Comparison of Zoonotic Bacterial Agents in Fleas Collected from Small Mammals or Host-Seeking Fleas from a Ugandan Region Where Plague Is Endemic

Fleas play critical roles in transmitting some infections among animals and from animals to humans. Detection of pathogens in fleas is important to determine human risks for flea-borne diseases and can help guide diagnosis and treatment. Our findings of high prevalence rates of B. elizabethae and R. felis in fleas in the Arua and Zombo districts of Uganda implicate these agents as potential causative agents of undiagnosed febrile illnesses in this area.

of cases from Africa. Northwestern Uganda, often referred to as the West Nile region, is a region where plague is endemic, with more than 2,400 suspect plague cases reported between 1999 and 2011 (1). Multiple studies on fleas have been conducted in this area (2)(3)(4)(5).
The genus Bartonella, with more than 30 species identified, has drawn increased attention as more and more reported clinical illnesses have been associated with Bartonella infections (6)(7)(8)(9)(10)(11). Rickettsia felis, the causative agent of flea-borne spotted fever, is considered an important neglected agent in sub-Saharan Africa (12). Both bartonelloses and rickettsioses are widely regarded as emerging/reemerging diseases (13). Many studies have reported Bartonella spp. in a variety of mammalian species and their ectoparasites, all over the world (14)(15)(16)(17)(18)(19)(20). In the Democratic Republic of Congo, Sackal et al. (21) reported the detection of Bartonella spp. and R. felis in Ctenocephalides felis and Pulex irritans collected from flea traps. Billeter et al. (22) identified Bartonella species in invasive and indigenous rodents and their fleas from Uganda. Surprisingly, the prevalence of Bartonella infections in invasive rats (Rattus rattus) reported in that study was very low (1.3%), a finding that stands in stark contrast to results of a study conducted in neighboring Kenya, in which the Bartonella prevalence was 13% in Rattus spp. (19). In addition, R. felis and "Candidatus Rickettsia asemboensis" were detected in fleas collected in Kenya, Tanzania, Ethiopia, and Uganda (13,(23)(24)(25).
Although Bartonella spp. and Rickettsia spp. have been reported in countries neighboring Uganda, information regarding the distribution of Bartonella spp., Rickettsia spp., and Yersinia spp. in fleas from Uganda is very limited and requires a better understanding, as each of these agents is likely to cause disease in humans. In this study, we evaluated the prevalence rates and characterized the genetic identities of Bartonella spp., Rickettsia spp., and Yersinia spp. in fleas collected from the West Nile region of Uganda.
Bartonella infections. Bartonella DNA was detected in 91 of the 632 tested fleas, with an overall prevalence of 14%. Among the positive fleas, 88 (97%) were collected from small mammals; only 3 were collected from Kilonzo traps. The Bartonella prevalence was 22% (88/407) in fleas collected from small mammals and 1% (3/225) in fleas collected from Kilonzo traps, showing a significant difference between the two sources ( 2 ϭ 38.4, P Ͻ 0.01). The Bartonella-infected fleas belonged to eight species (Table 1). Compared to other flea species, D. lypusus and C. calceatus fleas had a higher rate of Bartonella infection, with 51% (54/107) and 27% (12/44) of individuals infected, respectively. Of the most prevalent flea species found on rodents, X. cheopis, only 9% (18/202) of fleas of this species were found to be infected with Bartonella spp. Interestingly, only 3 of the 18 positive X. cheopis fleas were collected from black rats, which were the main hosts of X. cheopis in this study ( Table 2). In other words, only 3 of 148 X. cheopis fleas (2%) collected from black rats were infected with Bartonella spp. X. brasiliensis fleas also had a very low infection rate for Bartonella species, with only 1 positive flea among 43 tested fleas (2%). The three Bartonella-positive fleas that were collected from Kilonzo traps were of the species C. felis, S. torvus, and X. cheopis. Notably, only 1 of 182 C. felis fleas was infected with Bartonella. None of the 30 E. gallinacea fleas, all of which were collected from Kilonzo traps, was infected (Table 1).
