DNA metabarcoding, diversity partitioning and null models reveal mechanisms of seasonal trophic specialization in a Mediterranean warbler

Optimal Foraging Theory (OFT) predicts that a population's trophic niche expansion should occur in periods of food scarcity as individuals begin to opportunistically exploit sub‐optimal food items. However, the Niche Variation Hypothesis (NVH) posits that niche widening may result from increased among‐individual differentiation due to food partitioning to avoid competition. We tested these hypotheses through a DNA metabarcoding study of the Sardinian Warbler (Curruca melanocephala) diet over a year. We used null models and the decomposition of beta diversity on among‐individual dietary differentiation to infer the mechanisms driving the population's niche variation. Warblers fed frequently on berries, with a peak in late summer and, to a lesser extent, in autumn. Their diet also included a wide range of arthropods, with their prevalence varying among seasons. Consistent with OFT, the population's niche width was narrower in spring/summer when the population was strongly specialized in berries. In winter, the population's niche expanded, possibly reflecting seasonal declines in food abundance. As predicted by NVH, among‐individual differentiation tended to be higher in winter, but this was mainly due to increased differences in dietary richness rather than to the partitioning of resources. Overall, our results suggest that within‐individual niche does not increase in lean periods, and instead, individuals adopt either a more opportunistic or more specialized foraging strategy. Increased competition in periods of scarcity may help explain such patterns, but instead of showing increased food partitioning as expected from NVH, it may reflect OFT mechanisms on individuals with differential competitive ability to access better food resources.


