unique sequence found in a negative sample, including unused MID combinations, PCR negatives (n = 2) and extraction negatives (n = 6).
For each sequence, we identified the highest read number within a negative sample and removed this sequence from any sample where the read number was below this threshold (detailed in Appendix 3).
Five sequences had their highest read numbers in negative samples (n = 5; Appendix 3) and were thus discarded. Finally, we combined our 1,043 remaining sequences within each of 143 taxonomic units.
We briefly discuss the possible effects of faecal or plant inhibitors and secondary predation in the Supporting Information.
Where we had multiple faecal samples from two nestlings within the same nest (no nest contained more than two nestlings), we combined these into sampling units for subsequent analysis.

| Statistical analysis
For dietary overlap analyses and subsequent statistical analyses, we used the presence or absence of each taxonomic unit in each sampling unit. For morphometric analysis of nestlings at the level of the sampling unit, we averaged data from both nestlings to avoid pseudoreplication due to nonindependence of nestmates. All statistical analyses were carried out in R version 3.1.2 "Pumpkin Helmet" for Mac (R Core Team 2016) unless otherwise stated.

| Dietary breadth and overlap between columbid species
To determine whether species showed differences in the number of taxonomic units in their diet, we constructed a generalized linear model using the number of taxonomic units per sampling unit as the response variable and the columbid species as a fixed factor, allowing for a Poisson distribution corrected for overdispersion. We tested the significance of the species term by comparison of this model with a null model using likelihood ratio tests.
To calculate dietary overlap of each species pair at the taxonomic unit level, we calculated Pianka's measure of overlap (Pianka, 1986) in EcoSimR (Gotelli & Ellison, 2013) using the equation: where O jk is Pianka's measure of overlap between species j and species k, p ij is the proportion of total resources that resource i is for species j, and p ik is the proportion of total resources that resource i is for species k. O ranges from 0, where two species have no resources in common, to 1, where there is complete overlap in resource use. To portray dietary overlap between species, we constructed bipartite food webs using the BIPARTITE package (Dormann, Gruber, & Fruend, 2008).
Finally we assessed the diets of different columbid species at the level of both the taxonomic unit and the plant family. For each taxonomic unit (n = 129) or plant family (n = 34) where the taxonomic unit or family was found in the diet of more than one columbid species (taxonomic unit: n = 52; family: n = 19), we ran a binomial GLM corrected for overdispersion, comparing the proportions of diets from each family (calculated as the proportion of individuals within each columbid species whose diet contains each taxonomic unit and plant family separately), carrying out Tukey HSD post hoc tests to identify differences between turtle doves and other columbids.
As our sample sizes for nonturtle dove columbids is relatively small, we carried out rarefaction analysis using the package VEGAN (Oksanen et al., 2016) to estimate the proportion of total taxonomic units in the diet of each species that we are likely to have detected.
For our larger turtle dove sample, we created four subsets of our data, each with n = 13 and carried out rarefaction analysis on each subset separately to confirm differences in estimated numbers of taxonomic units between species.

| Associations between diet and condition in turtle doves
To identify whether relative proportions of taxonomic units in diet were associated with condition in adult or nestling turtle doves, we categorized dietary components into four broad categories according to likely source (detailed in Table 1): "fed" (eight taxonomic units) contained seeds likely to be found in the vicinity of bird tables and supplementary food sources such as game bird feeders or grain tailings; "cultivated" crop plants as well as those widely cultivated as components of seed mixes sown to provide seed for game or wild birds within our study area (16 taxonomic units; excluding wheat, as this was widely available as supplementary food at our study sites); "natural" contained any wild plant species (109 taxonomic units).
We considered "brassica" (Brassicaceae; 11 taxonomic units) as a separate category as this plant family forms components of provisioned bird seed as well as being widely cultivated within our study area and also contains several naturally occurring wild species.
We used residuals from a linear regression of mean nestling body mass on mean nestling tarsus length at 7 days old to give an index of mean nestling condition within each nest whilst controlling for the nonindependence of nestmates (Labocha & Hayes, 2012). We used tarsus length because wing length is not easily measured on nestlings with limited primary feather growth. To obtain an index of adult condition at capture, we used residuals from a linear regression of body mass on wing length (Labocha & Hayes, 2012). We then used the DIRICHLETREG package (Maier, 2015) to carry out Dirichlet regressions for compositional diet data (Sánchez & Dos Santos, 2015) to identify how the relative proportions of taxonomic units within each dietary category are associated with adult and nestling turtle dove condition separately.