Rickettsia infections. Rickettsia DNA was detected in 143 of the 632 tested fleas, giving an overall prevalence of 23%. Nearly all of the Rickettsia-infected fleas were of the species C. felis (88%; 126/143) or X. cheopis (10%; 14/143) ( Table 1). In contrast to Bartonella-infected fleas, the majority (88%; 126/143) of Rickettsia-positive fleas were collected from Kilonzo traps. Only 17 of 173 Rickettsia-positive fleas (12%) were collected from rodents. The prevalence of Rickettsia in off-host fleas was 56% (126/225), whereas it was 4% (17/407) in fleas collected from rodents. The prevalence of Rickettsia in off-host fleas was significantly higher than that in fleas collected from rodents ( 2 ϭ 129.9, P Ͻ 0.01). Of the 126 Rickettsia-positive fleas that were collected from Kilonzo traps, 125 were C. felis. The prevalence of Rickettsia in C. felis was also extremely high  (Table 1). Yersinia infections. Neither Y. pestis nor Y. pseudotuberculosis DNA was detected in any of the fleas tested.
Genetic identification of Bartonella and Rickettsia species. Sequencing analyses were performed based on gltA sequences for both Bartonella and Rickettsia. The 91 Bartonella-positive DNA samples represented 12 distinct genetic variants that have been deposited in GenBank under accession numbers MF443355 to MF443366. These variants were clustered in four major phylogroups (Fig. 1). Nine variants (GenBank accession numbers MF443355 to MF443363) accounted for 77 of the 91 sequences identified, and all of these were relatively close to each other and similar to a few variants previously described in rodents from Uganda (22). These variants are clustered with the so-called Bartonella elizabethae species complex (26). Sequences belonging to this phylogroup were obtained from fleas of all species found positive, with the majority (70%; 54/77) coming from D. lypusus fleas. In fact, all sequences obtained from D. lypusus fleas belonged to this group, showing a specific relationship between this flea species and Bartonella species. Bartonella sequences obtained from the three positive fleas that were collected from Kilonzo traps also belonged to this phylogroup; one variant (GenBank accession number MF443364) was similar to variant KM233490, which was described in an African giant shrew (Crocidura olivieri) from Kenya (19). This variant contained 10 identical sequences that were obtained from X. cheopis (n ϭ 8) and C. calceatus (n ϭ 2) and formed a separate phylogroup. Another variant (GenBank accession number MF443365) was similar to the variant KT881099 from a Mastomys sp. rat in Benin (27). This variant contained only two sequences, which were obtained from one C. calceatus flea (collected from A. niloticus) and one X. cheopis flea (collected from R. rattus), respectively. They are also quite distant from other identified Bartonella genotypes and formed a separate phylogroup. The last variant (GenBank accession number MF443366) contained two sequences, both of which were obtained from C. calceatus fleas that were collected from African grass rats. This variant was similar to Bartonella rochalimae (Fig. 1).
The 143 Rickettsia-positive DNA sequences represented six distinct genetic variants and have been deposited in GenBank under accession numbers MF459639 to MF459644. The six variants were clustered into three phylogroups. Variants MF459639 and MF459640, containing 15 and 30 Rickettsia sequences, respectively, belonged to the zoonotic pathogen R. felis; variants MF459641 and MF459642 were the most common and comprised 43 and 51 of all identified Rickettsia sequences, respectively. These two variants were closely related to "Candidatus R. asemboensis"; the last two variants, MF459643 and MF459644, containing three and one sequence, respectively, were close to "Candidatus Rickettsia sengalensis" (Fig. 2).

DISCUSSION
Using molecular approaches, we estimated the presence of multiple potential zoonotic pathogens, including Bartonella spp., Rickettsia spp., and Yersinia spp., in fleas collected from small mammals and those that were host seeking in human habitations in northwestern Uganda, where plague is endemic and occasionally causes outbreaks in the local human population (28).
