Testing the role of climate in the evolution of tickborne flaviviruses

If climate exerts such a strong influence on the successful transmission of tick-borne viruses, it is reasonable to hypothesize that climate has directed and constrained their evolution. One way to investigate whether climatic factors direct the evolution, rather than merely the distribution, of vector-borne flaviviruses is to test whether the eco-climatic spaces occupied by closely related viruses are more or less similar than those occupied by more distantly related ones. If they prove to be very similar one could conclude that climate has been a significant evolutionary constraint, whereas if they are significantly different one could conclude that viruses have been free to leap from one eco-space to another with no climatic constraint.

The idea is to test for matches between virus phylogeny and the environmental conditions in which each virus circulates by seeking correlations between two very different sorts of trees. First are the familiar molecular phylogenetic trees, constructed on the basis of genetic differences as a direct measure of evolutionary distance between related species. These describe evolutionary history. Second are conceptually novel eco-climatic trees, constructed from the statistical distances between the eco-climatic space in which each virus circulates. These trees are purely descriptive, containing no information on evolutionary history.

Constructing phenetic eco-climatic trees for viruses. The basis for these eco-climatic trees is the ability to define the conditions within which each virus can survive (i.e. exists), plotted in multivariate space (illustrated in Fig. 2 for just two variables, temperature and humidity). As made clear above, each virus only survives within a subset of the overlapping range of its hosts and vectors. The complex biological processes that determine the eco-climatic limits of vector-borne pathogens have not yet been completely quantified for any one let alone a whole cladeful. Therefore, statistical pattern-matching methods are currently the most reliable for identifying these limits (Randolph, 2000; Rogers & Randolph, 2000). Correlations are established between the spatial patterns of the distribution of organisms and spatial patterns of environmental conditions as revealed by meteorological satellite imagery (from the NOAA AVHRR sensor) (Hay et al., 2000). Of the variety of statistical methods available for seeking these correlations, discriminant analysis is the most biologically transparent (Green, 1978; Rogers et al., 1996). Furthermore, it has the particular virtue for the present purposes of measuring distance: the Mahalanobis distance is the co-variance-adjusted distance between the geometric centres of multivariate eco-climatic spaces (Fig. 2). It is a way of reducing the many differences in eco-climatic conditions to a single measure of separation between presence and absence, or between the presence of pairs of species. Using forward stepwise selection of up to ten variables, discriminant analysis also assigns relative importance to the variables that define the species distribution, thereby yielding both discrete and distance information.

The distributions of single diseases can now be captured routinely with high accuracy by these statistical methods (reviewed by Hay et al, 2000; Rogers et al., 2002). To create phenetic eco-climatic trees, Mahalanobis distances between species must be derived from the same set of variables. In theory, this could be achieved by putting all the species into a single analysis, but as each disease is added prediction accuracies are inevitably compromised. Nevertheless, to date it has proved possible to capture the distributions of all three most westerly tick-borne flaviviruses (LI, SSE and WTBE) in a single exercise with 80-95% accuracy [kappa index (Congalton, 1991) = 0.794]. When preliminary data for the Siberian subtype of TBE virus (TBEVs) were added [based on 500 locations within the western distribution of the tick I. persulcatus in Latvia, Estonia and western Russia, and from the Perm region of Siberia, with which TBEVs virus effectively, although not absolutely exclusively, coincides (Korenberg, 1994; Lundkvist et al., 2001)], the kappa index was still high, at 0.683. This means that the same ten predictor satellite variables may be used to distinguish areas of presence of each virus from each of the others, and from areas of absence. The identity of the ten predictor variables, however, changes somewhat as new species are added. As more species are added it will be necessary to select a consensus 'top-ten' set of variables that applies across the clade.

Statistically each of the four viruses analysed to date falls within a distinct eco-climatic space defined by factors that are all temporal Fourier variables (i.e. seasonal characteristics) of thermal and moisture conditions. To illustrate this on the page, two of the most significant variables are chosen as axes (Fig. 3). In addition, five points representing the observed locations for GGE and TSE are plotted on the same axes. Even in this simplified bi-variate space, there is clear separation between viral types. Furthermore, the evolutionary sequence from TBEVs to WTBE to GGE/TSE to SSE and finally to LI follows a progressive shift from high to low annual variance in air temperature, with associated changes in annual variance in vapour pressure deficit. Nothing can be deduced from the relative degrees of separation on this graph, partly because of the different numbers of observations and mostly because separation in multivariate space will be differentially increased by other variables. LI and WTBE appear to be close together on this graph, but just as the molecular phylogeny shows that LI virus did

Species 2 present

Species 2 present

absent

Species 1 present

/^ ยก"" Discriminant axis > / / \ J between presence of sp 1 and of sp 2

absent

Species 1 present

Variable 1 (e.g. temperature)

Fig. 2. An illustration of the principle of linear discriminant analysis used to distinguish between the abiotic conditions in which an organism is present or absent, and the conditions occupied by two different species. Mahalanobis distances are represented by the heavy double-ended arrows.

not evolve directly from WTBE virus, so the Fourier-processed satellite imagery of temperature and moisture conditions shows a barrier of seasonally distinct climate in France. Biologically relevant climate factors now explain why WTBE has not reached Britain through France (Randolph et al., 2000), but instead the distinct LI virus entered Britain apparently via Spain and Ireland (Gould et al., 2001; McGuire et al., 1998). These two viruses, however, are now approaching each other geographically via a northern route, as WTBE is spreading westwards through Scandinavia (International Scientific Workgroup on Tick-borne Encephalitis, 2002 - http://www.tbe-info.com/ epidemiology/index.html; Skarpass et al., 2002) and LI has been recorded in Norway.

Congruence between phylogenetic and eco-climatic trees? The above results are the first, very important steps towards deriving Mahalanobis distances between all the tick-borne flaviviruses, and so constructing phenetic ecological trees. In many ways, a clade that forms a cline is the least suitable for testing the effects of climate on evolution because of the co-variation between evolutionary change, geographic space and climate. Climate may match phylogeny simply because of the orderly march westwards in this case. The smaller scale of LI virus evolution within the British Isles on the one hand, and the more diffuse geographic patterns of the insect-borne flavi-viruses on the other, should provide more clearly interpretable results. Both of these are currently being studied by the Oxford Tick Research Group in Oxford. Whatever degree of congruence between the two types of tree emerges, it will direct the search for

40 0

Fig. 3. Each virus of the tick-borne encephalitis complex occupies a distinct 'eco-climatic' space, illustrated here in bi-variate space defined by two of the most significant climatic variables that predict the distribution of each virus. Annual variance is an indicator of the abruptness of the seasonal changes in these climatic variables, especially during spring and autumn. The phylogeny of these viruses, their geographic distribution and principal hosts are shown below.

Fig. 3. Each virus of the tick-borne encephalitis complex occupies a distinct 'eco-climatic' space, illustrated here in bi-variate space defined by two of the most significant climatic variables that predict the distribution of each virus. Annual variance is an indicator of the abruptness of the seasonal changes in these climatic variables, especially during spring and autumn. The phylogeny of these viruses, their geographic distribution and principal hosts are shown below.

the processes underlying the evolutionary patterns, leading to new interpretations and understanding of the forces that direct and constrain pathogen evolution. If the trees do not match, we shall have to conclude that forces other than climate must have been more significant in directing the evolution of the flaviviruses. If, on the other hand, the fit is better than random, it will already be clear which particular environmental factors are correlated with the origin of new viruses, and with the shifting incidence of existing vector-borne diseases under the forces of natural and anthropogenic environmental change.

0 0

Post a comment