| INTRODUC TI ON
Animals adjust their foraging behaviour to cope with fluctuations in the availability of trophic resources (Owen-Smith, 2008), which in turn greatly influences individual (e.g., fitness ;Trevail et al., 2019), population (e.g., survival rate; Boff et al., 2021), community (e.g., competition;Lancaster et al., 2022), and ecosystem (e.g., energy pathways; McMeans et al., 2016) processes.According to Optimal Foraging Theory (OFT), these fluctuations drive important variations in the diversity of resources used at individual and population levels (i.e., within-individual and total niche widths, respectively; Pyke, 1984;Roughgarden, 1972).When resources are abundant, within-individual and total niche widths are expected to be reduced, due to individuals specializing in a few optimal food items, while the reverse is expected in lean periods, due to individuals progressively becoming more generalist and incorporating sub-optimal items into their diet (Cachera et al., 2017;Pyke, 1984).A critical assumption of these predictions is that conspecifics exploit food resources in a similar manner (Bastille-Rousseau & Wittemyer, 2019), which may be untenable given the evidence that individuals within a population often vary widely in their resource use (reviewed in Bolnick et al., 2003Bolnick et al., , 2011).An alternative view, derived from the Niche Variation Hypothesis (NVH) (Bolnick et al., 2007), is that in periods of food scarcity individuals reduce competition by increasing the differences in the diet among individuals, or between-individual component (BIC) sensu Bolnick et al. (2003) and total niche width (TNW), but not necessarily within-individual component (WIC) trophic niche width.According to the model of Individual Specialization (Bolnick et al., 2003), this growth in variation is due to individuals specializing in subsets of food available to the population (i.e., niche partitioning: Pianka, 1981;Schoener, 1974), leading to a generalist population composed of specialist individuals.
Most approaches to estimate among-individual dietary differentiation measure the relationship of the individual trophic niche with the population's niche, either for dietary richness or composition (Costa et al., 2008;Costa-Pereira et al., 2018;Jesmer et al., 2020).
For example, indices of Individual Specialization, like the proportional similarity index (Ps i ), estimate the degree of among-individual niche partitioning from how frequently the resources consumed by each individual occur in the diet of the population (Bolnick et al., 2007).
Adaptations of this method have been widely used in trophic niche studies from isotope analyses since it compares few and the same dimensions between each individual and the population (Layman et al., 2012;Matich et al., 2021).However, overestimation of among-individual differences grows with the diet breadth of the actual population, being problematic for generalist or omnivorous species, and requiring multiple observations from each individual to confidently estimate the diet of individuals (Bolnick et al., 2007).
Therefore, direct comparisons of the degree of among-individual difference across spatio-temporal contexts are impracticable, and studies opt to use the deviation of the population from null models to track changes in among-individual differences (Costa et al., 2008;Matthews & Mazumder, 2004).Although this is generally considered adequate for estimating the degree of overlap between individuals and the rest of the population, it neglects the identification of the components of variation that contribute to the niche dissimilarity.This is because Individual Specialization can result from individuals with narrow diets, individuals foraging on rare resources, or a combination of both (Bolnick et al., 2007).In consequence, to accurately test the correlation of environmental pressures (e.g., resource availability or inter-specific competition) and niche partitioning, it is necessary to estimate the deviation of the expected overlap that is not explained by differences in richness between samples, and only due to individuals foraging on different resources.
In the last decades, ecologists identified two distinct phenomena that contribute to the dissimilarity between communities: species replacement (also called turn-over) and richness difference (or nestedness).Replacement occurs through the substitution of species, while richness difference is the result of species gains or losses across samples (Baselga, 2010;Podani & Schmera, 2011; Figure 1).
Although the field of trophic ecology has historically gained from conceptual transpositions from community ecology, like the relationship between TNW, WIC and BIC (Roughgarden, 1972) from measures of gamma, alpha and beta diversities (Whittaker, 1960), decomposition of among-individual dissimilarity has been scarcely done (to our knowledge only in Stewart et al., 2021) and only for the isotopic niche.Simultaneously estimating TNW, WIC and both components of BIC could allow us to assess more accurately the contribution of the individual diet to TNW, both through the average richness of the diet and its variation among individuals.Further, the decomposition isolates the effect of the identity of the foraged resources and tests if the reduction in overlap is due to an increase in niche partitioning (Figure 1).This distinction is not trivial, since both components presumably arise from different ecological phenomena (Legendre, 2014), being that richness difference signals a division in the optimal number of prey items and replacement of a segregation of the targeted prey items.Null models are critical for testing this hypothesis, since they allow to (1) estimate the probability that foraging patterns arise through random assignment of the population's diet (Götzenberger et al., 2012); (2) fix specific components of the dissimilarity (e.g., richness difference) and allow only other components vary (e.g., replacement vs overlap; Ulrich et al., 2017), facilitating the isolation of the niche dimension that explains the shift in among-individual variance; and (3) control the differences in the matrices dimensions across temporal or spatial instances (Gotelli & Ulrich, 2012;Matthews & Pomati, 2012).
This study provides a framework combining DNA metabarcoding, trophic diversity partitioning and null models to test the contribution of several trophic dimensions to the seasonal changes in diet specialization.For this, we used a detailed description of the annual diet of the Sardinian Warbler (Curruca melanocephala), a generalist passerine that is widespread and abundant around the Mediterranean Sea.The Sardinian Warbler feeds on a broad range of insects during the whole year and strongly targets fleshy fruits when available (Herrera, 1984;Tejero et al., 1983).Sardinian warblers are a good model species to study the effects of season on intraspecific trophic variation, mainly due to their rich omnivorous diet (Tejero et al., 1983), the past evidence of seasonality in their foraging behaviour (Herrera, 1984), and being present all year in the study area.
DNA metabarcoding was considered particularly adequate to test our hypotheses using this warbler species, because in contrast to stable isotopes, it provides detailed information on the taxonomic composition of the diet (Hoenig et al., 2022).It is also superior to conventional methods involving the visual detection of hard prey remains in scats due to its capacity of detecting soft-bodied prey and the possibility of a rapid and standardize processing of a large number of samples (Ando et al., 2020;Hoenig et al., 2022).These attributes are essential because we aimed to detect potentially subtle variations in a wide diversity of prey consumed by a large number of individuals during an entire annual cycle.Although resource availability was not measured in this study, previous research reveals a strong fluctuation in the number of arthropods, with a peak of biomass and richness in summer and a significant reduction in winter (da Silva, Heleno, et al., 2019;Welti et al., 2022).Following OFT, we expect to observe a seasonal fluctuation in TNW, with a narrower populational and individual niche during summer and an expansion in winter.However, if in line with NVH, the niche expansion should result from an increase in among-individual differentiation, and not due to an increase in the average diet richness.Further, if individuals are targeting a subset of the population's resources, replacement should be higher than expected, supporting among-individual niche partitioning and the framework of Individual Specialization.On the other hand, if there is no apparent segregation of the targeted resources, and richness difference is the component that explains the F I G U R E 1 Diagram of the partitioning of five dietary resources (squares) into the diet composition of two individuals.Diets shift from complete overlap (a) to 20% overlap through three different scenarios: increase of replacement and constant richness difference (b); equal increase of replacement and richness difference (c); and constant replacement and increase of richness difference (d).The three scenarios result in the same TNW (five resources), mean individual diet richness (3) and among-individual dissimilarity (4).Nomenclature of squares was adapted from the decomposition of the dissimilarity between communities of Carvalho et al. (2012) and Legendre (2014); and representation of among-individual diet as Gaussian curves is after the representation of individual specialization of Bolnick et al. (2003).