| Temporal variation in turtle dove diet
We carried out analyses of temporal variation in dietary importance   Denotes a family found exclusively in turtle dove diet. b Differences not tested statistically as the plant family was only found within one columbid species or in fewer than three individuals.
to small sample sizes). To determine the importance of each term within the model, we removed each term in turn and compared the fit of the model with and without each term using chi-squared tests.
We retained all terms in the final model from which we made predictions, to control for our unbalanced sampling design as not all sites were sampled in all years (Appendix 1). We then used Tukey HSD post hoc tests to identify where factor levels differed from each other.
We had data from nine nests where we also have data from one (n = 8 nests, n = 6 adults) or both (n = 1 nest, n = 2 adults) of the adults at the nest. However, all adults were caught a minimum of 27 days before their respective nestlings were sampled (mean ± SE: 45.8 ± 14.3 days). As there were temporal differences between adult and nestling samples, and between sequential nesting attempts from the same adult (n = 2 adults, two nesting attempts each), we treated these as independent data points for the purposes of the spatiotemporal analysis models described above as we had insufficient nonindependent samples to allow a mixed-effects model (including a "Family" term) to converge. However, to examine whether related adults and nestlings have more similar diets than unrelated adults and nestlings, we examined a subset of our data involving adults for whom we also had nestling samples and sampling units from sequential nesting attempts by the same adult where we did not have an adult faecal sample. We tested the effect of "Family" on the proportion of each dietary component category, as defined above, using a GLM with quasi-binomial error structure to allow for underdispersed proportion data.

| Diet composition and overlap between columbid species
We identified 55% of sequences to species ( (Table 2).
All taxonomic units were assigned to one of 34 plant families, and we examined differences in the mean proportion of diet comprised of each plant family between columbid species. Thirty-one families were found in turtle dove diet, of which 13 families were found exclusively in turtle dove diet ( Table 1). None of these families constituted more than 1% of taxonomic units in turtle dove diets.
We examined the proportion of diets from each columbid species that contained each family, and each taxonomic unit ( Epilobium sp.

| Dietary associations with turtle dove body condition
We found significant associations between diet composition and both adult and nestling turtle dove body condition ( Table 3). The proportion of fed taxonomic units in nestling diet was negatively associated with condition, with the diet of nestlings in the best condition containing half the proportion of fed items than those in the poorest condition (Table 3a; Figure 3a). On the contrary, the diets of nestlings in better condition contained a higher proportion of natural taxonomic units and a slightly (but significantly) lower proportion of brassicas (Table 3a; Figure 3a). (Table 3b; Figure 3b).

Adults in better condition had a higher proportion of both brassicas and cultivated taxonomic units in their diet
An increase in the proportion of fed taxonomic units was also associated with a marginally significant increase in adult condition (Table 3b; Figure 3b).

| Spatiotemporal variation in turtle dove diet
We found no evidence for differences in diet composition between adult and nestling turtle doves or between sites ( Table 4)

| Dietary overlap and composition in UK columbids
The high dietary overlap between all four columbid species suggests shared resources are important, although we also found significant differences in dietary composition. In contrast to the rapidly  (Murton et al., 1964). Competition between turtle doves and the recently colonized collared dove has been speculated as contributing to the turtle F I G U R E 3 Associations between diet composition (in terms of proportion of taxonomic units present) and condition for (a) nestling (n = 26 nests) and (b) adult (n = 26) turtle doves. Nestling condition indices are residuals from a linear regression of mean nestling body mass on mean nestling tarsus length at 7 days old for each nest, and adult condition indices are residuals from a linear regression of body mass on wing length at capture. Solid lines show trends significant at p < 0.05; dotted lines show marginally significant trends (p < 0.1). Statistical details are provided in the legend to Table 3 dove population decline (Rocha & Hidalgo De Trucios, 2000), but our data do not support this suggestion as collared doves showed the least overlap with all three other columbid species. Previous dietary studies have shown woodpigeons utilize green vegetation (as opposed to seeds alone; Murton, 1966;Ó hUallachain & Dunne, 2013) and can specialize on Brassicaceae crops when widely available (Inglis, Isaacson, Smith, Haynes, & Thearle, 1997). However, as this study shows relatively high dietary overlap between columbids, it is possible that different species may be feeding on different parts of the same plant species.
The concept of dietary competition relies on the assumption that shared food resources are limiting when in fact, species may be taking advantage of patchy but abundant resources (e.g., Pérez & Bulla, 2000), or using different foraging habitats (e.g., Emrich, Clare, Symondson, Koenig, & Fenton, 2014). Within our system, however, competition for seeds from limited and declining populations (Potts, Ewald, & Aebischer, 2010) of noncultivated plants remains likely (Browne & Aebischer, 2003). Here, it is important to look at diet as a whole, rather than examining the presence of individual taxonomic units or species groups: a single species may be present in a range of foraging situations or habitats, and taking diet as a whole (as we have done with our categorization of dietary components for turtle dove-specific analyses) may provide greater insight into foraging habitats. For example, during the breeding season, wheat or brassica seeds may be provided as a component of bird seed mixes in gardens or through supplementary feeding of songbirds or game birds.
Wheat and brassica seeds may also be found as a consequence of grain spillages during harvest or transportation. Wheat and brassica leaves may be taken year-round from growing crops, and, as crops ripen, fallen seeds may be acquired from the ground (or in situ from the standing crop-although turtle doves rarely use this method of foraging). All these sources would result in the same presence of wheat and brassica taxonomic units in faecal samples, but the source would have very different ecological implications in terms of resource availability and dietary competition.
We found a wide range of seeds in columbid diet that is likely to have originated from seed mixes provided for wild birds in gardens or on farmland. Whilst our more sensitive methodology might be able to detect and discriminate between a wider range of species than microscopic methods used by previous studies (Ando et al., 2013;Galimberti et al., 2016), seeds such as niger and hemp have a distinctive husk that should be readily detectable through microscopic analysis of faecal samples. Seed components such as hemp, niger and sorghum have not previously been recorded in turtle dove diet in the UK (Browne & Aebischer, 2003;Cramp & Perrins, 1994;Murton et al., 1964), but our findings concur with an increase in the feeding of birds with seed mixes that include these species, and  2007) but seed-rich habitat created within these schemes is usually aimed at providing forage for wintering birds (Henderson, Vickery, & Carter, 2004) or nectar for pollinating insects (Carvell, Meek, Pywell, Goulson, & Nowakowski, 2007) and often creates too dense a sward to be accessible by foraging doves in the breeding season . Despite this reduction in overall abundance of arable weeds (Potts et al., 2010), we found several species present within columbid diet, most notably within turtle and stock doves. Among the annual arable weeds commonly present in the diet of turtle doves (and other columbids), scarlet pimpernel and common chickweed are widespread but declining species on regularly tilled arable land within the UK and across Europe (Andreasen, Stryhn, & Streibig, 1996;Critchley et al., 2004;Fried, Petit, Dessaint, & Reboud, 2009;Sutcliffe & Kay, 2000;Walker et al., 2007 Overall, it appears that all four columbid species use similar foraging habitats although turtle doves have the greatest dietary range (as suggested by the results of our rarefaction analyses) and forage within a wider range of semi-natural habitats than their heterospecifics, but are more constrained by their inability to exploit green matter and in situ seed from tall vegetation. All four species eat anthropogenically fed seed probably sourced from gardens and farmyards: In the same way, high levels of dietary overlap were found in four co-existing columbid species in Venezuela, where Pérez and Bulla (2000) concluded that these closely related doves foraged opportunistically but randomly from the same available seed pool.
The same may occur within our system, especially early in the summer before natural seed resources become widely available: We do not know the degree to which dietary overlap is driven by food availability, and our data allow only limited insight into temporal variation in diet.

