Effects of diet, habitat, and phylogeny on the fecal microbiome of wild African savanna (Loxodonta africana) and forest elephants (L. cyclotis)

Abstract The gut microbiome, or the community of microorganisms inhabiting the digestive tract, is often unique to its symbiont and, in many animal taxa, is highly influenced by host phylogeny and diet. In this study, we characterized the gut microbiome of the African savanna elephant (Loxodonta africana) and the African forest elephant (Loxodonta cyclotis), sister taxa separated by 2.6–5.6 million years of independent evolution. We examined the effect of host phylogeny on microbiome composition. Additionally, we examined the influence of habitat types (forest versus savanna) and diet types (crop‐raiding versus noncrop‐raiding) on the microbiome within L. africana. We found 58 bacterial orders, representing 16 phyla, across all African elephant samples. The most common phyla were Firmicutes, Proteobacteria, and Bacteroidetes. The microbiome of L. africana was dominated by Firmicutes, similar to other hindgut fermenters, while the microbiome of L. cyclotis was dominated by Proteobacteria, similar to more frugivorous species. Alpha diversity did not differ across species, habitat type, or diet, but beta diversity indicated that microbial communities differed significantly among species, diet types, and habitat types. Based on predicted KEGG metabolic pathways, we also found significant differences between species, but not habitat or diet, in amino acid metabolism, energy metabolism, and metabolism of terpenoids and polyketides. Understanding the digestive capabilities of these elephant species could aid in their captive management and ultimately their conservation.


