Ecological Niche models of plague in Uzbekistan - PowerPoint Presentation

Ecological Niche models of plague in Uzbekistan: extracting biological information from multiple-species models vs. species specific models to understand hosts and vectors Jason K. Blackburn, PhD, Emerging Pathogens Institute & Department of Geography, University of Florida, Gainesville, FL USA Shalo Rakimova , PhD, Aminjan Nematov , PhD Center for GIS, Center for Prophylaxis and Quarantine for Most Hazardous Infections, Tashkent, Uzbekistan Christopher Shane Foster Department of Geography, California State University, Fullerton, CA USA

This Cooperative Biological Research project was funded by the United States Defense Threat Reduction Agency (DTRA) as part of the Biological Threat Reduction Program in Uzbekistan Acknowledgements

Sporadic human cases of plague were documented in 1979 and 1999 in Uzbekistan and plague remains a threat to public healthCPQMHI maintains an active surveillance effort for natural presence of the disease in rodent and flea populations throughout the known regions of historical cases Status of Plague in Uzbekistan Zoological surveillance remains a priority for plague, as both human and enzootic cases appear to be related to host/reservoir populations

Ecological niche modeling (ENM) predicts the potential geographic distribution of species’ through the analysis of non-random relationships between environmental variables (e.g. – temperature, precipitation, elevation, etc) and laboratory positive sample locations The idea is to model the distribution of the species in areas where surveillance may be lacking This study employed plague locations and species-specific occurrence data from field studies and several environmental variables to model the ecological niche for Y. pestis in Uzbekistan Ecological Niche Modeling

Fundamental Niche ENVIRONMENTAL PARAMETER 1 (e.g. vegetation) ENVIRONMENTAL PARAMETER 2 (e.g. soil pH) Realized Niche (Grinnell 1917, Hutchinson 1957) Ecological Niche Theory

Stockwell and Peters (1999) This study employed the GARP modeling system to predict the niche for plague Iterative modeling approach where rules define the distribution as PRESENT or ABSENT Once the rules are developed, they are mapped onto the geography to map the potential distribution GARP is an iterative modeling system that builds rules that relate points to ecological variables GARP is stochastic (like many approaches) and multi-model development and agreement are used to determine accuracy/relevance GARP: Genetic Algorithm for Rule-Set Prediction

Peterson et al. 2002 (Chagas) MODEL HOST SPECIES MODEL VECTOR SPECIES OVERLAP MODEL HOSTS AND VECTORS AND IDENTIFY OVERLAP

Peterson et al. 2004 (Ebola) MODEL HUMAN CASES OR MODEL ENTIRE HOST RANGE Williams et al. 2007 (Avian Influenza) Levine et al. 2007 (Monkey Pox)

Neerinckx et al. (2008) MODEL HUMAN CASES OF PLAGUE: Explore “Niche Space”

SPECIES OCCURRENCE DATA: PLAGUE CASES FROM 2000 - 2007 1 2 3 4 Plague Fleas Xenopsylla fleas Rodents (with or w/o plague) Rhombomys opimus (w or w/o plague) Meriones spp. (2 separate models)

Environmental Coverages Variable Name Description Spatial Resolution Source wd0103a0 mean MIR 1 km Hay et al. (2006) wd0103a1 MIR annual amplitude ( ◦ C) 1 km Hay et al. (2006) wd0103mn minimum MIR (◦C) 1 km Hay et al. (2006) wd0103mx maximum MIR (◦C) 1 km Hay et al. (2006) wd0107a0 mean LST (◦C) 1 km Hay et al. (2006) wd0107a1 LST annual amplitude (◦C) 1 km Hay et al. (2006) wd0107mn minimum LST (◦C) 1 km Hay et al. (2006) wd0107mx maximum LST (◦C) 1 km Hay et al. (2006) wd0114a0 mean NDVI (◦C) 1 km Hay et al. (2006) wd0114a1 annual amplitude NDVI (◦C) 1 km Hay et al. (2006) wd0114mn minimum NDVI (◦C) 1 km Hay et al. (2006) wd0114mx maximum NDVI (◦C) 1 km Hay et al. (2006) alt Altitude (m) 1 km www.worldclim.org Mean LST Mean NDVI

