Spatial distribution modelling of Culicoides (Diptera: Ceratopogonidae) biting midges, potential vectors of African horse sickness and bluetongue viruses in Senegal

African horse sickness (AHS) and Bluetongue (BT) are vector-borne viral diseases affecting equids and ruminants (mainly cattle and small ruminants), respectively. These two viruses are biologically transmitted by females of several species of Culicoides (Diptera: Ceratopogonidae) biting midges. Bluetongue is distributed worldwide, affecting almost all continents (except Antarctica) while AHS occurs in sub-Saharan Africa, with rare incursions into Europe and Asia [1–3]. The importance of these two arboviral diseases derives from their potential for rapid spread and their major economic impact due to direct mortalities, restriction of animal movements, surveillance and vaccination costs, as shown by the recent outbreaks of BT in Europe [4–6] and AHS in Africa [7, 8]. Their importance requires immediate notification to the World Animal Health Organization [9].

Following the important AHS outbreak in Senegal in 2007 [7, 8], several studies were initiated in the Niayes area (in the western part of the country) in order to better characterize Culicoides diversity and ecology [10–14], and the impact of environmental and climatic variables on the dynamics of major vector species (proven and suspected vectors) [15]. These studies recorded for the first time the presence of Culicoides oxystoma and showed that this species was extremely abundant in the Niayes, both in black-light and in horse-baited traps. Because of its abundance, close contact with horses and its known or suspected role in the transmission of other arboviruses (Akabane, epizootic haemorrhagic disease and bluetongue viruses), these studies concluded that it should be considered as a potential vector of AHS virus. These studies also showed that Culicoides population peak of abundance occurred at end of the rainy season in September and October.

In order to anticipate future events, prevent and better control Culicoides-borne disease outbreaks, it is essential to identify areas with higher risk of Culicoides-borne pathogens transmission at a national or regional level. This allows characterizing vector species distribution in Senegal. This is essential to determine if vectors are present throughout the country, if their abundance is highly variable or not and thus identify high-risk transmission areas. Combined with horse population distribution, this enables to target vaccination in these high-risk areas in the event of an outbreak. The first step towards this aim implies improving our knowledge on the distribution of vectors in Senegal. Thus, a nation-wide entomological trapping campaign was carried out to model the abundance of proven and potential vectors of AHS virus (AHSV) and/or BT virus (BTV) in Senegal and map their distribution.

Several studies demonstrated the ability to predict the presence and/or abundance of Culicoides midges using meteorological and environmental variables mainly derived from satellite imagery. These models were established using statistical techniques such as discriminant analysis [16–20], generalized linear models such as logistic regression [21–25] and more recently, data-mining techniques such as random forests [17, 26, 27].

The aim of this study is to model the spatial distribution of Culicoides species which are known or suspected to be vectors of AHSV and/or which are the most abundant in Senegal. The models could help stakeholders to make decisions for the control of these diseases. For this, entomological data were obtained through a nationwide trapping campaign and models for each species were developed to map their abundance. Three different modeling approaches were compared in order to find a simple yet robust method for each species. The first method consists in a spatial interpolation by ordinary kriging of Culicoides abundance data and does not necessitate environmental data. The other two consist of analyzing the relationship between Culicoides abundance, environmental and climatic data to model abundance in unsampled areas and were carried out by implementing generalized linear models and random forest models.

Between September, 17th and October, 14th 2012, among the 108 sites initially selected, 98 sites were sampled and 10 sites were not prospected because of logistical issues. In two of the 98 sampled sites the trapping results were not exploitable and thus were not considered (one conservation issue and one battery failure).

Each of the following five species represented more than 10% of the catch: Culicoides oxystoma, Culicoides enderleini, Culicoides imicola, Culicoides miombo and Culicoides trifasciellus. As found previously in the Niayes region, C. oxystoma (26.9% of catches) was the most abundant species and was present in 93 out of the 96 sites sampled (97% of sites). Percentages of individuals caught compared to the total (and frequencies of sites where the species was present) catches were 23.89% (95.83%) for C. enderleini; 14.38% (95.83%) for C. imicola; 12.87% (82.29%) for C. miombo and 10.76% (56.25%) for C. trifasciellus, respectively (Table 1).

