**4. Discussion**

We used fine-scale data on woody-plant and large-herbivore communities to evaluate local biodiversity linkages across trophic levels. The flora with the most abundant woody stems also supported the greatest taxonomic and phylogenetic diversity of woody plants. However, this abundance and diversity of plants did not necessarily translate into a greater abundance or diversity of herbivores. Across habitats, the low plain had a high abundance of both plants and large herbivores, but the slope had a comparably high abundance of plants with a markedly lower abundance of herbivores (Figure 2). Both within and across habitats, there was also a negative relationship between local plant and herbivore-species richness (Figure 3). While plant and animal communities differed across habitats, this pattern was not reinforced by phylogeny (Figure 4). In contrast with prior studies reporting phylogenetic associations between plants and insect herbivores [40,41], our results reveal how compositional turnover in some plant and animal assemblages may be random with respect to phylogeny, even when habitat filtering is highly nonrandom with respect to plant and animal functional traits across disparate evolutionary lineages [20].

Results were not generally consistent with the More Individuals or Resource Specialization hypotheses. The More Individuals Hypothesis posits that greater resource availability enables more individuals to establish, and hence more species of consumers to co-occur [16]. In contrast to positive correlations between the abundance and diversity of some plant and herbivore assemblages that have been studied at local-to-global scales [14,42,43], our results revealed landscape-level contrasts between the abundance and diversity of plants and large herbivores, as well as surprising negative correlations in their species richness across 33 sites. Under the Resource Specialization Hypothesis, plant diversity should enable more specialized consumer species to establish in the community. For example, insect-species richness may increase with the richness and phylogenetic diversity of grassland and forest plants [42,43]. However, ungulate herbivores that require relatively large quantities of food are unlikely to be attracted to or excluded from a local community based on similar mechanisms involving narrow feeding specializations. The low plain had relatively high plant abundance and species richness compared to the plateau, but it only revealed higher average herbivore abundance and not higher herbivore richness. Further contradicting both hypotheses, the slope and low plain had comparable levels of plant abundance and species richness, ye<sup>t</sup> the slope supported lower animal abundance and richness. We thus consider other non-mutually exclusive mechanisms to explain two striking patterns in our data: (*i*) the distinct flora and fauna across the gradient from heavy-clay vertisol ("black-cotton") soils of the plateau through the sloping transition to the sandy ("red") soils of the low plain and (*ii*) the negative correlation between plant and animal-species richness.

First, the flora of the plateau and low plain differed strikingly, but both were dominated by *Acacia* spp. (Figure 4). *Acacias* are a diverse suite of savanna trees that browsing ungulates eat extensively [19]. The clay soils of the plateau are dominated by *Acacia drepanolobium* whereas sandy soils of the low plain are dominated by *A. mellifera*, *A. etbaica*, and *A. brevispica* (Figure 1). Some herbivores were common in multiple habitats during the study period (e.g., giraffe, elephant, zebras, eland, and buffalo), whereas others were recorded almost exclusively in one habitat (hartebeest, oryx, and Grant's gazelle in plateau; dik-dik and waterbuck in low plain; Table S3, Figure S3) [44]. The subset of herbivores common across habitats spanned disparate lineages, whereas many of those with strong associations were close relatives from the Bovidae family (Figure 1). Thus, although the flora and fauna of each habitat is taxonomically different, taxa from diverse lineages share the functional abilities to occupy their shared habitats.

Second, there was a strong negative correlation between the richness of plants and animals together with contrasting patterns of plant and animal abundance and phylogenetic diversity across habitat types. Relatively high local plant abundance and diversity could be either a cause or consequence of relatively low herbivore abundance and diversity [45,46]. High plant abundance could cause of low animal abundance and diversity if thickets of woody plants are avoided by herbivores due to risk of predation. Many wild large

herbivores perceive risk associated with the presence of predators and prefer good sight lines in habitats with low tree density [35,44]. Topographic features, particularly slope and convexity, further interrupt sightlines such that steep habitats with high tree and predator densities, coupled with challenging terrain, could deter herbivores from steep slopes, thereby alleviating top-down pressure on plants [4,36,47]. Rapid changes in topography can impose particularly strong mechanical and energetic constraints on megaherbivores such as elephants and giraffes, which avoided the steep slopes and have been shown to inflict markedly less damage on trees in this habitat [36].