| INTRODUC TI ON
The animal gut is generally comprised of trillions of microorganisms collectively known as the gut microbiome. In some species, the gut microbiome has been described as an evolved mutualism in which microorganisms assist their hosts with ecological interactions, immunity, nutrient uptake, energy acquisition, and digestion of materials that would otherwise be impossible (Russell, Dubilier, & Rudgers, 2014;Stevens & Hume, 1998;Vavre & Kremer, 2014). However, this is not a universal phenomenon, as some animal taxa do not appear to exhibit microbial dependency. Studies have shown that some invertebrates (Lucarotti, Whittome-Waygood, & Levin, 2011;Shelomi, Lo, Kimsey, & Kuo, 2013;Whittome, Graham, & Levin, 2007), and specifically caterpillars (Hammer, Janzen, Hallwachs, Jaffe, & Fierer, 2017), lack a definitive resident microbiome. Additionally, Hammer et al. (2017) provided evidence of low bacterial abundance in some vertebrates, including the brant goose Branta bernicla and the little brown bat Myotis lucifugus. Given the complexity of these systems and the vast diversity of species, it is important to understand the factors that influence the composition of microbial communities and the roles those communities play in the evolution of species.
For species that have a resident microbiome, diet is considered one of the primary factors influencing the diversity and composition of the microbial community (Muegge et al., 2011). Different food substrates promote growth of microbial taxa with specialized metabolic functions, which can lead to variation in taxonomic abundance (De Filippo et al., 2010;Scott, Gratz, Sheridan, Flint, & Duncan, 2013;Wang et al., 2013). For example, drastic changes in diet that occur across life stages cause concomitant shifts in the gut microbiome (Ilmberger et al., 2014;Kohl, Cary, Karasov, & Dearing, 2013). Similarly, dietary differences among European and African human populations drive the evolution of microbiomes containing unique species (De Filippo et al., 2010). It has also been shown that habitat can have a considerable influence on the gut microbiome (Amato et al., 2013); however, this effect is most likely explained by the association between habitat and the food resources available (Barelli et al., 2015).
In addition to diet, host phylogeny has been found to influence the microbiome, often through genes associated with the immune system (Blekhman et al., 2015;McKnite et al., 2012;Zhang et al., 2010;Zhao et al., 2013). This leads to significant differences in the bacterial community structure among broad groups of animals with varying dietary strategies: herbivores, carnivores, and omnivores (Ley et al., 2008;Nishida & Ochman, 2017). Because diet and phylogeny are often interrelated (Hale et al., 2018), their individual effects may be complex and highly confounded. Microbiome composition may be very similar among some dietary specialists despite their distant phylogenetic relationships, such as anteaters, aardvarks, and aardwolves (Delsuc et al., 2014). Conversely, some closely related species may possess common microbiota even though they differ substantially in diet (Delsuc et al., 2014).
Our understanding of the gut microbiome's response to diet, habitat, and host phylogeny has been primarily derived from controlled experiments on humans (Turnbaugh et al., 2009) and other model species (Murphy et al., 2010). Despite the benefit of experimental manipulation in model organisms, animals that have been reared entirely in sterile laboratory environments are not representative of their wild counterparts (i.e., laboratory mice; Abolins et al., 2017;Beura et al., 2016;Leung, Budischak, The, Hansen, & Bowcutt, 2018;Reese et al., 2016;Rosshart et al., 2017). Thus, the results of controlled experiments cannot always be directly applied to ecological understanding or conservation. The literature on host-microbiome interactions in wild taxa has grown over the last decade (Delsuc et al., 2014), with foci ranging from the influence of ecosystem factors on the gut microbial composition of Neotropical primates (Amato et al., 2016) to the role of local environmental exposures in shaping the bacterial communities of Galápagos iguanas (Lankau, Hong, & Mackie, 2012). However, our knowledge of the community composition of host-associated microbiomes and the mechanisms driving variation among free-ranging individuals, populations, and species is still limited (Amato et al., 2016).
The two sister species of African elephant, the African savanna elephant (Loxodonta africana) and the African forest elephant (L. cyclotis), represent a unique system in which to study the influences of phylogeny, diet, and habitat on microbial communities in wild populations. These species diverged from a common ancestor 2.6-5.6 mya (Rohland et al., 2010) and are the two most closely related taxa within the extant Proboscidea. Although they are closely related and have been found to hybridize in regions of range overlap (Clemens & Maloiy, 1982), they have developed distinguishing ecological and morphometric differences. L. cyclotis have substantially smaller body sizes than L. africana, their ears are more oval-shaped, and their tusks point downward unlike the outward curved tusks of L. africana (Short, 1981), likely as an adaptation to traversing their thickly forested habitat. The diet of L. cyclotis is primarily made up of fruits, leaves, and bark of forest trees (Short, 1981), whereas the diet of L. africana consists primarily of grasses and woody browse (Codron et al., 2011). Differences in morphology and dietary variation suggest that there may be disparities in metabolic demands between the two species and that they may utilize unique communities of gut microbes to aid in digestion.
Although L. africana and L. cyclotis are generally associated with savanna and forest ecosystems, respectively, each species has been observed living in environments more typical of the other. This generality in habitat use can be partly explained by migratory behavior (Galanti, Preatoni, Martinoli, Wauters, & Tosi, 2006), but it is more often attributed to the continued expansion and encroachment of human populations, which force wild elephants into competition for resources (Balmford et al., 2001;Barnes, 1996;Galanti et al., 2006;Hoare, 2000;Hoare & du Toit, 1999;Naughton-Treves, 1998;Pimm, Russell, Gittleman, & Brooks, 1995;Woodroffe & Ginsberg, 1998). Reported incidences of crop-raiding, the consumption of human-grown grains, fruits, and vegetables, are on the rise as human populations expand, and natural habitat is converted to agriculture, largely for the purpose of farming cash crops such as corn (Rode, Chiyo, Chapman, & McDowell, 2006;Sitati, Walpole, Smith, & Leader-Williams, 2003). Chiyo and Cochrane (2005) found that L. africana crop-raiders obtained up to 38% of their daily forage from agricultural crops. In general, elephants are thought to resort to crop-raiding due to increasing proximity of their natural habitat to agricultural land, anthropogenic degradation of their food resources, and greater palatability and nutritional value of cultivated plants (Sach, Dierenfeld, Langley-Evans, Watts, & Yon, 2019;Sukumar, 1990). Additionally, Finch (2013) suggested that crop-raiding behavior may be associated with decreased parasitic loads in L. africana. Despite the potential ecological benefits to elephants, crop-raiding is a high-risk behavior, as the destruction of property can cost the elephants their lives. With increased pressure from human encroachment, human-elephant conflict is at the forefront of management concerns, so it is necessary to understand the environmental and physiological factors impacting the ecology and evolution of the African elephants.
In this study, we examined the effect of host phylogeny as well as diet, through habitat-based diet type and crop-raiding, on the microbiome of free-ranging L. africana and L. cyclotis. By comparing sequences of the 16S rRNA gene amplified from African elephant fecal samples, we compared the diversity, taxonomy, and predicted metabolic function of microbial communities between the two elephant species. We then tested for differences between crop-raiding and noncrop-raiding individuals from both savanna and forest habitats within L. africana.
We hypothesized that variation in habitat type and crop-raiding behavior between populations and individuals of L. africana would result in significant differences in their microbiota. Additionally, we predicted that L. cyclotis would have a more diverse microbiome as a result of living in the highly diverse tropical forests of Central Africa.

