The model showed clear enzootic and epizootic behavior for our two parameterizations ( respectively). In addition, model results for the prairie dog and ground squirrel parameterizations closely matched independent data (for detailed model results see Text S3
). Parameter values for California ground squirrels created enzootic behavior for prolonged periods (>3 years; ) with 90% of surviving hosts found to be resistant to plague. Similarly, a natural population of California ground squirrels showed evidence of antibody responses to previous plague exposure in 11 of 13 years, accounting for 93% of the total population 
. For the prairie dog parameter set, the model predicted high extinction probabilities similar to areas where epizootics have been observed on black-tailed prairie dog towns 
, as well as others specifically studied by us on the Pawnee National Grassland where all 12 confirmed plague epizootics on towns from 2003 to 2008 resulted in severe population declines or extinction (). The model also predicted short-lived epizootics with towns declining to near extinction after about 3 weeks with remnant hosts persisting for around 37.5 weeks (), a range inclusive of the observed 6–8 week window from first detection of plague to apparent town extinction 
Enzootic and epizootic plague dynamics.
Looking at the role of our three transmission routes during enzootic cycles, infection potential from the booster-feed infection cycle declines during the majority of the model run (i.e., negative growth rate of infection potential), while the infectious, questing flea reservoir increases almost throughout (). Infectious carcasses played little role in the enzootic cycle (). In contrast, infection potential from booster-feeds showed a sharp increase during the early-stages of an epizootic, but then quickly declined as the epizootic progressed (). Infection potential from the flea reservoir showed a similar pattern, although it continued to increase after infection from the booster-feed infection cycle crashed (). The role of infectious carcasses paralleled that of the flea reservoir although the magnitude of change was not as great ().
Systematic removal of transmission routes helped provide a clearer picture of each in plague dynamics, especially for epizootic behavior (). For the epizootic host parameterization, removing all flea-borne transmission (i.e., booster-feeds and the flea reservoir) resulted in a shift to enzootic behavior (). However, when only the infectious, questing flea reservoir is removed, model behavior is again dominated by epizootics (). Combined, these results suggest that booster-feed transmission plays an important role in epizootics. Removal of the carcass reservoir still results in primarily epizootics, but disease fade is more likely (). Removal of both reservoirs shows a significant increase in enzootic behavior with epizootics still dominating (). Overall, this supports the idea that booster-feed transmission dominates but at least one type of reservoir transmission is needed to reach epizootic levels. In the enzootic host, removal of all flea-borne transmission resulted in a shift to disease fade-out (). However, when booster-feeds were reinserted into the model and only the flea reservoir was removed, the shift to disease fade-out remained (). Removal of the carcass reservoir alone has little impact on plague dynamics (). Together, these results on enzootic probability suggest a consistent role for a flea reservoir in enzootic dynamics.
Transmission hypothesis testing.
Our multi-parameter sensitivity analysis was consistent with the relative importance of transmission routes described above and revealed that model results were sensitive to parameters influential to both the booster-feed infection cycle and the infectious, questing flea reservoir (; ). In particular, flea questing efficiency, a, was positively correlated with enzootic probability but had little effect on extinction probability. Increasing transmission efficiency from EP2, βL, increased extinction probability as did an increase in the transition rate between EP1 and EP2 for fleas taking non-infectious blood meals, θE.
Multi-parameter sensitivity analysis.
Population responses to plague infection were also sensitive to several host parameters in the model (). Among these, extinction probability was increased by higher rates of resistance loss, ϕ, and shorter host exposure periods (i.e., increased values of σ). However, increased host resistance, p, and increased host carry capacity, K, served to decrease epizootic behavior. In contrast, enzootic probability was increased by increasing host carrying capacity and declined with higher rates of resistance loss, shorter host exposure periods, and decreasing host connectance (i.e., increased values of our of spatial correction factor, B). Sensitivities that are not reported were not significant.