C. R. Stephens. Instituto de Ciencias Nucleares, Universidad Nacional Autónoma de México, A. Postal 70-543 México D.F. 04510, México. Tel: +52 555 4692; Fax: +52 555 622 4693; E-mail: email@example.com
We extend a recently developed method for constructing ecological networks to infer potential biotic interactions between species and to also include environmental factors, in particular land cover, thus permitting a simultaneous analysis of the interaction between environment and species distribution as well as inter-species interactions. We apply the method to the transmission and dispersal of leishmaniasis in Mexico. We find that the most important potential vectors and reservoirs can be classified into assemblages associated with different types of habitat. This in turn can be used to understand and map potential transmission risk, as well as to construct risk scenarios for the dispersal of disease from one geographical region to another.
• This study applies inferential ecological networks – a powerful tool with wide application to emerging disease studies – to the problem of leishmaniasis in Mexico. It shows that such networks can be used not only to understand the potential ecological interactions between species involved in the transmission of the disease but also to identify the potential role of environment in disease transmission and dispersal and infer corresponding geographical patterns.
• It provides information about risk scenarios of leishmaniasis transmission and dispersal, identifying both the species assemblages (bats, squirrels, opossums, rats and mice) and habitats that represent significant risk factors for this zoonotic disease.
• It provides a risk map for leishmaniasis in Mexico that combines both mammals and land cover distribution and highlights the need to conduct more specific studies in areas lacking prior information but with high risk potential for the disease.
Although it poses a significant health threat, there are relatively few studies available that allow us to better understand its complex transmission cycle. Of course, transmission cycles can potentially depend on a huge number of factors, both abiotic and biotic, that range from the microscopic to the macroscopic. For instance, the Leishmania parasite generically requires the presence of two wild hosts – a mammal reservoir and an insect vector – to complete its life cycle (Ashford, 1996; Sánchez-Saldaña et al., 2004; Chaves and Pascual, 2006). Unfortunately, little is known about these ecological components of the disease – reservoirs and vectors and their mutual interactions. For instance, there are a large number of possible reservoirs, as the vector can potentially feed from many different species. Furthermore, as each vector and reservoir species is strongly dependent on environmental factors that affect its distribution (Peterson, 2007; Reithinger et al., 2007), environmental changes can alter geographical distributions and therefore the dynamics of the disease (Ostfeld et al., 2006; Poteet, 2006; Sánchez-García et al., 2010). For instance, current habitat fragmentation can increase the risk of human exposure to leishmaniasis via its impact on species composition (Reithinger et al., 2007; González et al., 2010). So, to better understand the dynamics of disease transmission and dispersal, we need to potentially study many different factors, ranging from the distribution of potential reservoirs to climatic and land cover factors, as well as the potential interactions between them.
Fortunately, a great deal of geographical information associated with many of these factors is now available in public databases. Such data facilitate the implementation of spatial analyses that can then be used to better understand the geographical patterns of disease transmission (Peterson, 2007; Ready, 2008). Such studies can, for instance, be applied to evaluate the potential risk of zoonotic diseases (López-Cárdenas et al., 2005; Porcasi et al., 2005). However, such analyses have chiefly been limited to a small subset of species that have been previously recognized as being directly involved in the transmission of the disease (e.g. Peterson et al., 2002; Peterson and Shaw, 2003). In general, though, the number of potential species that are involved in the transmission of a disease will be more than the known ones (Acha and Szyfres, 2003; Reithinger et al., 2007; De Lima et al., 2008). In other words, the number of known reservoirs is usually much less than the number of potential reservoirs (Stephens et al., 2009). This is especially true of a neglected disease such as leishmaniasis. Consequently, there is a need to develop methodologies that can predict which organisms could be important reservoirs and infer their relationships with the environment (Ostfeld et al., 2006; Poteet, 2006). Therefore, one of the most important challenges in modelling disease transmission is to create mathematical models that can incorporate many different variables of different types that allow us to better understand the complex process of disease transmission by incorporating as many variables as possible to infer inter-species ecological interactions and the role of environment in determining the geographical zones with the greatest risk.
