Ecological Niche Modeling for Filoviruses: A Risk Map for Ebola and Marburg Virus Disease Outbreaks in Uganda
Ecological Niche Modeling for Filoviruses: A Risk Map for Ebola and Marburg Virus Disease Outbreaks in Uganda
Introduction: Uganda has reported eight outbreaks caused by filoviruses between 2000 to 2016, more than any other country in the world. We used species distribution modeling to predict where filovirus outbreaks are likely to occur in Uganda to help in epidemic preparedness and surveillance. Methods: The MaxEnt software, a machine learning modeling approach that uses presence-only data was used to establish filovirus - environmental relationships. Presence-only data for filovirus outbreaks were collected from the field and online sources. Environmental covariates from Africlim that have been downscaled to a nominal resolution of 1km x 1km were used. The final model gave the relative probability of the presence of filoviruses in the study area obtained from an average of 100 bootstrap runs. Model evaluation was carried out using Receiver Operating Characteristic (ROC) plots. Maps were created using ArcGIS 10.3 mapping software. Results: We showed that bats as potential reservoirs of filoviruses are distributed all over Uganda. Potential outbreak areas for Ebola and Marburg virus disease areas were predicted in West, Southwest and Central parts of Uganda, which corresponds to bat distribution and previous filovirus outbreaks areas. Additionally, the models predict the Eastern Uganda region and other areas that have not reported outbreaks before to be potential outbreak hotspots. Rainfall variables were the most important in influencing model prediction compared to temperature variables. Conclusions: Despite the limitations in the prediction model due to lack of adequate sample records for outbreaks, especially for the Marburg cases, the model outputs provide a risk map to the Uganda surveillance system on filovirus outbreaks. The risk maps for potential filovirus outbreaks will aid in identifying areas to focus the filovirus surveillance for early detection and responses hence curtailing a pandemic. The results from this study also confirm previous findings that suggest that Filoviruses are mainly limited by the amount of rainfall received in an area.
We are grateful for funding from Norwegian Agency for Development Cooperation (NORAD) through the Norwegian Programfor
Capacity Building in Higher Education and Research for Development (NORHED) project of Capacity Building in Zoonotic
diseases Management using integrated approach to Ecosystems health at the human-livestock–wildlife interface in Eastern and
Southern Africa. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the
Uganda has experienced eight filovirus outbreaks, five Ebola Virus Disease (EVD) and three Marburg virus disease (MVD),
between 2000 and 2016, more than any other country in the world.
The first outbreak in Uganda was caused by Ebolavirus of the species Sudan ebolavirus in 2000 in the Northern district of Gulu,
where 425 cases were registered with a case fatality rate (CFR) of 53%1. The second outbreak was caused by Bundibugyo
Ebolavirus in the western part of Uganda bordering with Democratic Republic of Congo (DRC), with 192 cases and a CFR of
34%2,3 . In 2011, another EVD outbreak occurred where only one case was involved in Luweero district Zirobwe village, 45 km
North of Uganda’s Capital City Kampala4. Two more EVD outbreaks were observed in 2012, one in June in the Western District
of Kibale and another in November, Luweero district in Central Uganda5.
Likewise, three outbreaks of MVD have occurred in Uganda; the first one was in Kamwenge district in 2007 associated with
mining activity in the Kitaka gold mine that is occupied by bats6. This outbreak was later linked to cave-dwelling Egyptian fruit
bats (Rousettus aegyptiacus) that occupy these mines, as they tested positive for Marburg virus by polymerase chain reaction
(PCR)7,8. Another outbreak of MVD was in 2012 where several districts were involved with a CFR of 58 % (15/26)9. This
outbreak was also traced back to the same gold mines in Western Uganda, and subsequent testing of the bats in the mines
revealed a spill over to human populations10. The latest MVD outbreak was in Kampala where the only fatal case was a health
worker, and no other cases were identified11.
It is hypothesized that distribution of filoviruses is limited by the distribution of the probable reservoirs bats. All the filovirus
outbreaks in humans have been reported to originate from Sub-Sahara Africa and only one species, Reston virus that is not
known to infect humans was detected outside Sub-Sahara Africa in the Philippines12. It is has been suggested that
transmission from the natural reservoir occurs when humans get into contact with the reservoir or its body fluids such as feces,
urine, and blood via activities such as hunting and consumption of bush meat13. Because previous outbreaks in Central Africa
have been linked to reports of bush meat consumptions and deaths of wildlife14, many hypotheses have been put forward to
suggest wildlife such as bats, primates, and antelopes as possible sources of infection. The debate on bats as potential
reservoirs of Ebolaviruses is still not concluded, as no Ebolavirus has been isolated from bats despite finding some bats
seropositive for Ebolavirus and others with viral RNA15. The role of non- human primates as reservoirs has been unconvincing
since these do die from infection with filoviruses16,17,18,19 . Other wildlife that has been reported to be infected by Ebolavirus
was one duiker, whose bone tested positive by PCR in Republic of Congo bordering Gabon19. Dogs and pigs are the only
domestic animals associated with ebolaviruses. Dogs were found to be IgG seropositive in Gabon20 whereas Reston virus has
been reported in pigs and have shown potential for infection with Ebola virus21,22,23. Unlike EVD, there is progress in research
trying to describe the reservoirs of Marburg virus. Bats of species Egypticaosu rouseatus, found in Kitaka gold mine and Python
cave from the Albertine region in Western Uganda have been described as potential reservoirs of Marburg virus in Uganda8,10,
24. The bats in these caves have been linked to three MVD outbreaks, where artisanal gold miners got infected with Marburg
virus6,9. Transmission of Marburg virus in human populations just like Ebolaviruses happen after a spillover event from the
natural reservoir wildlife. Lack of a clear reservoir and true source of infection or spill-overs into human populations has been a
call for alternative methods of heightening surveillance and developing risk maps is one of them.
Situated in the rich and complex ecological systems with high biodiversity in East Africa, Uganda is not only affected directly by
filovirus outbreaks but also vulnerable to outbreaks from neighboring countries such as DRC.
