Extraction.

Upon receiving colon samples from Sand Island, we dissected the samples to obtain fecal pellets for DNA extraction in a sterilized lab setting using flame-sanitized tools with 95% ethyl alcohol before each dissection. Fecal pellets were stored in 2 mL individually-labeled vials at -80°C. Genomic DNA was extracted with a DNeasy PowerSoil Pro kit (Qiagen, Carlsbad, USA) implemented on a QIAcube Instrument (Qiagen), following the manufacturer’s protocol. Samples were homogenized with a FastPrep-24 System (MP Biomedicals, Irvine, USA) for 40 s at 6 m/s.

Sequencing and library preparation.

Genomic DNA was prepared for sequencing with a two-stage amplicon sequencing workflow [50]. Initially, genomic DNA was PCR-amplified using two primer sets (independently). For arthropods, we used the fwhF2/fwhR2n primers to target a 205 bp region of cytochrome c oxidase I (COI) of the MT-CO1 gene [51, 52]. For plants, we used the UniplantF/UniplantR primers to target a 187–387 bp region of the internal transcribed spacer 2 (ITS2) region of nuclear ribosomal DNA (between genes 5.8S and 28S; [53]). The primers—fwhF2/fwhR2n and UniplantF/UniplantR—contained 5’ linker sequences compatible with Access Array primers for Illumina sequencers (Fluidigm, South San Francisco, USA). PCRs were performed in a total volume of 10 μL using MyTaq^™ HS 2X Mix (Bioline/Meridian Bioscience, Cincinnati, USA), with primers at 500 nM. We also attempted to amplify genomic DNA with several bird-specific primers but were unsuccessful owing to high host DNA amplification, despite the use of peptide-nucleic acid (PNA) oligonucleotide blockers and restriction enzymes. For more details about PCR conditions and outcomes, see S1 Appendix.

DNA extraction, library preparation, pooling, and sequencing were performed at the University of Illinois at Chicago (UIC) Genome Research Core within the Research Resources Center. For plant amplicons, DNA sequencing was performed on an Illumina MiSeq instrument (San Diego, California) employing V3 chemistry (600 cycles, 2 x 300 bp). For arthropod amplicons, DNA sequencing took place on an Illumina NovaSeq 6000 instrument with an SP flow cell type (2 x 250 bp), owing to high host DNA amplification.

Bioinformatics pipeline.

Bioinformatics analysis was performed by the UIC Research Informatics Core within the Research Resources Center. Forward and reverse reads were merged using paired-end read merger (PEAR) [54]. For sequencing trimming, we used cutadapt, and quality trimming was based on quality threshold and length parameters (p = 20, min length = 100 bp; [55]). Adapter/primer sequences were trimmed from reads; reads that lacked the adapter/primer sequences were also discarded. Additionally, ambiguous nucleotides (N) were trimmed from the ends and reads with internal ambiguous nucleotides were discarded. Chimeric sequences are artifacts of the PCR process and occur when portions of two separate amplicons fuse during the amplification process. The UIC analysis pipeline uses a standard chimera checking program to identify chimeric sequences and remove them from datasets; chimeric sequences were identified using USEARCH in a denovo fashion [56]. We then generated separate amplicon sequence variant (ASV) tables for plant and arthropod amplicons using the DADA2 pipeline [57] with a 97% identity (sequence similarity) threshold in order to increase accuracy of taxonomic assignment and exclude chimeric sequences. Sequences were matched to references (thereby producing taxonomic annotations) using a nucleotide BLAST search [58] via nt, the NCBI Genbank non-redundant nucleotide database [59]. We then exported these data along with raw sequence counts into tables of taxonomic annotations from phylum- to species-level results (separately for plant and arthropod ASVs). For more details about sequence data, see S1 Appendix.

Tissue collection.

