Niau Atoll (16°9'15"S, 146°21'20.4"W) (Tuamotu Archipelago, French Polynesia, South Pacific Ocean) (Figure 1), probably colonised by humans over 1000 years ago and regularly visited since 1820 (Butaud 2007), has a current resident population of ca. 200. Niau Island measures 26 km² and culminates at 6 m above sea level (Andréfouët et al. 2005). Its climate is tropical and oceanic (Mueller-Dombois and Forsberg 1998). The vegetation consists of three main formations: (i) feo forest, a dense forest growing on jagged fossilised limestone coral covering 1800 ha of the atoll, (ii) Cocos nucifera plantations covering 700 ha, of which only 10% is used for copra farming and (iii) wetlands dominated by shoreline purslane Sesuvium portulacastrum and Jamaica swamp sawgrass Cladium mariscus (Butaud 2007).

This endemic bird is confined to Niau and preferentially nests and forages within coconut groves located on the east side of the island, especially in semi-open and exploited groves (Gouni et al. 2006; Coulombe et al. 2011; Thibault and Cibois 2017) (Figure 1). Between 1970 and 1990, the Niau kingfisher was reported as common (Seitre and Seitre 1992), with a population estimated at 400–600 individuals (Holyoak and Thibault 1984). The reasons for its decline in the early 2000s remain unknown, but hypotheses include introduced mammalian predators (Kesler et al. 2012) and/or land-use changes (decreased coconut harvesting and use of fire to clear coconut plantations) (Coulombe et al. 2011; Thibault and Cibois 2017). Currently, the population has stabilised at ~140 individuals (100 mature individuals) (Thibault and Cibois 2017). To breed, the Niau kingfisher excavates the trunks of dead coconut trees, where it nests, tending to return to territories and partners over several years (Gouni and Sanford 2003; Holyoak and Thibault 1984). The breeding season lasts from October to February and both parents undertake parental duties until juveniles are independent (Kesler et al. 2012). To date, the few field observations available on the kingfisher’s diet suggest that it mainly forages on terrestrial arthropods (Araneidea, Coleoptera, Dermaptera, Anisoptera) and lizards (Scincidae and Gekkonidae) (Holyoak and Thibault 1984; Marie 2006).

Sample types and collection

Sampling was conducted at two periods yearly over two consecutive years (from November 2009 to March 2011): the first in November at the beginning of kingfisher reproduction (laying, incubating and hatching periods) and the second in February-March at the end of the breeding season (fledgling and feeding periods of juveniles). Predators’ diet samples and prey availability were sampled from six habitat types within surveyed kingfisher territories: two feo forests (low < 8 m and high > 8 m), three coconut plantations (abandoned, cultivated and intensively cultivated) and a wetland (Figure 1).

Rats were trapped in each habitat along a 320 m transect of 40 equally-spaced Victor (model BM201, Forest Stewardship Council, USA) snap-traps, set over five consecutive nights, baited with coconut flesh before dusk and checked early in the morning. Trapped rats were identified at the species level, weighed, sexed and dissected. Their guts were collected and stored in 90% ethanol during transport and then frozen until examination. Cat scats were collected across all island paths, stored in Ziploc bags (SC Johnson, USA) and frozen until analysis. Kingfisher pellets were collected during the final field session (March 2011; end of reproductive season) below nests or hunting perches. The entire set of analysed diet samples consisted of 186 kingfisher pellets, 578 cat scats and 295 rat digestive tracts (218 R. exulans and 77 R. rattus).

To evaluate the availability of the main potential prey groups, the density of 16 different taxa (i.e. rats, Scincidae, Amphipoda, Isopoda and twelve terrestrial arthropod Orders) was estimated within the six studied habitats in February and November 2010 (see Suppl. material 1 for a detailed description of sampling protocols).

Diet analysis

Morphological diet analyses were conducted on all the collected diet samples. Each rat’s stomach contents and three last non-expelled faeces were individually extracted, homogenised and the entire volume analysed. Kingfisher pellets were dried before analysis and cat scats were analysed by washing over a 0.5 mm sieve under a stream of hot water. All hard prey remains (e.g. hairs, feathers, bones, scales, chitin) were isolated and identified to the finest taxonomic level possible under a dissecting microscope by comparison with reference materials from field-collected specimens and via identification keys (for details, see Zarzoso-Lacoste et al. 2013, 2016). When possible, lizards were identified at the family level (Gekkonidae, Scincidae), arthropods at the order level (for Coleoptera, Hymenoptera, Orthoptera, Hemiptera, Dictyoptera, Dermaptera, Lepidoptera, Diptera, Odonata, Scorpiones, Araneae, Decapoda, Amphipoda and Isopoda) and Myriapoda at the subphylum level (containing Diplopoda and Chilopoda classes). Gastropoda and fish (Teleostei) were identified at the class and infra-class level, respectively. Birds were identified at the finest taxonomic level possible and assigned to the Aves class level in further statistical analyses. Mammalian prey corresponded only to the Rattus genus. Plants were only reported as presence/absence data for both omnivorous rat species.

