For the first time, a retrospective study of CanL in France has been carried out, based on cases reported between 1965 and 2007. The map of cases highlights a strong heterogeneity in the spatial distribution of the disease. Visually, the distribution of CanL in southern France is clustered, with higher case densities on the southern slopes of the Cévennes Mountains and two regions of the Maritime Alps (Figure 1). In addition to these Mediterranean records, this case map also shows a northern focus, corresponding to 13 cases detected by Houin et al.  in six different localities near Tours.
The case map is based on presence only, which does not takes into account the prevalence data obtained by some surveys. Some biases could not be avoided. Firstly, a single case report has the same value as a locality with high disease prevalence. Secondly, the clustering of presence spots might reflect the spatial distribution of the disease, and/or the sampling effort and strategy of the leishmaniasis teams from Montpellier, Lyon, Marseilles and Nice. Certainly, some areas were insufficiently reported, such as in the Pyrénées-Orientales département, where all the specific case localities were not noted in a publication giving an overall prevalence of 6.9% .
The statistical environmental analysis (PCA followed by a HAC) revealed the existence of two groups of leishmaniasis cases. The first group is located on the Cévennes slopes, characterized by relatively low average temperatures, high average rainfall and much forest cover. The second group is located on the Mediterranean coast, characterized by higher average temperatures, lower average rainfall and less forest cover (Figure 2). These two groups may correspond to the environments favoured by the two species of sandfly vectors in France, as previously shown in southern France and Morocco [22, 23]. Rispail et al.  identified, in Morocco, different associations between the distributions of the two vectors and Mediterranean bioclimatic zones, namely humid and sub-humid for P. ariasi, compared with sub-humid and semi-arid for P. perniciosus. Our environmental model also identified two distinctive profiles, with the two main classes matching the bioclimates associated with the two vector species: Class1 matches the bioclimates of P. ariasi, whereas Class 2 matches those of P. perniciosus.
According to the univariate correlation analysis, human and canine densities are, as we expected, significant variables for explaining the distribution of CanL in France. Their densities co-vary (Spearman r = 0.84), but both were retained to provide all relevant information about CanL distribution. This was possible because the Maxent procedure does not require independent variables. Average annual temperature and winter temperature also helped to define both environmental profiles, but in different ways: the number of CanL cases was negatively correlated with the temperature averages in the first profile (P. ariasi), whereas it was positively correlated in the second profile (P. perniciosus). Moreover, additional significant variables were selected for P. ariasi: the average summer rainfall, the proportion of coniferous forest and the elevation. On the other hand, the distribution of P. perniciosus was negatively correlated with the proportion of broadleaf forest. These differences are also consistent with the ecological niches of these two sandflies .
Our ecological niche modelling approach has produced the first risk map of CanL for France, highlighting the potential distribution of the disease. The new areas at risk are mostly located in western France, along the Atlantic coast, from the Pyrénées-Atlantique in the South to the Loire-Atlantique in the North. These areas correspond mainly to areas likely to be favoured by P. perniciosus, with only a few places having the ecological profile of P. ariasi (Haute-Vienne and Pyrénées-Atlantiques) (Figure 4). Some new areas at risk have ecological niches likely to be favourable for both species, but they are rare (Figure 4).
This risk map is consistent with the known distribution of P. perniciosus. Moreover, it should be noted that, in recent years, CanL cases have been reported in several locations outside the Mediterranean region, and always inside the Atlantic area where the risk map predicts emergence. These cases were reported around Limoges (Haute-Vienne) , near Cholet and Angers (Maine-et-Loire) (Bourdeau, 2009, personnal communication), and around Niort (Deux-Sèvres) (Kasbari, 2009, personal communication). In these places, a few imported CanL cases seem to be at the origin of local dog transmission, and horizontal dog transmission cannot be ruled out because cases were always grouped inside kennels. However, vectors were present as well.
Our risk map does not match well the range extensions of CanL mapped by Bourdeau . The latter is based on vet questionnaires, and it shows a more limited range extension of clinical autochthonous cases of CanL to the north and west of the enzootic Mediterranean region. Our ecological niche model predicts the environmental suitability for CanL, separating this into two classes that probably reflect the niches of the two vectors, but the realized niche may be smaller than the fundamental niche predicted by the model .
The areas identified at risk for the disease may be used for entomological or veterinary surveillance. However, our results have to be treated with caution. Indeed, the risk model is based on a retrospective database concerning all reported cases of CanL. The assignment of the case locations (often the centre of the municipality) is likely to introduce some errors. For example, some cases recorded from the littoral plain of the Languedoc could be related to hunters' dogs, which could have contracted the disease in the Cévennes Mountains, where they are taken for hunting . Subclass 2b could correspond to the leishmaniasis cases in this population of dogs (Figure 3).