For stable isotope analysis (SIA), we used mouse hair samples because they are simple to obtain and store but also because of the unique growth cycle of mouse hair. Unlike that of other mammals, mouse hair does not grow at a continuous rate. Instead, mice grow a pelt or coat (once—maybe twice—depending on longevity), which is not continuously growing. However, single hair follicles do undergo successive growth phases—namely growth (anagen), regression (catagen), and rest (telogen)—that may take up to seven weeks to complete [60, 61]. In addition, not all hairs are shed during these growth cycles, and up to four hairs (all of which may be in different phases) are retained in one follicle [60]. Thus, mouse hair follicles are in a continuum of growth phases [62], making it difficult to ascertain the time during which a hair was actively growing and—in turn—assimilating carbon and nitrogen isotopes from food. For our project, we used only paired colon and hair samples from mice identified as “adult” (i.e., >7 weeks old), meaning that they had completed at least one full hair cycle (and perhaps more). Accordingly, the use of mouse hairs for SIA allows us to understand broader trends in mouse diet because hair samples likely contain hairs in multiple growth phases representing several weeks to months of food assimilation.

Based on the plant and arthropod ASV tables generated from NGS, we collected samples of plant and arthropod taxa detected in ≥5% of the mouse fecal samples (sensu [32]). Plant samples were obtained between April 2020 to November 2021. We sought to collect ≥10 samples per taxon, including leaves, stems, inflorescences, fruits, and seeds. Plant samples were air-dried at room temperature for one day, then stored in individually-labelled coin envelopes in a plastic bag filled with silica beads. In addition to mouse trapping for this study, we also completed arthropod pitfall trapping simultaneously along the same traplines (from April 2018 to May 2020; see [63]). Arthropod specimens collected from these pitfall traps were used for SIA. As with plant taxa, we sought to collect ≥10 samples per arthropod taxon. For larger arthropods, such as Araneae and Blattodea, we used only the head and legs for SIA; however, for smaller arthropods, such as Isopoda and Hymenoptera, we used the entire body as well as one or more individuals in a sample to obtain an appropriate sample mass for SIA. Arthropods from pitfall traps were stored in 70% ethanol. We also collected mōlī feathers from January through November in 2019, based on the documented predatory and consumptive behavior of mice on Sand Island [10]. Feather samples were stored in individually-labeled coin envelopes at room temperature. Because mouse diet on Sand Island was unknown prior to our NGS work, it was not possible to sample food taxa for SIA until after mouse colon and hair samples had already been collected and analyzed; regardless, we attempted to collect food taxa from different habitats throughout the year to account for spatial and temporal variation in isotopic values.

Stable isotope analysis.

To prepare samples for stable isotope analysis, we soaked samples in a 2:1 mixture of methanol/chloroform for at least two hours in an analog orbital shaker at a low-shake frequency (200 rpm) to remove any contaminants or residue [64]. We then rinsed all samples twice in DI water and oven-dried them at 55°C for 72 hours. After drying, samples were homogenized and weighed to the nearest 0.002 mg. Prior to packing our samples into tin capsules, we used a mortar and pestle (which was cleaned before and after each sample) to crush arthropod and plant samples and we cut mouse hairs and mōlī feathers into the smallest pieces possible. We used approximately 0.8 mg of mouse hairs, 1.0 mg of mōlī feathers, 3.0 mg of plant tissue, and 1.0 mg of arthropod tissue to analyze for carbon and nitrogen elemental composition (C:N ratio) and carbon and nitrogen isotopes (δ¹³C and δ¹⁵N). We processed our samples in isotope-ratio mass spectrometers (Delta Plus Advantage Mass Spectrometer and Delta Plus XL Mass Spectrometer, Thermo Fisher Scientific, Waltham, USA) at Northern Illinois University (NIU) and the University of Tennessee–Knoxville (UTK). Stable isotope ratios are expressed in δ notation in per mil units (‰), based on the equation: where X is ¹³C or ¹⁵N and R is the corresponding ratio of ¹³C to ¹²C or ¹⁵N to ¹⁴N. R_standard values are based on Vienna PeeDee Belemnite (VPDB) for δ¹³C and atmospheric N₂ for δ¹⁵N. At NIU, samples were compared against NBS 22, USGS-25, IAEA-CH-6, IAEA-N1, and IAEA-N2 standards. At UTK, samples were corrected to the standard values of UT729, acetCost, and LGlu4510; isotopic values were calibrated against the international standard materials of USGS40 and USGS41. Sample precision based on repeated standard reference materials was ±0.1‰ for δ¹³C and ±0.2‰ for δ¹⁵N; sample precision between the NIU and UTK laboratories based on replicate tissue samples was ±0.5‰ for δ¹³C and ±0.4‰ for δ¹⁵N.

