Citation:

Silvestri, E., S. Douglas, V. Luna, C. Jean-Babtiste, D. Pressman-Mashin (née Harbin), L. Hempel, T. Boe, T. Nichols, AND D. Griffin. Spatially Integrating Microbiology and Geochemistry to Reveal Complex Environmental Health Issues: Anthrax in the Contiguous United States. Chapter 19, Geospatial Technology for Human Well-Being and Health. Springer Basel AG, Basel, Switzerland, , 355-377, (2018). https://doi.org/10.1007/978-3-030-71377-5_19

Impact/Purpose:

Soil composition and environmental conditions have been shown to influence the distribution of Bacillus sp. and Bacillus anthracis in surface soils. As demonstrated in the literature, soil characteristics such as Calcium (Ca) content and pH influence B. anthracis occurrence, however, large-scale geochemical and microbiological investigations are still needed to determine continental-wide constraints on distribution. In this study, Maximum entropy modeling (Maxent) was used to gain insight into the environmental factors that influence the occurrence of Bacillus sp. and B. anthracis in soils of the contiguous United States, and to identify regions that may be vulnerable to future anthrax outbreaks in livestock or wildlife. Soils with the presence of Bacillus sp. and B. anthracis were identified from a large, national soil dataset collected for the USGS North American Soil Geochemical Landscape Project (NASGLP), which screened samples for Bacillus sp. and B. anthracis using a multiplex polymerase chain reaction (PCR) assay, and from livestock outbreaks reported by the National Animal Health Reporting System (NAHRS). Maxent was run using two scales; the Outbreak State scale which included only states reporting animal anthrax outbreaks from 2001 to 2013, and the National scale which included all states in the contiguous United States. Three iterations of the data were also used and included; the Sample Location dataset, the Normalized dataset, and the Interpolated dataset. Two metrics were used to evaluate model performance, including the widely used Area Under the Curve (AUC) test and an additional metric, the True Skill Statistics (TSS). Results showed that most of the Maxent models in this study performed best when using Outbreak State scale. When the models were scaled up to the National scale, model performance declined, except for the Normalized variable dataset. At the Outbreak State scale, a large proportion of the area was predicted to be of higher probability and the statistical measures assumed the model was underfitting the data. The model with the highest AUC and TSS scores for this study was the Outbreak State scale using Sample Location dataset (AUC= 0.918 and TSS= 0.82). Some of the variables found to be closely related to the occurrence of B. anthracis in this study included pH, drainage potential, and concentration of elements including Na, Ca, Sr, and Mg, which have also been found to be related to outbreaks or to the occurrence of B. anthracis in previous studies. The models in the current study indicated possible regions that have not had recent outbreaks or might not typically be considered an “at risk” area for an outbreak, but may still be at risk (Michigan and Maine). This work provides an extension to use of ecological niche modeling to investigate distribution of B. anthracis in the United States because it utilizes additional soil geochemistry data and has shown that further validation techniques, such as the TSS, should be considered in addition to AUC. Results from this study could be used by animal and public health officials to identify areas with a higher risk of anthrax outbreak in wildlife, livestock, and humans, due to naturally occurring soil and environmental conditions.

Description:

Maxent models were run using the B. anthracis presence data and/or the animal outbreak presence data. Models run using the animal outbreak data alone utilized two scales: the Outbreak State scale which included only states reporting animal anthrax outbreaks from 2001-2013 and the National scale which included all states in the contiguous United States. Three iterations of the environmental data were used and included the Sample Location dataset which utilized the environmental variable data with assigned latitude and longitude locations from the USGS NASGLP project; the Normalized dataset which scaled the environmental variables so that the values fell between 0-1; and the Interpolated dataset which provided an interpolation of the environmental variables averaged for each county and assigned to a point for that county at the centroid (rather than using the NASGLP latitude and longitude location). Two metrics were used to measure model performance including the widely used area under the curve (AUC) and an alternative method, the True Skill Statistic (TSS). The AUC gives the probability that a randomly chosen presence location has been correctly ranked higher than the absence/background site. AUC values at 0.5 or lower mean the ranking is no better than random, while the AUC values nearer to 1 mean the model is a better predictor. The TSS provides a comparison of how well the background predictions made by the model match the model results at the test dataset (presence) locations. TSS values near +1 means the model approaches perfect agreement, while values near −1 indicate the model is no better than random. Maxent models to determine the influence of environmental factors on the B. anthracis distribution using the PCR data yielded a low TSS, which suggested the model might be underfitting the data. This was not surprising due to the difficulty in recovering B. anthracis in soil samples as well as the samples themselves being discrete in nature and only capturing a snapshot in time. Therefore, the distribution of B. anthracis and its niche in the contiguous United States could not be determined in this study. However, efforts to investigate environmental factors that would have a higher potential of supporting an anthrax outbreak in wildlife and livestock yielded better results. Results showed that most of the Maxent models in this study performed best when using the Outbreak State scale. When the models were scaled up to the National scale, model performance declined, except for the Normalized variable dataset. At the Outbreak State scale, a large proportion of the area was predicted to be of higher probability for wildlife/livestock anthrax outbreaks, and the statistical measures assumed the model was underfitting the data. The model with the highest AUC and TSS scores for this study was the Outbreak State scale using Sample Location dataset (AUC=0.918 and TSS=0.82). Some of the variables found to be closely related to the occurrence of B. anthracis in this study included pH, drainage potential, and concentration of elements including Na, Ca, Sr, and Mg, which have also been found to be related to animal outbreaks or to the occurrence of B. anthracis in previous studies. The models in the study indicated possible regions that have not had recent wildlife/livestock anthrax outbreaks but contained environmental conditions that could potentially support an outbreak if one were to occur (Michigan and Maine). This work provides an extension to the use of ecological niche modeling to outbreak potential in livestock/wildlife in the United States because it utilizes additional soil geochemistry data and has shown that further validation techniques, such as the TSS, should be considered in addition to AUC. Results from this study could be used by animal and public health officials to identify areas with a higher potential for anthrax outbreak in wildlife and livestock due to naturally occurring soil and environmental conditions.

URLs/Downloads:

Record Details: