Captivity reduces diversity and shifts composition of the Brown Kiwi microbiome

Captive rearing is often critical for animals that are vulnerable to extinction in the wild. However, few studies have investigated the extent to which captivity impacts hosts and their gut microbiota, despite mounting evidence indicating that host health is affected by gut microbes. We assessed the influence of captivity on the gut microbiome of the Brown Kiwi (Apteryx mantelli), a flightless bird endemic to New Zealand. We collected wild (n = 68) and captive (n = 38) kiwi feces at seven sites on the north island of New Zealand. Using bacterial 16 S rRNA and fungal ITS gene profiling, we found that captivity was a significant predictor of the kiwi gut bacterial and fungal communities. Captive samples had lower microbial diversity and different composition when compared to wild samples. History of coccidiosis, a gut parasite primarily affecting captive kiwi, showed a marginally significant effect. Our findings demonstrate captivity’s potential to shape the Brown Kiwi gut microbiome, that warrant further investigation to elucidate the effects of these differences on health.


Background
Diet, behavior, habitat type, and environmental species pools can all influence the composition and diversity of gut microbiomes [1][2][3][4]. However, few studies have investigated the impact of captivity, a severe lifestyle shift, on avian gut microbiota. Reports on nine species of parrots, red-crowned crane (Grus japonensis), and vultures (Gyps fulvus and Neophron percnopterus) demonstrate that captivity can impact gut microbiomes [5][6][7]. Fewer studies have compared wild and captive gut bacteria and fungi across spatially distinct sites that vary by climate and vegetation [8], which are expected to differ in microbial species pools, a potential source for gut microbes.
Captive rearing is often necessary to conserve populations of threatened wildlife. In the case of the Brown Kiwi (Apteryx mantelli), predation from introduced mammals has made it imperative for some chicks to be raised in captivity until individuals are large enough to defend themselves [9]. Although successful in increasing population size, consequences to kiwi health via modification of the gut microbiome remain largely unknown. Factors pervasive in captivity such as artificial diets, sterilized built environments, human interaction, and medical intervention [10][11][12] may cause changes to the microbiome, but such changes remain undescribed.
Altering microbial communities may have costs to host health as microbes continue to be recognized for their roles in immune function, pathogen defense, and digestion [13]. Coccidiosis, a gut parasite caused by protozoan Eimeria spp., is a common disease in captive kiwi [14]. However, the relationship between captivity, coccidia, and gut microbial communities has garnered little attention. We sought to compare gut bacteria and fungi between captive and wild kiwi. We tested the hypothesis that captivity status and history of coccidiosis would decrease diversity and modify composition of the gut microbiome.

Results
Fresh fecal samples were collected from seven sites on the north island of New Zealand (Fig. 1a [15] and fungal ITS genes [16] were amplified using DNA extracted from captive (n = 38) and wild kiwi fecal samples (n = 68). To ensure our findings were not an artefact of spatial autocorrelation, we conducted a Mantel test and found a weak association between the sites and the kiwi gut microbiome, however, it was not statistically significant (Mantel correlation, r = 0.138, p =0.308). PERMANOVA results (Supplementary Table 2) found a significant association of site in bacteria (r 2 = 0.13, p = 0.001) and fungi (r 2 = 0.183, p = 0.001), which shows a relationship due to site but not necessarily due to spatial proximity. Kiwi eggs prior to captive rearing were lifted from five locations (Fig. 1a) that span the range of natural sites, indicating that captive lifestyle is more influential than geographic origin.
To determine whether captivity influences kiwi gut microbiota, we used PERMANOVA, linear models, and NMDS analyses. Bacterial communities clustered by captivity across spatially independent sites with little overlap between the 95 % confidence interval ellipses (Fig. 1b, PERMANOVA, r 2 = 0.07, p = 0.001). We fitted separate linear models for bacteria and fungi, using log transformed principal coordinates axis 1 that explained 12.7 % of bacterial variation and 12.8 % of fungal variation, as a proxy for community composition, as our response variables. We found that captivity was a significant predictor of bacterial composition (r 2 = 0.30, p < 0.001) but not fungal composition (r 2 = 0.003, p = 0.301). Although bacterial phyla composition was variable within and across captivity status, Firmicutes was more prevalent in wild kiwi, while Proteobacteria dominated captive kiwi (Fig. 2b). Three fungal phyla, Ascomycota, Basidiomycota, and Mucoromycota, which contained nine classes were predominant in kiwi feces, and varied within and across captivity status (Supplementary Fig. 3).
Bacterial (Fig. 2a, ANOVA, p < 0.005) and fungal (Supplementary Fig. 1, ANOVA, p = 0.012) alpha diversity were significantly lower in captive kiwi by 33 and 74 % respectively. Using Shannon diversity (alpha diversity) as a response variable, we fitted a linear model to determine the relationship with captivity status and found captivity to be a significant predictor of bacterial (r 2 = 0.288, p < 0.001) and fungal (r 2 = 0.135, p = 0.012) alpha diversity. To assess the spread of variation among kiwi microbiomes in captive and wild conditions, we calculated distance to centroid, a metric for beta diversity. No discernible pattern was observed for bacteria (Supplementary Fig. 2a, ANOVA, p = 0.948), but a marginally significant pattern was detected in fungal communities where the mean value was higher in captive kiwi (Supplementary Fig. 2b, ANOVA, p = 0.051).
We also tested if site (a factor nested within captivity status) and history of coccidiosis (positive or negative) had an influence on variation in microbial communities using PERMANOVA (Supplementary Table 2). Site showed a significant effect on bacteria (r 2 = 0.129, p = 0.001) and fungi (r 2 = 0.183, p = 0.001). We ran a linear model using log transformed principal coordinates axis 1 as a proxy for community composition to test the influence of coccidiosis history, data only available for captive Teal colors correspond to captive sites, while brown colors correspond to wild sites. B NMDS plot using Bray-Curtis distance metric shows samples clustering by captivity status with little overlap between the groups (PERMANOVA, r 2 = 0.07, p = 0.001). Ellipses denote 95 % confidence level samples, on gut microbiota. We found a significant trend with bacteria ( Supplementary Fig. 4, linear model, r 2 = 0.118, p = 0.041) but not fungi (r 2 = 0.043, p = 0.204). This contested our PERMANOVA findings which found a weak relationship that was not significant in bacteria (r 2 = 0.048, p = 0.095) and fungi (r 2 = 0.074, p = 0.087).
Using a multinomial species classification method (clamtest) [17], we categorized OTUs into four classes: rare, generalist, wild specialist, and captive specialist. For bacterial OTUs, 10 % were classed as generalist, 53 % as rare, 20 % as wild specialist, and 17 % as captive specialist (Fig. 3a, Supplementary Table 3). For fungal OTUs, 0 % were classed as generalist, 47 % as rare, 27 % as wild specialist, and 27 % as captive specialist (Fig. 3b, Supplementary Table 3). We conducted a simper analysis [17,18] to determine the most influential OTUs that differentiate captive and wild kiwi samples for both bacteria and fungi. Thirteen bacterial OTUs and two fungal OTUs accounted for about 70 % of the differences between wild and captive samples (Supplementary Tables 4   and 5). Nine bacterial OTUs were more abundant in wild samples and five OTUs in captive kiwi (Fig. 3c). Two fungal OTUs were abundant only in wild kiwi (Fig. 3d).