We chose to model the distribution of five species: the first four most abundant (C. oxystoma, C. enderleini, C. imicola and C. miombo) and also Culicoides bolitinos as it is a proven vector of AHSV in South Africa [55]. Culicoides bolitinos was collected in low numbers (0.32% of catches) and in 57.29% of the prospected sites.

For the five Culicoides species of veterinary interest, observed abundance data (maximum abundance of the two consecutive nights) are shown in Fig. 1. Overall, three species had similar (yet not identical) patterns: C. oxystoma, C. imicola and C. enderleini. They were abundant in the west (groundnut basin), in the south-east and in some sites in the north-east (river delta). The distribution of C. miombo differed mainly in that it was very rare in the north-eastern sites while the distribution of C. bolitinos was much sparser.

Regarding C. oxystoma, the sites where abundances were the highest are located in west central (Fatick, Kaolack and Diourbel regions), in the far west (Dakar region) and south of Senegal (Kolda region) (Fig. 2). For C. enderleini, the highest abundances were observed in the south (Kolda and Kedougou), in the groundnut basin (specifically in Fatick and Kaolack) and northwest of the Saint-Louis. For C. imicola, the highest abundances were recorded in the south and southeast (Kolda, Tambacounda and Kedougou) and in the center-west (Thies, Diourbel, Fatick and Kaolack). Culicoides miombo and C. bolitinos were very rare in northern Senegal. The highest abundances of C. miombo were obtained in Kolda, Tambacounda and in the groundnut basin (Fig. 2). Culicoides bolitinos abundances were very low. Culicoides bolitinos highest densities were noticed in Kedougou region and in the groundnut basin (particularly in Fatick, Kaffrine and Kaolack).

The variogram results determined the maximal interpolation distances for each of the five species. These distances were: 20 km for C. oxystoma, 48 km for C. imicola, 13 km for C. enderleini, 25 km for C. miombo and 10 km for C. bolitinos (Fig. 3). Different models were fitted to the variograms: spherical models for C. oxystoma, C. imicola and C. bolitinos, a circular model for C. miombo and an exponential model for C. enderleini. Abundance maps by ordinary kriging were developed respecting the maximal interpolation distance for each species (Fig. 4). For C. oxystoma, C. enderleini, C. imicola and C. miombo, abundances were predicted to be very high in west-central and southern Senegal and decrease gradually towards the north. However, high abundances of C. enderleini were predicted in the northwest of Senegal and, to a lesser extent, for C. oxystoma. For C. bolitinos, the highest abundances were modelled in the extreme south of Senegal and south of the groundnut basin (specifically in Kaffrine region).

To produce abundance maps for the entire Senegalese territory, GLM and RF models were developed. Table 2 presents the 10 most important predictors for each species according to GLM models. Maximum NDVI was among the three most important variables for all species. Average rainfall was also among the 3 most important variables for 3 of the 5 species (C. oxystoma, C. enderleini and C. imicola). Other environmental variables that influence Culicoides abundances were day and night temperatures and percentage cover of water bodies for C. oxystoma and C. imicola; night temperatures for C. enderleini; day and night temperatures for C. miombo; livestock density and percentage cover of savannah for C. bolitinos. Abundance maps based on GLM modeling showed that for all five species of veterinary interest, predicted abundances are very low along the Senegal River Valley and very high in the south (Fig. 5). High abundances of C. imicola and C. oxystoma are predicted on the entire country although for C. imicola areas of high and low abundance are strongly interlinked (areas of high abundance are often close to areas of low abundance). Culicoides imicola and C. enderleini predicted abundances are particularly strong in the southern third of the country and medium in the middle third. The distribution of C. miombo in the south is even more pronounced: abundant in the southern third, it is less abundant in the central and almost absent in the northern third of the country. Not surprisingly, C. bolitinos predicted abundances are lower than those of the other species. The area where it is most abundant is the delta in Casamance in southeast Senegal and in southwest Senegal (Fig. 5).