For epidemic preparedness and response, Uganda’s health surveillance system needs to know where and when these
epidemics are likely to occur so that active surveillance can focus on those areas for early detection to avoid a pandemic but
also focused research in reservoirs. This can be achieved by applying spatial epidemiology modeling techniques. One such
technique is Ecological Niche Modeling (ENM) also known as Species Distribution Modeling (SDM), that has been used to
establish the relationship between species and their environment25,26,27,28. ENM has also been used to predict the ecology
and distribution of filoviruses before. Peterson et al used a Genetic Algorithm for Rule-Set Production (GARP) model to predict
suitable environments for filoviruses as being in afro-tropics with EVD being predicted more in the humid rain forest of Central
and West Africa and MVD was predicted to occur in the drier and more open areas of Central and East Africa29. More efforts
were made to improve the spatial prediction model for MVD for Africa using a Bioclimatic variable(Bioclim), with environmental
variables from World Climate Data 30, which predicted filoviruses mainly in Zimbabwe and abroad potential distribution across
the arid woodland regions of Africa31. Pigott et al developed zoonotic niche maps for Marburg and Ebola viruses in Africa using
species distribution models32,33. In these maps, they have predicted EVD at risk areas occupied by 22 million people while
MVD is predicted to occur in 27 countries across Sub-Sahara Africa. Enhanced vegetation index which corresponds to high
levels of rainfall was identified as the most important variable limiting the distribution of the Ebola virus in Africa32,33.
These predictions are not country specific, and they lack details of individual countries regarding vector and raster data. For
example, they used online databases that are not accurate especially in estimating environmental covariates and getting
coordinates of index cases, hence, affected countries find these maps limited for focused and targeted surveillance
A Maximum Entropy species distribution modeling environment(MaxEnt) has been used to predict the ecological niche for
various species. The MaxEnt algorithm uses presence-only occurrence records to estimate the actual or potential geographic
distribution of a species34 and has been known to outperform other species’ distribution modeling approaches such as Domain,
Generalized Additive Models(GAM), Generalized Linear Models, Genetic Algorithm (GARP) and Bioclim35.
Maximum entropy (MaxEnt) models have been used widely to predict ecological niches of different vectors and disease-causing
organisms36,37,38,39,40,41,42,43, but it has not been used for prediction of filovirus outbreaks in Uganda. Briefly, MaxEnt is a
multipurpose machine-learning technique and aims at estimating the probability of distribution of a species occurrence using the
environmental features. Our major aim was to develop a country-specific risk map for Uganda using updated data on EVD/ MVD
outbreaks and bat occurrence and environmental variables specific for Uganda using the MaxEnt modeling approach. The
model outputs will improve filovirus epidemic preparedness, surveillance and response, and in the search for a reservoir
especially in a disease prone country like Uganda
Materials and methods
DataEVD/ MVD/ Bat occurrence data
A total of 16 locations of the Ebolavirus outbreaks in Uganda since 2000 was obtained from published databases44. An
additional 27 occurrence points for Ebola and Marburg virus diseases outbreaks were collected from the field where these
outbreaks occurred especially for new outbreaks whose geographical points were not collected before. All locations where
confirmed cases of Marburg or Ebola viruses were reported were collected with Global Positioning System(GPS) receiver and
points, were entered into an Excel spreadsheet. The total of 43 filovirus outbreak occurrence points (30 for EVD outbreak and
13 for MVD outbreak) were used for this prediction model (Supplementary File 1). These filovirus occurrence points represent
households in villages where confirmed cases were residing. Due to the contagious nature of filoviruses, one household had
more than one cases hence the reason for not using all the 562 EVD cases and 20 MVD cases. A fruit bat location survey was
also done to determine the location of fruit bats in a cross-section of Uganda. We purposively selected districts to scout for bats
based on previous filovirus outbreaks and anecdotal reports of bats in trees. Using a snowballing approach, we collected 84 fruit
bat locations using a GPS receiver from different districts of the country. Here community members acted as informers of the
roosting locations of fruit bats and caves that contain bats.
An additional, 517 bat locations from all over Uganda were generously provided by Kityo Robert (Department of Zoology,
Makerere University Kampala Uganda) also published in Uganda Bat Atlas45, resulting in a total of 601 bat
coordinates(Supplementary File 1).
Ecologically suitable environmental covariates for filovirus outbreaks for Uganda were compiled from Africlim46, with a spatial
resolution of 1 km. The environmental covariates considered were moisture (mean annual rainfall, rainfall wettest month, rainfall
driest month, rainfall seasonality, rainfall wettest quarter, rainfall driest quarter, annual moisture index, moisture index arid
quarter, number of dry months, length of longest dry season) and temperature variables (mean annual temperature, mean
diurnal range in temperature, isothermality, temperature seasonality, maximum temperature warmest month, minimum
temperature coolest month, annual temperature range, mean temperature warmest quarter, mean temperature coolest quarter,
potential evapotranspiration). We used ENMTOOLs; a toolbox that facilitates quantitative comparisons of environmental niche
models47 to test for multicollinearity between the predictor variables and we ran a pairwise Pearson correlation, and only
variables with less than (+/-0.75) correlation were retained in the final prediction model (Supplementary Table 2). After this test,
only seven environmental variables were retained (Table 1); three moisture variables (Rainfall seasonality, Rainfall driest
quarter, and mean annual rainfall) and four temperature variables(Temperature seasonality, Mean diurnal range in temperature,
mean annual temperature and Isothermality )
Ecological Niche Model
We used Maximum Entropy Species Distribution Modeling (MaxEnt), Version 3.3k for modeling. Filovirus occurrence points and
Environmental covariates were imported into MaxEnt, and the default MaxEnt model parameters were used (Auto features,
convergence threshold =0.00001, the maximum number of background points=10,000, regularization multiplier=1). The output
of the model was a logistic format prediction map showing the relative probability of the presence of filoviruses survival on a
scale ranging between 0 and 148. The occurrence data was subdivided into k-folds where 25% was set aside for testing the
accuracy of the model, whereas 75% was used for training the model. However, there were few presence records (
) for the
Marburg cases and all the records were used in training model. The Receiver Operating Curve (ROC) was used to assess the
overall model predictive performance, a measure of the ability of the model to distinguish presence from the absence of a
species with a value of 1 indicating a perfect prediction while 0.5 is as good as a random prediction49,50. A jackknife test was
used to evaluate individual covariate importance in the model developments (Supplementary file 3). To improve model
robustness, 100 replicates were averaged for the final model outputs. MaxEnt outputs were imported into ArcGIS 10.3 mapping
software to develop final maps
The bat occurrence and filovirus outbreak locations.
As shown in Figure 1, bats are distributed all over Uganda, with a high distribution around water bodies which is a core need for
survival. Areas around Lake Victoria, River Nile, and Western Rift Valley have high numbers of bats. Their locality is in line with
regions that have reported filovirus outbreaks in Uganda.
From 100 bootstrap replicates, a bat distribution map was generated (mean AUC=0.80; SD=0.012). Compared to a random
prediction of AUC 0.5, our model was able to distinguish presence from the absence of bats within the geographic space with a
high accuracy51. The relative probability of presence (RPP) ranged from highly suitable areas represented by red to orange
colors to unsuitable areas represented by the green color in Figure 2A. The map shows that most areas in Uganda are suitable
habitats for bats (both insect and fruit bats) with high RPP occurring in the following districts; Mbarara, Bushenyi, Bundibugyo
and Kabale located in the western part of Uganda, around Lake Victoria (Kampala and Luweero districts) and in eastern region
of Mbale and Soroti districts. Moderately suitable regions largely cover most parts of Uganda. The RPP of bats were mainly
influenced by rainfall driest quarter with 24.7%, mean annual rainfall with 17.2%, mean diurnal range in temperature with 14.5%,
and isothermality with 11.5% (Table 2).
Ebola virus disease outbreak risk map
High RPP for EVD outbreak was predicted in more than half of the country with hotspots in Western Rift valley districts of
Bundibugyo, Masindi, Kibale and Hoima, Kasese, Kabarole, Kamwenge, Bushenyi and Ibanda as shown in Figure 2B (mean
AUC=0.90; SD=0.024). In Central Uganda, Luweero, Kayunga, Mpigi, Kampala, Mityana and Nakasongola districts are
predicted as potential areas for EVD outbreaks. In the eastern part of the country, it is mainly the Busoga region along River
Nile and Mbale district around Mt. Elgon that are potential EVD hot spots. Other places that have not recorded outbreaks before
but are predicted as potential probable areas for the spread of EVD include areas surrounding Lake Victoria and around Mount
Elgon. A low RPP for EVD outbreak was predicted in North Eastern Uganda (Karamoja region) and Northern Uganda in the
districts of Kitgum and Pader. Rainfall seasonality (33.2%), Mean annual rainfall (22.7%), rainfall of the driest quarter(20.8%)
and mean diurnal range in Temperature(9.9%) had the highest relative contribution to the MaxEnt model for Ebola virus
ecological suitability(Table 2).
Marburg virus disease outbreak risk map
The map in Figure 2C shows that Western, Southwestern and Central Uganda are potential areas for outbreaks of Marburg
cases(AUC=0.92). Unlike predicted potential areas for EVD, predicted areas for MVD are mainly in the western sub-regions of
Ankole, Tooro, Bunyoro, and Rwenzori region extending into DRC. Areas in the North and Eastern part of Uganda have a low or
no relative probability of presence for MVD outbreaks as shown by the green color in Figure 2C. Temperature seasonality
(68.2%) and rainfall seasonality (25.3%) contributed heavily to the model prediction (Table 2). Notably, temperature seasonality
had the highest influence in MVD model compared to other variable contributions in all the models. However, the occurrence
points were few in number to give us an accurate prediction for MVD occurrence points.
Combining Marburg and Ebola virus occurrence points (Figure 3), we see the range of the possible distribution of filovirus,
mainly in western, southwestern Uganda, Victoria basin districts and eastern Uganda (mean AUC=0.90; SD =0.023). Predictor
variables that contributed more than 75% in the model include; rainfall seasonality (29.6%), rainfall of the driest quarter (26.3%),
Temperature seasonality and mean annual rainfall (14.9%) (Table 2).
A: Rainfall driest quarter(BIO17) vs Relative probability of bat presence. B: Rainfall seasonality(BIO15) vs. Relative
probability of presence of Ebola virus outbreak; C: Temperature seasonality(BIO4) vs. Relative probability of presence of
Marburg virus outbreak; D: Rainfall seasonality(BIO15) vs Relative probability of presence of Ebola or Marburg virus
We used seven environmental variables in this model prediction. This was after assessing for collinearity in the model and
removing all the collinear variables. Variable contribution assessment as shown in Table 2 showed that rainfall variables were
the most important predictors. The importance of rainfall or precipitation and moderate to high temperature was highlighted by
Peterson et al(2004) when they modeled filovirus distribution in Africa using GARP model29,31. Rainfall is important for the
obvious reason that provides water which is very important for bats survival52,53. Rainfall also provides for the development of
fruiting trees that provide roosting areas for bats as well as food for fruit bats. Uganda is endowed by many water bodies and
several rainforests, and hence bat distribution tends to be all over the country as seen in Figure 2A. Bats are hypothesized to be
reservoirs for filoviruses; their distribution tends to correlate with that of filovirus predicted niches (Figure 3). Although we have
some progress with Marburg virus in trying to describe bats as a source of infection for humans7,8,10,54, more research needs
to be done especially on the reservoir for Ebola virus as these models can only give a clue as to the possible surveillance sites
and possible areas to focus the research and to identify other potential reservoirs for filovirus. Temperature and rainfall
seasonality were the most important environmental variables contributing to spatial prediction model for the Ebola and Marburg
viruses. Seasonality has been found to be key in outbreaks of filoviruses, especially MVD as was reported in an ecological study
by Amman et al. 20128. In this study, outbreaks of MVD are associated with the birthing seasons of adult juvenile bats when the
virus circulation was high. This is further validated by a high percentage contribution (68.2%) of temperature seasonality into the
MVD outbreak prediction model (Table 2). The relative probability of the presence of a Marburg outbreak is higher (80%) and at
very low-temperature seasonality, which is a standard deviation (SD) over monthly values (Figure 4C). Therefore, areas with
fewer variations in monthly temperature and rainfall are more like to experience MVD and EVD outbreaks and this has been
predicted by the models in Figures 2 & 3. The areas shown on the risk maps with a high relative probability of the presence of
an outbreak are mainly in the South, the West and Central Uganda that have minimal temperature and rainfall variations
compared to North Eastern Uganda that is not predicted for filovirus outbreaks except for bat presence. Bat presence model is
mainly influenced by the variable rainfall driest quarter (24.7%) and mean annual rainfall (17.2%) (Table 2). As these variables
increases, the relative probability of the presence of bats tends to increase. Areas of high rainfall are more likely to be forested
or with many fruiting trees that provide a suitable habitat for bats, and this is true for three-quarters (75%) of Uganda.
