Epomops franqueti
INTRODUCTION
Humans are affected by a large number of infectious diseases. Most of
these infectious diseases are zoonoses. Most of zoonoses have mostly
viral origin and are emerging and reemerging. The viral diseases of the
Filoviridae family, such Ebola, causes particularly hemorrhagic fevers.
Ebola hemorrhagic fever is indeed among the zoonotic viral disease with
high mortality rate. The Ebola virus causes severe morbidity and high
mortality in humans and wildlife (Bausch and Schwarz, 2014; Fiorilloet al. , 2018).
Number of Ebola virus disease outbreaks have occurred in humans last
forty years, with mortality rates reaching values up to 90 % (Feldmann
and Geisbert, 2011; Bausch and Schwarz, 2014). Most of Ebola hemorrhagic
fever outbreaks have historically occurred in Central Africa. The first
outbreak of Ebola virus in humans was registered in 1976 in Southern
Sudan (Report of a WHO/International Study Team, 1978), but likely
occurred as early as 1972 in Tandala, DRC. Until 2014, DR Congo had
already recorded seven outbreaks of Ebola (Maganga et al. , 2014).
There have been two other outbreaks, the latest of which is in
Beni-Butembo, which is still at the beginning of 2019. Thus, to date,
the DR Congo has already recorded nine Ebola outbreaks being hence one
of the countries that have experienced the most Ebola outbreaks.
Up to now, it is not well understood the way the Ebola transmission
between outbreaks is maintained (Peterson et al. 2004). However,
significant progress has been made in identifying the potential
reservoirs of Ebola viruses. Recent studies have identifiedHypsignathus monstrosus (H. Allen, 1861), Myonycteris
torquata (Dobson, 1878) and Epomops franqueti (Tomes, 1860),
three bat species, as the most likely to be the reservoirs of the Ebola
viruses (Leroy et al. , 2005; Groseth et al. , 2007;
Peterson et al. , 2007; Pourrut et al. , 2009; Haymanet al. , 2010; Olival and Hayman, 2014). Leroy et al. (2007) have
shown that in Africa, there is evidence of Ebola outbreaks in humans due
to exposure to bats. Based on this evidence, it can be assumed that the
distribution of Ebola viruses is limited by certain factors, including
the distribution of bats such H. monstrosus, M. torquata andE. franqueti (Nyakarahuka et al. , 2017).
It is believed that climate change will affect future distribution of
bats reservoirs of Ebola virus. A number of studies have reported the
impact of climate change on Ebola outbreaks in Africa and the migratory
patterns of bats (Newson et al. , 2009). Indeed, migratory species
are particularly likely to be affected by climate change at some point
in their life cycles, and there is already compelling evidence for
impacts on a wide range of birds, marine mammals, fish, sea turtles,
squid, bats, terrestrial mammals and insects (Robinson et al. ,
2009). Reduced precipitation, increasing temperatures and
desertification have caused a large number of fruit bats to migrate from
their ecological niches in the equatorial rain forest to other areas
where environmental conditions are more favorable for survival (Omolekeet al. , 2016).
Actually, the world is facing climate change and this has been for
decades. The effects that climate change is having on wildlife
populations are of increasing interest to ecologists (Adams and Hayes,
2008). The survival of species and integrity of terrestrial ecosystems
are threatened by the climate change. To adapt to climate change, a
thorough knowledge of its impact on plant and animal species is of
crucial importance (Fandohan et al. , 2013). Recent research
increasingly shows that climate change will significantly affect
biodiversity and species distribution. With the most recent research,
one of the hypotheses put forward is that in Africa, 25 to 42 % of
plant and animal species could be threatened and could thus lose up to
90 % of their geographical distribution areas by 2085 if global warming
exceeds 1.5 to 2.5 degrees Celsius (Busby et al. , 2012). Elithet al. (2010) and Guisan and Thuiller (2005) have enumerated
modeling tools among which species distribution models. The utilization
of species distribution models is spreading out (Martínez-Meyer et
al. , 2004; Papeş and Gaubert, 2007; Tsoar et al. , 2007; Jenningset al. , 2013). Species distribution models are numerical tools
that provide potential distribution, aid in conservation planning and
their project future behaviour in response to environment changes
(Peterson, 2003, 2006; Thorn et al. , 2009; Barve et al. ,
2011). The main objective of mapping spatial distribution of vectors and
reservoirs of diseases is to manage diseases impacts by providing
geographic information that enables decision-makers to make
evidence-based decisions (Benedict et al. , 2007; Lindsay et
al. , 2010) or the planning and targeting of surveillance and
interventions (Dicko et al. , 2014). An additional motivation of
the wide use of species distribution models techniques comes from the
high impacts of diseases on humans, animal and plant and the fact that
environmental drivers play a major role in their emergence (Chaveset al. , 2008; Jones et al. , 2008; Pautasso et al. ,
2010). During last decade, modeling techniques have been developed to
model ecological distribution. Maximum entropy (MaxEnt) approaches have
recently been introduced and used by many research work and have shown
to be very successful tool (Peterson et al. , 2007; Elith et
al. , 2010). Several studies comparing species distribution modeling
techniques indicated that MaxEnt modeling (Phillips et al. , 2006)
performed as well or better than the other techniques (Elith et
al. , 2006, 2011; Hernández et al. , 2006; Phillips et al. ,
2006; Phillips and Dudík, 2008; Baldwin, 2009). As such, considering the
new features integrated in the new release of MaxEnt program (Phillips,
2017), it should be a very useful and accurate tool for delineating
species distributions.