Ecological networks offer a powerful tool to visualize inter-species ecological and evolutionary interactions from geographical data (Strogatz, 2001; Montoya et al., 2006; McCann, 2007). These networks have important applications, such as in biodiversity studies and emerging diseases. For instance, using purely point collection data, Stephens et al. (2009) used ecological networks to predict potential reservoirs for leishmaniasis in Mexico by inferring potential biotic interactions between vectors and reservoirs. However, what was not considered in that article is how this novel methodology can also be used to incorporate variable types other than point collection data. In this contribution, we show how the methodology proposed in the study of Stephens et al., 2009 to construct ecological networks can also be used to model other important variables. In particular, we use land cover to infer the role of landscape in determining the dynamics of disease transmission and dispersal and consequently identify geographical patterns and focal species for leishmaniasis transmission.
Of course, we do not wish to imply that the presence of potential vectors and reservoirs is the only factor that plays a role in the transmission cycle. At a more microscopic level, for instance, pathogen transmission will depend on the realized reservoir competence of a given species. This in turn is likely to be specific to a particular reservoir–vector pair. Thus, one is left with the problem of how to model in the absence of detailed information about these factors. So, although the presence of potential vectors and reservoirs is not a sufficient condition to maintain a transmission cycle, it is necessary. Thus, the logic of the present methodology is to construct ‘first-order’ risk models for transmission and dispersal by assuming that in the absence of other information, competencies can be assumed to be equal. The fact that this approach can lead to predictive results is manifest in the study of Stephens et al. (2009), where the methodology correctly identified currently known reservoirs of Leishmania.
Materials and Methods
We used collection points for 145 mammals considered as the most important potential reservoirs for leishmaniasis (Stephens et al., 2009). Additionally, we included data for 312 mammals not presented in the previous list. The subsequent analysis allows us to explore whether, by habitat co-association, we can identify other potential reservoirs. Data are based on museum voucher specimens from national and international collections and public electronic databases (GBIF; http://www.gbif.org, and CONABIO; http://www.conabio.gob.mx). The mammal data set contains 58 040 unique point collections from georeferenced localities. Additionally, we built a geographical database for 11 species of Lutzomyia (sandflies) using 270 collection points taken from published literature and from national collections: Instituto de Diagnóstico y Referencia Epidemiológica (InDRE, Mexico City); Colección Entomológica Regional, Universidad Autónoma de Yucatán (UADY, Mérida); and the Laboratorio de Medicina Tropical, Universidad Nacional Autónoma de México (UNAM, Mexico City).
Land cover data
We used the map of the Inventario Nacional Forestal 2000 (Palacio et al., 2000; http://www.igeograf.unam.mx, http://www.inegi.gob.mx) as a base for current land use and vegetation types in Mexico. This is based on both LandSat satellite imagery interpretation and ground field validation of the main vegetation types and land use in Mexico and scaled at 1 : 250 000. This layer included seven majors kinds of natural and transformed land cover distributed in Mexico: tropical forests, temperate forests, xeric scrubs, wetlands, grasslands, crop fields and human settlement.
To build an ecological network, we adopt a nonparametric ‘data mining’ approach, using available geographical data (for more details, see Stephens et al., 2009). The first step was to determine the occurrence of species in any given land cover. This can be done in different ways. In this study, we divided up a geographical region of interest into spatial cells, xα, and then counted co-occurrences in the different cells of the grid. The geographical region of interest for the data of the present study is Mexico. We used 3337 square cells of linear size 25 km, which corresponds to an average of 20 point collections per cell. We consider Bi (xα) as a measure of the presence of the taxa i in the spatial cell xα. Our main object of interest is P(Bi(xα)|I(xα)), the probability that the distribution measure Bi(xα) takes a certain value in the spatial cell xα conditioned on I(xα), which is composed of all factors that affect species distributions corresponding to their niche (Soberón and Peterson, 2005).
To quantify the relationship between the species and habitats, first we are interested in the probability P(Bi (species)| I′ (habitat)) = NBiAND I′/NI′, where NBiAND I′ is the number of spatial cells where there is a co-occurrence of the taxon Bi and habitat I′, and NI′ is the number of cells where the habitat variables take their stated values. The habitat I′(xα) associated with a spatial cell xα then determines the probability of the distribution variable, Bi(xα), in that cell, and one now has a predictive model. The associated statistic we used to determine the potential relationship between species and habitat is a common one in data mining applications and is a signal-to-noise measure given explicitly by
which measures the statistical dependence of Bi on Ik relative to the null hypothesis that the distribution of Bi is independent of Ik and randomly distributed over the grid, that is, , where is the number of grid cells with point collections of species Bi and N is the total number of cells in the grid. The sampling distribution of the null hypothesis is a binomial distribution where, in this case, every cell is given a probability P(Bi) of having a point collection of Bi. The numerator of equation (1) then is the difference between the actual number of co-occurrences of Bi and Ik relative to the expected number if the distribution of point collections was obtained from a binomial with sampling probability P(Bi). As we are talking about a stochastic sampling, the numerator must be measured in appropriate ‘units’. As the underlying null hypothesis is that of a binomial distribution, it is natural to measure the numerator in standard deviations of this distribution, and that forms the denominator of equation (1).
