Zoonotic vector-borne disease systems are ecologically complex due to the numerous interactions between pathogens, vectors, and hosts, in the context of the environment, that ultimately determine the risk of disease in humans. The presence of competent arthropod vectors is sometimes used as a proxy for pathogen exposure and transmission risk, particularly when transmission cycles and other risk factors are poorly understood or unknown (Lippi, Gaff, White, & Ryan, 2021; Lippi, Ryan, et al., 2021). This is the case for many tick-borne pathogen transmission systems, where there can be major knowledge gaps concerning wildlife reservoirs, environmental associations, or even the etiology of disease.

Leveraging tick presence to estimate geographic risk has logistical advantages, as existing data sources (e.g., surveillance programs, museum collections, online data repositories) can provide ample occurrence points for spatial risk modeling and are typically easier to obtain than protected human health data. Nevertheless, vector presence alone is not sufficient to maintain transmission cycles. Many medically important ticks are often habitat generalists, with broad geographic distributions exceeding known disease risk areas (Lippi, Gaff, White, St. John, et al., 2021). Additionally, ticks with expanding ranges can be associated with increases in the burden and diversity of pathogens in newly established locations compared with their historic range (Fornadel et al., 2011; Nadolny & Gaff, 2018; Ogden et al., 2013). These instances highlight the need to understand other components of transmission cycles that drive the risk of exposure beyond vector presence. Reservoir hosts are an important, yet in many instances underexplored, component of zoonotic tick-borne transmission cycles (Tomassone et al., 2018). When considering geographic risks of pathogen transmission, the distribution of wildlife hosts may serve as another important proxy for exposure risk. Therefore, while the use of ticks as proxies of exposure may be advisable in the absence of additional data, determining the distributions of reservoir hosts needed for pathogen amplification and spillover may serve as more nuanced indicators of risk and provide a focus for public health messaging and resources.

Tick-borne diseases have garnered increased attention from the medical community and public health practitioners over the past decade (Lippi, Ryan, et al., 2021; Rochlin & Toledo, 2020). This is largely a result of the increasing incidence of diseases such as Lyme borreliosis and spotted fever group (SFG) rickettsioses (CDC, 2021). Tick-borne infections are often underdiagnosed and go unreported, in part due to nondescript clinical presentations (e.g., febrile illness) and a lack of accurate diagnostic testing (Madison-Antenucci et al., 2020; Rosenberg et al., 2018). Therefore, public health advisories aimed at preventing exposure to potentially infective tick bites are an important aspect of managing these diseases. The Gulf Coast tick (Amblyomma maculatum) is a species of medical and veterinary importance in the Americas (Paddock & Goddard, 2015). Amblyomma maculatum is a competent vector of Rickettsia parkeri, the causative agent of Rickettsia parkeri rickettsiosis, a SFG disease (Paddock et al., 2004; Paddock & Goddard, 2015). Amblyomma maculatum is a habitat specialist found in open fields dominated by grasses and shrubs, requiring high rainfall, temperature, and humidity (Nadolny & Gaff, 2018; Paddock & Goddard, 2015). In the United States, A. maculatum has been undergoing a northward range expansion from its historic range along the Gulf Coast, most recently establishing populations in Connecticut, Illinois, and New York (Molaei et al., 2021; Phillips et al., 2020; Ramírez-Garofalo et al., 2022). Importantly, studies have shown that the prevalence of R. parkeri infections is higher in A. maculatum collected from some newly established locations, a possible result of higher pathogen burden in new vertebrate host populations (Fornadel et al., 2011; Nadolny & Gaff, 2018). Therefore, identifying competent reservoir hosts that underpin transmission cycles and delineating their distributions could provide better estimates of risk in the face of vector range expansion.

The marsh rice rat (Oryzomys palustris) is a suspected reservoir host for R. parkeri (Cumbie et al., 2020) and a vertebrate host for A. maculatum (Nadolny & Gaff, 2018; Wilson & Durden, 2003). Recent field studies have indicated that O. palustris may be involved in the pathogen transmission cycle of R. parkeri; the rats themselves have tested positive for infection, and A. maculatum collected from rats in the field have also tested positive for the pathogen (Cumbie et al., 2020). The distribution of O. palustris has been previously described with observation-based range maps, which are limited to the extent of field observations (Wolfe, 1982). Coarse species range maps are useful in certain contexts but are prone to localized errors and provide no information on spatial variation in the probability of species occurrences (Maréchaux et al., 2017). Further, recently proposed taxonomic changes to the O. palustris species complex are relevant to our understanding of where potential reservoirs are located, yet are available only as range maps (Hanson et al., 2010). To our knowledge, species distribution models (SDMs) have not been used to estimate the suitable geographic range of marsh rice rats. Frequently used in ecology and conservation research, SDMs have become an increasingly utilized methodology in medical geography and public health for estimating the geographic distribution of vector species. A typical SDM workflow entails using occurrence records for the species of interest with environmental predictor layers as inputs for an SDM algorithm, which produces spatially continuous outputs of potential habitat suitability throughout the study area. Here, we apply this technique to estimate the geographic range of the marsh rice rat to assess its potential role in tick-borne pathogen transmission.