Not surprisingly, Bartonella spp. were prevalent in the fleas, with an overall prevalence of 14%. The prevalence in fleas collected from small mammals was much higher Each genetic variant is indicated by its GenBank accession number in boldface, followed by the flea species and number of identical sequences obtained from the flea species in brackets and parentheses. The Rickettsia variants formed three phylogroups (boxed clades). Two variants fell into the cluster of R. felis; other variants clustered with either "Candidatus R. asemboensis" or "Candidatus R. sengalensis." The phylogenetic tree was constructed by the neighbor-joining method, and bootstrap values were calculated with 1,000 replicates. than in off-host fleas (22% versus 1%). In fact, two of the three Bartonella-positive off-host fleas (S. torvus and X. cheopis) commonly parasitize small mammals (Crocidura sp. and Rattus spp. rats, respectively) and are likely to be found only temporarily off their mammal hosts. It is not clear whether these Bartonella spp. cause pathogenic damage in their animal hosts; it is less likely these fleas seek blood meals from humans. Given the context of assessing human risk, studies of vector competence by various flea species might be necessary. The off-host fleas were mostly comprised of C. felis (80%). In the Democratic Republic of the Congo, Sackal et al. (21) similarly found that only 3.6% of off-host fleas (mainly C. felis and Pulex irritans) were infected by Bartonella spp. The absence of Bartonella in the C. felis fleas in the present study might be related to their off-host origin. Bartonellae are known hemotropic parasites (29) and might not be able to survive in fleas that fail to take frequent blood meals, and starving fleas might be quickly cleared of any Bartonella infection. Another possible explanation is that these C. felis fleas may rarely become infected by Bartonella because of the low level of infection in their presumptive hosts. As common ectoparasites of cats and dogs, C. felis has been reported to carry B. henselae (the causative of Cat scratch disease), B. clarridgeiae, and other Bartonella species (30,31). In northwestern Uganda, however, C. felis feeds primarily on people, goats, and occasionally on chickens, sheep, guinea pigs, or pigs (32). Previous studies have reported that Bartonella bacterium was not cultured in goats (33) and was not detected in chickens from Kenya (Y. Bai, unpublished data).
Among the fleas collected from rodents, D. lypusus demonstrated an extremely high prevalence (Ͼ50%) of Bartonella infection, suggesting that this species of flea may play a major role in transmitting Bartonella infection between rodents and potentially from rodents to humans. Again, the vector competence should be studied for this flea species. Although the prevalence of Bartonella in fleas collected from rodents was relatively high, the prevalence in X. cheopis fleas was quite low, with only 9% of this flea species being found infected. Notably, the X. cheopis fleas collected from black rats, which are the main hosts of the fleas, showed an even lower prevalence, with only 2% infected. Typically, high percentages (Ͼ40%) of Rattus spp. animals are infected by Bartonella spp. (14,15). However, Billeter (22) reported that only 2% of R. rattus animals were infected with Bartonella spp. in another study in Uganda. R. rattus invaded the West Nile region of Uganda recently (34) and may have not had ample time to establish a niche in which Bartonella bacterium can amplify. The competence of X. cheopis is not clear, but field investigations have suggested that Bartonella infection is most likely transmitted between rats by X. cheopis (35). In our study, the low number of X. cheopis fleas that were found to be infected with Bartonella spp. might help explain the very low prevalence of this infection in R. rattus from the West Nile region, previously reported by Billeter (22).
Sequencing analysis demonstrated that at least four Bartonella species or phylogroups are circulating in the flea communities in the region of Uganda where plague is endemic. Most of the Bartonella sequences (85%) obtained from the fleas in this study belonged to the B. elizabethae species complex (26), and several species and strains of this complex have been linked to human illnesses recently (6)(7)(8)(9)(10). Importantly, B. elizabethae was found in all flea species that were positive for Bartonella in the present study. Our findings suggested that B. elizabethae might be the causative agent for undiagnosed illnesses in local residents, and it is recommended that further investigations be conducted to verify whether B. elizabethae is indeed a human pathogen in this region of Africa. Remarkably, D. lypusus fleas, a very common species found on rodents, were all infected with B. elizabethae, showing a very specific relationship between this flea species and the Bartonella species. D. lypusus is a known vector of Y. pestis in Kenya (36). Our results suggest that they may also transmit B. elizabethae to humans in this region. B. rochalimae, another human pathogen (11) which has been also reported in carnivorous species and their fleas (Pulex simulans) (37)(38)(39) and in rodents (40), was also detected in two fleas in this study, suggesting that the presence of this bacterium in local flea communities should not be ignored.