| Study area
The

| Laboratory procedures
DNA was extracted from bird droppings using a custom protocol, which consisted of an initial incubation period using a lysis buffer (0.1 mTris-HCl, 0.1 mEDTA, 0.01 mNaCl, 1% N-lauroylsarcosine, pH 7.5-8; Maudet et al., 2002), followed by inhibitor removal using inhibitEX tablets (QIAGEN), DNA precipitation and washing using E.Z.N.A. Tissue Kits (Omega).The extraction protocol started by adding 800 uL of lysis buffer to the dropping.Samples were homogenized with a spatula, vortexed and left in a dry bath at 70°C for 30 min.Afterwards, samples were short spun for 10 s in a bench-top (500 cycles) along with dietary samples from other bird species.

| Bioinformatic analysis
Bioinformatic processing of generated Illumina reads was done using standard bioinformatic packages for metabarcoding data.First, paired reads were merged with Flash (Magoč & Salzberg, 2011), followed by primer removal and sample tagging of reads with the command 'ngsfilter' of Obitools (Boyer et al., 2016).Reads were then dereplicated per sample using 'obiuniq', and singletons of each sample were also removed with the command 'obigrep'.Afterwards, samples were merged and the reads denoised using the command '--cluster_unoise' of VSEARCH (Rognes et al., 2016).Resulting ZOTUs were further inspected for chimeras using the command '--uchime3_ denovo' and then clustered at 99% similarity using '--cluster_size'.
Reads were then mapped back to the retained OTUs using the command '--usearch_global' with an identity level of 99%.Finally, LULU (Frøslev et al., 2017) was used to merge similar OTUs (identity >84%) with high co-occurrence levels (>95% of samples).This greatly re- (e.g., Ferreira et al., 2020;Pauperio et al., 2023;Sousa et al., 2021), thereby reducing the likelihood of erroneous assignments.Whenever an OTU matched several species, genus or families at similar identity levels, we selected the most inclusive taxonomic rank.OTUs assigned only to a genus, family, order or class, which matched the same group of sequences and usually diverged less than 2%, were further clustered with the help of a neighbour-joining tree (Mata et al., 2019) into distinct OTUs (e.g., Carabidae 1, Carabidae 2, etc.).Each OTU was also categorized as either being "diet" (i.e., most arthropods and plants with fleshy fruits) or "not diet" (e.g., birds, fungi, human, internal parasites).
For plant taxa, because there have been intensive flora surveys in the region, we discarded all plant taxa not recorded in those surveys or during fieldwork (Pereira et al., 2016).Furthermore, for each plant taxon, we evaluated whether it produced fleshy fruits and in which months the fruits were available (González-Varo et al., 2021), and categorized each bird-plant interaction per month as "likely" (i.e., plant taxon produces fleshy fruits and interaction was observed in fruiting season) or "unlikely" (i.e., plant taxon does not produce fleshy fruits or the interaction was observed outside the fruiting season).This was done to avoid incorporating secondary detections in ecological analysis, a problem known to occur frequently in omnivorous bird species (da Silva et al., 2020;Tercel et al., 2021).Also, interactions potentially arising from flower-visitation behaviour were not considered, as no evidence of pollen or nectar was found on the sampled birds (da Silva et al., 2014(da Silva et al., , 2017)).Samples that did not have at least 100 reads belonging to dietary items were considered to have failed and were discarded.Finally, we filtered from each sample all diet OTUs that represented less than 1% of the total number of dietary reads of that sample (Mata et al., 2019).This filter was applied only after discarding all the non-dietary OTUs as the number of non-target reads can vary considerably across samples, potentially creating biases in the proportion of discarded OTUs.The number of reads and OTUs that remained in samples used for analysis after each filtering step is presented in Table S1.