Whereas Pigott et al (2015) used environmental covariates with a spatial resolution of 5km in their models55,56 , we used
Africlim data with 1km spatial resolution. High-resolution data increases the accuracy of the models, and this was observed in
our study by a high AUC greater than 0.8 recorded in all models.
The predictions show that a big part of Uganda, a country of 34 million people is at risk of a filovirus outbreak. This is more so in
the Lake Victoria basin districts and in the Albertine Rift region districts and the areas that occur in between(Figure 2 & 3). The
Albertine Rift region provides a variety of habitats characteristic of the East African savannahs and the West African rain forests
that are suitable for reservoirs of filoviruses. According to Uganda National Meteorological department, these are the areas that
receive near or above normal seasonal rainfall, and seasonal temperature variations are minimal57. Moreover, we see from
variable contribution(Table 2), response curves(Figure 4) and Jackknife test(Supplementary material S3 ) that rainfall and
temperature seasonality were the most important variables in predicting outbreaks. The lower the variability in rainfall and
temperature, the high the relative probability of presence and vice versa and an increase in mean rainfall variables increases
relative probability of having a filovirus outbreak (Figure 4). Indeed, six filovirus outbreaks have happened in this region, one
caused by Bundibugyo ebolavirus in Bundibugyo district in the plains of Rwenzori mountains2, Sudan Ebolavirus in Kibale
district5 and four outbreaks of Marburg virus all linked to Python cave and Kitaka gold mines in Kamwenge, Ibanda, and Rubirizi
districts6,9,58,59. This remains a high-risk area with cross-border movement between Uganda and DRC where another EVD
outbreak happened in 2012 in the neighboring Isiro region60 The Albertine Rift of East Africa needs to remain under heightened
surveillance especially that oil exploration will be taking place bringing an invasion of virgin lands by humans and interaction of
wildlife and humans. Important to note also in this region has six national parks of Uganda (Queen Elizabeth National Park,
Murchison Falls National Park, Kibale Forest National Park, Semiliki National Park, Bwindi Impenetrable National Park and
Mgahinga National Park) on Uganda side and several other national parks on the DRC and Rwanda side as well as several
forest reserves all of which harbor various species of bats and other possible reservoirs of filoviruses. All outbreaks of Marburg
virus disease in Uganda have been investigated, and all originate from the old gold mines found in Ibanda and Kamwenge
district6,9 in the Western Rift Valley which validates MVD distribution model in Figure 2C as it shows these as high-risk areas for
filovirus outbreaks. A similar finding was obtained by Peterson and Samy 2016 in a recent model using MaxEnt as they
predicted Sudan Ebola virus to occur in North Western Uganda between Lake Albert and Lake Vitoria61. We also see areas that
have not had EVD outbreaks before such as West Nile region being predicted potential areas for EVD outbreak. These include
areas along River Nile and areas boarding South Sudan and DRC (Figure 2B). From Table 2, we see that rainfall variable
contribute a higher percentage of the relative probability of presence for filovirus habitants. These areas receive average annual
rainfall between 100-120mm and are endowed with high vegetation cover and water bodies all of which make the region
conducive for reservoirs of filoviruses
Another area of high concern predicted by this model is Lake Victoria basin and districts in Nile River basin in Central districts of
Uganda. Uganda has reported three outbreaks of filoviruses previously detected in these regions in the districts of Luweero4,5
and Mpigi11,62. This also can be attributed to the variety of habitats provided by water bodies, forests, swamps and high
presence of fruit bats and other wildlife in this region. For example, the Kasokero cave that is the habitat of many Egyptian fruit
bats that known to harbor Marburg virus is found just on the banks of Lake Victoria in Masaka district, and several pathogens
have been isolated from this cave 63. This is at the same time a highly-populated region with Uganda’s capital in the middle and
needs to be heightened surveillance. We also predicted other regions that have not heard outbreaks of filoviruses in the past
such as the Eastern region of Mbale, Busia and Tororo districts near the Mt. Elgon regions bordering with Kenya. This also still
attributed to by the presence of suitable conditions for survival of putative reservoirs of Ebola and Marburg viruses. An outbreak
happened in neighboring Kenya in Kitum cave64,65. These newly detected hotspots need to be kept under surveillance for early
outbreak detection and response.
We build on filovirus risk mapping efforts by Pigott et al32,33,56 and Peterson et al29,31,61 all of which have been done at the
continental level of Africa. Their work was more of ecologically oriented and more focused on identifying the ecological niche of
species, they lacked country specific details that we bring in this publication with a bias in public health surveillance and
outbreak detection rather that ecological niche identification. For public health surveillance of a country like Uganda, all filovirus
species (Marburg virus, 5 Ebola virus species) are of public health importance. This makes our models more sensitive as
opposed to specific risk map and hence more useful tools to the surveillance activities. There is already enough evidence of
filovirus outbreaks in Uganda, especially areas predicted by our models. Focused surveillance needs to be done in these areas
and bring additional surveillance in other new predicted areas where we have not heard outbreaks before. So we think modeling
the map at a genus level (filovirus) level as opposed to species level is more informative for surveillance but not be the best for
ecological studies for which is not the purpose of this study. We know that disease outbreak is a combination of very many
factors, not only suitable environmental covariates. However, we were not able to include as many factors as possible in this
model because of lack of or poor quality data for Uganda specifically. We did not use bats as a predictor in our model because
of their widespread distribution all over Uganda, otherwise doing this would lead to misleading interpretation and bias of
potential outbreak hotspots as being the whole country. Another point would have been good to include in the prediction model
are socio-economic factors since they play a big role in the outbreak of filoviruses.