C. oxystoma		C. enderleini		C. imicola		C. miombo		C. bolitinos
Predictors	VarImp	Predictor	VarImp	Predictor	VarImp	Predictor	VarImp	Predictor	VarImp
AvRaina	83	MaxNDVIa	100	AvRaina	74	MaxNDVIa	100	MaxNDVIa	72
MaxNDVIa	74	AvRaina	64	MinNDVIa	58	MinNDVIa	98	AvHosta	64
MinDlsta	62	MinNDVIa	58	MaxNDVI	34	MaxNlsta	87	lcSavanhaa	63
MaxNlsta	57	MaxNlst	39	MaxNlst	32	MinDlsta	68	MaxDlst	47
MinNlst	37	MinNlst	33	MinDlst	32	AvHost	37	MaxNlst	36
lcWater	35	MaxRain	28	lcWater	28	AvRain	36	avDlst	30
MinNDVI	22	lcWater	17	avNlst	27	AvDlst	21	lcForest	16
MaxRain	18	AvDlst	13	avDlst	27	MaxRain	15	lcWater	14
AvDlst	14	lcForest	11	AvHost	25	AvRain	12	AvRain	10
AvHost	12	MinDlst	8	MaxRain	23	lcWater	8	MinDlst	6

Abbreviations: AV average, Min minimum, Max maximum, Dlst day land surface temperature, Nlst night land surface temperature, lc landcover, NDVI normalized difference vegetation index, Host livestock density

^a Predictors for which VarImp are greater than 50%

Predicted Culicoides abundance maps provided by RF model are shown in Fig. 6. For the five species, an increasing abundance gradient from north to south is predicted. For C. oxystoma, C. enderleini and C. imicola, predicted abundances are very high in the southern third of the country, and for C. miombo abundances are high to very high in the southern and middle thirds. For all the species, predicted abundances are very low in the north of the country, particularly in the north of Ferlo and in the Senegal River valley. Culicoides oxystoma is predicted to be particularly abundant along the western coast from Dakar to Ziguinchor and in The Gambia along the river. This species also has very high predicted abundances in the south, southeast and center-west of Senegal (Fig. 6). Culicoides enderleini and C. imicola exhibit a similar pattern with very high abundances predicted in Casamance (Ziguinchor, Sedhiou and Kolda), eastern Senegal (Tambacounda and Kedougou) and the center-west of the groundnut basin (Fatick and surrounding). Regarding C. miombo and C. bolitinos, very low abundances were predicted in the Niayes’ area. Culicoides bolitinos predicted abundances are lower than those of other species; the area where its abundance is predicted to be the highest is the south and southeast, particularly in Kedougou region.

The identification of suitable areas for five targeted Culicoides species in Senegal was performed with three different modelling techniques: a spatial interpolation method, kriging and two statistical modelling methods using meteorological and environmental variables, GLM and RF models. Interpolation by kriging yielded incomplete prediction maps. To obtain complete maps, more intensive entomological survey would be necessary, especially for C. bolitinos and C. enderleini. Given the logistical efforts required to conduct such survey and identify Culicoides species, this option seems unrealistic.

The comparison of predicted abundance maps from each model showed consistent spatial structures. Although the predictions of maximum abundances provided by kriging were limited by extrapolation distances, abundance maps by kriging and RF models identified similar structures whereas GLM models provided abundance maps slightly different from those resulting from kriging and RF models. GLM outcomes differed for C. oxystoma particularly in Tambacounda and Kedougou, for C. enderleini in Fatick, Kaffrine, Kaolack and western Saint-Louis, for C. imicola in Thies, Fatick, Kaolack, for C. miombo in Dakar, Kaffrine and Tambacounda and for C. bolitinos in Kaolack, Kaffrine and Fatick.

Prediction errors between observed and predicted abundances were lower for the RF model than for the GLM model. Recent studies have also shown that RF approaches provide good modelling results both for regression [26] and classification [17, 26, 27] trees. The effect of climatic and environmental variables will be discussed for each species based on the results from RF models since they provided better estimates of abundances.