Sequence data analysis.

For our plant ASV table, we filtered out all ASVs assigned to any phylum other than Streptophyta (land plants and all green algae except Chlorophyta). Similarly, for the arthropod ASV table, we filtered out all ASVs assigned to any phylum other than Arthropoda. For both plants and arthropods, we used only ASVs identified at least to genus in our subsequent analyses. For plant ASVs, we removed Arachis sp. and Avena sp. because we used peanut butter and oats as bait in house mouse traps, which occasionally were detected in mouse fecal samples (but neither species occurs on Kuaihelani).

Because of biases that occur during PCR, sequencing, and bioinformatic analysis processes, sequence read counts provide—at best—a coarse estimate of how much a given ASV contributes to diet. In other words, the percentage of sequence reads belonging to each food taxon is not a reliable proxy for relative biomass consumed [41, 65]. Thus, we converted ASV sequence read counts to occurrence data (i.e., presence/absence of food taxa). Although this transformation can introduce biases—such as omitting poorly amplified taxa [66]—it is a more conservative, semi-quantitative diet summary. Although we sought to retain as many ASVs as possible (including those with low read counts), we used both sequence read normalization and read thresholds to exclude artifacts. We obtained the relative read abundance (RRA—proportional summaries of sequence read counts; [66]) for each food taxon by dividing the number of reads per ASV by the total count of ASVs for each sample, separately for both the plant and arthropod ASV tables. We considered a taxon to be present if it contributed ≥1% of the sequence reads per sample [66, 67]; 1% is considered a suitable threshold for many diet studies, but lower thresholds may be required when diets are extremely diverse or large and variable differences occur in amplification of food taxa [68]. We used these transformed occurrence datasets (%FOO—percent frequency of occurrence; [66]) to summarize diet taxa and for additional statistical analyses.

We conducted multivariate analyses using PRIMER-e v7 [69] to examine spatial and temporal differences in plant and arthropod consumption by mice. First, we computed two resemblance matrices of occurrence data (plant and arthropod ASVs) using the Sørensen dissimilarity index [70] and visually examined the output using principal coordinates analysis (PCoA; [69, 71]). We then used permutational multivariate analysis of variance (PERMANOVA) tests with habitat type and trapping session as fixed effects in a two-way crossed design using a Type III SS (with 9,999 iterations). Here, the estimated sizes of components of variation can be used to compare the relative importance of different effects in a model to explain overall variation [71]. Significant results from a PERMANOVA test indicate differences in plant or arthropod consumption among groups (i.e., centroid location effect) and/or variability within groups (i.e., dispersion effect; [72]). As a follow-up, we used the PERMDISP test to check for differences in the homogeneity of multivariate dispersion among habitat type and trapping session groups [71]; in other words, PERMDISP tests the variability of plant and arthropod composition in diet among groups. Significant results from PERMDISP help to elucidate if plant or arthropod consumption differs owing to distinct diet composition and/or variability in diet composition.

Dietary mixing model analysis.