The quantitative values of ε(Bi |Ik) can be interpreted in the standard sense of hypothesis testing by considering the associated P-value as the probability that |ε(Bi |Ik)| is at least as large as the observed one and then comparing this P-value with a required significance level. In the case where then a normal approximation for the binomial distribution should be adequate, in which case ε(Bi|Bk) = 2 would represent the standard 95% confidence interval. When a normal approximation is not accurate, other approximations to the cumulative probability distribution of the binomial must be used. Note that such a statistical association does not necessarily prove that there is a direct ‘causal’ interaction between taxa and habitats. Rather, it allows for a statistical inference to be made or a hypothesis to be formulated that may be validated subsequently.
Constructing predictive models
Probabilities P (Bi |I′), where I′ is of high dimension, can be constructed using different classification models, such as neural networks or discriminate analysis. A particularly transparent, simple and effective approximation is the naive Bayes approximation (Hand et al., 2001)
where, in the first equality, Bayes rule has been used and, in the second, it has been assumed that the niche variables Ik are independent. The product here is over the N niche variables under consideration as conditioning factors for Bi. In the case of the relationship between species (vectors and mammals) and habitats, N represents the different types of habitat. A score function that can be used as a proxy for P(Bi |I′) is
where is the complement of the set Bi. For example, if Bi is the set of cells with the presence of taxon Bi, then represents the set of cells without presence. S(Bi | I′) is a measure of the probability to find the distribution variable Bi when the niche profile is I′. It can be applied to a spatial cell xα by determining the niche profile of the cell, I′(xα). The score function allows us to build a distribution model for Lutzomyias sp. using mammal point collection records only and another one that combines both mammal point collections and land cover layers. The intuition is that these latter, more complete models would lead to more informative maps that show the relative importance of the different variables that can then be used to better understand and explain the distribution of any given species.
As our data mining method allows us to integrate different types of geographical information, for instance both point collections and environmental layers, we can use it to determine the statistical associations between vector and reservoir species, as represented by point collections, and land cover, as represented by environmental layers. We will first show results for those potential reservoirs previously identified (Stephens et al., 2009) by their significant co-association with Lutzomyias. After that, we will consider results for those mammals that presented a significant association with those habitats in which species of Lutzomyia are associated, but which have not been previously identified as a risk through their direct co-association with Lutzomyias. In other words, we will make an indirect association by identifying mammals that might present risk by being associated with habitats that, in their turn, are associated with sandflies.
As a measure of statistical association, we consider the statistic ε(Bi |Ik), defined above, where Bi represents the ith species and Ik the jth habitat occupied. For the 145 mammals identified in the study of Stephens et al. (2009) as having the most statistically significant co-occurrence with sandflies, there are 1092 habitat–species pairs and hence different statistical associations represented by ε(Bi |Ik). In Fig. 1, we show a network that consists of the 331 (25%) most important positive associations (highest values of ε) between land cover and species of mammals and sandflies. The different habitats are marked as green nodes, while the sandflies are marked as red; known reservoirs are yellow, and potential reservoirs are marked as blue. We identified eight species groupings according to their connectivity across the ecological network.
The connectivity of the network shows which species are associated with which habitats. For instance, in grouping D in Fig. 1, the topmost mammal species is Scirius aureogaster (squirrel), which, we can see, has significant co-association with the habitats – crops, human settlement, temperate forest and tropical forest. These co-associations allow us formulate hypotheses about the way in which the disease could be transmitted from one geographical region to another. Additionally, the network allows us to identify those species that represent a higher risk of disease transmission as well as those that could promote disease dispersion owing to their ability to occupy both natural and transformed habitats.
Assemblages can also be deduced by clustering via the similarity in their connectivity and number of links with different types of habitat, for example species linked to natural habitats only or species linked to natural habitats and crop fields. Essentially, we can identify three main risk scenarios for leishmaniasis transmission associated with the different species groupings. Below is a brief description of the identified assemblages and their potential role in disease transmission; the full list of species is given in Table 1.