Modelling H. monstrosus, M. torquata and E. franquetispecies distribution using MaxEnt modelling technique could essentially
improve our understanding of the spatial distribution of current and
future risk of increased or decreased distribution of these species.
In order to better understand the nature of Ebola viruses risk, this
study aims to define areas of DR Congo where zoonotic transmission of
Ebola viruses can occur, currently and in the future. Thus, by studying
the spatial distribution of H. monstrosus , M. torquata andE. franqueti , potential reservoirs of the Ebolavirus, this study
seeks to identify their current and future favorable and suitable areas
in DR Congo. This study illustrates the climate change risk assessment
of the spatial distribution of these species according to the climate
scenario in 2050 (2041 - 2060) and 2070 (2061 - 2080), and to seek
attention of their future favourable habitats in DR Congo.
MATERIALS AND METHODS
Occurrence records
In this study, due to the lack of sufficient occurrence data of DR
Congo, we used occurrence data from the entire African continent to
build the model and make the prediction only in DR Congo. The
geographical coordinates (longitude and latitude) of H.
monstrosus, M. torquata and E. franqueti were collected online
from the GBIF database (http://www.gbif.org). All occurrence
records were checked for accuracy in ArcGIS prior to use. The data were
quality controlled in order to eliminate or remove suspicious or
duplicate records (Lobo et al. , 2008; Warton and Shepherd, 2010;
Stigall, 2012). A total of 123 observation points of H.
monstrosus , 79 M. torquata and 201 E. franqueti in Africa
continent were used for the modelling. In DR Congo, only 10 occurrence
records of E. franqueti , 9 of H. monstrosus and 7 ofM. torquata were found. It was not possible to model the
distribution of these species in DR Congo using only these data, since
they were insufficient. This is why we came up with the idea of using
data from all of Africa’s occurrence records to build the models and do
prediction only on the extent of DR Congo. The map representing the
points of occurrence is illustrated in Figure 1.
Environmental variables
In this study, we used elevation data together with bioclimatic data.
Elevation data (Digital Elevation Model) was obtained from USGS database
(https://earthexplorer.usgs.gov) and the current and future
climate data were collected from WorldClim database
(http://www.worldclim.org) and used to build the species
distribution model in order to find the suitable areas for H.
monstrosus, M. torquata and E. franqueti . Bioclimatic data
collected from the WorldClim database are obtained from interpolations
of monthly averages of precipitation and temperature taking into account
climate data collected over long periods of time (Fick and Hijmans,
2017). The 19 bioclimatic variables (bio1; Mean Annual Temperature,
bio2: Mean Diurnal Range, bio3: Isothermality, bio4: Temperature
Seasonality, bio5: Maximum Temperature of Warmest Month, bio6: Min
Temperature of Coldest Month, bio7: Annual Temperature Range, bio8: Mean
Temperature of Wettest Quarter, bio9: Mean Temperature of Driest
Quarter, bio10: Mean Temperature of Warmest Quarter, bio11: Mean
Temperature of Coldest Quarter, bio12: Annual Precipitation, bio13:
Precipitation of Wettest Month, bio14: Precipitation of Driest Month,
bio15: Precipitation Seasonality, bio16: Precipitation of Wettest
Quarter, bio17: Precipitation of Driest Quarter, bio18: Precipitation of
Warmest Quarter and bio19: Precipitation of Coldest Quarter) have a high
biological significance, are widely used to explain the adaptation of
species to environmental factors and have been widely used in modelling
species distribution. In the Worldclim database, the current period
represents interpolations from monthly average precipitation and
temperature data collected from 1950 to 2000. All environmental
variables were in raster format with a 2.5-arc minute resolution (≈ 4.5
km2). The 20 environmental variables (19 bioclimatic
variables and Elevation) were subject to the correlation test using the
R software (R Development Core Team, 2018). Consequently, Pearson
correlation coefficients belonging to the interval ]-0.8,0.8[
(|r| < 0.8) with only a subset of variable
were included in order to eliminate the problem of collinearity in
environmental predicators (Elith et al. , 2010).