2x2 Niche Space Predicted geography from niche models was plotted in 2x2 variable space to illustrate differences in realized niche space for each model developed in this study

Predicted distributions of “plague”: Using all Y. pestis positives Metric Plague n to build models 76 n to test models (independent) 25 Total Omission 4 Average Omission 14 Total Commission 40.13 Average Commission 27.8 AUC (z, SE) 0.7638 (7.74**, 0.056) Total % area predicted by 6 or more 39.1

Predicted distributions of plague: Fleas Metric Fleas n to build models 22 n to test models (independent) 7 Total Omission 14.3 Average Omission 2.4 Total Commission 49.19 Average Commission 33.35 AUC (z, SE) 0.6945 (3.41**, 0.115) Total % area predicted by 6 or more 50.4

Predicted distributions of plague: Plague vs Fleas Metric Plague Model/Flea metrics n to build models 76 n to test models (independent) 29 Total Omission 17.2 Average Omission 16.5 Total Commission NA Average Commission NA AUC (z, SE) 0.72 (8.2**, 0.05) Total % area predicted by 6 or more NA

Metric Rodent n to build models 133 n to test models (independent) 44 Total Omission 4.5 Average Omission 8.7 Total Commission 41.8 Average Commission 33.8 AUC (z, SE) 0.7778 (11.6**, 0.0415) Total % area predicted by 6 or more 42.0 Predicted distributions of plague: Rodents Model (all rodent species)

Predicted distributions of plague: Rhombomys opimus Metric Rhombomys opimus n to build models 158 n to test models (independent) 52 Total Omission 0 Average Omission 14 Total Commission 34.08 Average Commission 15.48 AUC (z, SE) 0.8595 (11.9**, 0.0328) Total % area predicted by 6 or more 32.3

Predicted distributions of plague: M. erythrourus (taxonomic review?) Metric M. erythrourus † n to build models 6 n to test models (independent) NA Total Omission NA Average Omission NA Total Commission 11.31 Average Commission 8.73 AUC (z, SE) NA Total % area predicted by 6 or more 11.0

Predicted distributions of plague: M. meridianus Metric M. meridianus n to build models 46 n to test models (independent) 15 Total Omission 0 Average Omission 0.7 Total Commission 27.93 Average Commission 20.99 AUC (z, SE) 0.8763 (7.08**,0.058) Total % area predicted by 6 or more 26.8

Predicted distributions of plague: WHICH NICHE? Metric R. opimus Model/M. meridianus metrics R. opimus Model /M. erythrourus metrics n to build models 158 158 n to test models (independent) 61 6 Total Omission 0 0 Average Omission 7.7 10 Total Commission NA NA Average Commission NA NA AUC (z, SE) 0.87 (13.0**, 0.03) 0.88 (3.5**, 0.09) Total % area predicted by 6 or more NA NA

Predicted distributions of plague: WHICH NICHE? Metric Plague Model / R. opimus metrics n to build models 76 n to test models (independent) 210 Total Omission 8.6 Average Omission 23.1 Total Commission NA Average Commission NA AUC (z, SE) 0.73 (22.4**, 0.02) Total % area predicted by 6 or more NA

Mixed mammal models are more reflective of the larger sample sizeWhile successfully predicting test data, these models confound ecological signatures and are therefore not hugely informative Flea models predicted the greatest geographic area, again confounding the ecological signature (niche definition) of any given host species The “plague model” under predicted the R. opimus model, suggesting the “disease distribution” is a subset of host distributions Overall, mixed species models are difficult to interpret and can either over or under predict any given component of the transmission cycle. While the geography may be captured, it may not be biologically meaningful when compared to specific models What do the models mean?

GARP can be interpreted as a “fundamental niche” modeling tool – but this relies on a Grinnellian definition of a single niche per species; modeling multi-species data sets conflates this definition and confuses the ecological space occupied by any given species of the enzootic transmission cycle of plague in Uzbekistan Extrapolating potential geography to niche space (here 2x2 plots) illustrates the ecological space occupied by any individual species within the data set and compares each to the “plague” definition Summary Evaluating each species individually both adheres to the traditional ecological niche theory and provides an opportunity to evaluat e possible differences in regional ecologies that comprise the “plague foci” of Uzbekistan