Table 1. List of the 145 mammals reported previously by Stephens et al., 2009 as having significant co-association with Lutzomyia. The list includes the code used in the ecological network, species common name, assemblage code in the network and the habitats to which the species are linked
Scenario 1: High risk of leishmaniasis transmission to humans
The mammals in assemblages A and B potentially play an important role in disease transmission to humans, because they can be present both in natural habitats (tropical forest and wetlands) and in human settlements. However, these species do not favour disease dispersion as they are restricted to human settlements near tropical and wetlands habitats. Sandflies with similar connectivity to these mammal assemblages are Lutzomyia olmeca olmeca, Lutzomyia cruciata and Lutzomyia shannoni.
Scenario 2: High risk of leishmaniasis transmission to humans and dispersion of the disease
Assemblages C, D and E contain mammals that inhabit both natural habitats and transformed areas (crop fields and human settlements), without being restricted to tropical habitats. We can then make the hypothesis that transformed habitats, such as crop fields, might be functioning as dispersal corridors for vectors and reservoirs, thereby permitting the dispersion of the disease out of tropical regions. Sandflies with similar connectivity to these mammal groupings are Lutzomyia longipalpis and Lutzomyia diabolica.
Scenario 3: Low risk of leishmaniasis transmission to humans:
The last three assemblages (F, G and H) are the ones considered to be of lowest risk of disease transmission to humans. As these species are more restricted to natural habitats, transmission risk only appears when people work in such natural areas and are therefore in direct contact with these species. Additionally, these assemblages contain the highest biodiversity of mammals (about 80 species), and as mentioned in other studies, higher diversity could buffer the transmission of the disease (LoGiudice et al., 2003; Ostfeld et al., 2006). In this scenario, we find the largest number of potential vectors with six associated species (L. anthopora, L. panamensis, L. ovallesi, L. gomezi, L. ylephielor and L. evansi).
To view the geographical distribution associated with the distinct risk scenarios identified in the ecological network, we constructed a potential distribution map using the score function S(Bi|I′), where Bi is associated with the ith vector and I′ represents the presence of mammal assemblages. Thus, Scenario 1 (assemblages A and B) represents potentially high risk in the south-east and Pacific slope of Mexico (Fig. 2). On the other hand, Scenario 2 (assemblages C, D and E) showed a much different potential distribution, with high-risk areas located in the central and northern regions of Mexico. This scenario would favour dispersion of the disease (Fig. 2).
The second ecological network (Fig. 3) is based on first looking at co-occurrences between all mammals and those habitats identified in the first network as being associated with the presence of sandflies. As the latter are associated with the habitats – tropical forests, temperate forests, wetlands, grasslands, crop fields and human settlement – this corresponds to 2730 links. We then identified the 25% of links with highest values of ε(Bi |Ik). As some of these links are the same as those identified in Fig. 1, we removed them from consideration. The resulting network contains 136 mammals (Table 2) that have significant associations with the same habitats as Lutzomyias, but without having a significant direct co-association with them according to collection records. We then performed a cluster analysis on the network, identifying similar and new assemblages of mammals by their connectivity and number of links with different types of habitats. The species composition of this new list, seen in Fig. 4, contains principally species of bats (25%), mice (24%) and squirrels (10%). However, the majority of these mammals are found in assemblages considered to be of low risk (F′, G′, H′ and J) as they are most associated with natural habitats. For the assemblages that represent a high risk, only assemblages D′ and E′ showed an increase in the number of species, and most of these have a northern distribution, where the disease has not been registered up to now (e.g. the Baja California Peninsula).
Table 2. List of the 136 mammals with significant co-association with habitats where Lutzomyia spp. are present. The list includes the code used in the ecological network, species common name, assemblage code in the network and the habitats to which the species are linked
The new assemblages, as identified by their network connectivity, were the groups I, J and K. Assemblage J has low risk, with species mainly associated with grassland and temperate forests, while groups K and I would represent a higher risk, because these species can be distributed near to human settlements and crops. Yet, as we mentioned above, most of these species have a northern range; hence, their potential role in leishmaniasis transmission needs to be considered with caution.
The next goal was to determine which species types present higher risk as potential reservoirs. To do this, the first step was to associate each species’ taxonomic name with an animal type. For example, Sigmodon hispidus is a rat. The results of this matching are given in Table 1. The reason we do this is that we believe that animal type is more informative and relevant than the taxonomic name, being a better proxy for the phenotypic characteristics of the animal that are relevant for its interactions with the vectors and the habitat. Thus, our analysis of the ecological network has led us to the discovery of general patterns. The most general insight is the recognition of a similar species composition in all the groupings. These were characterized by a great number of bats, followed by mice, opossum, rats and squirrels (Fig. 5). These mammals can be considered to present high risk, and consequently, we recommend conducting surveys for these specific species.