Regarding C. oxystoma, this is, to our knowledge, the first spatial distribution model published. Studies on Culicoides dynamics in Senegal show that this species is very frequent and abundant in the Niayes area [13, 14]. Our study shows that this species is widespread and abundant over the entire country, with an increasing gradient from north to south, and high predicted abundance in the west, near the coast and around river deltas.

Average rainfall was the most important variable influencing C. oxystoma, C. enderleini and C. imicola abundance, and the second most important variable for C. miombo. This variable seems to limit the abundance of these species in the north, more arid (see isohyets representation of Senegal in Fig. 2). In addition to rainfall, day and night temperatures strongly influenced C. oxystoma abundance. Its highest abundance was observed in the groundnut basin and southern Senegal. Although ecological drivers of C. oxystoma remain poorly described, this species is known to be abundant in South Korea [56] and in Kagoshima, southern Japan [57]. These two areas are characterized by a humid subtropical climate. Culicoides oxystoma was collected from May to October in the south of Korea [56] and from May to November in southern Japan [57], i.e. in the warmer and rainy period of the year, showing that rainfall and temperature appear to influence the distribution of this species in both Japan and in Senegal.

Livestock density seems to have a weak influence on C. oxystoma abundances. Generally, livestock density was not identified as a variable with a high impact on Culicoides abundance except for C. bolitinos. This is surprising because for many other Culicoides species, especially those vectors of pathogens to animals, the presence of hosts and their density are key factors known to affect their catch [58–60]. Several reasons might explain this outcome. First, host densities data was available at a low spatial resolution (100 ×100 km). Secondly, these consisted in 2005 predictions, and densities could have changed (increased or decreased) since then. Thirdly, host densities included only three ruminant species (cattle, sheep and goats) but did not include horse densities. To our knowledge there is no map of horse densities available for Senegal. The lack of data on horse densities could be an important hindrance since in the field, in Senegal, Fall et al. [61] found high abundances of C. oxystoma on horses. Finally, some Culicoides species are able to feed on other hosts, including birds and wildlife, and it is possible that the densities of these other host species impact the distribution of that species.

As for C. oxystoma, little information in the literature is available on climatic and environmental variables that could impact C. enderleini. In this study, C. enderleini was very abundant in the southern third of the country. RF models showed that its abundance is significantly associated with rainfall, vegetation index, day and night temperatures and, to a lesser extent, livestock density. This is coherent since in the south of Senegal, rainfall is important, average NDVI, temperatures and livestock density are high.

The distribution of C. imicola and C. enderleini followed a similar pattern with an increasing north-south gradient. Again, average rainfall could be a limiting factor in the north, explaining higher abundances in the south (Ziguinchor, Kolda and Kedougou) and center-west (in the groundnut basin) regions that record the highest rainfall in Senegal. The importance of rainfall on the distribution and abundance of C. imicola was highlighted in Morocco [33] and Europe [22, 26, 37, 62]. The study of Culicoides dynamics in the Niayes area also showed that peaks of abundance of C. imicola occurred during the end of the rainy season (September-October) [13].

Maximum and minimum NDVI were the second and third most important variables associated with C. imicola abundance. The link between the distribution of C. imicola and NDVI is consistent with previous studies showing that high NDVI values were associated with increased C. imicola abundance in the Niayes area in Senegal [15], in north Africa and in Europe [16, 24, 30–33]. In Senegal, areas where the vegetation index is the highest are also those where rainfall is highest, as both are linked [63]. Studies have also highlighted the link between NDVI and soil moisture [64]. Soil moisture could be an important factor influencing larval development of this species. Temperatures (day and night) were also found associated with C. imicola abundance. This finding has been confirmed by several other studies [16, 18, 19, 25, 29, 32, 35].