The predicted geographic distribution of O. palustris primarily spans the southeastern United States and along the eastern Atlantic coastline. Mean landscape predictions for the three models (Figure 3B,E,H) primarily occurred in four US regions: the West Coast, the Southeast, the Gulf Coast, and the East Coast. The West Coast, including Washington, Oregon, and California, was primarily represented in the full dataset model and eastern clade model with maximum probabilities of 0.817 and 0.821, but the western clade model projected lower probabilities (maximum probability = 0.379) on the West Coast. The western clade model also predicted a smaller area of suitability on the West Coast than the other models. In all three models, the predicted Southeast range spanned Tennessee, Arkansas, Mississippi, Louisiana, Alabama, and Florida, extending north to the southern tip of Illinois and West Virginia, and extending west into Texas, Oklahoma, and Kansas. The full dataset (maximum probability = 0.834) and western clade (maximum probability = 0.826) models had the highest predicted suitable habitat in southeastern states compared with the eastern clade (maximum probability = 0.744). Although the western clade model prediction probability does reduce eastward, the other models maintain predictive consistency across the region. Maximum predicted probabilities for the full dataset (0.917) and eastern (0.916) models are the same for the Gulf Coast and East Coast regions. These models also have consistent prediction probabilities from the Gulf Coast of Texas to Maryland and taper off further north. The western clade model predicts the Texas and Louisiana sections of the Gulf Coast with a maximum probability of 0.928, but probabilities are not consistent towards Florida as in the other models, and the maximum probability for the East Coast is 0.708.

All three models had similar responses to the covariates consistent with occurrence data partitioning and overlap between models (Figure 4); however, differences in the magnitude of relationships between predicted probabilities and environmental predictors were apparent, especially between the western clade and the other two models (Figure 5). Elevation was the top contributing environmental predictor in all three models; prediction probability was highest at sea level and rapidly decreased with increasing elevation. The western clade model plateaued in prediction probability reduction earlier than the other models. Precipitation of the wettest month and annual mean temperature were the next highest contributors. All models generally exhibited positive relationships with increasing precipitation and temperature beginning around 100 mm and 10°C, respectively. There were some distinctions in the relationship with soil volumetric water content between the eastern and western models, as the former maintained a negative relationship while the latter had a positive relationship. However, this variable contributed minimally to the eastern model. The BRT simplification process excluded soil volumetric water content from the full dataset model and soil organic carbon stock from the western model (Figure 4). Temperature seasonality was excluded from the full dataset and eastern models (Figure 4).

All three models performed well with cross-validation, internal AUC values, and AUC values from external test data that remained above 0.9. The mean value for the full dataset model cross-validation AUC was 0.957 (SD = 0.007), internal test data AUC was 0.959 (SD = 0.015), and external test data AUC was 0.939 (SD = 0.005). The mean AUC of the eastern clade for cross-validation AUC was 0.960 (SD = 0.008), internal test AUC was 0.958 (SD = 0.013), and external test data AUC was 0.983 (SD = 0.004). The western clade model did have marginally higher standard deviations with AUCs of 0.965 (SD = 0.009) for cross-validation, 0.964 (SD = 0.018) for internal test data, and 0.989 (SD = 0.011) for external test data. The accuracy metrics for all three models were similar, although the AUC value for external testing data from the full dataset model was the lowest of the three models.

Dichotomized probabilities (at 0.5, 0.6, and 0.7) of mean landscape predictions for rice rat habitat suitability indicate the East Coast region extending from Maryland to Florida, peninsular Florida, the lower Mississippi River Basin, and the Gulf Coast region extending from Florida to Texas have the highest potential for rice rat species occurrence in all models (Figure 6; Appendix S1: Figures S1 and S2). The full dataset and eastern clade models predict occurrence across this entire area, with predictions of suitability extending further inland to Arkansas, Tennessee, Missouri, and Kentucky, and a small area on the West Coast (Figure 6; Appendix S1: Figures S1 and S2). Modeled distribution of the western clade is more restricted, by comparison, with predicted suitability primarily centered in Louisiana, Texas, and along the Mississippi River (Figure 6; Appendix S1: Figures S1 and S2). Distribution models for each type of data partitioning (full dataset, eastern clade, and western clade) overlap in mean prediction along the western portion of the Gulf Coast, primarily in Texas, Louisiana, and southern peninsular Florida (Figure 6; Appendix S1: Figures S1 and S2).

