Discussion
There has been a robust debate about whether there is a generalizable
effect of changes in biodiversity (as a consequence of habitat loss
and/or fragmentation) on the emergence and prevalence of infectious
diseases (Levi et al., 2016; Ostfeld, 2013; Ostfeld & Keesing, 2013;
Randolph & Dobson, 2012; Wood et al., 2014; Wood & Lafferty, 2013).
This debate has been stymied by a lack of landscape-scale empirical data
across land-use gradients and observational approaches that are unable
to deduce the mechanistic underpinning to changing disease risk. This
can be especially complex for vector-borne pathogens where land-use
change can differentially influence hosts, vectors, and pathogens
(Burkett-Cadena & Vittor, 2018). We addressed this debate with a
landscape scale epidemiology approach across a forest habitat loss
gradient within the world’s largest tropical deforestation frontier
induced by large-scale agricultural commodity production.
By using DNA metabarcoding of sandflies and their vertebrate bloodmeals,
we were able to link medically important hosts and vectors to
deforestation at large scales (56,775 sandflies collected from 27 light
traps set for 3 days at each of the 39 sites). Metabarcoding allowed us
to identify thousands of sandflies at the species or genus level so that
the ecology of vectors and non-vectors could be investigated. Species
level sandfly data is usually only possible with painstaking
morphological identification of sandfly species, which requires rare
taxonomic expertise. In contrast, previous research has shown that DNA
derived from sandflies are a good measure of the vertebrate diversity in
this landscape and requires significantly less resources and
identification effort (Massey et al., 2022). Ideally, sample pools would
have included fewer individuals, or we would have a larger sample size,
so that direct vector to host comparisons and interaction networks could
be quantified without data contamination from multiple sandflies
sequenced as part of the same pool. However, metabarcoding of sandfly
pools allowed for sufficient cost reduction to allow sequencing of over
50,000 individual sandflies to taxonomically identify both sandfly
species and any vertebrate bloodmeal remnants.
Fundamentally, sandfly responses to deforestation reported here were
nuanced. Although total sandfly abundance did not vary with
deforestation, the relative abundance of vector species ordinated in the
direction of greater forest cover with the most significant positive
response to forest cover found with Psychodopygus davisi , a known
vector of Leishmania braziliensis (Fig. 4). Further, the
probability of a sandfly pool containing any vector or the dominant
sandfly vector genus, Nyssomyia spp., was higher at sites
surrounded by less cattle pasture (Fig. 4). Psychodopygus davisiand Nyssomyia spp. are vectors known for transmitting species ofLeishmania responsible for cutaneous leishmaniasis in Brazil,
which while treatable, causes disfiguring and painful skin lesions. We
found no significant positive responses of sandfly density or the
probability of finding a vector species to increasing deforestation
which runs counter to the hypothesis that vector amplification (as a
consequence of increased host density) occurs in response to
deforestation.
However, disease risk is a product of both vector and host ecology.
Despite the heavily deforested nature of this region, we found that this
landscape supports a large diversity of terrestrial and arboreal
vertebrate species. The majority of sylvatic vertebrate taxa we detected
using sandflies as a source of iDNA are known host species forLeishmania parasites. The responses of sylvatic vertebrate taxa
in sandfly bloodmeals were driven by the high prevalence of armadillos
in the genus Dasypus, particularly the disturbance-tolerant
nine-banded armadillo, D. novemcinctus , which was by far the most
common source of bloodmeals (Fig. 3). These data suggest that sandflies
strongly select for armadillos, which are among the most important hosts
for Leishmania spp., which cause leishmaniasis in humans
(Christensen et al., 1982; Kocher et al., 2017; Lainson & Shaw, 1989).
Armadillos were the only prevalent vertebrate taxa to show a significant
relationship to the deforestation gradient with increased probability of
finding D. novemcinctus in sandfly bloodmeals as forest cover
decreased (Fig. 4). Bipartite networks, which directly measured sandfly
× vertebrate interactions, and the Leishmania -positive samples
suggest that armadillos drive the feeding ecology of sandflies andLeishmania transmission dynamics across a degraded forest
landscape. However, we excluded domesticated species from our analyses
because (i) we were primarily concerned with how land use change
influences wild vertebrate communities, and (ii) possible routine lab
contamination. Even with conservative read thresholds, we found that
domestic dog (Canis lupus familiaris ) occurred in just over 20%
of the sandfly samples and was ubiquitous across the landscape (Appendix
SI: Fig. S1), including in some samples that also containedDasypus (Appendix SI: Fig. S3). Given that pathogen spillover to
humans increases when domesticated species are in close proximity to
sylvatic hosts, domestic dogs may play a key role in peridomestic
transmission of Leishmania to humans. Our findings support this
potential mechanism of zoonotic disease transmission primarily through
the observed pattern of armadillo and dog co-occurrences from sample
pools across the study area.
While our sampling scheme allowed us to sample across a landscape-scale
deforestation gradient, it is important to note that we did not sample
the extremes of vast tracts of continuous, undisturbed forests (compared
with Kocher et al. 2022) nor entirely deforested landscapes lacking
forest remnants. Consequently, while there was a gradient of
deforestation across our study region, it is likely that the current
forest landscape structure has not resulted in the same level of
extirpation of vertebrate species as other studies have documented.
Instead, our study system represents a landscape-scale deforestation
gradient resulting from rapid and recent forest conversion into seed
crop agriculture. As discussed previously, this expansion of human
activity into tropical forests can alter ecological communities and
species interactions particularly at transition zones between forests
and peri-urban areas (Aguirrea & Taborb, 2008; Roque & Jansen, 2014),
thereby potentially increasing the risk of infectious disease emergence
from wildlife reservoirs and vectors into domestic vertebrate hosts
and/or humans (Lambin et al., 2010; T. Lima et al., 2017; McCauley et
al., 2015; Vanwambeke et al., 2007). Our findings support this given the
high diversity of vertebrate hosts and sandfly vectors found across the
landscape and the lack of localized extinctions at even the most forest
degraded sites.
In summary, we found that it was most important to examine the responses
of individual species even when investigating the generality of
biodiversity and disease risk to land-use change. While overall sandfly
abundance (including non-vectors) was unrelated to deforestation,
sandfly vectors were more strongly associated with more intact forest
landscapes (either more forest or less pasture), which was largely
driven by the response of the dominant vector taxa (Nyssomyiaspp. ). Likewise, changes in the relative abundance of sylvatic
hosts (namely armadillos) were apparent despite no significant response
of the aggregate vertebrate host community across the deforestation
gradient. Additionally, samples that tested positive for the presence ofLeishmania species also failed to show any response to forest
cover, pasture cover, or distance to the urban center, suggesting thatLeishmania transmission can occur across both intact and degraded
forests in this system. In conclusion, the complex combined responses of
vectors and hosts within the context of partly deforested landscapes did
not support the generality of the ‘dilution effect’ hypothesis. However,
patterns of individual species responses to deforestation and vector ×
host interactions across the deforestation gradient show disease risk
consequences of forest conversion and increased human encroachment into
Amazonian primary forest.