Prevalence and speciation of brucellosis in febrile patients from a pastoralist community of Tanzania

Brucellosis is an endemic zoonosis in sub-Saharan Africa. Pastoralists are at high risk of infection but data on brucellosis from these communities are scarce. The study objectives were to: estimate the prevalence of human brucellosis, identify the Brucella spp. causing illness, describe non-Brucella bloodstream infections, and identify risk factors for brucellosis in febrile patients from a pastoralist community of Tanzania. Fourteen (6.1%) of 230 participants enrolled between August 2016 and October 2017 met study criteria for confirmed (febrile illness and culture positivity or ≥four-fold rise in SAT titre) or probable (febrile illness and single SAT titre ≥160) brucellosis. Brucella spp. was the most common bloodstream infection, with B. melitensis isolated from seven participants and B. abortus from one. Enterococcus spp., Escherichia coli, Salmonella enterica, Staphylococcus aureus and Streptococcus pneumoniae were also isolated. Risk factors identified for brucellosis included age and herding, with a greater probability of brucellosis in individuals with lower age and who herded cattle, sheep or goats in the previous 12 months. Disease prevention activities targeting young herders have potential to reduce the impacts of human brucellosis in Tanzania. Livestock vaccination strategies for the region should include both B. melitensis and B. abortus.

Study eligibility. All individuals seeking care at the outpatient department of Endulen Hospital were eligible for screening, which was integrated into the patient triage processes and performed by Endulen Hospital staff. All individuals aged ≥two years with reported fever within the past 72 hours and/or a tympanic temperature of ≥38.0 °C at presentation were eligible for enrolment. After initiation of the outpatient visit and routine clinical assessment by Endulen Hospital staff, eligible patients were approached by a member of the study team for informed consent to participate in the study.
Sample collection, malaria testing and questionnaire data collection. After cleaning the participant's skin with isopropyl alcohol and povidone iodine, blood was drawn from study participants for culture, Brucella serology, and malaria testing. The target blood volume at enrolment was 40 mL, enabling distribution of two 10 mL volumes into BacT/ALERT (BioMérieux, Durham, NC, USA) aerobic blood culture bottles for automated culture, 10 mL into Castañeda media 32 (media prepared at the Animal and Plant Health Agency (APHA), Weybridge, UK), and 10 mL into a plain vacutainer (BD, Franklin Lakes, NJ, USA) for serology. For study participants weighing less than 25 kg, target blood volumes were determined based on weight 33 and paediatric BacT/ ALERT bottles were used for automated culture. For all study participants, a fill order algorithm was followed, with sample collection prioritised as follows: first automated aerobic blood culture bottle, plain vacutainer, Castañeda culture bottle, and second automated aerobic blood culture bottle. Malaria rapid diagnostic testing was performed directly from the sample collection syringe using the SD BIOLINE Malaria Ag P.f/Pan rapid diagnostic test (Standard Diagnostics/Abbott, Abbott Park, IL, USA) or CareStart Malaria HRP2 (Pf) (ACCESS BIO, INC. Somerset, NJ, USA).
A member of the study project team administered a structured, closed-ended questionnaire for each study participant (Supplementary Methods). Question topics included: demographic data, symptoms of current and recent illness, dietary practices, and animal-related activities. Data collected from study participant clinical records included the the initial clinical diagnosis recorded and any drug treatments prescribed on the day of hospital presentation (i.e., before any study diagnostic results were available).
All participants were approached for convalescent-phase blood sampling at their homes four to six weeks after enrolment. Up to 10 mL blood was collected into a plain vacutainer for convalescent-phase serology.