Ecological niche modeling techniques have been widely used in predicting where disease outbreaks are likely to occur, more
specifically where species have suitable living conditions depending on their environmental factors. The MaxEnt modeling
algorithm uses presence only occurrence data and has been useful to estimate species’ niche in environmental space where
absence records for a species are not available as it is the case with filoviruses. Given the public and global importance of
filoviruses, developing models that predict where they are likely to occur is very important, and efforts in this direction have been
done focusing on the African continent. In this paper, however, we focus on Uganda as one of the affected countries; and
develop a country-specific prediction map. We show which places in Uganda that are hot spots for filovirus disease outbreaks
and hence a focus on surveillance for early detection. Until now, no verified true reservoir for Ebola virus has been identified,
and studies in this direction are still ongoing. In the absence of a known reservoir, these risk maps will help in early focused
surveillance and early detection to avoid a global catastrophe like it happened in West Africa in 2014. Minimal seasonal
variations in temperature and rainfall were important predictors of a filovirus outbreak. We believe these risk maps will important
in targeted surveillance, research and epidemic preparedness for Uganda. The results from this study also confirm previous
findings that suggest that Filoviruses are mainly limited by the amount of rainfall received in an area.
The authors have declared that no competing interests exist.
All data is available in the paper and supporting files which can be found on Figshare as follows: S1: Occurrence dataset used
(Filovirus and Bats Occurrence coordinates)(10.6084/m9.figshare.5306875 <https://doi.org/10.6084/m9.figshare.5306875>); S2:
Results of the quantitative comparisons of environmental variables to test for multicollinearity(10.6084/m9.figshare.5306908
<https://doi.org/10.6084/m9.figshare.5306908>); S3: A jackknife test result to evaluate individual covariate importance in the
model developments(10.6084/m9.figshare.5306914 <https://doi.org/10.6084/m9.figshare.5306914>); S4: The response curves
of all the predictor variables in all the four models (10.6084/m9.figshare.5306932
We thank Dr. Robert Kityo of Makerere University for bat occurrence data he provided us. We thank the Viral hemorrhagic
fevers program and staff of Uganda virus Research Institute in collaboration with United States Centers for Disease Control and
Prevention(CDC) for the surveillance activities and outbreak Investigations which generate some of the data we were using
such as filovirus outbreak occurrence points. The funders had no role in study design, data collection and analysis, decision to
publish, or preparation of the manuscript.
Virus from Monkeys Imported to USA. Lancet. 1990;335(8688):502-5. doi: Doi 10.1016/0140-6736(90)90737-P. PubMed PMID:
Trop Med Hyg. 2006 Jul;75(
):9-15. PubMed PMID:16837700.
1. Okware SI , Omaswa FG , Zaramba S , Opio A , Lutwama JJ , Kamugisha J , et al. An outbreak of Ebola in Uganda . Trop Med Int Health . 2002 ; 7 ( 12 ): 1068 - 75 . PubMed PMID: 12460399 .
2. Wamala JF , Lukwago L , Malimbo M , Nguku P , Yoti Z , Musenero M , et al. Ebola hemorrhagic fever associated with novel virus strain , Uganda , 2007 - 2008 . Emerg Infect Dis. 2010 ; 16 ( 7 ): 1087 - 92 . doi: 10 .3201/eid1607.091525. PubMed PMID : 20587179; PubMed Central PMCID : PMCPMC3321896
3. Nyakarahuka L , Kankya C , Krontveit R , Mayer B , Mwiine FN , Lutwama J , et al. How severe and prevalent are Ebola and Marburg viruses? A systematic review and meta-analysis of the case fatality rates and seroprevalence . BMC Infect Dis . 2016 ; 16 ( 1 ): 708 . doi: 10 .1186/s12879-016 -2045-6 . PubMed PMID: 27887599
4. Shoemaker T , MacNeil A , Balinandi S , Campbell S , Wamala JF , McMullan LK , et al. Reemerging Sudan Ebola virus disease in Uganda , 2011 . Emerg Infect Dis . 2012 ; 18 ( 9 ): 1480 - 3 . doi: 10 .3201/eid1809.111536. PubMed PMID : 22931687; PubMed Central PMCID : PMCPMC3437705 .
5. Albarino CG , Shoemaker T , Khristova ML , Wamala JF , Muyembe JJ , Balinandi S , et al. Genomic analysis of filoviruses associated with four viral hemorrhagic fever outbreaks in Uganda and the Democratic Republic of the Congo in 2012 . Virology. 2013 ; 442 ( 2 ): 97 - 100 . doi: 10 .1016/j.virol. 2013 . 04 .014. PubMed PMID : 23711383 .
6. Adjemian J , Farnon EC , Tschioko F , Wamala JF , Byaruhanga E , Bwire GS , et al. Outbreak of Marburg hemorrhagic fever among miners in Kamwenge and Ibanda Districts , Uganda, 2007 . J Infect Dis . 2011 ; 204 Suppl 3 : S796 - 9 . doi: 10 .1093/infdis/jir312. PubMed PMID: 21987753; PubMed Central PMCID : PMCPMC3203392 .
7. Towner JS , Amman BR , Sealy TK , Carroll SA , Comer JA , Kemp A , et al. Isolation of genetically diverse Marburg viruses from Egyptian fruit bats . PLoS Pathog . 2009 ; 5 ( 7 ):e1000536. doi: 10 .1371/journal.ppat.1000536. PubMed PMID : 19649327; PubMed Central PMCID : PMCPMC2713404 .
8. Amman BR , Carroll SA , Reed ZD , Sealy TK , Balinandi S , Swanepoel R , et al. Seasonal pulses of Marburg virus circulation in juvenile Rousettus aegyptiacus bats coincide with periods of increased risk of human infection . PLoS Pathog . 2012 ; 8 ( 10 ):e1002877. doi: 10 .1371/journal.ppat.1002877. PubMed PMID : 23055920; PubMed Central PMCID : PMCPMC3464226 .
9. Knust B , Schafer IJ , Wamala J , Nyakarahuka L , Okot C , Shoemaker T , et al. Multidistrict Outbreak of Marburg Virus DiseaseUganda , 2012 . J Infect Dis . 2015 ; 212 Suppl 2 : S119 - 28 . doi: 10 .1093/infdis/jiv351. PubMed PMID: 26209681 .