Further, we can identify specific species with any given assemblage. For instance, in the assemblages A, D and F, we can highlight the presence of squirrels as species that could be particularly relevant for leishmaniasis. This is a significant result, as these mammals are closely associated with human settlements and would represent a high risk of leishmaniasis transmission. In the case of assemblages C and D, we observed the presence of opossums. These mammals have been considered to be important in public health, owing to the fact that they are reservoirs for many diseases and also have a broad distribution throughout Mexico. Therefore, they can also be disease dispersers.
Predictive models for disease risk
An important goal is to construct a predictive model for disease risk in any given geographical region. Here, we take as risk measure the presence probability for potential vectors, while the prediction itself is based on a model that includes the presence of both potential reservoirs and land cover. This model was constructed using the mapped score function S(Bi | I′), where Bi is associated with the ith vector and I′ represents the presence of mammals and land cover.
In Fig. 6, we see the results of a predictive model that includes both mammal species and land cover. This result shows that it is possible to integrate biotic variables of different types into the construction of risk maps for leishmaniasis. For instance, this map shows us areas of high risk in the Pacific slope where no records of vectors currently exist. Our results highlight the need to conduct studies in these zones (e.g. the Pacific slope in Fig. 4). Finally, it is important to emphasize that vegetation type might be forming corridors for disease dispersion by encouraging the presence of vectors and reservoirs that may inhabit areas transformed into crop fields.
Although leishmaniasis represents an important public health problem in Mexico, there is a notorious lack of rigorous studies about it. Therefore, there is little information about which vectors are involved in the transmission of the parasite in different geographical regions and also which mammal species are the most important reservoirs. Although there are only eight mammal species that have been found to be infected with Leishmania parasites (Canto-Lara et al., 1999; Van Wynsberghe et al., 2000), it is known that this is still a very small number when compared to the total number of potential reservoirs (Ready, 2008; Stephens et al., 2009). In this study, we expanded the previous analysis of Stephens et al. (2009), first by taking their previous list of the 145 most important potential reservoirs and identifying their relationships with land cover to create an associated ecological network. In this way, we could determine both geographical risk patterns and the particular species or animal type that represents a higher risk of leishmaniasis transmission.
One of the major short-term aims in epidemiology is to conduct specific studies on potential agents involved in zoonotic diseases. The construction of an inferential ecological network allowed us to identify eight species assemblages that clustered into three main risk scenarios for the transmission and dispersal of leishmaniasis. On the one hand, we identified species with a high risk of maintaining and dispersing leishmaniasis according to their connectivity with natural and modified habitats (Scenarios 1 and 2). On the other hand, we found species that were more restricted to natural habitats and therefore represented lower risk of leishmaniasis transmission and dispersal. In this case, the risk only appears when people work in these natural areas (Scenario 3). These types of results can help us select specific species for conducting more in-depth future studies. Our present analysis suggests that priority be given to the species of Scenarios 1 and 2. For instance, bats and squirrels have not been considered previously as possible reservoirs in Mexico, although bats have been found with the parasite in South America (Lampo et al., 2000; De Lima et al., 2008).
A relevant result was the recognition of a similar structure of species composition in all the different assemblages in the network. This biotic structure was characterized by a higher proportion of bats in all cases; therefore, these mammals should be considered as target species in future surveys. Additionally, mice, opossums and rats are also important components in the species’ assemblages. Some species of these types of animals have been previously reported as carriers of the parasite (Davies et al., 2000; Calvopina et al., 2004; Gramiccia and Gradoni, 2005).
An unexpected result was the identification of squirrels in several of the assemblages. These have not been considered previously as potential reservoirs, but could represent a high risk of transmission in human settlements. Co-distributions of these species’ assemblages and the Lutzomyia potentially reflect the strong biotic relation that exists between reservoirs and vectors. Therefore, we would consider that this species composition in biotic communities could favour the presence of vectors in any associated geographical region. Likewise, with mammals, we may determine those vectors that represent a greater dispersal risk given their pattern of connectivity across different habitats, the relevant species being L. olmeca olmeca, L. cruciata, L. shannoni, L. longipalpis and L. diabolica. We would expect that these species have a high probability of being associated with human settlements. Significantly, L. olmeca olmeca, L. cruciata and L. shannoni have been found to be infected with parasites (Biagi et al., 1967; Rebollar-Téllez et al., 1996; Sánchez-García et al., 2010).