Blood culture. Inoculated culture bottles were packed for transportation to achieve a target temperature range of 4-10 °C 34 . Tinytag Transit 2 Temperature Data Loggers (Gemini Data Loggers Ltd, Chichester, UK) were used to monitor transport temperature. Inoculated culture bottles were transferred to the Kilimanjaro Clinical Research Institute (KCRI) in Moshi, Tanzania, for laboratory processing the day after inoculation. Time and date of inoculation and receipt at the laboratory were recorded. Prior to incubation, the bottom of the BacT/ ALERT blood culture bottle was visually assessed to confirm no colorimetric change. BacT/ALERT bottles were then loaded into the BacT/ALERT 3D instrument and incubated for up to 5 days. Castañeda bottles were incubated in a CO 2 incubator at 5-10% CO 2 and 37 °C. Bottles were examined for growth every 72 hours for up to 35 days. For two periods during the study (first from 30 April 2017 to 16 June 2017 inclusive and second from 13 July 2017 to 30 September 2017 inclusive) BacT/ALERT bottles were also processed by manual culture methods due to technical malfunction with the BacT/ALERT system. Standard methods were used for identifying isolates 35,36 . Isolates of gram-negative coccobacilli that had positive reactions for urease, catalase and oxidase were classified as presumptive Brucella and stored on Microbank beads (Pro-Lab Diagnostics, Bromborough, UK) at −70 °C. Culture bottles were classified as adequately filled if the blood volume added met the supplier's recommended volume +/− 20%. The following organisms were considered likely contaminants: non-anthracis Bacillus spp., Corynebacterium spp., Escherichia vulneris, Micrococcus spp., coagulase-negative Staphylococcus spp., viridans streptococci. Additionally, Pantoea spp. was identified in three bottles of Castañeda media and was also classified as a likely contaminant in our analysis. The results of positive blood culture were shared immediately with the clinical team at Endulen Hospital to inform clinical management.
Brucella speciation. Presumptive Brucella spp. isolates were shipped on dry-ice to APHA, UK, for confirmatory testing. Identification of Brucella species by culture was performed according to an established typing scheme 37 . Crude lysates for use as a template for PCR were prepared by suspending a single colony from solid media in 500 µL of nuclease-free water and heating (100 °C for 10 minutes). Molecular confirmation of culture-based species typing was performed using multiplex conventional PCR 38 , quantitative PCR 39 and application of a nine locus Brucella spp. multilocus sequence typing scheme 40,41 . Brucella serology. Filled vacutainer tubes were inverted five times immediately after blood collection and kept at ambient temperature for 45-60 minutes to ensure clotting prior to centrifugation at 1300-1500 g for 10 minutes. Serum was pipetted into cryovials and stored temporarily at 4 °C at Endulen Hospital before being transported at 4-10 °C to KCRI. Tinytag Transit 2 Temperature Data Loggers monitored transport temperature of serum samples. Sera were stored at −80 °C on arrival at KCRI. Sera were shipped on dry ice to the APHA, for serological testing using the Serum Agglutination Test (SAT). Standardised antigen for the detection of antibodies to B. abortus, B. melitensis and B. suis was used (RAA0054, APHA, Weybridge, UK) at a working strength to give 50% agglutination with a 1/650 dilution of the International Standard Serum to B. abortus. Samples were screened in a microtitre plate at serum dilutions of 1/5, 1/10, 1/20 and 1/40, with total volume per well of 200 µl. Any sample showing agglutination in any screening titre was then retested using the tube test format to confirm the final titre.
Brucellosis case definition. Brucellosis cases were defined as meeting the study criteria for febrile illness of reported fever within the past 72 hours, and/or a tympanic temperature of ≥38.0 °C at hospital presentation, plus laboratory evidence of infection. Laboratory evidence of confirmed brucellosis was defined as a blood culture positive for Brucella spp. or a four-fold or greater rise in Brucella antibody titre between acute and convalescent serum samples. Laboratory evidence of probable brucellosis was defined as a SAT antibody titre of ≥160 in either acute or convalescent-phase serum 42 .

Data management and statistical analyses.
A sample size calculation was performed to estimate the number of enrolled febrile hospital participants required in order to (i) detect a minimum number of blood culture positive individuals and (ii) estimate brucellosis case prevalence (based on culture and serologically identified cases). Assuming that 3% of presenting febrile individuals would be Brucella blood culture positive as in a previous study in Egypt 43 , it was estimated that a sample of 348 individuals was needed to detect a minimum of six Brucella blood culture positive cases with 95% probability. Based on an assumed prevalence of 9% 18 , it was estimated that a sample of 126 individuals would be required to estimate brucellosis case prevalence with precision of 5% and 80% power.