10. Amman BR , Nyakarahuka L , McElroy AK , Dodd KA , Sealy TK , Schuh AJ , et al. Marburgvirus resurgence in Kitaka Mine bat population after extermination attempts , Uganda. Emerg Infect Dis . 2014 ; 20 ( 10 ): 1761 - 4 . doi: 10 .3201/eid2010.140696. PubMed PMID : 25272104; PubMed Central PMCID : PMCPMC4193183 .
11. Nyakarahuka L , Ojwang J , Tumusiime A , Balinandi S , Whitmer S , Kyazze S , et al. Isolated Case of Marburg Virus Disease, Kampala, Uganda , 2014 . Emerg Infect Dis . 2017 ; 23 ( 6 ): 1001 - 4 . doi: 10 .3201/eid2306.170047. PubMed PMID : 28518032 .
12. Jahrling PB , Geisbert TW , Dalgard DW , Johnson ED , Ksiazek TG , Hall WC , et al. Preliminary-Report - Isolation of Ebola 13 . Pooley S , Fa JE , Nasi R . Ebola and bushmeat . New Sci . 2014 ; 224 ( 2989 ): 31 -. PubMed PMID: WOS:000342656200017
14. Leroy EM , Rouquet P , Formenty P , Souquiere S , Kilbourne A , Froment JM , et al. Multiple Ebola virus transmission events and rapid decline of central African wildlife . Science . 2004 ; 303 ( 5656 ): 387 - 90 . doi: 10 .1126/science.1092528. PubMed PMID : 14726594 .
15. Leroy EM , Kumulungui B , Pourrut X , Rouquet P , Hassanin A , Yaba P , et al. Fruit bats as reservoirs of Ebola virus . Nature . 2005 ; 438 ( 7068 ): 575 - 6 . doi: 10 .1038/438575a. PubMed PMID: 16319873 .
16. Jaax NK , Davis KJ , Geisbert TJ , Vogel P , Jaax GP , Topper M , et al. Lethal experimental infection of rhesus monkeys with Ebola-Zaire (Mayinga) virus by the oral and conjunctival route of exposure . Archives of Pathology & Laboratory Medicine . 1996 ; 120 ( 2 ): 140 - 55 . PubMed PMID: WOS: A1996TV90500005 .
17. Geisbert TW , Pushko P , Anderson K , Smith J , Davis KJ , Jahrling PB . Evaluation in nonhuman primates of vaccines against Ebola virus . Emerging Infectious Diseases . 2002 ; 8 ( 5 ): 503 - 7 . PubMed PMID: WOS: 000175415300010 . 18. Formenty P , Boesch C , Wyers M , Steiner C , Donati F , Dind F , et al. Ebola
18. Formenty P , Boesch C , Wyers M , Steiner C , Donati F , Dind F , et al. Ebola virus outbreak among wild chimpanzees living in a rain forest of Cote d'Ivoire . Journal of Infectious Diseases . 1999 ; 179 : S120 - S6 . doi: Doi 10.1086/514296. PubMed PMID : WOS: 000078400300022 .
19. Rouquet P , Froment JM , Bermejo M , Kilbourn A , Karesh W , Reed P , et al. Wild animal mortality monitoring and human Ebola outbreaks , Gabon and Republic of Congo , 2001 - 2003 . Emerg Infect Dis. 2005 ; 11 ( 2 ): 283 - 90 . doi: 10 .3201/eid1102.040533. PubMed PMID : 15752448; PubMed Central PMCID : PMCPMC3320460 .
20. Allela L , Boury O , Pouillot R , Delicat A , Yaba P , Kumulungui B , et al. Ebola virus antibody prevalence in dogs and human risk . Emerg Infect Dis . 2005 ; 11 ( 3 ): 385 - 90 . doi: 10 .3201/eid1103.040981. PubMed PMID : 15757552; PubMed Central PMCID : PMCPMC3298261 .
21. Weingartl HM , Embury-Hyatt C , Nfon C , Leung A , Smith G , Kobinger G . Transmission of Ebola virus from pigs to nonhuman primates . Sci Rep . 2012 ; 2 : 811 . doi: 10 .1038/srep00811. PubMed PMID: 23155478; PubMed Central PMCID : PMCPMC3498927 .
22. Team E. Ebola Reston Virus Detected Pigs in the Philippines . Eurosurveillance . 2009 ; 14 ( 4 ): 11 -. PubMed PMID: WOS:000267921400004.
23. Marsh GA , Haining J , Robinson R , Foord A , Yamada M , Barr JA , et al. Ebola Reston Virus Infection of Pigs: Clinical Significance and Transmission Potential . Journal of Infectious Diseases . 2011 ; 204 : S804 - S9 . doi: 10 .1093/infdis/jir300. PubMed PMID : WOS: 000295991400009 .
24. Towner JS , Amman BR , Sealy TK , Carroll SAR , Comer JA , Kemp A , et al. Isolation of Genetically Diverse Marburg Viruses from Egyptian Fruit Bats . Plos Pathogens . 2009 ; 5 ( 7 ). doi: ARTN e1000536 10.1371/journal.ppat.1000536 . PubMed PMID : WOS: 000269224500012 .
25. Peterson AT . Ecological niche modelling and understanding the geography of disease transmission . Vet Ital . 2007 ; 43 ( 3 ): 393 - 400 . PubMed PMID: 20422515 .
26. Batista TA , Gurgel-Goncalves R . Ecological niche modelling and differentiation between Rhodnius neglectus Lent, 1954 and Rhodnius nasutus Stal, 1859 (Hemiptera: Reduviidae: Triatominae) in Brazil . Mem Inst Oswaldo Cruz . 2009 ; 104 ( 8 ): 1165 - 70 . PubMed PMID: 20140378 .
27. Hawlitschek O , Porch N , Hendrich L , Balke M. Ecological niche modelling and nDNA sequencing support a new, morphologically cryptic beetle species unveiled by DNA barcoding . PLoS One . 2011 ; 6 ( 2 ):e16662. doi: 10 .1371/journal.pone.0016662. PubMed PMID : 21347370; PubMed Central PMCID : PMCPMC3036709 .