To maximise the detection and identification of Niau kingfisher DNA in cat and rat diet samples, we implemented a PCR-based method (see Zarzoso-Lacoste et al. 2013, 2016) targeting the 5’ end of the multi-copy mitochondrial cytochrome c oxidase subunit I gene (Cox1), a standard barcode sequence that enables species-level discrimination in birds (Hebert et al. 2004). This molecular method includes the combined use of three bird-specific primer pairs that were selected for their high power of detection, their specificity and their sensitivity towards bird DNA in cat and rat diet samples (for details, see: Zarzoso-Lacoste et al. 2016). This protocol notably deals with cases where bird soft tissue (e.g. vitellus, embryo or flesh) were consumed: our protocol is able to detect and identify bird DNA (including the DNA of the Niau kingfisher) at concentrations as low as 0.01 ng.μl^-1 (the lowest concentration tested in vitro using dilution series; see Zarzoso-Lacoste et al. 2016). Furthermore, we previously demonstrated that the molecular protocol systematically performs better in both the detection and the resolution of the taxonomic identification of birds in the diet samples of their predators than the morphological method (including samples where no hard remains of prey were found; see Zarzoso-Lacoste et al. 2016).

Data analyses

All analyses were performed using the Statistical Software R version 3.5.1 (R Development Core Team 2018) – see Suppl. material 2 for the original data used to perform the following analyses.

Diet descriptors

To describe each predator’s overall diet and for each of the above prey taxa, several indices were calculated from (i) number of Prey Occurrences (PO) and (ii) Minimum Number of Individuals (MNI; White 1953). Prey Occurrence Frequency (POF = PO/S) indicates the frequency of a prey category in predator diet samples, “S” being the total number of samples per predator. Mean Prey Number (MPN = MNI/S) indicates the mean abundance of a prey category in predator diet samples. To more accurately evaluate the intensity of predator consumption on a particular prey category “i”, we assessed (i) P_iPN = MNI_i/∑MNI, the proportion of this prey relative to the total number of animal prey ingested by each predator and (ii) MPN_i = ∑(MNI_i/PO_i) reflecting the predation pressure on this prey. Finally, we constructed a bipartite network to visualise links between predators and their shared or exclusive prey, using MNI data and the bipartite R package (Dormann et al. 2008).

Diet comparison based on identified animal prey

We performed all subsequent analyses using MNI data for the 21 identified animal prey taxa (excluding plants and unidentifiable lizards and terrestrial arthropods). Abundance-based diet data were square-rooted prior to analysis to reduce the influence of the most abundant taxa (Clarke and Warwick 2001).

Sampling representativeness, diet richness and diversity

We used sample-size-based Hill numbers (orders q = 0, 1 and 2), plus interpolated and extrapolated accumulation curves to estimate (i) the sampling representativeness of each predator diet and prey availability based on taxonomic richness (q = 0) and (ii) predator diet diversity using the exponential Shannon’s entropy index (giving more weight to rare species, q = 1) and the inverse of Simpson’s concentration index (giving more weight to abundant species, q = 2) (Chao et al. 2014). We then compared their diet diversity up to the same base sample size (here, m = 1473 prey MNI, see Chao et al. 2014 for more information), using the package iNEXT (Hsieh et al. 2016). Corresponding 95% confidence intervals were obtained by a bootstrap method based on 999 replications.

Identification of indicator prey in predators’ diet

To identify the prey or combination of prey either included in the diet of a particular predator and/or contributing most to niche overlaps, we conducted “indicator species analyses” using the multipatt and strassoc functions of the indicspecies package (De Caceres and Legendre 2009). We used the signassoc function with 999 permutations and Sidak’s correction for multiple comparisons to compare the abundance of prey taxa in predator diets.

