Our sampling was focused in Arua and Zombo districts in the plague-endemic West Nile Region of northwestern Uganda. The majority of human plague cases in these districts have been reported from localities situated above the Rift Valley escarpment, which roughly bisects our study site [6], [11]. We established 10 small mammal sampling sites that were evenly spaced with respect to elevation along a transect spanning from approximately 725 (site 1) to 1630 m (site 10) in elevation (average difference in elevation between neighboring sites: 99 m, range: 73–127 m). Half of the sites were situated in areas identified by our previous risk model as posing an elevated risk for plague (sites 6–10) [6] and four of the highest elevation sites (sites 7–10) were above the 1300 m elevation threshold that was previously identified as delineating differences between areas of low or elevated plague risk (Figure 1) [6], [11]. Our delineation of areas that pose an elevated risk was based on approximately 10 years of epidemiological surveillance data collected from clinics above and below the 1300 m elevation threshold [6], [11]. Animal-based surveillance for plague activity is not routinely conducted in Uganda, in part due to the cost but also because seropositivity and culture recovery efforts are extremely low, especially during inter-epizootic periods. Thus, although humans are incidental hosts of Y. pestis, the epidemiological dataset provided the best available estimate of plague activity in the region. It is possible that Y. pestis may circulate in small mammal communities outside of the identified epidemiological focus. However, surveys designed to assess differences in human demographics or behaviors that we believed to be important risk factors for plague (e.g., household size, food storage and agricultural practices and vector and rodent control methods) revealed no obvious differences between high (>1300 m) and low elevation sites. Thus, lack of human infections at low elevation sites is likely attributable to either a lack of bridging vectors between zoonotic cycles and human hosts, or lack of Y. pestis. Sites located above 1300 m were generally wetter (median cumulative annual rainfall [range]: 1644.8 mm [1500.2–1889.2 mm]) and cooler (median average maximum monthly temperature [range]): 26.1°C [25.3–27.0°C]) than sites below this elevation (1080 mm [864.2–1483.3 mm]; 30.6°C [28.3–32.7°C]). Site-specific rainfall and temperature data were derived from high-resolution atmospheric model simulations for 1999–2008, described below.

Small mammals and their associated fleas were collected from 40 homesteads situated within the 10 trap sites described above. Timing of the five collections coincided primarily with the rainy seasons. Collections were conducted in February–March 2010 and 2011, which mark the end of the dry season and the start of the secondary rainy season, and October 2010 and November 2010, which represents the peak of the primary rainy season, and April 2011, which falls in the middle of the secondary rain season.

For each homestead, which is comprised of a small cluster of traditional homes constructed of mud waddle and thatch roofing (mode: 5 huts, range 2–6 huts per homestead; median numbers of huts included per elevation were similar), 2 Tomahawk and 2 Sherman traps were set inside of each hut, 8 Tomahawk and 8 Sherman traps were set in the family compound outside of the huts, and 64 traps comprised of 32 Tomahawk and 32 Sherman traps were set in the peridomestic setting extending 40 m away from the huts. One Tomahawk and one Sherman trap was placed per trap station and trap stations were set 10 m apart radiating from the homestead in the 4 cardinal and 4 inter-cardinal directions (figure 2). Traps were baited with a mixture of 1/3 corn, 1/3 ground nuts, and 1/3 dried fish. For each trap session, trapping was conducted for two consecutive nights with rodents retrieved each morning and traps reset at dusk. We acknowledge that our sampling is biased towards nocturnally active small mammals and abundances of diurnal rodents, such as Arvicanthis niloticus may be underrepresented. Upon capture, small mammals were anesthetized using an inhalation anesthetic (halothane). Sedated animals were measured, weighed, identified to genus or species and combed for fleas. For host species that could not be identified to species because of a lack of morphological differences that were detectable in the field, individuals were identified to genus (e.g., Crocidura spp., Mastomys spp. and Praomys spp.). Collected fleas were preserved in 70% ethanol for later identified to species following published taxonomic keys [22], [23], [24], [25]. All animal procedures described in the study were approved by the Centers for Disease Control and Prevention Division of Vector Borne Diseases Institutional Animal Care and Use Committee (protocol 09-023).