When selecting sources for our mixing model analysis, we retained only plant and arthropod taxa detected in ≥5% of mouse fecal samples (i.e., %FOO ≥5%), so as to avoid overparameterizing our model with rare food sources (sensu [32, 73]). Additionally, even though we were unable to detect mōlī DNA in mouse fecal samples, we decided to include mōlī as a source as well (given mouse predatory behavior on Sand Island; [4, 10]). In total, we identified 25 food taxa from ASV data as sources for our model; however, with so many sources—and only two isotopic tracers—a mixing model is unlikely to yield precise estimates of source proportion in diet [73]. To reduce the number of sources initially, we a priori categorized food taxa into three biologically-meaningful and taxonomically-relevant groups: C₃ plants, C₄ plants, and mōlī. For the remaining arthropod taxa, we used Ward’s hierarchical cluster analysis (“ward.D2” from the “hclust” function in R package ‘stats’; [74]) and multivariate tests (PERMANOVA and PERMDISP in PRIMER-e; [69]) to determine the most appropriate grouping of arthropods into source groups. Through cluster analysis, we sought to categorize arthropods based on feeding ecology [75] while minimizing isotopic overlap among groups. However, many of the arthropod taxa consumed by Sand Island’s mice are generalist omnivores, detritivores, and scavengers and share similar isotopic signatures (S3 Table). Five source groups emerged via cluster analysis, namely: 1) two Lepidopteran taxa (phytophagous larvae and adults); 2) Megaselia scalaris (Diptera; zoophagous, detritophagous, and mycetophagous larvae; predator, parasitoid); 3) Periplaneta sp. (Blattodea; detritophagous larvae and adults); 4) Trachyzelotes jaxartensis (Araneae; predator); and 5) an array of isotopically-similar taxa from Diptera, Ixodida, Hymenoptera, and Isopoda with diverse feeding ecologies (S3 Table). For each of these arthropod source groups, within-group isotopic variation was smaller than among-group variation. Thus, our stable isotope mixing model had eight main source groups: C₃ plants, C₄ plants, Diptera-Ixodida-Hymenoptera-Isopoda, Lepidoptera, M. scalaris, mōlī, Periplaneta sp., and T. jaxartensis (S1 Fig). Although the discriminatory power of mixing models declines with seven or more prey sources [73], Bayesian mixing model frameworks such as MixSIAR can still estimate the posterior distributions of source proportions for under-determined mixing systems [42]. While we initially specified eight sources in our model, we combined sources a posteriori (sensu [42]) into four groups to summarize results: arthropods, C₃ plants, C₄ plants, and mōlī (Fig 2). Doing so preserves the covariation structure among source proportions while narrowing the distribution of possible solutions for each source [73]. In addition to our main stable isotope mixing model, we also ran an additional model excluding mōlī as a source, and a posteriori combined our sources into three groups: arthropods, C₃ plants, and C₄ plants (see S2 Appendix).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. δ¹³C and δ¹⁵N values of house mice and four main source groups.

Source groups include arthropods, C₃ plants, C₄ plants, and mōlī; mean values of δ¹³C and δ¹⁵N indicated by source group symbol and bars represent ± 1 SD. House mouse stable isotope values were adjusted by subtracting diet-tissue discrimination factors for hair tissue (δ¹³C = 1.7 ± 1.3‰; δ¹⁵N = 3.2 ± 1.1‰).

https://doi.org/10.1371/journal.pone.0293092.g002

Because consumer tissues contain slightly elevated isotopic concentrations of C and N owing to discrimination during assimilation and excretion processes, it is necessary to correct isotopic data before analysis via a diet-tissue discrimination factor (DTDF; also commonly called a trophic discrimination factor [TDF] or trophic enrichment factor [TEF]; [76]). Before applying our mixing model, we evaluated several different DTDFs to determine the most appropriate fit given our consumer (mouse) and food taxa data sets; in other words, we sought to find the most suitable DTDF that would result in the highest probability that our proposed mixing models would explain the isotopic signature of Sand Island’s mice. Some of the DTDFs we tested were specific to house mouse hair tissue, whereas other DTDFs were more broadly applicable to various taxa [77–84]. We used a Monte Carlo procedure from Smith et al. [85] to construct many possible mixing polygons for each DTDF to test if all consumer isotopic data fell within the 95% mixing region. Based on this approach, we selected DTDF values estimated from the R package ‘SIDER’ [81] as the most appropriate fit; in the process, we also identified three outliers in our consumer data set (i.e., outside of the 95% mixing region) and excluded these data points from our mixing model (sensu [85]).