Another result that should be highlighted is that in Scenarios 1 and 2, we could identify consistent geographical patterns in the ecological network. For instance, Scenario 1 (assemblages A and B) represents high risk to humans near to tropical forests and wetlands, principally in the south-east and Pacific slope of Mexico (Fig. 2). Scenario 2 (assemblages C, D and E), on the other hand, has a broader potential distribution than Scenario 1. The corresponding distribution includes not only tropical regions but also significant areas of risk in the central and northern regions of Mexico. This situation indicates a risk that the disease can move out of tropical regions (Fig. 2).
Our second approach to understanding the relationship between potential vectors and reservoirs for leishmaniasis was by considering co-association between mammals and habitats where potential vectors are located. In this case, the analysis was not restricted to areas where there was a significant co-association between sandflies and mammals through their collection points. The logic here is that there may exist the potential for vector–reservoir interactions that are not reflected in the current point distributions. This second ecological network allowed us to identify 136 new mammals that occupy habitats where sandflies are found. Interestingly, most of these species have a northern distribution where information about leishmaniasis is limited. However, it is worth pointing out that leishmaniasis has been recorded in northern states of Mexico, such as Coahuila (Díaz, 1971), Sinaloa (Ochoa-Díaz et al., 2010) and Durango (Pérez-Vega et al., 2009), but it is not known what potential species could be involved. We hope that this new list could be a first step to filling in information gaps that exist in this region. For this new list, we can also highlight those genus that have previously been identified as reservoirs, both in Mexico and in other countries, such as Neotomas, Peromyscus, Sigmodon, Sylvilagus, Conepatus and Sciurus (Davies et al., 2000; Van Wynsberghe et al., 2004; González et al., 2010; WHO, 2010).
There is no doubt that habitat co-association can allow us to better understand the role of environment in zoonotic diseases (Alonso et al., 2008). Habitat destruction is the primary cause of change in both diversity and species composition. However, some species appear relatively insensitive to habitat fragmentation and could play a specially important role for any given zoonoses (Poteet, 2006; Reithinger et al., 2007; Sánchez-García et al., 2010). Previous research, in fact, has highlighted that change in natural habitats can modify the dynamics of disease (Poteet, 2006; González et al., 2010; Sánchez-García et al., 2010) and is therefore an important risk factor (Desjeux, 2001; Ostfeld et al., 2006). In the present case, species with a significant association with modified habitats (e.g. mice and bats) have been implicated in the presence of leishmaniasis and an increased risk of human exposure to the disease. Additionally, habitat modification can be greatly impacted by climate change, with consequences that could benefit disease dispersal. Several papers have assessed the impacts of climate change for current and future distributions of different diseases (Peterson and Shaw, 2003; Patz et al., 2005; McMichael et al., 2006). Although in this work we do not consider climate change, it is an important element to return to in the future.
One of the main short-term goals in the public health sector is to create risk maps for zoonotic diseases. Preferably, these maps should also give us information about the specific factors that favour the presence of disease in any given geographical region. In this study, we showed how these risk maps can be obtained by combining the presence of mammals and land cover to predict the potential distribution of leishmaniasis. Our methodology also allowed us to identify which factors are more important for this disease. As a result, we observed that transformed habitats, such as crop fields, might be functioning as corridors for the dispersal of both vectors and reservoirs and could permit the migration of the disease from one geographical region to another. Additionally, we found that bats, opossums, mice, squirrels and rats are important potential wild reservoirs and that their relation with land cover can help us to focus the attention of future epidemiological studies on certain areas.
We believe that our results are an important step forward in the application of ecological networks to the study of emerging disease. Undoubtedly, our understanding of the dynamics of zoonotic disease transmission can be greatly enhanced by determining the specific roles of the different agents involved – reservoirs, vectors and landscape. In this article, we showed how the construction of biotic interaction networks can be used, not only to better understand potential ecological interactions between species but also to help identify the role of environment (land cover) in disease transmission systems.
We are grateful to Conacyt grant number 80156 for financial support and also to the Conacyt network Ciencia, Complejidad y Sociedad. We are grateful to Camila Gonzalez and Eduardo Rebollar for help with the Lutzomyia data.