Prey selection

We computed the Jacobs’ electivity index (D; Jacobs 1974) to investigate whether these four predators consumed prey proportionately to their relative densities in the environment. This index ranges between -1 (negative selection) and +1 (positive selection), indicating whether a prey species is selected, respectively, less or more than proportionate to its availability. We performed chi-square tests with Bonferroni adjustment to test for the significance of prey selection by the studied predators.

Diet dissimilarity, breadth and overlap

We measured interspecific niche separation and intraspecific variability amongst predator diet samples using a Bray-Curtis dissimilarity matrix. We calculated the mean dissimilarity (MD) of diet composition between and within predator species using the meandist function of the vegan package (Oksanen et al. 2018). We then compared predator diets using a permutation-based test with the betadisper and permutest functions, corrected with Tukey’s ‘Honest Significant Difference’ method for multiple comparisons. To visualise patterns in dietary dissimilarity within and amongst species in a low dimensional space, we performed Nonmetric Multidimensional Scaling (NMDS) using the metaMDS function. Finally, we calculated predator Diet Breadth (DB; from 0-specialised diet-to 1-generalist diet-) and Diet Overlap (DO; from 0 -no overlap- to 1 -complete overlap -), using the nichevar and nicheoverlap functions of the indicspecies package. Corresponding 95% confidence intervals were obtained by a bootstrap method based on 999 replications.

Kingfishers almost exclusively prey on terrestrial arthropods (PiPN = 50%; see Suppl. material 3), lizards (40%) and Decapoda (9%) (Figure 2[1]). The terrestrial arthropod taxa Coleoptera (23%), Hymenoptera (10%, of which 92% were Formicidae) and Dictyoptera (9%) were the most frequently and abundantly preyed upon. Scincidae were the most consumed lizards (23%) but Gekkonidae were also widely eaten (11%). Of these prey, Scincidae, Gekkonidae, Decapoda, Araneae and Coleoptera were strongly and significantly associated with the kingfisher diet (component A > 0.63) (Tables 1, 2).

The cat diet mainly consisted of rats (49%), followed by terrestrial arthropods (28%) and lizards (10%) (Figure 2[1], see Suppl. material 3). Dictyoptera were the terrestrial arthropods most frequently and abundantly consumed (8%), followed by Dermaptera (5%) and Coleoptera (2%). Gekkonidae were the most consumed lizards (4%). Rats and fish were significantly associated with cat diet samples (Tables 1, 2).

The rat diet was mainly plants (mainly coconut flesh, POF = 89% and 94% for R. exulans and R. rattus, respectively), but included a large proportion of animal prey (Figure 2[1], see Suppl. material 3). Terrestrial arthropods were the animal prey consumed most by both R. exulans and R. rattus (PiPN = 82% and 75%, respectively), followed by lizards (11% and 7%, respectively). R. rattus preyed more frequently on lizards, Gastropoda and crustacean than R. exulans. For both rats, Scincidae were the most consumed lizards (6% and 3% for R. exulans and R. rattus, respectively). Of the identified terrestrial arthropods, R. exulans and R. rattus mainly consumed Dictyoptera (11% and 8%, respectively), Hymenoptera (25% and 20%, respectively, all Formicidae), Diptera (11% and 8%, respectively, mainly larvae) and Coleoptera (7% and 4%, respectively). Hemiptera, Isopoda and Amphipoda were indicator prey particularly present in the R. rattus diet (Tables 1, 2).

Morphological and PCR-based methods, used in combination, allowed the detection of 28 bird individuals in cat (n = 20) and rat (n = 3 and 5 for R. rattus and R. exulans, respectively) diet samples and the identification of 24 of them as belonging to the following seven species; Gygis alba, Ptilinopus coralensis, Anous stolidus, Gallus gallus, Sterna bergii, Puffinus lherminieri and Accrocephalus atyphus (for details, see Zarzoso-Lacoste et al. 2016). None was assigned to the Niau kingfisher.

Rarefied and extrapolated species-accumulation curves (See Suppl. material 4) tended to saturate, indicating accurate sampling of predators’ diet and mean prey availability in the environment. Diversity accumulation curves revealed that kingfishers presented the lowest prey richness (q₀D = 12.9), although their prey abundance was more equally distributed amongst categories (q₁D = 6.9 and q₂D = 5.8) (See Suppl. material 4). Conversely, cats exhibited intermediate-level prey richness (q₀D = 17), consuming prey categories more unevenly (q₁D = 5.3 and q₂D = 2.9). The two rat diets presented the highest prey richness (q₀D = 21.9 and 20.1 for R. rattus and R. exulans, respectively), but differed in distribution of prey abundance (q₁D = 9.2 and q₂D = 6.5 for R. exulans, q₁D = 7 and q₂D = 4.7 for R. rattus).