We then used the MixSIAR Bayesian mixing model framework [42] in R v4.1.2 [74] to quantify the proportion contributed to mouse diet by source group. We fitted our MixSIAR models with raw isotopic data of mouse hair samples, raw isotopic data of prey source groups (Fig 2 and S1 Fig), and the estimated DTDF for mouse hair (δ¹³C = 1.7 ± 1.3‰; δ¹⁵N = 3.2 ± 1.1‰). Because we had ASV data for all of our sources except for mōlī, we used “uninformative”/generalist Dirichlet priors for our mixing model as a conservative approach. We included habitat type and trapping session as random effects (S2 and S3 Figs) to assess the relative importance of these factors in explaining variation in source contribution via variance partitioning (see [42]). Additionally, we used a process*resid error structure to account for variability in consumer isotopic data [86] and incorporated elemental concentration dependence as well [87]. Our models consisted of three Markov Chain Monte Carlo (MCMC) chains of 1,000,000 iterations, thinned by 500, and with a burn-in of 500,000 [42]. We checked model convergence and fit by plotting the posterior predictive distributions and assessing Gelman-Rubi diagnostic values (i.e., Gelman-Rubin statistics < 1.1; sensu [19, 88]). A posteriori, we combined the original eight sources into four source groups for summarizing and reporting source proportion in diet; similarly for our model excluding mōlī as a source, we combined the original seven sources into three source groups.

Arthropods.

We detected 56 genus-level arthropod ASVs from 14 orders and 41 families in mice, with a mean of 4.55 ASVs per mouse (SD = 3.07, range = 0–11). Most arthropod ASVs were rare, with 43 arthropod ASVs (77%) detected in <5% of samples (Table 1). However, several arthropod ASVs were commonly detected, with eight arthropod ASVs present in ≥30% of samples and 13 ASVs in ≥5% of samples. Nearly all arthropods consumed by mice are non-native. Of 13 commonly-detected arthropod ASVs, all are non-native species—except for Ornithodoros capensis, which has a worldwide distribution as a seabird ectoparasite. Additionally, of the 43 arthropod ASVs detected in <5% of samples, 39 ASVs (91%) are likely non-native; only two ASVs (dermestid beetles, Dermestes ater and D. maculatus) are native to the Hawaiian archipelago.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Arthropod ASVs detected in house mouse fecal samples through next-generation sequencing (NGS); n = 318 samples.

https://doi.org/10.1371/journal.pone.0293092.t001

Plants.

We detected 53 genus-level plant ASVs from 14 orders and 23 families in mice, with a mean of 2.71 ASVs per mouse (SD = 1.48, range = 0–9). Like arthropods, most plant ASVs were rare, with 42 ASVs (79%) detected in <5% of samples (Table 2). Eleven ASVs were detected in ≥5% of samples; only three plant ASVs occurred in ≥30% of samples. Most plants consumed by mice are non-native, and the two most frequently-occurring plant ASVs (in ~50% of samples) are highly invasive: Eleusine indica and Casuarina equisetifolia. Of the remaining nine plant ASVs found in ≥5% of samples, four are non-native (Lobularia maritima, Tournefortia argentea, Phyla nodiflora, and Stenotaphrum secundatum) and four are native species found across Pacific Islands and/or within the Hawaiian archipelago (Chenopodium album [likely C. oahuense / ʻāweoweo, present on Kuaihelani], Ipomoea pes-caprae / pōhuehue, Boerhavia repens / alena, and Solanum nigrum [closely related to S. americanum / pōpolo, present on Kuaihelani]). One ASV identified to genus has both native and non-native congeners present on Kuaihelani (Eragrostis sp. = E. variabilis / kāwelu [native], E. paupera [native], and E. tenella [non-native]). Of the 42 ASVs detected in <5% of samples, 32 ASVs (76%) are non-native, seven are native, and three have native and non-native congeners on Kuaihelani.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Plant ASVs detected in house mouse fecal samples through next-generation sequencing (NGS); n = 318 samples.

https://doi.org/10.1371/journal.pone.0293092.t002

Habitat types.