| Study sites and sample collection
We collected approximately 20 g of fecal sample from elephant dung piles within 12 hr of excretion (based on sample moisture and odor) in Kenya and Gabon, in areas where historical or contemporary genetic hybridization between L. africana and L. cyclotis was unlikely ( Figure 1a). All samples were collected during the wet season at all sites to minimize the effect of environmental variability on sample quality. Using a plastic tape measure, we also took up to three bolus circumference measurements from sampled dung piles to serve as a proxy for age using the standards established by Morrison, Chiyo, Moss, and Alberts (2005) for L. africana and Schuttler, Whittaker, Jeffery, and Eggert (2014) for L. cyclotis. In compliance with requirements under USDA-APHIS permit #128686, the tubes containing samples were boiled to prevent the transmission of pathogens. They were stored in Queen's College Preservation Buffer (20% DMSO, 0.25 M EDTA, 100 mM Tris, pH 7.5, saturated with NaCl; Amos, Whitehead, Ferrari, Payne, & Gordon, 1992) at room temperature in the field and exported to the United States where they were stored at −20°C prior to DNA extraction. In a comparison of storage methods for fecal samples, Kawada, Naito, Andoh, Ozeki, and Inoue (2019) found that the fecal microbiota detected in samples stored in this buffer clustered with fresh samples in a PCoA and found no effect of storage buffer on alpha or beta diversity. (n = 20) were categorized as forest habitat as this region is characterized by semideciduous, dry deciduous, and acacia woodlands (Sitati et al., 2003). Samples collected from the Loita Plains of the Narok District (n = 15) were classified as savanna habitat as the area is characterized by dwarf shrub and whistling thorn (Acacia drepanoligium) grasslands (Serneels & Lambin, 2001). In addition, we collected samples from crop-raiding events to compare the effects of agricultural products (primarily maize) and natural vegetation on the gut microbiome. Our field team was notified the morning after crop-raiding events by a World Wide Fund for Nature scout team who worked with local farmers on crop-raiding issues. On the same day as the reported incident, we collected elephant fecal samples from raided fields and classified them according to surrounding natural habitat type to allow comparative F I G U R E 1 Locations of the study areas including Loxodonta africana (red) and L. cyclotis (blue) distributions across Africa (a); the Narok and Transmara Districts surrounding Maasai Mara National Reserve in Kenya (b); and Lope National Park in Gabon. Alberts (2003) that include increased volume of starting material as well as increased proteinase K volume and digestion time to maximize the final DNA concentration. They were then genotyped at 10 (L. africana) and 12 (L. cyclotis) microsatellite loci (Finch, 2013;Schuttler, Whittaker, et al., 2014). To determine the sex of each genotyped sample, we amplified two Y-specific fragments (SRY1 and AMELY2) and one X-specific fragment (PLP1), a sexing technique described by Ahlering, Hailer, Roberts, and Foley (2011). For all extractions and PCRs, we used both negative controls to detect contamination of the reagents and positive controls to standardize allele scoring. We computed pairwise relatedness between samples using ML-RELATE (Kalinowski, Wagner, & Taper, 2006); when comparisons yielded a coefficient of relatedness >0.25, one of the two samples was omitted. Exception was given to subadult and adult males that were sampled in different habitats than related females, as males are the dispersing sex. Subadult and adult males found in different habitats were assumed to have dispersed; after dispersal, we assumed they were making dietary choices independent of their family groups. For each sample, sex and age of elephant were determined as a possible variable for microbiome differentiation.

| Microbial extraction and sequencing
For microbial DNA extraction, we used bead beating (Yu & Morrison, 2004) with modification to accommodate double the starting material (0.50 g). From each sample, the hypervariable V4 region (253 bp) of the bacterial and archaea 16S rRNA gene, recommended by Liu, Chen, Wang, Oh, and Zhao (2005), was amplified using PCR forward primer 515F (5′-GTG CCA GCM GCC GCG GTA-A3′) and reverse primer 806R (5′-GGA CTA CHV GGG TWT CTA AT-3′; Caporaso et al., 2010;Gilbert et al., 2010). We modified each primer to include Illumina forward, reverse, and multiplex sequencing primers and added a unique 6 bp barcode to each reverse primer (Bartram,

| Microbial bioinformatic pipeline
Raw sequencing reads were processed using qiime 2 v.2019.1 (Caporaso et al., 2010). Metadata files were verified for formatting using keemei (Rideout et al., 2016). Paired-end data were joined using vsearch (Rognes, Flouri, Nichols, Quince, & Mahé, 2016), and quality scores filtered within quality-filter (Bokulich et al., 2013) plugins using default parameters. Joined sequences and their corresponding quality scores were visually inspected on the qiime2 interactive viewer to obtain optimal trim length. Sequences were filtered using these quality scores, singletons and chimeras were removed, and resulting sequences were length trimmed with the plugin deblur (Amir et al., 2017)  rarefied to an even sampling depth using the package qiime2R v. 0.99.11 (Bisanz, 2018).