Data were entered using the OpenText TeleForm system (OpenText, Waterloo, Ontario, Canada) into an Access database (Microsoft Corporation, Redmond, WA, USA). Data manipulation and statistical analyses were performed using R 44 . Proportions were reported within exact binomial confidence intervals. Generalised linear models were used to evaluate associations between brucellosis case status (positive or negative, combining confirmed and probable cases) and clinical data. Univariable logistic regression models were used to investigate pairwise associations between brucellosis case status and candidate risk factors for human brucellosis. Questionnaire variables were selected for evaluation based on the previously identified or biologically plausible risk factors. The following variables were evaluated: age (years), sex, and a series of livestock-related risk variables occurring within the previous 12 months. In each case, livestock refers to cattle, sheep and/or goats. Variables evaluated were: assisting livestock parturition, occurrence of livestock abortion or still-born offspring in flock/herd, milking livestock, herding livestock, contact with livestock waste (e.g., cleaning animal enclosures or in the construction of buildings), slaughtering or butchering livestock, consumption of raw meat (including offal and raw animal blood) from cattle, sheep and/or goats, and consumption of raw dairy products from cattle, sheep and/or goats. Univariable models were evaluated using likelihood ratio tests (LRT), with a significant p value ≤0.05. Given the small number of positive brucellosis cases and high potential for data overfitting, variable selection was performed using lasso regression using the R package glmnet 45 . Variables retained in the lasso regression were fitted in a multivariable logistic regression model and evaluated using LRT. Odds ratios (OR), adjusted odds ratios (aOR), and 95% confidence intervals (CI) were estimated.
Research clearance and ethics. Approval to conduct the the study was granted by the Tanzania

Results
From 15 August 2016 to 11 October 2017, 3,473 patients were screened for enrolment into the study. Of the 3,473 screened patients, 435 (12.5%) were eligible for inclusion in the study. A total of 230 participants enrolled and contributed data for the study analyses ( Fig. 1).
Blood culture and Brucella speciation. Fourteen (6.1%) of 228 participants with one or more inoculated blood culture bottle had a bloodstream infection. Brucella spp. was the most common bloodstream infection. Eight (3.5%) of 228 participants were Brucella spp. culture positive. Six other pathogenic bacterial species were identified, each in a single participant (Table 1). Brucella spp. isolates from seven (3.0%) participants were identified as B. melitensis while B. abortus was isolated from one participant (0.4%). The results of molecular speciation assays were congruent with phenotyping. Multilocus sequence typing (MLST) identified all seven B. melitensis isolates as sequence type (ST) 12, while the single B. abortus isolate was identified as ST32.
In total, 601 culture bottles were inoculated, with one or more bottles inoculated for 228 (98.3%) of 232 participants. Of the 601 culture bottles, 531 (88.4%) bottles collected from 215 participants were adequately filled and 21 (3.5%) of 601 were contaminated. The median time interval between sample collection and processing at the laboratory was 25.9 (interquartile range 24.8 to 27.0) hours. A total of 77 (85.0%) of 91 shipments of inoculated culture bottles stayed within the target temperature range, all inoculated culture bottles were processed in the laboratory. The comparison of Brucella spp. isolation success from adequately filled culture bottle types is given in Table 2. Brucella spp. was identified by culture from 16 bottles from the eight culture positive participants, twelve of which had adequate fill volumes. Brucella spp. was identified in the Castañeda media bottle for all eight of the culture positive individuals. Brucella spp. was identified in one or more of the inoculated BacT/ALERT bottles for five of the eight culture positive individuals including adequate and non-adequate fill volumes. The details of Brucella spp. isolation success from the individual culture bottles used for the eight Brucella spp. positive individuals is given in Supplementary Fig. S1.The combination of BacT/ALERT bottles used for participants varied throughout the timeline of the study, largely governed by stock and supply variation ( Table 2). www.nature.com/scientificreports www.nature.com/scientificreports/ Brucella serology. Of 232 participants, 230 had acute and/or convalescent serum collected. Paired acute and convalescent phase samples were collected for 154 (70.0%) of 230 participants: 70 participants had acute serum collected only; and six had convalescent serum only, due to an insufficient blood volume collected at acute hospital presentation. Eleven (4.8%) of 230 participants met the case definition for probable brucellosis with one or more SAT antibody titre ≥160. One additional participant met the case definition for confirmed brucellosis based on serology data alone, showing a four-fold rise in antibody titre.