Mean dissimilarities (Table 3) and intra-species dispersion (Figure 2[2]) were highest in R. rattus and R. exulans diets (but lower for R. rattus), while intermediate values were observed for the Niau kingfisher, with the lowest values for cats. Differences were significant for all pairs of predators (permutation test, p_adjusted < 0.001-TukeyHDS-), except between R. exulans and R. rattus (p_adjusted = 0.13) (Table 4). These results are consistent with calculated predator diet breadths (DB, see Suppl. material 3), the two rat species showing the widest diet breadth (0.43, CI = 0.42–0.44), closely followed by the kingfisher (0.42, CI = 0.41–0.42), while the cats showed the narrowest (0.35, CI = 0.33–0.35).

Jacobs’ electivity index (D) confirmed that the kingfisher positively selected its main prey (i.e. Scincidae, Coleoptera, Hymenoptera, Dictyoptera and Dermaptera) (Table 5). The kingfisher negatively selected Araneae and strongly avoided almost all the other taxa. Of the main kingfisher prey, Dictyoptera and Dermaptera were strongly and positively selected by both rat species and cats (D > 0.92, p < 0.001; and D = 1, p < 0.001, respectively). Scincidae, Coleoptera and Hymenoptera were also positively selected by R. exulans (D > 0.76, p < 0.001) and R. rattus (D > 0.62, p < 0.001), while cats slightly positively selected Coleoptera and tended to shun Hymenoptera (D = 0.48 and -0.26, respectively, p_adjusted < 0.001).

Interspecific niche separation was highest between cats and the three other predators (MD = 0.92, 0.94 and 0.95 for R. rattus, kingfisher and R. exulans, respectively), intermediate between R. exulans and both kingfisher and R. rattus (0.81 and 0.79, respectively) and lowest between kingfisher and R. rattus (0.76) (Table 3, Figure 2[2]). These results are consistent with the calculation of diet overlap between predators. Diet overlap was almost complete between the two rat species (DO = 0.98, CI = 0.96–0.99), intermediate between kingfishers and rats (0.63 CI = 0.56–0.70 and 0.57 CI = 0.48–0.69 with R. exulans and R. rattus, respectively) and low between cats and the three other predators (0.21 CI = 0.15–0.27 with both Niau kingfisher and R. rattus and 0.17 CI = 0.13–0.20 with R. exulans). The prey taxa contributing most to diet overlaps between the Niau kingfisher and both rat species were Dictyoptera and Hymenoptera (components A > 0.86, B > 0.33), while R. rattus strongly shared two additional prey with this bird (Scincidae and Coleoptera; components A > 0.86, B > 0.33) (Table 1). Diptera, Myriapoda, Lepidoptera and Orthoptera were strongly shared by the two rat species (components A > 0.85), while R. rattus only significantly shared Gastropoda with cats (components A > 0.86).

Our study offers the first detailed diet analysis of the Niau kingfisher during its chick-rearing period, adding to the limited existing data. Our findings are crucial for the accurate conservation and management of this critically endangered bird.

First, the Niau kingfisher consumes a narrow range of prey taxa, but in regular abundances (i.e. low diet richness but relatively high diversity). Dissimilarity of diet samples is low, suggesting a relatively homogeneous diet. These results support a narrow diet breadth and specialised diet at a population level that make the kingfisher highly vulnerable to exploitative competition for its few main prey.

Second, Gekkonidae, Scincidae, terrestrial arthropods (principally Coleoptera, Dyctioptera and Araneae) and small Decapoda represent crucial resources for adult, nestling and fledgling kingfishers. In particular, Scincidae, Coleoptera, Hymenoptera and Dictyoptera are positively selected (i.e. consumed more than proportionate to their availability in the environment), suggesting that these scarce prey may be potentially at risk for exploitative competition with introduced predators.