| Data analysis
All statistical analyses were performed on R v4.0.3 (R core Team, 2020).Statistical significance was considered at an alpha value of 0.05.Dietary analyses were based on presence/absence of prey items on each sample at the family and OTU levels.We carried out most analyses at both levels, because previous research has shown that they can yield complementary information on species dietary patterns (da Silva et al., 2020;da Silva, Heleno, et al., 2019;Gleason et al., 2023).On the one hand, while analysis at the family level involves a relatively low taxonomic detail, it can reveal predator preferences for certain functional prey types, as families often represent species with similar morphology and other life history traits (Laiolo et al., 2020;Lovell et al., 2007).On the other hand, while analysis based on OTUs often face data sparsity constraints (i.e., many OTUs have a very small representation in the diet), they can still yield relevant information regarding temporal and spatial variations on the species most relevant to a predator's diet (e.g., Hacker et al., 2021;Tercel et al., 2021).Months were grouped in pairs for analyses to guarantee an adequate sample size per level of factor.
The overall richness of ingested prey items, or TNW, was estimated using rarefaction curves based on Hill numbers (q = 0) with the function 'iNEXT' of the package 'iNEXT', that were extrapolated for two times the observed sample size (Chao et al., 2014).TNW was estimated across equal levels of sampling coverage by selecting the lowest value observed in each variable at extrapolation with the function 'estimateD' of the package 'iNEXT' and 9999 bootstrap replications (Hsieh et al., 2016).The mixture of extrapolating the smaller samples and rarefying the larger samples to a fixed sampling coverage maximizes the utilization of data, in contrast with traditional rarefying that does not extrapolate samples (Chao & Jost, 2012).For comparing the richness of ingested prey items between variable categories, we considered that they differed significantly if the 84% confidence intervals did not overlap.Although it is commonly assumed that 95% confidence interval is equivalent to 5% Type 1 error, simulation studies show it is overly conservative and favour an 84% interval for capturing the error more accurately Differences in dietary composition between sexes, ages and months were tested using a permutational multivariate analysis of variance (PERMANOVA; 9999 permutations) with the function 'adonis' of the package 'vegan' (Dixon, 2003) based on a 'Jaccard' dissimilarity matrix calculated with the function 'vegdist' of the same package.
Multivariate homogeneity of group dispersion was assessed using the function 'betadisper' of the package 'vegan' and then tested if distance to centroid was significantly different between categories with the functions 'anova' and 'TukeyHSD' of the package 'stats' (R core Team, 2020).To identify the prey items that contributed the most to differences in dietary composition, the function 'simper' of the package 'vegan' was used with 9999 permutations.
To test for differences among individuals, or Between Individuals

| RE SULTS
We made 292 captures of 165 Sardinian warblers, collecting a total of 273 samples (droppings).From these, we successfully amplified and sequenced DNA from 234 samples belonging to 146 individuals, which were used in dietary analysis (Table 1).For each pair of months, we obtained 39 samples ±6.8 (SD), corresponding to 20.7 ± 2.6 males and 12.5 ± 4.7 females.There was also a total of 35 samples from birds with undetermined sex collected between June and August.For each pair of months, we analysed 25.5 ± 8.6 samples of first-year birds (1y) and 13.5 ± 8.1 samples of second-year or older birds (2y+).After all filtering steps, samples had an average of 9515 ± 17,416 reads for the plant, and 13,026 ± 22,966 reads for the animal component of the diet.

| Diet description
We identified 617 OTUs, belonging to 179 families and 38 orders.

| Total dietary richness
The total number of families and OTUs detected in the diet varied significantly across months, but not between age classes and sexes (Figure 3; Tables S4 and S5).

| Among-individual dietary dissimilarity
The average Jaccard distance between pairs of samples fluctuated along the year, reaching the lowest value during August-September (0.829 ± 0.005) and then increasing from December to March (0.897-0.927 ± 0.002-0.008).The replacement component explained less of the dissimilarity from December to January (0.613), and it increased constantly across the year until its peak in October and November (0.707).Richness difference was inversely proportional to replacement, and therefore followed the opposite pattern, with the highest richness occurring during December-January, and then decreasing until October-November (Figure 5).
The average dietary dissimilarity between individuals varied significantly over time, considering expectations from both null models.
Regarding null model r1 that reflects the replacement component, dissimilarity was lower than expected in August-September (observed dissimilarity = 0.829 ± 0.005 vs expected dissimilarity = 0.863 ± 0.03; p-value = .011),and did not differ from expectations in other months.

| Diet of the Sardinian warbler
Building on the power of DNA metabarcoding, our study provides detailed and highly resolved taxonomic information on the diet of the Sardinian Warbler in northeast Portugal, along with its seasonal