Long-term herbivore-exclusion experiments at Mpala highlight the ability of megaherbivores to exert strong top-down effects on plant abundance and diversity. For example, elephants can reduce the availability of forbs that comprise a substantial portion of browsing mesoherbivore diets [19,48,49] while also increasing visibility and the availability of grasses for grazers [19,50]. The defensive strategies employed by abundant tree species on the slope sugges<sup>t</sup> plant-herbivore sensitivity to top-down regulation that could contribute to their differential abundance across habitats. For example, some tree species that were relatively abundant on the steep slope invest heavily in antiherbivore defenses, including *Croton dichogamus* (Euphorbiaceae; produces noxious latex) and *Euclea divinorum* (Ebenaceae; produces tough, fibrous, and tannin-rich leaves) [19,51]. These defenses could deter consumption by the smaller-bodied ruminants that are abundant on the slope (e.g., dik-dik), but could be less effective against megaherbivores such as elephants and giraffes that otherwise frequently consume them elsewhere [52]. Whereas many recent studies focus on behaviorally mediated fear responses and trophic cascades [4,47], our results reinforce the important and spatially heterogeneous indirect effects that megaherbivores can have on vegetation [12,44,48–50,53].

The same physical habitat features that obstruct herbivore sightlines and could lead to avoidance of thickets can have methodological implications for camera-trap studies. We assumed approximately equivalent 25-m<sup>2</sup> radii detection ranges for camera traps at each grid site. However, reduced sightlines could lead to underestimates of the abundance and diversity of large herbivores in dense vegetation or on slopes [54]. This possibility suggests that the contrast between animal abundance and diversity observed in the plateau and low plain habitats could be conservative, since the low plain had both higher stem density and higher herbivore RIA (Table 1). However, the slope habitat had especially high stem density and complex topography, coupled with comparatively low herbivore RIA (Table 1), suggesting that further camera-trapping efforts may reveal additional animal use of plots in this habitat. It would be interesting to model seasonal variation in both detection rates and ungulate habitat associations across the plot: habitat associations might become stronger in the dry season when ungulates have the greatest need to monopolize the specific resources for which each is best able to compete, or alternatively these associations may be dampened if food depletion forces them to access the same reserves of riskier or less-preferred resources that accumulate on slopes and in thickets [4,5,55].

Our analyses combined a comprehensive woody-plant survey with DNA barcoding, which was methodologically useful both for refining plant identifications and modeling their phylogenetic relationships [23]. Data that support detailed investigations into ecological linkages between plants and herbivores—whether these linkages amount to spatial cooccurrence, trophic interactions, or both—are needed because these integrations have strong influences on the structure and function of both terrestrial and aquatic ecosystems worldwide [56,57]. Yet while the ecological and evolutionary processes that generate patterns of phylogenetic signal in relatively specialized species interactions have received substantial recent attention (e.g., host–microbiome [58], host–parasite [59]), less attention has been paid to phylogenetic structure in broader types of species interactions (e.g., plant– mycorrhiza [60], seed disperser networks [61], vertebrate trophic networks [19]) or habitat associations [62]. Because all species interactions and co-occurrence networks are subject to environmental and biological filters that act on species' functional traits, patterns of habitat use may generate phylogenetic signal in cooccurrence networks across trophic levels more