The other two Bartonella species or phylogroups (variants MF443364 and MF443365) were found in small numbers of fleas. Interestingly, variant MF443364 was quite different from other identified genotypes in our study but was similar to the variant KM233490 previously described in an African giant shrew (Crocidura olivieri) from Kenya (12). This variant was found in 10 fleas which were collected from rodents (A. niloticus, Mastomys spp., and R. rattus) but not from shrews (Crocidura spp.) in the current study.
Although Bartonella DNA was detected in fleas collected from four Crocidura in the present study, all of the Bartonella genotypes belonged to the B. elizabethae species complex. Further investigations are needed to illustrate the relationship between shrews and Bartonella spp. in Africa, but it should be noted that we have found shrews infested with X. cheopis, X. brasiliensis, D. lypusus, and C. calceatus fleas, providing a possible means for these animals to become infected with B. elizabethae-like bartonellae. Like many previous reports, we found that Rickettsia infection is commonly circulating in the local flea community, with an overall prevalence of 23%. In contrast to Bartonella, Rickettsia-infected fleas were mainly C. felis fleas collected from Kilonzo traps. Only a very small portion of Rickettsia-infected fleas was obtained from rodents. Such results suggest that Rickettsia infections in the West Nile region are carried primarily by C. felis fleas.
Sequencing analysis demonstrated three Rickettsia species were circulating within the local flea community in the region of Uganda where plague is endemic. R. felis, a frequently identified zoonotic pathogen that causes cat-flea typhus in humans, was commonly detected in the C. felis fleas we tested. Due to its worldwide distribution, R. felis infection has been considered an emergent threat to human health (13). Because C. felis fleas in Uganda commonly feed on humans in the West Nile region and C. felis is a well-known competent vector of R. felis (41), our results suggest that humans might be exposed to cat flea-transmitted rickettsioses caused by R. felis or another closely related rickettsial species. The other two species identified in this study were "Candidatus R. asemboensis" and "Candidatus R. sengalensis." In fact, "Candidatus R. asemboensis" was the most common species detected in the study. Although it is not clear whether this species can cause disease in humans or animals, further investigations are needed regarding its wide distribution. The number of E. gallinacea fleas collected from Kilonzo traps was the second largest, after the number of C. felis fleas taken from these traps. Interestingly, neither Rickettsia DNA nor Bartonella DNA was detected in these fleas. Although these fleas occur on a wide range of bird and mammal hosts, they are commonly known as poultry fleas, a fact that could explain the absence of both bartonellae and rickettsiae in populations of these fleas.
Neither Y. pestis nor Y. pseudotuberculosis was detected in the present study. Notably, the villages tested were those lacking a history of plague cases, although all were within the portion of the West Nile region considered at risk for plague. Even in villages where plague has been reported repeatedly, it is very uncommon to find Y. pestis in fleas collected from rodents or to find this bacterium in years when there is no evidence of epizootic activity among rodent populations. Some researchers have suggested that Y. pestis circulates at very low and difficult-to-detect levels in fleas and rodents or is maintained in another reservoir, such as soil, during the interepizootic periods, but the true reservoir is yet to be determined (42,43).
In summary, the relatively high prevalence rates of different genotypes of Rickettsia and Bartonella in fleas from the West Nile region where plague is endemic suggest that flea-borne diseases other than plague are likely present in this part of Uganda. The potential importance of these flea-borne diseases is also supported by the high abundance of certain species of fleas in this area, including C. felis, X. cheopis, and X. brasiliensis, each of which is known to bite people. Nevertheless, the vector competence of these flea species should be studied in order to gain understanding of whether these fleas pose a risk to humans. In addition, an animal host likely would not die from a Bartonella or Rickettsia infection; therefore, the relevant flea species may not leave their natural hosts. Further research is needed to determine the extent to which humans in this region become infected with not only Y. pestis but also with bacteria of the genera Rickettsia and Bartonella and whether the people infected with these two arthropod-associated bacteria suffer clinically recognizable illnesses.

