Phylogeographic Analysis of Tick-borne Pathogen - Recent Advances and Issues in Environmental Science - page 79

World Ecoregions and Association with A. marginale Strains
Ecoregions are used herein to classify the world across dynamic environmental fac-
tors. We assumed that (i) ecoregions could be delineated using quantitative abiotic
characters based on well-recognized and repeatable attributes and (ii) A. marginale
strains are associated with each ecoregion and subjected to different environmental
conditions that could be analyzed by multivariate geographic clustering [22]. Mul-
tivariate geographic clustering involves the use of standardized values for selected
environmental conditions in a set of raster maps. Those values serve as coordinates in
the environmental data space, in which environmental conditions are further clustered
according to their similarities. The feature selected to put together the clusters was the
monthly Normalized Difference Vegetation Index (NDVI). The NDVI is a variable
that reflects vegetation stress, a feature that summarizes information about the eco-
logical background for tick populations [23]. We obtained a 0.1° resolution series of
monthly NDVI data for the period 1986-2006. The 12 averaged monthly images were
subjected to Principal Components Analysis (PCA) to obtain decomposition into the
main axes representing the most significant, non-redundant information. The strongest
principal axes were chosen using Cattell's Scree Test [22]. It has been found that the
first principal component derived from NDVI typically represents the greenness of
the surveyed area [24]. Component 2 is interpreted as a change component, taken to
represent a winter/summer seasonality effect. Components 3 and 4 are also essentially
seasonal, but represent areas where the timing of green-up is different from that in
component 2. Our PCA analysis retained three principal axes, explaining the 92% of
total variance. These three axes were related to the mean NDVI values, annual am-
plitude, and NDVI values in the period May to August, respectively. We then used a
hierarchical agglomerative clustering on PCA values to classify multiple geographical
areas into a single common set of discrete regions. Mahalanobis distance was used as
a measure of dissimilarity and the weighted pair-group average was used as the amal-
gamation method. A value of 0.05 was used as the cut-off probability for assignment
to a given ecoregion. All the procedures adhered to methods previously described [25].
The decision about the number of ecoregions to retain without any prior detail
about the information they contain is a problem to which a solution has not yet been
found. The main goal is to defi ne unambiguously the A. marginale strains recorded
mostly in a single ecoregion cluster and present in the highest number of geographical
sites belonging to that cluster. The result is to refi ne the degree of clustering that gives
the optimal degree of association between A. marginale strains (“species”) and ecore-
gions (“sites”). This analysis was done using the “indicator species” method [26], a
previously published multivariate statistics procedure to defi ne “sites” as a function of
their faunal composition (“species”). We began the agglomerative process described
above with an unrealistic high number of ecoregions. At every step of the agglom-
erative process, pathogen strains and ecoregions were ordered by a correspondence
analysis, and then analyzed using the “indicator species” method. The procedure runs
iteratively, trying to improve the association with further clustering of ecoregions.
However, the method does not force a cluster if specifi cations of cut-off probabilities
for ecoregions are violated, and does not assume any a priori condition about the
geographical range of any cluster. The procedure is only an indicator that stops when