Richness difference (c0) Among−individual difference (BIC)
variation during an entire year.We found significant fluctuations in the trophic ecology of this generalist species over the year, from periods when the population as a whole specialized on a few dietary items to months of increased niche width and high variability in the diet richness of the individuals.During summer, when fleshy fruits and arthropods were abundant (da Silva, Heleno, et al., 2019;Welti et al., 2022), Sardinian Warblers specialized on few prey times, significantly increasing the degree of dietary overlap between individuals and reducing the niche width of the population.In contrast, during the winter period of berry and arthropod scarcity (da Silva, Heleno, et al., 2019;Welti et al., 2022), the population's niche size increased considerably, and dietary richness became more variable among individuals.Additional differences in diet composition and individual richness were found between age categories, with older birds having a significantly richer diet.In contrast to other studies on passerine trophic ecology (da Silva et al., 2020;Temeles et al., 2017), we found no differences between sexes in TNW, dietary composition and individual diet richness.
The overall dietary composition of the Sardinian Warbler observed in our study is similar to that reported in previous studies based on foraging behaviour (Debussche Isenmann, 1983) and stomach content analysis (Tejero et al., 1983), with Hymenoptera, Hemiptera, Coleoptera, Lepidoptera, Diptera and Stylommatophora dominating the diet.We found evidence of fruit consumption throughout the entire year except in April and May when there were no ripe fruits in the study area.Fruit consumption, mainly that of R. ulmifolius peaked in August and September, followed by a steady decrease until February and March.Compared to studies in Spain, the consumption of R. ulmifolius was higher but the overall proportion of berries consumed annually was lower in northeast Portugal (Herrera, 1984;Jordano, 1982).These differences are likely related to the availability of fruits with adequate size, which are known to vary greatly among years and sites (e.g., González-Varo et al., 2021;Molina et al., 2011).The differences in fruit availability across sites are possibly a major driver of the foraging patterns, and therefore, of the degree of specialization attained during summer by this population of Sardinian Warblers.We also found major differences in the consumption of different arthropod groups over the year, which likely reflect fluctuations in their availability.However, apart from a general decline in arthropod abundance during winter (da Silva, Heleno, et al., 2019;Welti et al., 2022), little is known about the seasonal availability of different arthropod groups during the year.Therefore, a greater understanding of Sardinian warbler prey selection patterns would require future studies combining dietary metabarcoding with the analysis of arthropod prey availability along the annual cycle.

| Temporal variation in trophic niche
The strong annual fluctuation in the consumption of prey items by Sardinian warblers was associated with temporal variation in the population's and individual's diet richness, as well as changes in among-individual diet variation.The TNW increased from spring (April 2018) to late winter (March 2019), except for a steep decrease in OTU richness during August and September.The average richness of the diet was lower during the breeding period (April-July) than in the post-breeding season (August-March), which probably contributed to the increase of the population's niche size.However, individual richness remained constant during the strongest increase in the population's niche width (from late summer to winter).This is partially in line with the tenets of the NVH, which posits that the increase in population niche width during periods of food scarcity is mainly a consequence of higher dietary differentiation between individuals, rather than resulting from the increase in individual's diet richness (Bolnick et al., 2007;Svanbäck & Bolnick, 2006).It may be argued, however, that this result is a consequence of DNA metabarcoding only detecting organisms consumed within a few hours of the collection of the sample (Drinkwater et al., 2021), which might constrain the estimates of individual dietary richness and thus their contribution to the TNW of the population.Nevertheless, we expect minimal variation in the temporal detection window of DNA metabarcoding for the same type of samples throughout the year.
Moreover, in our study, DNA metabarcoding allowed a wide range of individual diet richness estimates, with particularly high values in winter reaching up to 25 different prey items per individual, while the average richness always remained within about 5.3-8.2prey items per individual.This suggests that, despite its limited detection time window, DNA metabarcoding could have detected much higher values of average diet richness than those obtained in our study.
Consequently, we believe that our comparative estimates of the individual diet should not be severely limited by the methodology.
Although the traditional decomposition of TNW into individual richness and among-individual differences provides support to the NVH, a close inspection of the components of compositional dissimilarity among individuals reveals a more complex picture.Since r1 simulations maintain the same richness per sample in the simulated communities (Patterson & Atmar, 1986), the reduction in the observed among-individual dietary dissimilarity during August and September is solely explained by the identity of the prey items.
We hypothesize that individuals are eating more similar food items during late summer than in other seasons, probably because it coincides with R. ulmifolius ripeness peak, which was consumed by 78% of the individuals during those months.A high proportion of individuals target these profitable resources to maximize energetic intake (Sargeant, 2007), thus supporting the OFT (Pyke, 1984).In contrast, during the population's niche expansion in winter, c0 simulations registered a higher among-individual dissimilarity than expected, suggesting a higher variation in diet richness among individuals (Jonsson, 2001;Figure 4).In fact, a disproportionately high number of individuals showed extremely high values of prey families (>10) during winter in comparison to the rest of the year (χ 2 = 18.809, pvalue < .001),while lower values were equally observed throughout the year.This pattern does not seem to have been driven by juvenile individuals, as the proportion of juveniles and adults with extreme values of prey family richness in winter did not differ (χ 2 = 0.730, p-value = .393).NVH and Individual Specialization have been used to explain the increase of among-individual differences during lean periods, as an indicator of increased resource partitioning due to among-individual competition (Bolnick et al., 2003(Bolnick et al., , 2007).Yet, we did not find partitioning of the targeted prey items among individuals, which would be expected if individuals were specializing in different subsets of the population's resources.Instead, it appears that the population became a mixture of specialist and generalist individuals.Overall, our results point out major temporal fluctuations in the trophic niche of the Sardinian warbler, from periods of resource abundance when the population as a whole specializes in a few and highly rewarding food items such as berries to lean periods when some individuals still feed on a few prey items while other greatly increase their trophic niche width.