This work is the first effort to apply SDM techniques to estimate the geographic distribution of the O. palustris species complex. The predicted distributions of all three taxonomic groupings were found to be predominantly restricted to coastal areas in the southeastern United States. Additionally, the estimated suitable habitat for both O. palustris taxonomic groupings (i.e., the full dataset and the eastern subset) extends into the southern portion of the Mid-Atlantic region, with a discontinuous, limited area of suitability in coastal California. In contrast, habitat suitability for the western clade is primarily highest along the Gulf Coast, extending west into coastal Louisiana and Texas. Further, output across models indicates a corridor of suitable habitat that extends north, throughout Louisiana, and into Arkansas, Mississippi, Tennessee, Kentucky, and southeastern Missouri, an area that coincides with the lower Mississippi River Basin. These spatial trends in estimated habitat suitability are perhaps unsurprising, as O. palustris is a semiaquatic rodent species that is primarily associated with both freshwater and estuarine wetlands, including tidal marshes, watersheds, and emergent wetlands (Eubanks et al., 2011; Kruchek, 2004). Rice rats are heavily dependent on the vegetation typically associated with coastal wetlands, including common wetland plants such as sedges (Carex spp.), rushes (Juncus spp.), cattails (Typha spp.), panic grass (Panicum spp.), and salt grass (Spartina alterniflora) (Eubanks et al., 2011; Rose & McGurk, 2006; Svihla, 1931). Suitable vegetation is a critical microhabitat feature for O. palustris, directly providing essential resources as both cover and food (Eubanks et al., 2011; Kruchek, 2004; Rose & McGurk, 2006). We were limited in this study by the inability to directly include environmental predictors related to land cover, as few O. palustris collections were made after the year 2000 in our datasets. Future work that ties contemporary rice rat observations with current land cover will enable us to assess reliance on vegetation type with greater resolution.

Elevation was the environmental predictor that contributed the most to estimated rice rat distributions, regardless of taxonomic clade. The highest probabilities of occurrence coincide with areas of low elevation at or near sea level, predominantly along coastlines in the Southeast and Gulf states. Precipitation of the wettest month and annual mean temperature were also influential factors in all three models, where higher precipitation values and temperatures exceeding ~13°C were associated with increasing habitat suitability. Together, elevation, precipitation, and temperature are likely capturing qualities of low-elevation marsh and wetland habitats that are necessary to maintain wetland plant communities and support rice rat populations. When considering other potential environmental drivers of rice rat populations, we start to see differences between the models for different taxonomic groupings. Low precipitation seasonality and soil organic carbon stock exceeding ~100 tons/ha are associated with increased probability of occurrence in the full O. palustris and eastern clade models. Again, these are factors that may support stable and robust plant communities in wetlands. In contrast, precipitation seasonality did not cause the same suitability drop-off in the western clade model. Instead, soil volumetric water content and temperature seasonality become important environmental predictors, indicating that water availability and temperature are more limiting factors for western rice rat populations.

Estimating the range of O. palustris is an important step forward in delineating the general distribution of potential hosts for A. maculatum and the rickettsial pathogens they transmit. Reservoir hosts involved in tick-borne transmission cycles are targets for control efforts and may serve as better indicators of risk than vector presence alone, particularly for ticks with widespread distributions. However, vertebrate hosts involved in many tick-borne transmission cycles are poorly understood or unknown. This is particularly true for SFG pathogens, many of which have only recently garnered increased scrutiny from medical and public health authorities, relative to other tick-borne diseases such as Lyme disease, where pathogen–host–vector dynamics are extensively studied (Halsey et al., 2018; Ostfeld, 2012). Components of the transmission cycle for R. parkeri are not currently known, but several small mammal species are suspected to play a role in the pathogen’s ecological maintenance, based on field surveys and laboratory experiments (Cumbie et al., 2020; Moraru et al., 2013). Investigation of the potential role of O. palustris as a reservoir host for R. parkeri is warranted. In addition to detection of R. parkeri in field-collected O. palustris, other closely related species of the genus Oryzymys have been established as competent hosts of rickettsial pathogens (Krawczak & Labruna, 2018; Padilha et al., 2013). Although the marsh rice rat has suspected involvement, further work is needed to establish the competency of O. palustris as a host capable of supporting the R. parkeri transmission cycle.