Concerning C. miombo, maximum NDVI was by far the most important variable influencing its abundance. Its highest abundances were recorded in areas where the vegetation index is the highest, i.e. in the groundnut basin and in southern Senegal. In the north of the country where the vegetation index is the lowest, C. miombo is almost absent. In the Niayes area (particularly in Dakar and Thies), the abundance of C. miombo was low, confirming the results of a previous study [13]. This is probably due to the low vegetation index in this area. In this study, rainfall and temperatures were found associated with the abundance of C. miombo, in coherence with Meiswinkel’s study, which revealed that this species is sensitive to rainfall and high temperatures [65].

Regarding C. bolitinos, no single variable was a dominant driver of its abundance. Indeed, the contribution of the best environmental predictor (maximum NDVI) on its abundance remained very low (variable importance = 6.31). The most important variables impacting its abundance were the NDVI, livestock density and temperature. These three types of variables have also been linked to C. bolitinos abundance in South Africa [66]. Livestock density was an important variable which greatly impacted model accuracy. Studies conducted by Fall et al. [61] showed that C. bolitinos is particularly aggressive on horses with a blood-feeding rate of over 75%. This species is also known to feed on large mammals [67]. Other studies describe the larval habitats of C. bolitinos as being the dung of wild and domestic Bovidae in Africa [68]. The ecology of this species is therefore closely linked to its hosts (through feeding and through its breeding media) and it is thus coherent to find a strong link to livestock densities and weaker environmental drivers of its abundance.

Overall, the highest abundances of these five species, proven or suspected vectors of AHS and BT viruses, were recorded in southern Senegal. If C. oxystoma seems present rather in the southwest of the country, C. enderleini, C. miombo and C. imicola have very high abundances in the center-south (Kolda), in the south-east (Kedougou) and in the groundnut basin. Culicoides abundance maps from RF models were compared with those of AHS outbreaks in 2007 in Senegal [8], which show significant mortality of horses in the groundnut basin. According to Diouf et al. [8], the greatest risk of introduction of AHS virus in Senegal is through the northeastern border because of important commercial trade movements and numerous markets to and from which carters transport goods. These authors suggested focusing surveillance of AHS virus introduction in that area in order to prevent the virus from reaching the groundnut basin where the dense network of markets could substantially amplify disease transmission and diffusion. Diouf et al. [8] considered the south and southeastern part of the country to be less at risk of AHS because of lower horse densities due to the presence of tsetse flies. Because low densities of Culicoides were found in the north-eastern part of Senegal, this study strengthens the hypothesis that the main driver of AHS introduction and spread in 2007 was horse movements. It confirms that the groundnut basin is an area at high risk of AHS transmission, because it combines important vector, host and market densities, which could lead to important epidemics. The very important densities of Culicoides in the south suggest that if the virus was present in the neighboring countries (Mali, the Republic of Guinea and Guinea Bissau), it could be introduced through the spread of infected Culicoides and then be amplified in donkeys.

The two main knowledge gaps to better assess AHS transmission risk in Senegal are (i) vector competence of suspected vectors and of very abundant species; and (ii) a map of equid (horse and donkey) densities. Assessing the vector competence for ASHV of species such as C. oxystoma, C. enderleini and C. miombo is essential to evaluate their roles in disease transmission. Mapping not only equid densities but also equid movement networks in Senegal would be of great help for decision makers, since combining vectors abundance maps with maps of equid densities and movements of equids would enable evaluation of AHS transmission risk. In a second step, combining transmission risk with AHS virus introduction risk would enable making recommendations in terms of early warning systems and vaccination policies.

To our knowledge, this study is the first to map the distribution of five major species of Culicoides of veterinary interest in Senegal. The highest abundances of Culicoides were observed in the south and in the groundnut basin, whereas abundances were the lowest in northern Senegal. Abundance maps were produced using three different modelling approaches. RF models provided better estimates of abundances than GLM models and were not limited by interpolation distances contrary to kriging. Environmental and climatic variables of importance that influence the spatial distribution of species abundance were identified. It is now crucial to evaluate the vector competence of major species and then combine the vector densities with densities of horses and ruminants to quantify the risk of transmission of AHS and BT virus across the country.