| Core microbiome characterization
To identify the taxonomic level that was most appropriate to characterize the core microbiome, we calculated the proportion of unique OTUs that were successfully classified at the previously described 99% certainty to any given taxonomic level (kingdom, phylum, class, order, family, genus, or species). We then compared the proportion of classified OTUs at each taxonomic level with average confidence to determine the optimal taxonomic level at which to characterize the microbiome of all samples, which we defined as the most specific taxonomic level that maintained a high average confidence of classification.
We characterized the core microbiome based on habitat, pres-

| Diversity analyses
We calculated alpha diversity as the Shannon diversity index, in vegan v. 2.5.4 (Oksanen, Blanchet, Friendly, Kindt, & Legendre, 2017) for habitat and diet (Subset A) and phylogeny (Subset B).
We compared average Shannon Diversity values among habitat, diet, and phylogeny treatments using GLMM in the package Lme4, including elephant sex in each model as a random effect.
For analysis of diet and habitat, we first tested for the presence of a diet-habitat interaction before assessing main effects in a separate GLMM.
We evaluated beta diversity between diets using data Subset A and phylogeny using Subset B by calculating Bray-Curtis dissimilarity and running 9999 permutations in PERMANOVA (Anderson, 2001) vegan v. 2.5.4 (Oksanen et al., 2017). For all comparisons, we included sex as a stratum that is akin to setting a random effect in a linear mixed model; for diet and habitat, we first tested for a diet-habitat interaction before assessing main effects in a separate PERMANOVA. We visualized beta diversity with nonparametric multidimensional scaling (NMDS) in vegan 2.5.4. Finally, we assessed differences in within-population variability between diet and phylogeny using the centroid method of the beta dispersion test in vegan 2.5.4.

| Sequencing results
Sequencing of the 16S rRNA gene from these 48 samples produced
Within all phyla, we found a total of 58 microbial orders at varying proportions within African elephants; however, only 18 of these orders were shared by all individuals (Table S2). An additional six orders were shared at 100% frequency among L. cyclotis individuals, and two additional orders were shared among all L. africana individuals (Table S2). When evaluating the effect of diet, 17 orders were shared between crop-raiders and noncrop-raiders. While four additional bacterial orders were shared among all crop-raiders, no additional orders were shared among all noncrop-raiders (Table S2).
When evaluated by habitat type, there were 17 shared microbial orders, and while savanna habitat had another six orders found in all individuals, forest habitat had none (Table S2).
The most abundant orders were assessed within the three phyla with the highest average relative abundance across all samples: Bacteroidetes (

| Alpha diversity
Shannon diversity for all samples ranged from 2.047 to 6.500, where the average for L. africana was 4.681 ± 0.996 and the average for L. cyclotis was 4.186 ± 1.49. Shannon diversity did not differ significantly between species (GLMM; p = .449; Figure S2A). In our twoway mixed-effect models for diet types, we found no significant interaction between habitat and diet types (GLMM; p = .757). Within diet types, the effects of crop-raiding (GLMM; p = .126; Figure S2B) and habitat (p = .239; Figure S2C) were also not significant.

| Beta diversity
We found significant differences in beta diversity between species (PERMANOVA; p = .001; Figure 3a). Within diet types, the effects of crop-raiding (p = .007; Figure 3b) and habitat (p < .001; Figure 3c) were also significantly different between groups within L. africana.