There is considerable geographic overlap with O. palustris and A. maculatum in the southeastern United States. Comparing the model of our full O. palustris dataset to an SDM for A. maculatum produced by Flenniken et al. (2022), most of the overlap seen in estimated distributions occurs along the Gulf Coast, peninsular Florida, and along the East Coast into Virginia, Maryland, and New Jersey, with some incursion along the lower Mississippi River Basin (Figure 7; Appendix S1: Figures S3 and S4). The availability of suitable hosts may be an important element of successful range expansion of A. maculatum, which has drastically expanded the northern extent of its range in recent years (Molaei et al., 2021; Nadolny & Gaff, 2018; Wojan et al., 2022). Notably, much of this northward expansion has occurred along the East Coast, a potential indicator of the role that underlying host species like O. palustris play in the habitat suitability of tick vectors. The ranges of both O. palustris and A. maculatum also coincide with much of the estimated distribution of R. parkeri in North America, as described in Moo-Llanes et al. (2021). Moo-Llanes et al. built SDMs with records of ticks, comprised of several Amblyomma species, that tested positive for R. parkeri bacteria. The resulting model shows high predicted suitability in North America along the Gulf of Mexico and portions of the East Coast, with moderate to high predicted suitability extending inland throughout the eastern United States.

In addition to O. palustris, two other rodents in the family Cricetidae have also been implicated as hosts of A. maculatum, with potential involvement in the transmission cycle for R. parkeri: the hispid cotton rat (Sigmodon hispidus) and the eastern meadow vole (Microtus pennsylvanicus) (Cumbie et al., 2020). The geographic range of S. hispidus overlaps with much of our estimated distribution for O. palustris, extending further north into Nebraska and west into California, whereas the range of M. pennsylvanicus has some overlap with the northern extent of the O. palustris distributions, but extends north into Canada and Alaska (Jackson & Cook, 2020; USGS, 2018). Although S. hispidus has been documented as an important blood-meal host for immature A. maculatum, lab experiments have demonstrated that this species quickly clears R. parkeri infection, making them unlikely reservoir hosts (Baker et al., 2003; Johnson et al., 2022; Moraru et al., 2013). Likewise, M. pennsylvanicus is not a likely reservoir host of R. parkeri. While infected A. maculatum have been collected from M. pennsylvanicus, the ranges of the rodent and tick have minimal historical overlap (Cumbie et al., 2020; Nadolny & Gaff, 2018). Nevertheless, these three rodent species collectively span a vast area, potentially facilitating the expansion of ticks that utilize these species as hosts and providing opportunities for pathogen spillover events.

The underlying distribution of reservoir hosts is suspected to play a role in the geographic risk of tick-borne disease transmission. Tick-borne pathogens can exhibit geographically restricted ranges that are “nested” within the larger distribution of competent vectors. For example, within the eastern portion of the United States, foci for Borrelia burgdorferi, the spirochete that causes Lyme disease, are restricted within the range of its vectors, primarily Ixodes scapularis (Fleshman et al., 2021). The existence of an “infected” niche of ticks that tested positive for R. montanensis infections within the full geographic range of its vector, Dermacentor variabilis, has also been supported with SDMs (Lippi, Gaff, White, St. John, et al., 2021). The absence of pathogen transmission can be due to many factors, and certainly, vector presence alone is not sufficient for transmission. Absence of pathogens could be a result of undersampling or a lack of robust surveillance for tick-borne disease. Differences in suitable environmental conditions for tick presence and pathogen transmission may also contribute to range discrepancies (Galletti et al., 2016). However, the importance of reservoir hosts in maintaining vector-borne transmission cycles is widely acknowledged, yet in the case of many tick-borne diseases, not well understood. Even so, there is increasing evidence to support the key role that reservoir and amplifying hosts play in the localized density of infected ticks, and risk of transmission for an array of tick-borne diseases (Krawczyk et al., 2020). This has notably been the case with Lyme disease, where increasing burden and range expansion of the disease have been linked to range shifts and population changes in vertebrate hosts (Levi et al., 2012; Thompson et al., 2001).

Conclusive identification of reservoir and amplifying hosts is still needed to establish transmission cycles for many tick-borne diseases, including the majority of those caused by rickettsial agents. Given that many species of medically important ticks have generalized habitat requirements, or are actively expanding their geographic ranges, identifying areas where tick presence overlaps with suitable reservoir habitat may be necessary for refining risk maps and public health advisories derived solely from tick presence. In this study, we estimated the geographic range of O. palustris, a suspected reservoir host of R. parkeri, and found considerable overlap with the distribution of A. maculatum. Though future work is needed to investigate the definitive role of O. palustris in transmission cycles, our findings highlight the potential need for extended surveillance efforts in those overlapping areas.