| Associations between diet and condition, and spatiotemporal variation in diet
We predicted that the consumption of anthropogenic food resources such as cultivated crops, and food provided for game and songbirds, would be associated with poor condition in both adult and nestling turtle doves, which have evolved to exploit other types of seed. This hypothesis was supported in nestlings by a negative association between the proportion of fed and brassica taxonomic units and body condition, and a positive effect of natural taxonomic units.
Contrary to our predictions, adult condition was positively associated with brassica and cultivated taxonomic units; anthropogenically fed taxonomic units showed a marginally significant positive association.
Given the higher calorific value of seeds such as hemp and sunflower (Hullar, Meleg, Fekete, & Romvar, 1999), this may be a beneficial side effect of a forced change in foraging ecology resulting from the background decline in availability of alternative, natural, food sources. However, any potential benefits of provisioned seed need to be balanced with potential negative impacts (e.g., increased risk of predation or parasite transmission) where high densities of DUNN ET AL.
We found no evidence for systematic geographic variation in diet. Given the relative landscape-scale homogeneity across our study sites, this is not surprising and adds validity to our examination of dietary overlap at multiple sites within our study area when we were not always able to sample from multiple species at each site. We predicted that diet would show both inter-and intra-annual variation with anthropogenic food resources more important early in the breeding season. We did find that brassica consumption decreased sharply from mid-May to mid-June, possibly reflecting a reduction in availability of oilseed rape tailings at our sites over this time period. We found no evidence for systematic trends in diet composition between years, although interannual differences in diet are likely to represent variability in seed abundance driven by Samples from adults prebreeding and their chicks, or multiple nests from the same adult, showed a tendency for consistency in the proportion of cultivated food within their diet. This may be a consequence of adults specializing on certain foraging habitat types as adult and nestling samples, as well as samples from consecutive nesting attempts, were temporally separated, although larger sample sizes would be required to test this rigorously.
Our findings of positive associations between a higher proportion of dietary components from natural arable plants and turtle dove nestlings in better condition and a higher proportion of anthropogenically provided seed and adults in better condition are ecologically important. They suggest that habitat management providing additional sources of fed seeds for adults early in the breeding season, coupled with habitat rich in accessible seeds of arable plants  once chicks are present, may be crucial to conserving the species.

ACKNOWLEDGEMENTS
Faecal sample collection was funded jointly by the Royal Society for

AUTHOR CONTRI BUTIONS
This work was part of a wider study led by J.C.D. and overseen by A P P E N D I X 4 Predicted species accumulation curves for each columbid species based on the accumulation of taxonomic units. Predicted points, denoted by "+," are overlaid by confidence intervals (grey shading) and barplots from raw data based on 100 permutations of adding samples in a random order.