| The elephant microbiome
The core microbiome among mammalian lineages varies a great deal in the relative abundance of each microbial phylum; however, across mammalian taxa, four bacterial phyla appear to be key drivers: previously, but only in captive juveniles, a three-week-old Asian elephant (Ilmberger et al., 2014), and a seven-month-old L. africana (Ley et al., 2008). However, all of our samples for L. cyclotis were from adults, leaving us to hypothesize that this may be a result of the dietary difference in L. cyclotis, which is higher in fruit, and therefore has a higher proportion of simple carbohydrates and a lower fiber content (Moermond & Denslow, 1985) than the primarily woody browse and grasses that compose the L. africana diet.
Our ability to predict functional differences in the microbial communities of the elephant species was limited by the fact that the KEGG database primarily reflects information gleaned from studies of humans and model organisms. There have been very few studies of the microbiome in wild species, especially those such as L. cyclotis, whose habitats are remote and inaccessible. Thus, we are only able to make limited inferences. In our dataset, L. africana was significantly higher for metabolism of terpenoids and polyketides (p = .006) and amino acid metabolism (p = .006). Although structurally diverse, the dominant types of polyketides are widespread as antibiotics, antifungals, and antiparasitics. L. cyclotis was significantly higher for energy metabolism (p = .006). These functional differences may indicate a difference in energy allocation and metabolic capabilities and illustrate the need for metabolomic comparisons between the species.

| Habitat and diet
While both diet and body size can affect alpha diversity in the microbiome of a broad variety of taxa, the most important predictor of alpha diversity has been found to be gut physiology (Reese & Dunn, 2018). Animals with simple guts, such as carnivores, typically have lower microbial richness than foregut ruminants or hindgut fermenters. As large-bodied hindgut fermenters, African elephants would be expected to have the relatively high alpha diversity we found in this study ( Figure S2). We were surprised, however, that the Shannon diversity index revealed no significant differences in the abundance or evenness of microbial taxa between species, diets, and habitats. If microbiome diversity is correlated with diet diversity, African elephants that live in the highly diverse tropical forest environments would be predicted to have higher alpha diversity.
Our finding that differences in diet and habitat were not associated with differences in alpha diversity is in agreement with the results of Kartzinel, Hsing, Musili, Brown, and Pringle (2019), who found that in general species with diverse diets did not have the most diverse microbiomes. Taken together, our studies suggest that we are only beginning to learn about the mechanisms that underlie the diversity of the microbiome.
Within L. africana, we found overall differences, as reflected by beta diversity, between the microbiota of crop-raiding and noncrop-raiding elephants, as well as between individuals that live in savanna and forest habitats. Although there were few statistically significant differences found within microbial orders between groups, there may be biologically significant differences.
For instance, one of the secondary metabolites of a species of Wautersiella, which differed between Laf:S + CR and all other groups, has been found to be active against nematodes in laboratory tests (Chen, Wang, Zhang, & Li, 2015). In addition, savanna elephants that live near human populations encounter different environmental conditions than those that live in less anthropogenically affected habitats. To the extent that the microbiome is involved in the stress response (Maloney, Desbonnet, Clarke, Dinan, & Cryan, 2014), small differences in the encounter rate of different bacterial strains may affect elephants. For instance, the genus F I G U R E 4 Mean metabolic contribution, calculated as the proportion of an individual sample's metabolic profile comprised by a given pathway, and standard deviation of KEGG metabolic pathways that were significantly different between African elephant species. The primary function of the represented pathway is listed on the x-axis. (*p < .05) Sporosarcina, one of the genera within phylum Firmicutes that was trending toward significance (GLMM; p = .094) between L. africana and L. cyclotis, includes a species (S. urease) that is found in high densities in soils that are subject to animal urine (Pregerson, 1973).

| Implications for species conservation
To our knowledge, this study is the first to characterize differences between the African elephant species and to incorporate the more recently described L. cyclotis in a microbiome study. We also exam-  (Clauss, Loehlein, Kienzle, & Wiesner, 2003). To our knowledge, no metagenomic or metabolomic studies have been conducted on wild populations.
As we continue to advance our understanding of host-microbiome associations, we need to increase our focus on how current and new information will aid in the conservation of threatened taxa (Redford, Segre, Salafsky, del Rio, & McAloose, 2012). Studies of humans and model organisms have established links between the microbiome and health, not only through the effects of pathogens but also through alterations in the composition of the microbiota. In this study, we provide evidence that phylogeny, diet, and habitat all independently influence the gut microbiome of the two African elephant species, both of which are considered endangered. Furthermore, the dietary effect we observed in these elephants is in part attributed to crop-raiding, a behavior that often occurs as a major component of human-elephant conflict. Cultivated crops may be more palatable and, in some cases, more nutritious than analogous wild plants (Sukumar, 1990). As African elephants continue to adapt, both in diet and in habitat tolerance, to expanding human populations, conservation management will be essential to coexistence. For elephants and other wildlife species, conservation management may well depend on gaining a better understanding of the effects of alterations of the microbiome on reproduction and survivorship.

ACK N OWLED G M ENTS
We thank the WWF scout team for data collection in the field; Ryan

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interest to declare. Writing-review & editing (equal).