frequently than currently realized. Our analysis focused on a uniquely fine-scale example of habitat associations across trophic levels, including phylogenetically diverse communities of woody plants and large herbivores. Results sugges<sup>t</sup> no strong phylogenetic structure to the filtering of plant and animal traits that determine their habitat associations, since closely related species had divergent habitat associations and members of divergent clades often co-occurred.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/d14030219/s1, Figure S1: Maps of the study plot showing fine-scale major topographic characteristics, including (**a**) elevation, (**b**) convexity, (**c**) slope, and (**d**) topographic wetness index (TWI); Figure S2: Sample-based rarefaction of (**a**) woody-plant and (**b**) large-herbivore communities to compare total species richness within habitats based on our 33 grid sites (the actual numbers of sampling units in each habitat are shown as large points); Figure S3: Indicator species analysis for (**a**) woody plant and (**b**) large herbivore species across habitats. Points are colored by habitat and closed circles correspond to habitat associations (positive) or avoidance (negative) that differ significantly from 0; Table S1: Summary of plant species and DNA barcodes. The table matches names of species in our phylogeny with names used in the ForestGEO dataset, includes information on new DNA barcodes presented in this paper, descriptions of how species were added to the existing phylogeny [23], and the abundance of species across the plot as well as within our 33 grid sites; Table S2: Summary of plant abundances across 33 grid sites in this analysis. The table includes grid information (corresponding camera ID number, habitat, location in UTM) with site-specific plant abundance (total stems, individual trees) and diversity values (richness, *ses*MPD, *ses*MNTD); the *p*-values for site-specific phylogenetic diversity metrics are provided (significant positive values indicate phylogenetic overdispersion and negative values represent clustering). The site × species matrix based on the count of individual trees is then provided; Table S3: Summary of large-herbivore RAI values across 33 camera sites. The table includes grid information (camera ID number, habitat, location in UTM) with site-specific animal abundance (summed RAI) and diversity values (richness, *ses*MPD, *ses*MNTD), and the *P*-values for site-specific phylogenetic diversity are provided (significant positive values indicate phylogenetic overdispersion and negative values represent clustering). The site × species matrix based on RAI values is then provided; Table S4: Raw large-herbivore cameratrap records for the study period. For each photo, the table includes grid location, herbivore species, camera deployment date, photo-capture date and time, season (binned "wet" and "dry"), and the total camera-trap deployment days (in both wet and dry seasons). File S1: Nexus-formatted phylogeny for the 47 woody plant species present at the grid sites and used in analyses. References: [23,63–65] are cited in Supplementary Materials.

**Author Contributions:** Conceptualization, P.T.F. and T.R.K.; methodology, P.T.F., P.M.M. and D.K. (David Kenfack); P.T.F. and T.R.K.; investigation, P.T.F., R.O.A., D.K. (Duncan Kimuyu), P.L.E., M.M. and T.R.K.; data curation, P.T.F., B.A.G. and T.R.K.; writing—original draft preparation, P.T.F. and T.R.K.; writing—review and editing, all authors; funding acquisition, D.K. and T.R.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by NSF DEB-1930820.

**Institutional Review Board Statement:** Not applicable.

**Data Availability Statement:** New plant DNA barcodes are available on Genbank (Accessions: *trn*L-F, OM238064-OM238067; *rbc*La, OM238062-OM238063), the full plant DNA barcode library is available on BOLD ('DS-UHURUR3- , https://doi.org/10.5883/DS-UHURUR3), and the plant phylogeny is available in File S1.

**Acknowledgments:** We thank the Government of Kenya for granting permission to conduct this research, Mpala Research Centre for its staff and logistical support, and the Institute at Brown for Environment and Society for supporting student participation in the work. The Mpala ForestGEO plot is a collaborative project of the National Museums of Kenya and the Mpala Research Centre in partnership with the Forest Global Earth Observatory (ForestGEO) of the Smithsonian Tropical Research Institute (STRI). Funding for the first census was provided by ForestGEO. Smithsonian Tropical Research Institute, ForestGEO, and International Foundation for Science (D/5455-2) provided additional funding support for this work.

**Conflicts of Interest:** The authors declare no conflict of interest.