28. Mudenda NB , Malone JB , Kearney MT , Mischler PD , Nieto Pdel M , McCarroll JC , et al. Modelling the ecological niche of hookworm in Brazil based on climate . Geospat Health . 2012 ; 6 ( 3 ): S111 - 23 . doi: 10 .4081/gh. 2012 .129. PubMed PMID : 23032277 .
29. Peterson AT , Bauer JT , Mills JN . Ecologic and geographic distribution of filovirus disease . Emerg Infect Dis . 2004 ; 10 ( 1 ): 40 - 7 . doi: 10 .3201/eid1001.030125. PubMed PMID : 15078595; PubMed Central PMCID : PMCPMC3322747 .
30. WorldClim. WorldClim - Global Climate Data , Free climate data for ecological modeling and GIS 2016 [cited 2017 16.January 2017]. Available from: http://www.worldclim.org/bioclim. REFERENCE LINK
31. Peterson AT , Lash RR , Carroll DS , Johnson KM . Geographic potential for outbreaks of Marburg hemorrhagic fever . Am J 32 . Pigott DM , Golding N , Mylne A , Huang Z , Henry AJ , Weiss DJ , et al. Mapping the zoonotic niche of Ebola virus disease in Africa . Elife. 2014 ; 3:e04395 . doi: 10 .7554/eLife.04395. PubMed PMID : 25201877; PubMed Central PMCID : PMCPMC4166725 .
33. IPigott DM , Golding N , Mylne A , Huang Z , Weiss DJ , Brady OJ , et al. Mapping the zoonotic niche of Marburg virus disease in Africa . Trans R Soc Trop Med Hyg . 2015 ; 109 ( 6 ): 366 - 78 . doi: 10 .1093/trstmh/trv024. PubMed PMID: 25820266; PubMed Central PMCID : PMCPMC4447827 .
34. Phillips SJ , Dudík M . Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation . Ecography . 2008 ; 31 ( 2 ): 161 - 75 .
35. P Anderson R , Dudík M , Ferrier S , Guisan A , J Hijmans R , Huettmann F , et al. Novel methods improve prediction of species' distributions from occurrence data . Ecography . 2006 ; 29 ( 2 ): 129 - 51 .
36. Chamaille L , Tran A , Meunier A , Bourdoiseau G , Ready P , Dedet JP . Environmental risk mapping of canine leishmaniasis in France . Parasit Vectors . 2010 ; 3 : 31 . doi: 10 .1186/ 1756 -3305-3-31. PubMed PMID: 20377867; PubMed Central PMCID : PMCPMC2857865 .
37. Kantzoura V , Kouam MK , Feidas H , Teofanova D , Theodoropoulos G . Geographic distribution modelling for ruminant liver flukes (Fasciola hepatica) in south-eastern Europe . Int J Parasitol . 2011 ; 41 ( 7 ): 747 - 53 . doi: 10 .1016/j.ijpara. 2011 . 01 .006. PubMed PMID : 21329694 .
38. Mischler P , Kearney M , McCarroll JC , Scholte RG , Vounatsou P , Malone JB . Environmental and socio-economic risk modelling for Chagas disease in Bolivia . Geospat Health . 2012 ; 6 ( 3 ): S59 - 66 . doi: 10 .4081/gh. 2012 .123. PubMed PMID : 23032284 .
39. Zeimes CB , Olsson GE , Ahlm C , Vanwambeke SO . Modelling zoonotic diseases in humans: comparison of methods for hantavirus in Sweden . Int J Health Geogr . 2012 ; 11 : 39 . doi: 10 .1186/ 1476 -072X- 11 -39. PubMed PMID: 22984887; PubMed Central PMCID : PMCPMC3517350 .
40. Chikerema SM , Murwira A , Matope G , Pfukenyi DM . Spatial modelling of Bacillus anthracis ecological niche in Zimbabwe . Prev Vet Med . 2013 ; 111 ( 1-2 ): 25 - 30 . doi: 10 .1016/j.prevetmed. 2013 . 04 .006. PubMed PMID : 23726015 .
41. Mweya CN , Kimera SI , Kija JB , Mboera LE . Predicting distribution of Aedes aegypti and Culex pipiens complex, potential vectors of Rift Valley fever virus in relation to disease epidemics in East Africa . Infect Ecol Epidemiol . 2013 ; 3 . doi: 10 .3402/iee.v3i0.21748. PubMed PMID : 24137533; PubMed Central PMCID : PMCPMC3797365 .
42. Nogareda C , Jubert A , Kantzoura V , Kouam MK , Feidas H , Theodoropoulos G . Geographical distribution modelling for Neospora caninum and Coxiella burnetii infections in dairy cattle farms in northeastern Spain . Epidemiol Infect . 2013 ; 141 ( 1 ): 81 - 90 . doi: 10 .1017/S0950268812000271. PubMed PMID: 22370223 .
43. Mughini-Gras L , Mulatti P , Severini F , Boccolini D , Romi R , Bongiorno G , et al. Ecological niche modelling of potential West Nile virus vector mosquito species and their geographical association with equine epizootics in Italy . Ecohealth. 2014 ; 11 ( 1 ): 120 - 32 . doi: 10 .1007/s10393-013-0878-7. PubMed PMID: 24121802 .
44. Mylne A , Brady OJ , Huang Z , Pigott DM , Golding N , Kraemer MU , et al. A comprehensive database of the geographic spread of past human Ebola outbreaks . Sci Data . 2014 ; 1 : 140042 . doi: 10 .1038/sdata. 2014 .42. PubMed PMID : 25984346; PubMed Central PMCID : PMCPMC4432636 .
45. Kityo R , Howell K , Nakibuka M , Ngalason W , Tushabe H , Webala P. East African Bat Atlas . Makerere University and University of Dar es Salaam, Kampala, Uganda. 2009 .
46. Platts PJ , Omeny PA , Marchant R. AFRICLIM : high?resolution climate projections for ecological applications in Africa . African Journal of Ecology . 2015 ; 53 ( 1 ): 103 - 8 .
47. Warren DL , Glor RE , Turelli M. ENMTools: a toolbox for comparative studies of environmental niche models . Ecography . 2010 ; 33 ( 3 ): 607 - 11 .
