Phylogenetic and biogeographical traits predict unrecognized hosts of zoonotic leishmaniasis

Abstract

The spatio-temporal distribution of leishmaniasis, a parasitic vector-borne zoonotic disease, is significantly impacted by land-use change and climate warming in the Americas. However, predicting and containing outbreaks is challenging as the zoonotic Leishmania system is highly complex: leishmaniasis (visceral, cutaneous and muco-cutaneous) in humans is caused by up to 14 different Leishmania species, and the parasite is transmitted by dozens of sandfly species and is known to infect almost twice as many wildlife species. Despite the already broad known host range, new hosts are discovered almost annually and Leishmania transmission to humans occurs in absence of a known host. As such, the full range of Leishmania hosts is undetermined, inhibiting the use of ecological interventions to limit pathogen spread and the ability to accurately predict the impact of global change on disease risk. Here, we employed a machine learning approach to generate trait profiles of known zoonotic Leishmania wildlife hosts (mammals that are naturally exposed and susceptible to infection) and used trait-profiles of known hosts to identify potentially unrecognized hosts. We found that biogeography, phylogenetic distance, and study effort best predicted Leishmania host status. Traits associated with global change, such as agricultural land-cover, urban land-cover, and climate, were among the top predictors of host status. Most notably, our analysis suggested that zoonotic Leishmania hosts are significantly undersampled, as our model predicted just as many unrecognized hosts as unknown hosts. Overall, our analysis facilitates targeted surveillance strategies and improved understanding of the impact of environmental change on local transmission cycles.

Fig 1. The ranges of zoonotic Leishmania and their known and predicted hosts extend throughout the Americas.

The number of overlapping ranges of known Leishmania hosts are depicted in (a & c), with the range of human cases of L. (Viannia) and L. (Leishmania) outlined in grey. The number of overlapping ranges of newly predicted hosts are outlined in (b & d). Hylaeamys megacephalus and Calomys callosus are unrecognized hosts with the highest and second highest predictions for L. (Viannia) according to Shapley values. Their ranges are outlined in dark green and light green, respectively, in (b). Dasypus novemcinctus and Leopardus wiedii are hosts with the highest and second highest Shapley values for L. (Leishmania). Their ranges are outlined in dark blue and light blue, respectively, in (d). An exhaustive list of newly predicted hosts is listed in S2 Table. Base maps were created in R using open source shapefiles from Natural Earth (admin-0 countries: https://www.naturalearthdata.com/downloads/10m-cultural-vectors/; https://www.naturalearthdata.com/about/terms-of-use/) [51]; species ranges were created in R using open source shape files from IUCN (https://www.iucnredlist.org/resources/spatial-data-download; https://www.iucnredlist.org/terms/terms-of-use) [27].

Fig 2

The number of known and predicted hosts per order for L. (Viannia) (a) and L. (Leishmania) (b). Grey bars represent the number of known hosts within each order, colored bars represent the number of newly predicted hosts per order. Animals were classified as newly predicted if > 95% of mean SHAP values were greater than 0 (a-b). Bars in (c) represent the proportion of newly classified hosts within each order; colors match (a-b).

Fig 3. Biogeography, life-history, and phylogenetic traits all significantly contributed to model predictions for host status.

Biogeographical traits (minimum longitude of the range, maximum longitude of the range, % cover of land-use/land cover in the species range, average temperature in the warmest quarter in the species range, range area) are colored in blue, life-history traits (population density, gestation length, litter size) are colored in red, phylogenetic distance (location along PCoA ordination axes) is colored in orange, and study effort is colored in grey. Points are the absolute value of the mean Shapley importance (mean |SHAP value|) for the trait across all mammals (i.e., global feature contribution), bars represent the absolute values of the 0.05–0.95 percentiles. Only features with 0.05 percentiles > 0 are shown.

Fig 4. Hosts have fast-paced life-histories and live in proximity to humans.

Shapley partial dependence plots showing the effect of each feature on L. (Viannia) (a) and L. (Leishmania) (b) host status after accounting for the average effect of the other features in the model. Colored lines represent the average effect across model iterations, while grey lines show each individual model iteration (model fit with 80% of data) (blue = biogeographical traits; red = life-history). Features with global mean feature contribution scores > 0 for > 95% of model iterations are shown. Rug plots on the x-axis indicate the distribution of the data.