Brucellosis case classification.
A total of nine (3.9%) of 230 participants met criteria for confirmed brucellosis; eight based on blood culture and one by seroconversion. Six (75.0%) of the eight blood culture positive participants had an acute phase SAT titre ≥160, with four of the six also having a convalescent phase SAT titre ≥160. In total, 14 (6.1%) of 230 participants, met the study definition for a probable or confirmed brucellosis case. The correspondence between culture and serology results for study participants and details of the dates of culture positives are shown in Supplementary Fig. S2. participant characteristics, clinical presentation and management. Study participants had a median (range) age of 27 (2, 78) years and brucellosis cases had a median (range) age of 11 (7,20) years. Ten (71.4%) of 14 brucellosis cases were male and 13 (92.9%) reported their tribe as Maasai. Of 232 study participants, 226 (97.4%) reported residence in the NCA with 170 (73.3%) from Endulen village. Residence was also reported in the larger Arusha region and adjacent Simiyu region (Fig. 2). All brucellosis cases were participants living within the NCA.  The relationship between clinical symptoms and brucellosis case status, as well as details of participant clinical management are shown in Table 3. Brucellosis cases were significantly more likely to report night sweats and less likely to report back pain as compared to non-cases (Table 3). Brucellosis case status was significantly associated with receiving a presumptive diagnosis of brucellosis at hospital presentation (LRT χ 2 = 4.82, df = 1, p = 0.03, n = 188) and admission to the inpatient ward of the hospital (LRT χ 2 = 4.97, df = 1, p = 0.03, n = 191). Sixteen (7.0%) of 230 participants, were prescribed a drug regimen consistent with treatment for brucellosis at their initial visit. All of the participants treated for brucellosis had a presumptive diagnosis of brucellosis and treatment for brucellosis was significantly associated with case status as defined by this study (LRT χ 2 = 6.71, df = 1, p = <0.01, n = 230). Data on treatment regimens prescribed after the provision of blood culture results (e.g., after the date of the initial visit) are not included in our dataset. Six (2.6%) of 230 participants had a test positive for malaria. No brucellosis cases were malaria positive.
Risk factors for brucellosis. The univariable associations between the selected risk variables and brucellosis case status are shown in Table 4. Risk factor variables significantly associated with brucellosis in febrile hospital participants in univariable models were: age of participant (LRT χ 2 = 20.05, df = 1, p = <0.01, n = 229), with probability of brucellosis declining with increasing age in years; sex (LRT χ 2 = 4.76, df = 1, p = 0.03, n = 230), with a greater probability of brucellosis in male participants; and herding livestock in the last 12 months (LRT χ 2 = 14.09, df = 1, p = <0.01, n = 219), with participating in herding cattle, sheep and/or goats associated with increased probability of brucellosis. The lasso regression indicated that herding livestock and age variables should be retained. Variables significantly associated with brucellosis in febrile hospital participants in the final multivariable model were: age of participant (LRT χ 2 = 18.17, df = 1, p = <0.01, n = 219), with probability of brucellosis declining with increasing age in years; and herding livestock in the last 12 months (LRT χ 2 = 11.71, df = 1, p = <0.01, n = 219), with participating in herding cattle, sheep and/or goats associated with increased probability of brucellosis (Table 4).

Discussion
We found that 3.9% of febrile participants had confirmed brucellosis and 6.1% met critieria for confirmed or probable brucellosis in a pastoralist area of Tanzania. Brucella spp. was the most commonly identified bloodstream infection in this population and both B. melitensis and B. abortus were isolated, with a predominance of B. melitensis. Risk factor analysis indicated that young livestock herders were at particular risk of infection.