48. Young N , Carter L , Evangelista P. A MaxEnt model v3. 3.3 e tutorial (ArcGIS v10) . Fort Collins, Colorado. 2011 .
49. Townsend Peterson A , Pape? M , Eaton M. Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent . Ecography. 2007 ; 30 ( 4 ): 550 - 60 .
50. Anderson RP , Lew D , Peterson AT . Evaluating predictive models of species' distributions: criteria for selecting optimal models . Ecological modelling . 2003 ; 162 ( 3 ): 211 - 32 .
51. Swets JA . Measuring the accuracy of diagnostic systems . Science . 1988 ; 240 ( 4857 ): 1285 - 93 .
52. Russo D , Cistrone L , Jones G . Sensory ecology of water detection by bats: a field experiment . PLoS One . 2012 ; 7 ( 10 ):e48144. doi: 10 .1371/journal.pone.0048144. PubMed PMID : 23133558; PubMed Central PMCID : PMCPMC3483877 .
53. Adams RA , Hayes MA . Water availability and successful lactation by bats as related to climate change in arid regions of western North America . J Anim Ecol . 2008 ; 77 ( 6 ): 1115 - 21 . doi: 10 .1111/j.1365- 2656 . 2008 . 01447 .x. PubMed PMID: 18684132.
54. Amman BR , Jones ME , Sealy TK , Uebelhoer LS , Schuh AJ , Bird BH , et al. Oral shedding of Marburg virus in experimentally infected Egyptian fruit bats (Rousettus aegyptiacus) . J Wildl Dis . 2015 ; 51 ( 1 ): 113 - 24 . doi: 10 .7589/2014-08-198. PubMed PMID: 25375951; PubMed Central PMCID : PMCPMC5022530 .
55. Pigott DM , Golding N , Mylne A , Huang Z , Weiss DJ , Brady OJ , et al. Mapping the zoonotic niche of Marburg virus disease in Africa . Trans R Soc Trop Med Hyg . 2015 ; 109 ( 6 ): 366 - 78 . doi: 10 .1093/trstmh/trv024. PubMed PMID: 25820266; PubMed Central PMCID : PMCPMC4447827 .
56. Pigott DM , Millear AI , Earl L , Morozoff C , Han BA , Shearer FM , et al. Updates to the zoonotic niche map of Ebola virus disease in Africa . Elife. 2016 ; 5 . doi: 10 .7554/eLife.16412. PubMed PMID : 27414263; PubMed Central PMCID : PMCPMC4945152 .
57. UNMA. seasonal forecasts 2016 [cited 2017 18th May 2017 ]. Available from: https://www.unma.go.ug/index.php/climate/seasonal-performance.
58. Centers for Disease C , Prevention . Imported case of Marburg hemorrhagic fever - Colorado , 2008 . MMWR Morb Mortal Wkly Rep . 2009 ; 58 ( 49 ): 1377 - 81 . PubMed PMID: 20019654 .
59. Timen A , Koopmans MP , Vossen AC , van Doornum GJ , Gunther S , van den Berkmortel F , et al. Response to imported case of Marburg hemorrhagic fever, the Netherland . Emerg Infect Dis . 2009 ; 15 ( 8 ): 1171 - 5 . doi: 10 .3201/eid1508.090015. PubMed PMID : 19751577; PubMed Central PMCID : PMCPMC2815969 .
60. Kratz T , Roddy P , Tshomba Oloma A , Jeffs B , Pou Ciruelo D , de la Rosa O , et al. Ebola Virus Disease Outbreak in Isiro, Democratic Republic of the Congo , 2012 : Signs and Symptoms, Management and Outcomes . PLoS One . 2015 ; 10 ( 6 ):e0129333. doi: 10 .1371/journal.pone.0129333. PubMed PMID : 26107529; PubMed Central PMCID : PMCPMC4479598 .
61. Peterson AT , Samy AM . Geographic potential of disease caused by Ebola and Marburg viruses in Africa . Acta Trop . 2016 ; 162 : 114 - 24 . doi: 10 .1016/j.actatropica. 2016 . 06 .012. PubMed PMID : 27311387 .
62. WHO. Marburg virus disease - Uganda: WHO; 2014 [cited 2016 5- 09 -2016]. Available from: http://www.who.int/csr/don/10- october-2014-marburg/en/.
63. Kalunda M , Mukwaya LG , Mukuye A , Lule M , Sekyalo E , Wright J , et al. Kasokero virus: a new human pathogen from bats (Rousettus aegyptiacus) in Uganda . Am J Trop Med Hyg . 1986 ; 35 ( 2 ): 387 - 92 . PubMed PMID: 3082234 .
64. Johnson BK , Ocheng D , Gitau LG , Gichogo A , Tukei PM , Ngindu A , et al. Viral haemorrhagic fever surveillance in Kenya, 1980 - 1981 . Trop Geogr Med . 1983 ; 35 ( 1 ): 43 - 7 . PubMed PMID: 6684336 .
65. Smith DH , Johnson BK , Isaacson M , Swanapoel R , Johnson KM , Killey M , et al. Marburg-virus disease in Kenya . Lancet . 1982 ; 1 ( 8276 ): 816 - 20 . PubMed PMID: 6122054 .
66. Leendertz SA , Gogarten JF , Dux A , Calvignac-Spencer S , Leendertz FH . Assessing the Evidence Supporting Fruit Bats as the Primary Reservoirs for Ebola Viruses . Ecohealth . 2016 ; 13 ( 1 ): 18 - 25 . doi: 10 .1007/s10393-015-1053-0. PubMed PMID: 26268210 .
67. Mari Saez A , Weiss S , Nowak K , Lapeyre V , Zimmermann F , Dux A , et al. Investigating the zoonotic origin of the West African Ebola epidemic . EMBO Mol Med . 2015 ; 7 ( 1 ): 17 - 23 . doi: 10 .15252/emmm.201404792. PubMed PMID : 25550396; PubMed Central PMCID : PMCPMC4309665 .
68. Omilabu SA , Salu OB , Oke BO , James AB . The West African ebola virus disease epidemic 2014-2015: A commissioned review . Niger Postgrad Med J. 2016 ; 23 ( 2 ): 49 - 56 . doi: 10 .4103/ 1117 - 1936 .186299. PubMed PMID : 27424613 .