Arthropods are a core food source for mice across habitat types, composing 42–86% of mouse diet (Fig 3B, S4 Table). Specifically, Periplaneta sp. are the main arthropod source in mouse diet in all habitats, except the wetland; in the herbland, in particular, the majority of mouse diet comes from Periplaneta sp. and T. jaxartensis (S5 Table). C₃ and C₄ plant contributions vary substantially among mice (Fig 3B); C₃ and C₄ plants are minimally consumed in the herbland, whereas C₃ plants are an important source in both shrub and forest habitats (31% and 19% respectively), and C₄ plants contribute the greatest proportion to mouse diet in wetlands (45%) (S4 and S5 Tables). Mōlī contribution to mouse diet is relatively low (between 7–13%)—but consistent—among all habitat types (Fig 3B), with a slightly elevated proportion in shrub habitat (S4 and S5 Tables). Similar results were found across habitat types with the model excluding mōlī as a source group (S2 Appendix).

Multivariate analyses reveal that mice consume significantly different compositions of plants and arthropods among habitats (Table 3). Habitat type has a stronger effect on plant diet, in particular; mice are 36% dissimilar in their plant consumption among habitats, but only 9% dissimilar in arthropod consumption (see “Component” [components of variation estimates] in Table 3). Indeed, habitat effects are visibly evident in plant diet composition (Fig 4A). In Fig 4A, mice cluster into loose groups that largely correspond to habitat types; this is especially apparent in contrasts of mice from the herbland with those from the forest across PCoA axis 1. This difference in plant diet composition is also evident in ASV data (S4 Fig). For example, over 90% of mice captured in the forest consumed Casuarina equisetifolia, compared to only 28%-38% of mice from other habitats (S4 Fig). Differences in arthropod consumption by mice, however, are relatively weak among habitat types (i.e., mice overlap in diet among habitats; Fig 4A). Although variability in plant composition of mouse diet differs among habitats, it does not for arthropod composition (Table 3, S5 Fig); pairwise comparisons further support these findings (S6 and S7 Tables). In summary, plant consumption by mice differs significantly in terms of composition and variability thereof among habitat types, whereas arthropod consumption differs (weakly) only among habitats. Notably, most of the variation in plant and arthropod diet composition is at the individual level; here, mice are 42–49% dissimilar in their plant and arthropod consumption, respectively (see “Component” in Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. PCoA of plant and arthropod consumption by house mice (n = 318) based on Sørensen dissimilarities by A) habitat type and B) trapping session.

Points located closer together represent more similar groups of plants or arthropods consumed by mice, and points farther away represent less similar groups.

https://doi.org/10.1371/journal.pone.0293092.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. PERMANOVA partitioning and analysis and PERMDISP test results of plant ASVs (53 taxa) and arthropod ASVs (56 taxa) detected in house mouse fecal samples, based on Sørensen dissimilarities.

https://doi.org/10.1371/journal.pone.0293092.t003

Trapping session.

Plant and arthropod composition of mouse diet differs significantly among trapping sessions (Table 3). However, this seasonal effect was not clearly apparent and much overlap occurred in both plant and arthropod diet composition across sessions (Fig 4B, Table 3). While variability in plant composition of mouse diet differs significantly among trapping sessions, variability in arthropod composition does not (PERMDISP results in Table 3, S6 and S7 Tables). Compared to variation in diet owing to habitat type, mice captured during different sessions are only 17% dissimilar in their plant diet composition and hardly vary in arthropod diet composition (only ~5%; see “Component” in Table 3, S6 and S7 Figs).

Although we cannot reliably interpret source contribution among trapping sessions from our stable isotope mixing model, all four source contributions remain largely even among sessions (S4 Table). Based on the estimated variation in diet for habitat type and trapping session in our model, habitat contributes nearly four times more to diet variation than does trapping session, which also mirrors our multivariate statistical findings.

Habitat type and trapping session interaction.

We detected an interaction between habitat type and trapping session for plant consumption only, but this interaction was very weak compared to these two factors independently. This effect is likely driven by differences in plant consumption (and/or variability thereof) by mice among habitats for each trapping session, rather than a temporal effect within each habitat (S6 Table).