Discussion
Our results indicate that captivity explains bacterial and fungal community differences in the Brown Kiwi gut. Bacterial composition clustered by captivity (Fig. 1b), suggesting that kiwi from the wild are more similar to each other than their captive counterparts, even across geographically distinct sites. Bacterial and fungal alpha diversity were significantly lower in captive kiwi (Fig. 2a,  Supplementary Fig. 1). The consequences of reduced microbial diversity between wild and captive kiwi remain unclear, but several studies have linked dysbiosis to higher disease prevalence in a variety of animals, including ostriches and chickens [19,20]. Coccidiosis history, data only available for captive samples, showed a marginally significant effect ( Supplementary Fig. 4). However, these results may be affected by small sample size. Our results suggest a potential link between differences in the microbiome to disease states that requires further exploration. Overall, our results suggest captivity simplifies the kiwi gut microbiome.
The shift in dominant bacterial phyla, Firmicutes to Proteobacteria, from wild to captive samples may be caused by microbially depauperate captive facilities, antibiotic treatment and post hoc probiotic supplementation. Frequent surface disinfection [11] and probiotic treatment [21] have been shown to increase Proteobacteria in human subjects. Lactobacillus (OTU 49), a common genus in probiotics, was grouped as a captive specialist and is overrepresented in captive kiwi (Fig. 3c). Other captive-associated taxa include Corynebacterium (OTU 62), which has been found in the cloaca of penguins and the preen gland of turkeys [22], and Bacteroides (OTU 544), normally found in animal hosts but can include potential pathogens [23]. Wild taxa such as, Ruminococcaceae (OTU 556) and Lachnospiraceae (OTU 570) (Fig. 3c), were overrepresented in wild kiwi. These two bacterial families were found in broiler chickens challenged with Clostridium perfringens [24], also a predominant wild taxa (OTU 13), suggesting a relationship between these taxa. Blautia (OTU 290), also common in wild kiwi, is a genus found in the human gut and associated with visceral fat accumulation [25]. Faecalitalea cylindroides (OTU 687), a butryrate producing microbe, has been detected in chicken [26]. These taxa may be indicative of nutrient acquisition in the wild, where food may be less available.
No fungal OTUs were categorized as generalists suggesting fungi in kiwi reflect their local environment. Some captive specialists include Cladosporium (OTU Fig. 3 Distinct microbial taxa are classed by their representation in captive and wild kiwi. A multinomial species classification method (clamtest) categorized bacterial and fungal OTUs into one of four classes: rare, generalist, wild specialist, and captive specialist. A For bacterial OTUs, 9.9 % were classed as generalist, 53 % as rare, 19.7 % as wild specialist, and 17.4 % as captive specialist. B For fungal OTUs, 0 % were classed as generalist, 46.7 % as rare, 26.7 % as wild specialist, and 26.7 % as captive specialist. Simper analysis detected several OTUs that explained 70 % difference between captive and wild kiwi. OTUs that were classed as either wild specialist or captive specialist in the clamtest were also represented in the same condition with simper. C Nine bacterial OTUs were significantly represented in wild kiwi and four bacterial OTUs in captive kiwi (FDR adjusted p < 0.05). D Two fungal OTUs were significantly represented in wild kiwi (FDR adjusted p < 0.05) 151) and Aureobasidium (OTU 2), both associated with indoor environments and plant material [27,28], implicating the contribution of soil and ferns added to enclosures. Trichosporon (OTU 171), another captive specialist, is a common human skin taxa [29], suggesting close human interaction may shape kiwi fungi. One wild specialist that is abundant in wild samples (Fig. 3d), Rhizopogon luteolus (OTU 159), has been identified as a dietary component of small mammals, suggesting kiwi may be consuming and dispersing these fungi [30]. Preussia (OTU 181) and Saitozyma podzolica (OTU 37), both associated with soil and litter, were grouped as wild specialists [31,32].

Conclusions
In captivity, artificial diet, sterilized built environments, and human interaction are key factors that can shape gut microbial communities [10,11]. Further detailed investigation of how gut microbes establish in developing kiwi chicks can elucidate how these factors inherent to captivity contribute to the kiwi gut microbiome. Overall, our data suggest that captivity explains differences in the gut microbiome of the Brown Kiwi with potential for health and disease assessment for captive-reared individuals.

Study system Captive
Samples were collected (n = 38) from two captive sites (Fig. 1a). The National Kiwi Hatchery is located at the Rainbow Springs Nature Park in Rotorua, New Zealand. It is the leading facility in kiwi husbandry, egg incubation, and kiwi rearing. The facility has hatched and reared nearly 2000 kiwi eggs. Otorohanga Kiwi House is located in Otorohanga, New Zealand. Both facilities are a part of the Operation Nest Egg (ONE), a program in which kiwi eggs laid in the wild are transported to a hatchery and reared in a captive environment. Coccidiosis information was only available for the captive samples, where diagnosis is determined using a fecal flotation assay [33]. Kiwi were housed in brooder boxes wooden boxes with soil, food, and water. Captive kiwi eat a diet mainly consisting of ox heart, cat biscuits, and rolled oats [34]. If positive for parasites or infections, individuals are administered antiprotozoal or antibiotic treatment. After antibiotics, kiwi are given probiotics that include a combination of Lactobacillus spp. and Bifidobacterium lactis.

Wild
Samples were collected (n = 68) from five natural sites with established wild kiwi populations (Fig. 1a). Ponui Island is located 30 km east of Auckland, New Zealand.
14 Brown Kiwi were introduced to the island by the New Zealand Wildlife Service in 1964, where populations have been increasing, establishing one of the densest populations of kiwi at an estimated 1500 individuals. Motuarohia Island is located in the Bay of Islands, 4 km northeast of Russell, New Zealand. Moturua Island is east of Motuarohia in the Bay of Islands. Puketi Forest is located in the Northland region of New Zealand. Rakaumangamanga is located near the Bay of Islands. These sites differ in climate and vegetation [8] and home to a number of Brown Kiwi individuals.

Sample collection
Fresh fecal samples (n = 108) were collected using sterile spatulas. Supplementary Table 1 details the quantity of samples collected per site. The interior of the fecal pellet was collected to ensure minimal environmental exposure. Due to the unique scent of kiwi excreta, we used the sample's scent to confirm the feces was of kiwi origin [35]. Fecal samples were stored in 5 mL Eppendorf tubes suspended in molecular grade (100 %) ethanol and stored in -20ºC. DNA was extracted using MN NucleoSpin Soil Kit (Macherey-Nagel, Duren, Germany) on Janus extraction robot (PerkinElmer, Waltham, United States), suspended in TE buffer, and stored in -20ºC until PCR amplification.

