Using iDNA to determine impacts of Amazonian deforestation on Leishmania hosts, vectors, and their interactions

Abstract

Background: There is debate concerning whether there exists a generalizable effect of land-use change on zoonotic disease risk. Strong data informing this debate are sparse because it is challenging to establish direct links between hosts, vectors, and pathogens. However, molecular methods using invertebrate-derived DNA (iDNA) can now measure species composition and interactions from vector samples at landscape scales, which has the potential to improve mechanistic understanding of the effects of land-use change on zoonotic disease risk.

Methodology/principal findings: We used iDNA metabarcoding of sandflies to disentangle the relationships between Leishmania parasites, sandfly vectors, and vertebrate hosts. We paired these samples with iDNA metabarcoding of carrion flies to survey vertebrates independent of sandfly feeding preferences. We collected sandflies and carrion flies at forest sites across a deforestation gradient in the southern Amazon 'Arc of Deforestation', which exemplifies global patterns of deforestation due to agricultural expansion. We used a series of models to test whether sandflies and the vertebrate they feed upon were influenced by deforestation, which we measured using percent forest cover, percent pasture cover, and distance to the major urban center. We found that vectors were encountered less frequently in forests surrounded by pasture. We also found that the probability of a Leishmania host/reservoir being detected in sandfly bloodmeals was quadratically related to local forest cover, with the highest probability found at sites with intermediate levels of deforestation. Hosts were also detected most often with carrion flies at sites with intermediate forest cover, suggesting that increased host availability rather than feeding preferences was responsible for this result. Domestic dogs and the nine-banded armadillo, Dasypus novemcinctus, were the most prevalent hosts found in the sandfly iDNA data.

Conclusions/significance: Our results did not support the generality of the 'dilution effect' hypothesis. However, important vectors and hosts showed consistent responses to deforestation and our findings suggest that interactions between domestic dogs and sylvatic hosts are a pathway for zoonotic disease transmission in human impacted tropical forests.

Fig 1. Map of the study area located near Sinop, Mato Grosso, Brazil.

Sinop is an urban area (shown in dark gray on the map) with more than 160,000 people in 2015. The study region is located at the southern fringes of the Amazon forest biome, just north of the northern boundary of the Cerrado biome, and the landscape is described by a mosaic of closed-canopy forest, cerrado scrublands, croplands, and cattle pastures. Our 39 study sites and a 2500-m circular buffer are shown as solid black circles. The base layer map and the land-use, land cover classifications were sourced from MapBiomas – Collection 5 of annual series of maps of land cover and land use of Brazil, in the year 2015, accessed on 11/3/2020 through the link: https://plataforma.brasil.mapbiomas.org/.

Fig 2. Deforestation at each site.

Percentage cover of forest, cropland agriculture, and cattle pasture within a 2500-meter buffer around each trapping site. The variability of forest cover and agriculture+pasture cover showcases the deforestation gradient across the entire study landscape. Example sites from across the deforestation gradient are shown (A4, B8, H32, I33) in order of decreasing amount of forest cover.

Fig 3. DNA metabarcoding results.

Sandfly (top) and vertebrate (bottom) species diversity as revealed by metabarcoding of sandfly DNA extractions. The relative abundance index (RAI) was calculated as the total number of occurrences for species i divided by the total number of pooled samples from across the entire study landscape.

Fig 4. Model results for vertebrate species from carrion fly iDNA dataset.

(A) The predicted probabilities for the significant models for all species (left panel), host species (middle panel; defined as host species in common with host species from the sandfly data), and domestic dogs (right panel) revealed by carrion fly DNA metabarcoding. (B) Vertebrate species diversity as revealed by metabarcoding of carrion fly DNA extractions. The relative abundance index (RAI) was calculated as the total number of occurrences for species i divided by the total number of pooled samples from across the entire study landscape.

Fig 5. Model results for sandfly species.

(A) Regression coefficients and their 95% confidence intervals for the effects of percentage forest (quadratic term), percentage pasture, distance to urban center, and Julian day on sandfly density (quasipoisson model) and the probability of encountering (binomial models each with a random effect for site) a vector species, Psathyromyia aragaoi, Psychodopygus davisi, and Nyssomyia spp. in a pooled sample of sandflies. Significant results are bolded and colored. (B) The significant model predicted probabilities for finding a vector, Pa. aragaoi, and Nyssomyia spp. (the significant results) in a sandfly pool across increasing pasture cover.

Fig 6. Model results for vertebrate groups from sandfly iDNA dataset.

(A) Regression coefficients and their 95% confidence intervals for the effects of percentage forest (quadratic term), percentage pasture, and distance to urban on the probability of encountering (binomial models each with a random effect for site) a host, a sylvatic host, a domestic host, a probable host, and a non-host from a pooled sample of sandflies. Significant results are bolded and colored. (B) The predicted probabilities for the significant models of finding host species, domestic hosts species, and probable host species across increasing forest cover.

Fig 7. Model results for vertebrate species from sandfly iDNA dataset.

(A) Regression coefficients and their 95% confidence intervals for the effects of percentage forest (quadratic term), percentage pasture, and distance to urban on the probability of encountering (binomial models each with a random effect for site) a target species from a pooled sample of sandflies. Significant results are bolded and colored. (B) The predicted probabilities for the significant models of finding Canis lupus familiaris (domestic dog) from a pooled sample of sandflies across increasing forest cover and Bos taurus (domesticated cattle) from a pooled sample of sandflies with increasing distance from the urban center of Sinop.

Fig 8. Leishmania results.

(A) Locations of Leishmania positive sites across our study landscape. Size and color of each circle indicate the proportion of samples at that site that tested positive for Leishmania species with either the kDNA1 primer or the L. braziliensis kDNA3 primer. The black circles indicate sites where there were no samples that tested positive for Leishmania. The base layer map and the land-use, land cover classifications were sourced from MapBiomas – Collection 5 of annual series of maps of land cover and land use of Brazil, in the year 2015, accessed on 11/3/2020 through the link: https://plataforma.brasil.mapbiomas.org/. (B) Regression coefficients and their 95% confidence intervals showing the effect of percent forest, percent pasture, distance to the urban center, and Julian Day on the probability of a pooled sample of sandflies testing positive for the presence of Leishmania. (C) Of 1130 samples, 42 were positive for the presence of at least one Leishmania species, 35 of which containing either sandfly and/or vertebrate data from DNA metabarcoding. The samples that tested positive for Leishmania are organized on the horizontal axis by amount of forest cover at their corresponding site. The sandfly and vertebrate taxa that were found in these samples are shown with either orange (sandfly) or purple (vertebrate) color showing the relative abundance of sequence reads for each taxon in that sample.