| Trophic specialization, DNA metabarcoding and diversity metrics
Besides detailing the natural history of the Sardinian Warbler, our study shows the value of combining DNA metabarcoding with beta diversity decomposition and null models to test more accurately well-known theories of trophic ecology, like OFT (Pyke, 1984), NVH (Bolnick et al., 2007) and Individual Specialization (Bolnick et al., 2003).We believe that this framework will find extensive plicability across various study systems, though this requires considering carefully the limitations and potential biases associated with DNA metabarcoding.First, metabarcoding cannot provide reliable information on the number or biomass of different prey ingested (Deagle et al., 2019), and so we focused on parameters that can be estimated using presence/absence in each sample, including diet richness, composition and frequency of occurrence of different prey items.Such parameters are particularly informative for generalist species such as the Sardinian warbler, which feed on a very wide variety of relatively small prey items, and for which the richness and type of prey items ingested varies markedly over time and across individuals.Our framework may be less effective for analysing species with diet variation primarily driven by proportional changes in the consumption of a small number of prey items, though further research would be needed to evaluate this idea.Second, the application of this framework needs to be based on large sample sizes such as ours, which is essential to obtain robust estimates of diet diversity and composition metrics (Mata et al., 2019).Third, primer selection needs to consider the objectives of the study and previous information on the diet of the target species, which in our case resulted in the use of primers targeting the COI and ITS2 regions.Several concerns have been raised regarding the use of protein-coding genes like COI for DNA metabarcoding studies (e.g., Clarke et al., 2014;Collins et al., 2019;Govender et al., 2022), due to the absence of conserved regions for primer design and, consequently, strong amplification biases.Nonetheless, this problem has been reduced in recent years with the development of highly degenerate primers (Collins et al., 2019;Elbrecht et al., 2019;Piñol et al. 2019), like the one used in this study.Even though amplification biases cannot be fully avoided, with some species being better amplified than others, some studies have shown species recovery rates >95% in complex mock communities (Elbrecht et al., 2019;Jusino et al., 2019).The problems previously reported for COI may probably be overcome by using alternative genes like 16S, but achieving at least comparable rates of OTU identification would still require the development of more comprehensive barcode libraries, which at present are far more complete for COI than any other gene (Andújar et al., 2018;da Silva, Mata, et al., 2019;Machida et al., 2017).Finally, while having high taxonomic resolution is one of the key advantages of DNA metabarcoding, it is also a potential source of problems in some statistical analysis due to data sparsity.In fact, a generalist predator such as the Sardinian warbler can feed on more than one thousand arthropod species during the annual cycle, which implies that the frequency of occurrence of most prey species is low and difficult to estimate accurately unless using large sample sizes (Cuff et al., 2022;Mata et al., 2019).Because of this, consistent patterns of variation can be difficult to estimate for all but the most frequently consumed prey items, while the incidence of the least consumed prey may be largely random.This can be circumvented to some extent by carrying out analysis at both OTU and family levels, because, while the first can reveal selection patterns of the most consumed prey species (e.g., Hacker et al., 2021;Tercel et al., 2021), the latter is useful to reveal broader preferences regarding taxonomic and functional groups (Gleason & Rooney, 2017;Laiolo et al., 2020).
Building on the strengths of DNA metabarcoding, our approach to understand the mechanisms of trophic specialization and to test OFT and NVH focused on the variation in diversity metrics across temporal periods with different food availabilities.This differs from previous studies assessing the degree of niche partitioning between individuals, or Individual Specialization, which tended to use metrics that summarize into one number the multivariate compositional differences between samples (Bolnick et al., 2003;Thiemann et al., 2011;Villsen et al., 2022).That method is especially useful for isotope analyses, since they compare the same number of resource axes in each sample, thereby facilitating temporal and spatial comparisons, albeit at the cost of having coarser dietary information (Bolnick et al., 2003).Here, we argue that for DNA metabarcoding studies, or any dietary analysis providing taxonomically resolved information on the prey items consumed, more detailed information can be gained using our analytic framework building on diversity analysis from community ecology.Specifically, we suggest an approach evaluating how gamma, alpha and beta diversities, and their replacement and richness difference components, change in relation to food availability, and how these changes are more or less pronounced than expected by chance using null models.First, we use gamma diversity (i.e., TNW) to evaluate whether the trophic niche of the population widens in periods of scarcity, which is compatible with both OFT and NVH predictions (Bolnick et al., 2007;Pyke, 1984).Second, we analysed alpha diversity (i.e., WIC) to evaluate changes in individual trophic niche width, for which an increase during periods of food