To our knowledge, this is the first study to characterise bloodstream infections in a pastoralist community of Tanzania. Brucella spp. was the most commonly identified bloodstream infection in this population, and this contrasts with previous studies which have typically identified Brucella spp. as a relatively rare bloodstream infection in sub-Saharan Africa 25 . Several more commonly reported bloodstream infections, including non-typhoidal serovars of Salmonella enterica, Staphylococcus aureus, Streptococcus pneumoniae, and Escherichia coli were also identified in this study population, but at lower relative frequency as compared to Brucella spp. 25,26 . Malaria was less prevalent than brucellosis in our study participants, as has been reported previously for northern Tanzania 46 .  6 . Our study identified B. melitensis as the predominant cause of brucellosis, and B. abortus was also identified in one participant, confirming the role of both Brucella species as causes of human illness in this setting. The Brucella genotypes identified represent sequence types that are almost exclusively associated with Africa, but widespread within the continent. B. melitensis ST12 has been identified previously in human cases associated with Somalia, Ethiopia, Eritrea, Kenya, Uganda, Nigeria, and Egypt, and in animal infections from Zimbabwe, South Africa, and a single Tanzanian livestock isolate 47 . Fewer isolates of B. abortus ST32 than B. melitensis ST12 have been reported from livestock and humans, in Kenya, Chad, and Rwanda 47 .
The isolation of both B. melitensis and B. abortus indicates that the host species for brucellosis in northern Tanzania are likely to include several of the key livestock species kept in this study area, such as cattle, sheep, and goats. Camels are not common within the study area. Consistent with the predominance of B. melitensis in this study, previous work in the region has identified small ruminants specifically as the most likely sources of human Brucella exposure 48,49 . The NCA is a wildlife conservation area and common species including buffalo (Syncerus caffer) and wildebeest (Connachaetes taurinus) are known hosts of Brucella spp. 50 . Transmission from livestock to humans is likely to account for the majority of human infections, but transmission from wildlife to humans in this setting cannot be ruled out.
The overall prevalence of brucellosis identified in this study is consistent with the findings of previous febrile surveillance studies conducted in Moshi, Tanzania, providing additional evidence of the health impacts of brucellosis in northern Tanzania. Using similar diagnostic approaches these previous studies have detected serologically confirmed brucellosis at between 3.5% and 6.9% 18,19 . All confirmed cases in these previous studies were defined by seroconversions 18,19 . In this present study, the majority of confirmed cases were blood culture positive and the number and proportion of confirmed cases defined by seroconversion was lower than expected. This difference in the ratio of culture to seroconversion-defined confirmed cases may relate to the timing of participant presentation at hospital. Brucella spp. isolation is typically more likely early in acute infection 21 . Confirmation of cases based on demonstration of seroconversion is most likely when an acute phase sample is collected very soon after initial infection and a convalescent phase sample is also collected several weeks after initial infection. Interestingly, the majority of blood culture-confirmed brucellosis cases identified in this study also had acute phase sample SAT titre ≥160, indicating that they had already seroconverted by the time of presentation and also had bacteraemia that was still detectable by blood culture (Supplementary Fig. S2). The majority of participants in this study lived close to Endulen Hospital, whereas earlier studies in Moshi were conducted at a referral hospital 18,19 . All confirmed cases in the previous studies were defined by seroconversion and it is plausible that longer time to presentation explain the absence of culture-confirmed cases in these studies 18,19 . Two Brucella spp. isolates have been obtained from similar ongoing studies in Moshi (Crump JA & Rubach MP, unpublished data). Patients attending an urban referral hospital may have had increased access to antimicrobials prior to hospital presentation, again   www.nature.com/scientificreports www.nature.com/scientificreports/ reducing chances of positive culture. Additionally, our study was designed to preferentially detect acute brucellosis in acutely febrile individuals seeking healthcare, and is thus less likely to detect chronic brucellosis cases. It has been estimated that 40% of brucellosis cases may persist with chronic disease for two years following disease onset, and 10% may persist with clinical manifestations after six years 51 . Further investigations of the combined health impacts of acute and chronic human brucellosis in this population are warranted to evaluate the combined burden of disease 52 .
