Nature | Vol 584 | 20 August 2020 | 401A trait-mediated explanation is also supported by our finding that
differential host and non-host species responses to land use are most
clearly detected when comparing across large clades with a wide
diversity of life histories—such as rodents, passerines and, notably,
mammals overall (Extended Data Fig. 5). By contrast, clades that
are generally longer-lived and larger-bodied (for example, primates
and carnivores) show more idiosyncratic or negative responses to
landscape disturbance (Fig. 3 ).
Overall, our results indicate that the homogenizing effects of land
use on biodiversity globally^8 have produced systematic changes to
local zoonotic host communities, which may be one factor underpin-
ning links between human-disturbed ecosystems and the emergence
of disease. By leveraging site-level survey data, our analyses reflect
community changes at the epidemiologically relevant local-landscape
scale^21 , negating the need to ignore community interactions or general-
ize ecological processes to coarser spatial scales (a typical limitation
of global studies that can confound or mask biodiversity–disease
relationships^29 ). Our results reflect potential zoonotic hazard, because
proximity to reservoir hosts is not sufficient for spillover^30 and emer-
gent disease risk will depend on contextual factors (for example,
pathogen prevalence, intermediate host and vector populations,
landscape structure, socioeconomics) that may synergistically or
antagonistically affect transmission dynamics and exposure rates^11.
Nonetheless, land use also predictably affects other factors that
can amplify within-species and cross-species transmission^31 (such
as resource provisioning^10 and vector diversity^32 ), and increases the
potential for human–wildlife contact^12 : for example, human popula-
tions are consistently higher at disturbed sites in our dataset (Extended
Data Fig. 8). The global expansion of agricultural and urban land that is
forecast for the coming decades—much of which is expected to occur
in low-and middle-income countries with existing vulnerabilities to
natural hazards^17 —thus has the potential to create growing hazardous
interfaces for zoonotic pathogen exposure. In particular, the large
effect sizes but sparser data availability for urban ecosystems (espe-
cially for mammals; Extended Data Fig. 4) highlight a key knowledge
gap for anticipating the effects of urbanization on public health and
biodiversity. Our findings support calls to enhance proactive human
and animal surveillance within agricultural, pastoral and urbanizing
ecosystems^33 ,^34 , and highlight the need to consider disease-related
health costs in land use and conservation planning.Psittaciformes Carnivora PrimatesPasseriformes Chiroptera Rodentia CetartiodactylaSecondaryManaged Urban SecondaryManaged Urban SecondaryManaged UrbanSecondaryManaged Urban–1000100200–1000100200Land useAbundance difference from primary land (%)Non-host
HostspeciesTo tal Zoonotichosts Sites126 36 1,13358 23 82733 11 604177 36 34638 17 5831,868 62 4,30277 22 2,500Abundance difference from primary land (%)SecondaryManaged Urban SecondaryManaged Urban SecondaryManaged UrbanLand use Land useFig. 3 | Effects of land use on species abundance of mammalian and avian
zoonotic hosts and non-hosts. Points, wide and narrow error bars show
average difference in species abundance (posterior median, 67% and 95%
quantile ranges, respectively, across 500 bootstrap models to account for host
status uncertainty) in secondary, managed and urban sites relative to a primary
land baseline (dashed line). Differences are estimated across all host and
non-host species in each mammalian or avian order. For mammals, zoonotic
host status was defined strictly (direct pathogen detection, isolation or
confirmed reservoir status), and urban sites were excluded owing to sparse
urban sampling (only two studies; in addition, no non-host primates were
recorded in managed land, and urban 95% quantile range for Psittaciformes is
not shown owing to high uncertainty). Abundance differences were predicted
using a hurdle-model-based approach to account for zero-inf lation (combining
separately fitted occurrence and zero-truncated abundance models; see
Extended Data Fig. 5, Methods). The table shows per-order numbers of species
in the dataset (between 8% and 35% of the total described species in each
order), known zoonotic hosts (before bootstrap) and sampled sites.
Silhouettes obtained from PhyloPic (http://phylopic.org/).