scarcity has been interpreted as supporting OFT (Pyke, 1984) and an absence of significant differences being part of the framework of NVH (Bolnick et al., 2007).Finally, we use beta diversity (i.e., BIC) to understand whether declines in food availability are associated with increased differentiation between individuals, which is usually considered sufficient to test NVH (Bolnick et al., 2007;Svanbäck & Bolnick, 2006).However, we argue that such a test requires the partitioning of beta diversity in its replacement and richness difference components, because an increase in BIC can be due to either: individuals consuming fewer prey items in common while maintaining differences in diet richness (i.e., niche partitioning); individuals having diets with larger differences in richness while maintaining species replacement; or individuals having diets with fewer prey items in common and larger differences in diet richness (Legendre, 2014).Only an increase in replacement during periods of food scarcity can be used to support the NVH, and only higher replacement than expected by chance can be used to support Individual Specialization, as they indicate that there is increased segregation among individuals in their selection of prey.In contrast, the richness difference component only informs on the degree of heterogeneity in diet richness among individuals, and thus it does not provide information on niche partitioning, or specialization on subsets of resources as described in the framework of Individual Specialization.An increase in richness difference during periods of food scarcity suggests that the population becomes composed of a mixture of more specialized and more generalist individuals, which supports the idea of OFT operating at the level of individuals with different competitive abilities.To the best of our knowledge, this idea has been hardly explored in trophic ecology, though it may be an important mechanism whereby changes in food availability affect prey selection in animal populations.Testing this and other ideas in trophic ecology should be the subject of further research, benefiting from the ability of DNA metabarcoding to analyse large sample sizes with high taxonomic resolution at ever lower costs.
-individual variation, it can be interpreted as individuals in the population adopting a mixture of more opportunistic and more specialized foraging strategies.
microcentrifuge, and up to 700 uL of supernatant was transferred to a new tube containing one-quarter of an inhibitEX tablet.Samples were then vortexed for 1 min and centrifuged at 8000 rpm for 30 s.Up to 500 uL of supernatant was transferred to a new tube, and 25 uL of OB Protease was added.The remaining steps followed the E.Z.N.A. kit recommendations, except that DNA was eluted two times in 50 uL into different extracts.This was done as a backup mechanism, but only the first elution was used for diet amplification.DNA was extracted in batches of 23 samples plus one negative control in which no dropping was added.Extracted DNA was distributed in 96-well plates, including extraction negative controls, and the last well was left empty for PCR negative control.DNA extracts were further purified using Agencourt AMPure XP beads (Beckman Coulter) with a ratio of 1:1.6, before PCR amplification.Invertebrate prey and consumed plants were independently amplified using the FwhF2-R2n(Vamos et al., 2017) and UniPlantF-R (Moorhouse-Gann et al., 2018) primer sets, respectively, modified with Illumina overhang adapter sequences.These primers target the COI and ITS2 regions, respectively, and have been shown to amplify a wide diversity of invertebrates and plants, while providing a high taxonomic resolution of the amplified taxa(Elbrecht et al., 2019;Moorhouse-Gann et al., 2018).PCR reactions consisted of 5 uL of Qiagen Multiplex Master Mix, 0.3 uL of each 10 nM primer, 3.4 uL of water and 1 uL of DNA extract.Cycling conditions were similar for both primers consisting of a 15 min period at 95°C, 35 cycles of 30 s denaturation at 95°C, 30 s annealing at 50°C, 30 s extension at 72°C and a final extension period of 10 min at 72°C.PCR products were diluted 1:4 and went through a second PCR reaction to incorporate 7 bp long indexes and P5 + P7 Illumina adaptors.PCR reactions and cycling conditions were similar to the first PCR except that KAPA HiFi HotStart ReadyMix (Rocher) was used, and only eight cycles of denaturing, annealing and extension were done, with annealing at 55°C.PCR products were purified using Agencourt AMPure XP beads (Beckman Coulter) with a ratio of 1:0.8, and subsequently quantified using Nanodrop and diluted to 15 nM.Purified and normalized PCR products were pooled per marker.These two libraries were then individually quantified using qPCR (KAPA Library Quant Kit qPCR Mix, Rocher) and diluted to 4 nM.Finally, libraries were pooled equimolarly and sequenced in a HiSeq Rapid SBS Kit v2 the number of retained PCR artefacts, sequencing errors, as well as nuclear copies of the mitochondria, that tend to artificially inflate the number of OTUs present in each sample.OTUs without the expected length (202-208 bp for Fwh2; 187-387 bp for UniPlant) were finally removed, and the number of reads observed per OTU present in extraction and PCR blanks was further subtracted to the respective samples associated with each extraction batch and PCR plate, to remove potential lab contaminations.OTUs were then compared to online databases (BOLD and NCBI) and identified to the lowest taxonomic rank possible, taking into account the likelihood of occurrence in Portugal and Iberia in the case of arthropods, and in the study area in the case of plants.Strong efforts have been made to generate barcode libraries of insects in the region