Risk factors associated with brucellosis in the univariable models in this study included younger age, male sex and herding livestock. Lasso regression selected herding livestock and age variables, which were retained for the multivariable model. The fact that the sex variable was not retained in the final model is likely to be due to correlation between male sex and having herded livestock. Higher risk of Brucella spp. exposure at an early age has been recognised for nomadic groups, with adults often more likely suffering from chronic brucellosis infection 2 . Within the largely Maasai community of the NCA, young boys are preferentially given responsibility for herding livestock (RFB personal communication during NCA community meetings). Herding can involve multiple activities that may put individuals at higher risk of exposure to Brucella spp. as well as many other zoonotic pathogens. These activities may include: frequent contact with livestock, assisting in animal parturition, butchering livestock and consuming specific organs raw or eating undercooked meat and blood 53 , and drinking raw milk directly from animals in the herd or flock 53 . Some of the more common risk factors for human brucellosis reported in East Africa such as: consumption of raw animal products 27,54 , assisting in animal parturition events and contact with aborted animal materials 31,55,56 and slaughtering 57 were not identified in this study. The small number of brucellosis cases identified and the high proportions of the study population participating in perceived high risk activities, complicate efforts to disentangle the relative importance of specific risk activities and transmission pathways in this population. Further investigation into activities conducted whilst herding that may increase exposure to Brucella spp. may identify additional specific risk factors for infection in the young herders identified as high risk in our study.
Diagnosis of brucellosis by clinical symptoms alone is unreliable due to varying clinical manifestations of infection 10 . In the present study brucellosis case status was positively associated with night sweats but negatively associated with back pain. These findings reinforce the challenges of using clinical symptoms for brucellosis case identification. In spite of this, brucellosis was a relatively common presumptive diagnosis made by clinicians at the Endulen Hospital, indicating high clinical awareness of the disease in this setting. There was also a significant association between presumptive diagnosis of brucellosis and brucellosis case status. Not all participants with a presumptive diagnosis of brucellosis were started on a brucellosis consistent treatment on the day of hospital presentation. However, the data presented on treatments refer to treatments prescribed at initial presentation only and do not capture later decisions based on additional findings, including the subsequent blood culture results provided through the study. Overall, these data highlight the challenges faced by clinicians in areas where brucellosis is endemic but access to high quality diagnostics including blood culture and SAT serology is limited. There is a need for improved diagnostic tools and diagnostic guidelines for the management of acute brucellosis and febrile illness more generally in Tanzania, and East Africa more widely 11 .
Brucellosis has been identified as one of six priority zoonotic diseases in Tanzania 58 , prompting the development of a national strategy for brucellosis prevention and control in humans and animals 59 . These study findings, including human prevalence estimates, identification of both B. melitensis and B. abortus as causes of human illness and risk factors for acute human illness, can inform the development of evidence-based control strategies for brucellosis in Tanzania.   www.nature.com/scientificreports www.nature.com/scientificreports/ conclusions We found that brucellosis was the cause of 6.1% of febrile illness among study participants presenting for outpatient hospital care, and that Brucella spp. are the most commonly identified bloodstream infection in this predominantly pastoralist community from the NCA, Tanzania. Our findings show that human brucellosis is caused by both B. melitensis and B. abortus consistent with transmission from multiple livestock host species. Within this pastoralist community, young livestock herders are at high risk of brucellosis infection. Brucellosis control activities in Tanzania which focus on prevention of transmission to individuals at high risk of infection, and livestock vaccination campaigns directed at both cattle and small ruminants have potential for substantial impacts in reducing human brucellosis risk.

Data availability
The datasets generated during and/or analysed during the current study are available in the Enlighten research data repository of the University of Glasgow (http://dx.doi.org/10.5525/gla.researchdata.978). The sequence typing data for the isolates from this study are available in the Brucella MLST Databases https://pubmlst.org/ brucella/.