Third, some of the Niau kingfisher’s main prey are cryptogenic species, probably introduced from South-East Asia by Polynesians over the last centuries (Fisher 1997; Austin 1999; Ineich et al. 2007; Hamilton et al. 2010): Emoia cyanura, E. impar, Lipinia noctua, Cryptoblepharus poecilopleurus (Scincidae), Lepidodactylus lugubris, Gehyra oceanica, G. insulensis (Gekkonidae) or more recently (late 1980s) for Hemidactylus frenatus (Case et al. 1994; Ineich et al. 2007). The latter species, generally considered as aggressive, has the potential to strongly compete and negatively impact local gecko populations (Case et al. 1994). The Niau kingfisher also widely consumes the cosmopolitan Dyctioptera Pycnoscelus surinamensis, Blatella germanica and Periplaneta australasiae, probably more recently introduced in the Tuamotu, at least partly from South Asia (Cochereau 1966; Parker et al. 1977). It is worth noting that the kingfisher likely shifted its nesting and foraging habits with the development and intensification of copra farming on the island, from patches previously dominated by the native palm tree (Pritchardia mitiaroana) to areas widely planted with the cryptogenic coconut tree on which this bird currently depends (Coulombe et al. 2011; Thibault and Cibois 2017). Hence, our study illustrates the positive function (e.g. alternative food or habitat resource) which exotic species can perform for island wildlife, particularly in man-transformed landscapes (e.g. Schlaepfer et al. 2011, 2012). Such complex interactions between native and introduced species, often scientifically neglected, need to be considered in restoration and conservation projects to prevent unexpected cascade effects from alien species removal or control.

No support for high predation pressure on Niau kingfisher population

Although our study used two complementary approaches (morphological and PCR-based methods) to analyse a large number of cat and rat diet samples (578 cat scats and 295 rat digestive tracts), collected within kingfisher territories during the critical incubating and rearing periods, no bird remains were formally identified as Niau kingfisher. Our results suggest that, if predation by cats and rats does occur, it is much less frequent than suggested by Kesler et al. (2012).

To explain the decline of the Niau kingfisher, Kesler et al. (2012) hypothesised that the low survival of adult females (ca. 40%) might be due to nocturnal rat predation of nests during incubation. This hypothesis is not consistent, however, with the aggressive and territorial behaviour reported for Niau kingfishers, observed chasing rats from nesting trees (Gouni et al. 2006; pers. obs.). Since eggs are rarely left unattended (parents take turns brooding), unlike nestlings (parents often out foraging; G. Coulombe pers. comm.), the nestling stage could be the most vulnerable to rat predation. However, our results do not support this alternative hypothesis either. Kesler et al. (2012) also suggested that cat predation may be responsible for the high mortality of kingfisher juveniles (approximately 90% each year). Although cats are efficient bird predators (Doherty et al. 2016; McCreless et al. 2016) and the behaviour of juvenile kingfishers increases their predation risks (e.g. perching near the ground, calling loudly; Gouni et al. 2006; Kesler et al. 2012; pers. obs.), the absence of kingfisher remains in cat scats do not support this hypothesis either.

Diet overlaps and potential exploitative competition between native and introduced predators

An extensive overlap in diet and food habits may indicate either a high potential for competition between species or a very abundant resource (Cupples et al. 2011). Evaluating whether the interacting predators select their shared prey more than proportionate to their availability in the habitat (positive selection) is therefore critical to untangle these two hypotheses.

On Niau Island, cats presented the narrowest diet breadth and the lowest variability in intraspecific diet composition, suggesting relatively homogeneous and specialised trophic behaviour (low prey richness with few abundantly preyed taxa) of individuals. Although cats and Niau kingfishers shared positively-selected prey (mainly Gekkonidae, but also Coleoptera, Dyctioptera and Dermaptera), their diets only marginally overlapped (DO = 0.21), making competition or competitive exclusion unlikely.

Conversely, both rat species presented generalist trophic behaviour, with the widest diet breadths and inter-individual variability in diet composition. Our study revealed a substantial niche overlap between the Niau kingfisher and both rat species (DO = 0.63 and DO = 0.57 for R. exulans and R. rattus, respectively). While Dictyoptera and Hymenoptera constitute the main diet overlap between kingfishers and both rat species, Scincidae and Coleoptera are also highly shared by R. rattus and kingfishers. Importantly, of the Niau kingfisher’s prey, all but Dermaptera were significantly positively selected by both rat species (and more intensively by R. exulans), indicating possible exploitative competition with the kingfisher for these highly nutritive and relatively scarce prey (see Table 5). This exploitative competition may reduce the density and availability of Niau kingfisher critical prey (Towns 2002; Rankin et al. 2018), thereby possibly affecting survival at different life-stages (adults, nestlings and fledglings) and/or breeding success.