Metabarcoding
We used a metabarcoding approach with a two-stage amplification process. During the first stage, we amplified the V4 region of the bacterial 16 S rRNA gene using 515 F/806R primers [15] and the fungal ITS gene [16]. The following PCR parameters were applied: denaturation at 95ºC for 2 min, followed by 35 cycles at 95ºC for 20 s, 50ºC for 20 s, and at 72ºC for 30 s, and final extension at 72ºC for 1 min. We used the resulting PCR products as template DNA in the second-stage PCR. Barcoded Fusion primers were used with the following PCR parameters: initial denaturation at 95ºC for 2 min, followed by 8 cycles of 95ºC for 20 s, 50ºC for 20 s, and 72ºC for 50 s, and final extension at 72ºC for 10 min [15]. We purified the second-stage PCR products using SeraMag magnetic beads to remove primer dimers and normalize concentration [36]. Qubit (dsDNA HS Assay Kit, Invitrogen, Carlsbad, United States) was used to quantify DNA concentration and libraries were diluted to 4 nM prior to final pooling. We pooled the libraries according to the concentration determined by Qubit, equimolar based on number of samples per library, and amplicon length. We used LabChip GX Touch Nucleic Acid Analyzer (PerkinElmer, Waltham, United States) to determine DNA concentration and assess quality of final pooled library. Samples were sequenced using Illumina MiSeq platform at Auckland Genomics Facility (University of Auckland), phiX spike 10 %, 250 × 2 cycles. Bioinformatics pipeline Claident was used to demultiplex raw sequences [37]. PEAR evaluated all possible paired-end read overlaps and merged sequences [38]. VSEARCH filtered noisy reads, removed chimeras, and clustered sequences into operational taxonomic units (OTUs) [39]. Claident clustered sequences into OTUs at 97 % similarity and assigned taxonomy with RDP classifier using the following databases, 16 S rRNA training set 16 (bacteria) and UNITE fungal ITS train set 07-04-2014 (fungi). Bioinformatic analysis were performed on the NeSI HPC environment.

Statistical analysis
We calculated Shannon diversity index (R package phyloseq version 2.5-7) [40] to test for a relationship between microbial alpha diversity and captivity. We calculated beta diversity using a multivariate version of Levene's test for homogeneity of variances (betadisper in R package vegan version 2.5-7) [17]. We reported the distance to centroid value. To test for spatial autocorrelation among sites, we conducted a Mantel test (R package ade4 version 1.7-16) using microbiome data, and site location data (latitude and longitude) [41].
We used non-metric multidimensional scaling (NMDS) with Bray-Curtis dissimilarity matrices to reduce multivariate data and spatially visualize microbial communities. NMDS was used to visualize clustering trends across captivity status. We used permutational analysis of the variance (PERMANOVA) also with Bray-Curtis distance matrices to determine whether different factors, such as captivity status (wild/captive), site (geographic area), microsite (i.e. in brooder box, soil, etc.), age (days old of captive individuals), weight (mass in grams for captive individuals), collection date, and history of coccidiosis (positive/negative) can explain microbial community variance. In addition, we ran linear models to determine whether captivity status or coccidiosis history were good predictors of both bacterial and fungal alpha diversity and community composition. We used Shannon Diversity Index values as a response variable for alpha diversity. In our community composition models, we used principal coordinates analysis axis values as our response variable. To fit the assumptions of the model and accommodate negative values, we added the minimum value plus one and log transformed the data. We ran a mixed effects model using site as a random effect, however, Akaike information criterion (AIC) and ANOVA confirmed that it did not improve the fit.
We used clamtest [17] to categorize bacterial and fungal OTUs into the following groups: generalist, too rare, and group specialist (wild-, captive-, positive-, negative-). Positive and negative correspond to individual kiwi who have had a history of coccidiosis. We conducted a simper analysis [17,18] to determine which OTUs explain over 70 % of the differences between groups and to identify OTUs that are overrepresented.
Abbreviations OTU: Operational taxonomic unit; PCR: Polymerase chain reaction; DNA: Deoxyribonucleic nucleic acid; rRNA: Ribosomal ribonucleic acid; PERM ANOVA: Permutational analysis of the variance; NMDS: Non-metric multidimensional scaling history of coccidiosis into rare, generalist, positive specialist, and negative specialist.