Host-seeking fleas were collected from each of the same huts described above for 2 consecutive nights during each small mammal trapping session by placing 1 modified Kilonzo pan trap [26], [27] on the floor in locations deemed to be the most protected from animals that are frequently in the huts (e.g., chickens, goats, dogs) and were placed away from areas where residents slept. The pan traps measured 20.0 cm in diameter and 3.0 cm in depth. A flashlight was suspended above the pans in a solution of 2% saline with Tween 80 (5 drops of Tween 80 added to 1.0 cm to 1.5 cm in depth of saline). Petroleum jelly was applied to the rim of the pan to prevent fleas from escaping. After each of the 2 nights, fleas were collected from the pans and stored in 70% ethanol until species identification could be performed.

Meteorological observation networks are sparse in sub-Saharan Africa, and no observation stations were established within our study area. As a result, to assess the relationship between flea diversity and climatic variables, we used ten-year averages (1999–2008) for cumulative monthly and annual precipitation, minimum temperature during the coolest month (June) and maximum temperature during the hottest month (February) of the year derived from a suite of atmospheric simulations performed specifically for the West Nile region with the Weather Research and Forecasting Model (WRF) [28]. WRF is a fully compressible conservative-form nonhydrostatic atmospheric model suitable for both research and weather prediction applications; it has demonstrated ability for resolving small-scale phenomena and clouds [29]. Monaghan et al. [28] demonstrated that the resulting WRF-based climate dataset better resolves the spatial variability and annual cycle of temperature, humidity, and rainfall in West Nile compared to satellite-based and in situ records, and therefore is suitable for use in this study.

Relative host abundance, measured as the number of hosts per trapping effort, was determined for each elevation point by dividing the numbers of hosts of a particular genus or species collected within an elevation point (e.g., sites 1–10 shown in figure 1) by the number of trap nights per site. Host diversity was estimated for each sampling site using Simpson's Index of Diversity [30], described as:where n is the total number of hosts of a particular species or genus and N is the total number of hosts of all species. Flea diversity, based on on-host fleas, was estimated using the same equation. Hosts and fleas used in these calculations correspond with those listed in Table 1. The Simpson's index of diversity ranges from 0 to 1 with the greater value indicating greater diversity within the sample. Median numbers of hosts per trap, total number of fleas collected from host, fleas per host, total numbers of host-seeking fleas, and diversity indices were compared between sites above or below 1300 m elevation using Wilcoxon rank sum tests with chi square approximations.

The relationship between elevation and flea diversity was explored using linear regression. To evaluate the association between flea diversity and rainfall, forward stepwise regression was used to screen average monthly site-specific values of rainfall. Rainfall in this region follows a bimodal distribution with the reliable heavy rain falling from August through November during the primary rainy season. December through February marks the dry season, which is followed by the less reliable and often lighter rainy season from March–June. June and July are characterized as the interval season with variable amounts of rain [28]. Spatial variation in rainfall between sites for any particular month can be considerable [21], therefore we used a forward stepwise screening process that included each average monthly value for rainfall to identify the best statistical fit to the flea diversity data. Regardless of the month, temperatures consistently decreased with increasing elevation and temperature showed only a single peak with highest temperatures observed in February and the lowest temperatures in June. Therefore, average minimum temperature in June (lowest temperature during the coolest month) and average maximum temperature for February (highest temperature for the warmest month) were included in models of flea diversity. These temperature extremes were selected because flea survival is often temperature dependent and average values may have obscured the true range in temperature experienced by the fleas. For both the rainfall and the temperature models, the best model was selected based on comparisons of the corrected Akaike Information Criterion. Because only 10 sites were included in the model, we restricted the number of predictive variables to one.