(
MacGregor-Fors & Payton, 2013;Payton et al., 2003).For estimating differences in individual diet richness, or the WIC, we used a GLM with negative binomial distribution using the function 'glm.nb' of the package 'MASS'(Venables & Ripley, 2002), with the number of prey items per sample as response variable and pair of months, sex and age category as independent variables.A likelihood ratio test comparing the negative binomial model with the Poisson model supported that the former fitted better the distribution of the data, for both OTU (p-value < .001)and family (p-value < .001).
Component (BIC), across months, the mean 'Jaccard' dissimilarity values in prey composition of warbler individuals were compared to the simulated dissimilarity values of 9999 null models.Z-scores were calculated by subtracting the mean dissimilarity values of the null models from the observed dissimilarity values, and then dividing it by the standard deviation of the null models (Z-score = (Observed dissimilarity -Average-nullmodels)/ SD-nullmodels).Positive Z-score values indicate dissimilarity values higher than expected and negative Z-scores values lower than expected, that is, more dissimilar or more similar diets, respectively.BIC was calculated using only prey families because the OTU sampling coverage per pair of months was low (mean = 0.58, range = 0.51-0.72)due to the breadth of the monthly diet as indicated by the asymptotic richness of OTUs.Only one sample per individual for each pair of months was randomly included in this analysis to avoid pseudo-replication.For this, models were simulated, and Z-scores were calculated for all possible combinations of samples while including only one sample per individual per pair of months.Null models were created using the functions 'nullmodel' of the package 'vegan' and 'simulate' of the package 'stats'.Comparisons of the observed community with the null models and Z-score were calculated with the function 'oecosimu' of the package 'vegan'.Methods 'r1' (RANDOM1:Patterson & Atmar, 1986) and 'c0' (RANDNEST:Jonsson, 2001) were used to simulate the null models for testing differences in replacement and richness, respectively.Method r1 maintains sample richness and selects prey items using marginal column sums as probabilities.The obtained null models have a fixed sample richness in relation to the observed community, but the degree of overlap and replacement vary inversely because of the probabilistic selection of the prey item identity in each individual diet (FigureS1).In contrast, c0 maintains prey item frequency and simulates a Gaussian distribution of diet richness in each null community, while maintaining the mean richness of the observed community.Therefore, the degree of overlap is fixed, and differences due to richness and replacement vary inversely.The first method tests if the identity of the foraged resources is different than expected by the availability of the resources, and therefore indicates if the population is focusing on similar prey items or partitioning resources among individuals.The second method tests if there are differences in prey richness among individuals, showing months when richness variation was lower or higher than expected by chance.

4
Violin plots of individual diet richness for family level along the sampling year for the Sardinian Warblers.White dots represent the average richness of each pair of months.Value of Jaccard dissimilarity explained by both components of beta diversity for each pair of months.Dissimilarity Z-scores comparing the observed diet dissimilarity per pair of months with the simulated diet dissimilarity of 9999 null models with methods r1 and c0, representing differences in replacement and richness, respectively.Asterisks (*) signal months with significant deviations from the null models (p < .05).Whiskers signal the standard deviation among the Z-scores calculated with different combinations of samples.