We used eleven environmental variables in this model prediction. This
was after assessing for collinearity in the model and removing all the
collinear variables. The result from the correlation analysis identified
eleven bioclimatic variables and elevation as contributing to the
environmental variation across the study area: Precipitation of Driest
Quarter (bio17), Annual Temperature Range (bio7), Elevation, Mean
Diurnal Range (bio2), Precipitation Seasonality (bio15), Precipitation
of Warmest Quarter (bio18), Precipitation of Wettest Quarter (bio16),
Mean Temperature of Wettest Quarter (bio8), Mean Temperature of Warmest
Quarter bio10), Min Temperature of Coldest Month (bio6), Precipitation
of Coldest Quarter (bio19) and Mean Temperature of Driest Quarter
(bio9).
To determine what the distribution of these species might be in the
future and thus assess the potential impacts of climate change on their
distribution, we used the model built from current data to make the
prediction using bioclimatic future prediction data obtained using the
future HadGEM-CC projection model. The impacts of climate change
strategies on greenhouse gas emissions are considered more in the RCPs
scenarios, and the projection of future climate change is more
scientifically described. RCP4.5 (medium greenhouse gas emission
scenario) and RCP8.5 (maximum greenhouse gas emission scenario) for the
near future: 2050 (2041 - 2060) and the middle century: 2070 (2061 -
2080) were selected for the future model prediction of H.
monstrosus , M. torquata and E. franqueti distribution in
DR Congo.
Distribution modeling
We used MaxEnt software (Phillips et al. , 2006, 2017) to build
model and predict suitable habitat distribution of H. monstrosus,
M. torquata and E. franqueti in DR Congo. We used Presence-only
data to model the suitable habitat of the three species. The MaxEnt
model was built as a function of environmental variables, and it is
consistently among the highest performing species distribution models
(SDMs) (Bradie and Leung, 2017). Response curves indicate the
relationships between climatic variables, and the predicted probability
of the presence of each species was determined by MaxEnt. The percent
contribution and permutation importance of environmental variables were
calculated, and jackknife procedures were executed in MaxEnt. These
analysis methods are all useful to measure the importance of the
environmental variables. MaxEnt estimates the probability a species will
be present based on presence records and randomly generates background
points by finding the maximum entropy distribution. An estimate of
habitat suitability for a species was exported from MaxEnt, and its
range generally varied from 0 (lowest) to 1 (highest). For each species
we ran 100 submodels each trained to a randomly selected bootstrap of
the occurrence dataset. Prediction map form each submodel has been
generated in order to calculate the mean prediction and standard
deviation of each pixel for each species. Model predictions were
imported into a geographic information system (GIS), and maps were
generated using ArcMap 10.6. Five arbitrary categories of habitat
suitability for the three species of bats were defined as not suitable
habitat (0.00 - 0.05), slightly suitable habitat (0.05 - 0.25),
moderately suitable habitat (0.25 - 0.50), suitable habitat (0.50 -
0.75) and highly suitable habitat (0.75 - 1.00) based on predicted
habitat suitability.
In this study, the ROC curve method was utilized to assess the model’s
explanatory power (Peterson et al. , 2007). The AUC (area under
roc curve) is an effective threshold-independent index that can evaluate
a model’s ability to discriminate presence from absence (or background).
For reducing the bias of estimation, jackknife method has been used.
This method can estimate parameters and adjust the deviation without
assumptions of distribution probability. In SDM, the jackknife method
was used to analyze the effects of environmental variables on model
results to choose dominant factors. The specific process involves:
- Calculating the training gain for the model with only variable. Higher
training gain indicates that the variable has high prediction power
and contributes greatly to species distribution;
- Calculating the training gain for the model without a specific
variable and analyzing the correlation between the removed variable
and the omission error. If the removal of an environmental variable
leads to a significant increase in the omission error, it indicates
that the variable has a significant effect on the model’s prediction;
- Calculating the training gain for the model with all variables.
RESULTS
Model performance and contributions of variables
In this study, from the ROC curves, AUC values were used to evaluate the
performance of the MaxEnt models. Many studies showed that an AUC of
high values led to better results that significantly differed from the
random predictions. The accuracy of prediction of H. monstrosus,
M. torquata and E. franqueti during the current period was found
to be excellent (Mean AUC ≈ 0.96, Figure 2) according to the identified
evaluation criteria.
Among the environmental variables, the Precipitation of the driest
quarter (43.1% - 64.3 %) played major role in the spread of H.
monstrosus, M. torquata and E. franqueti (Table 1 and Figure 3).
In addition, the Temperature Annual Range played also a major role in
the spread of E. franqueti (12.7 %) and H. monstrosus(14.6 %) while the Precipitation of Coldest Quarter has also showed to
play also a major role in the spread of M. torquata (36.4%) andH. monstrosus (10.4 %).
Tableau 1: Estimates of contribution and permutation importance of
environmental variables in MaxEnt modeling