During a total of 7,832 trap nights spanning from 22 February 2010 to 10 April 2011 and run within 40 homesteads at 10 elevation points, a total of 1,503 small mammals comprised of at least 14 species were captured. Together, the four most commonly collected hosts (Rattus rattus, Arvicanthis niloticus, Crocidura spp. and Mastomys spp.) accounted for 77% of all hosts examined (Table 1). From all hosts examined, a total of 1,707 fleas comprised of 12 species were collected. Five species were extremely rare, each representing less than 1% of the sample (Table 1). Subsequent analyses in this sub-section focus on the five species of fleas that were commonly associated with the four most abundant host species identified above: Xenopsylla cheopis, X. brasiliensis, Ctenophthalmus cabirus, and Dynopsyllus lypusus, and Stivalius torvus. Although X. nubica was more abundant than D. lypusus and S. torvus, they were collected primarily from gerbils (belonging to the genus Tatera or Taterillus), which represented only 10% of all hosts collected (Table 1). Xenopsylla cheopis and X. brasiliensis were collected commonly from each of the four key host species, but X. cheopis was rare above 1300 m (only 1 X. cheopis was recovered above 1300 m) and X. brasiliensis was absent on sampled hosts below this elevation threshold (Figure 3). By contrast, with the exception of a single C. cabirus collected at site 5, C. cabirus, D. lypusus, and S. torvus were not observed in sites 1–5, were observed in low numbers at site 6 (n = 4 D. lypusus and 1 each of C. cabirus and S. torvus) and were common at sites 7–10 (figure 3). It is noteworthy that site 6, although lower than 1300 m elevation, marks the first site to appear within the predicted elevated risk area (figure 1).

During approximately 1,790 Kilonzo trap nights, a total of 836 host-seeking fleas were collected. Ctenocephalides felis and Echinophaga gallinacea represented 61% (n = 510) and 37% (n = 311) of the captures, respectively. Other species captured included Tunga penetrans (n = 2), X. brasiliensis (n = 1) and X. cheopis (n = 12). The total numbers of fleas collected, and total numbers of the nearly cosmopolitan flea species, C. felis and E. gallinacea, were similar between sites located above compared with below 1300 m elevation.

For each host species or genus, we compared abundances between sites situated above or below 1300 m elevation. With the exception of Mus minutoides, which was more abundant above 1300 m (median 0.04 M. minutoides/trap: range: 0.03–0.04 M. minutoides /trap) than below (0.01 [0–0.03]; Wilcoxon rank sums test with chi square approximation χ² = 5.5 d.f. = 1 P = 0.02), host abundances were similar between high and low elevation sites. In addition, host diversity was similar between sites above and below 1300 m (figure 4). Likewise, median numbers of fleas and fleas per host were similar between sites above and below 1300 m. Thus, it is unlikely that differences in plague risk, which are associated with this elevation threshold, result from differences in host or flea abundance or from differences in host diversity. By contrast, flea diversity was significantly higher for sites above 1300 m (Simpson's index of diversity median 0.73; range: 0.73–0.75) than below (0.20 [0–0.65]; Wilcoxon rank sums test with chi square approximation χ = 6.55 d.f. = 1 P = 0.01) (figures 3 and 4). Flea diversity was positively associated with elevation and elevation explained 79% of the observed variation in flea diversity (Flea diversity = 0.0009 m⁻¹* elevation−0.55; F = 30.17, d.f. = 1,9, P = 0.0006).

Site-specific cumulative rainfall during the month of July explained 90% of the variation observed in the flea diversity index (F = 68.40, d.f. = 1,9, P<0.0001; Flea diversity = 0.006 mm⁻¹ * Cumulative rainfall in July – 0.276). Overall, rainfall amounts in July were significantly higher for high elevation sites (median 178.46 mm, rang: 162.76–193.71 mm) relative to low elevation sites (81.99 mm [63.46–132.98 mm]; χ = 6.55 d.f. = 1, P = 0.011). It should be noted that with the exception of May, October and December, rainfall within each month yielded a significant positive association with the site-specific flea diversity index. These months may represent time periods when rainfall was similarly above (during the heavy rains) or below (during the dry season) critical rainfall thresholds for sites above or below 1300 m. Flea diversity was negatively associated with the average monthly maximum temperature in the warmest month of the year (February) and this relationship explained 83% of the observed variation in site-specific flea indices. (F = 39.1, d.f. = 1,9, P = 0.0002; Flea diversity = −0.105 K-1 * Tmax (Feb)+32.53). Thus, diversity was lower in areas below 1300 m that are the hottest during the February heat. Average maximum temperature for February was significantly lower above 1300 m (301.96 K [301.05–302.80 K]; 29.8°C [28.9–30.7°C]) than below (306.34 K [304.24–308.37 K]; 34.2°C [32.1–36.2°C]).