MATERIALS AND METHODS
Study sites and flea collections. Study sites and flea and small mammal collection methods were described previously (3). Briefly, 15 villages, including 10 case villages and 5 control villages, were selected within portions of the Arua and Zombo districts that were considered to have increased plague risk, as classified by geographic information system-based statistical models that included variables for elevation (Ͼ1,300 m), precipitation, amounts of vegetative growth, and bare soil during the dry month of January (44). In other words, the study villages were believed to have conditions ecologically conducive for plague activity. Within each village, 10 homesteads were randomly selected for sampling. The fleas screened in this study were derived from the five control villages within which no human plague cases had been reported from 1999 to 2011. The fleas were collected either from small mammals that were trapped inside huts and surroundings of the selected homesteads or from modified Kilonzo flea traps (45) that were set inside huts. Collected fleas were stored in 70% ethanol and shipped to the CDC laboratory in Fort Collins, CO, for species identification based on morphological characteristics using appropriate taxonomic keys (46)(47)(48)(49) and bacterial testing.
DNA extraction and PCR detection. DNA was extracted from individual fleas. The fleas were first homogenized using a Bullet Gold blender homogenizer (Next Advance, Averill Park, NY). The homogenates were then transferred to a KingFisher platform (Life Technologies, Inc., Grand Island, NY) for DNA extraction following the tissue protocol provided by the manufacturer. A blank well was included as a negative control to ensure no cross-contamination occurred during the extraction. The DNA was then analyzed for the presence of Bartonella spp., Rickettsia spp., Yersinia pestis, and Y. pseudotuberculosis by using molecular approaches. Real-time PCR was first performed in a CFX96 PCR detection system (Bio-Rad, Hercules, CA) that was set up to target ssrA, gltA, YPO2088, and opgG for Bartonella spp., Rickettsia spp., Yersinia pestis, and Y. pseudotuberculosis, respectively. ssrA for Bartonella spp. and gltA for Rickettsia spp. have been described previously and have been widely applied in many studies (50,51). The YPO2088 and opgG assays, which were developed for detection of Y. pestis and Y. pseudotuberculosis, respectively, were applied for the first time in this study. Samples showing an amplification curve with a cycle threshold (C T ) value of Ͻ36 were considered positive. A one-step or nested conventional PCR was further performed in a C1000 Touch thermal cycler (Bio-Rad, Hercules, CA) for real-time PCR-positive samples, targeting Rickettsia-specific gltA (52) and Bartonella-specific gltA (18,53), respectively. The PCR products were analyzed for the presence of amplicons of the expected sizes via electrophoresis on 1.5% agarose gels containing GelGreen stain (Biotium, Hayward, CA). Positive and negative controls were included in each PCR to evaluate the presence of appropriately sized amplicons and to rule out potential contamination, respectively. The primers and probes used for the targets mentioned in this study are listed in Table 3.
Sequencing and phylogenetic analysis for Bartonella and Rickettsia spp. The PCR amplicons from conventional PCRs for Bartonella and Rickettsia were purified using a QIAquick PCR purification kit (Qiagen, Valencia, CA) according to the manufacturer's instructions and then sequenced in both directions by using an ABI 3130 genetic analyzer (Applied Biosystems, Foster City, CA). Forward and reverse sequences were assembled using the SeqMan Pro program in Lasergene v.12 (DNASTAR, Madison, WI). A BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) search of the GenBank database was  performed with all assembled sequences to confirm their Bartonella or Rickettsia identity. Positive sequences were compared between themselves and with Bartonella and Rickettsia reference sequences available in GenBank, after alignment using the Clustal algorithm in the MegAlign program in Lasergene. Phylogenetic trees were constructed for Bartonella or Rickettsia separately, using the neighbor-joining method. Branch support was estimated using 1,000 bootstrap replicates.
Statistical analyses. The chi-square test was performed to compare the prevalence of Bartonella and Rickettsia infections between fleas collected from small mammals and fleas collected from Kilonzo traps. Results were considered significant if P was Ͻ0.05.
Accession number(s). The novel Bartonella variants and Rickettsia variants obtained in the study have been deposited in GenBank under accession numbers MF443355 to MF443366 and MF459639 to MF459644, respectively.