The role of human vs. natural dispersal

Throughout history, humans moving animals has repeatedly led to the emergence and spread of rabies in susceptible populations, either through deliberate translocation (e.g. the relocation of animals, some of which may be incubating virus (Nettles et al. Reference Nettles, Shaddock, Sikes and Reyes1979)) or inadvertent movement (e.g. raccoons on garbage trucks (Wilson et al. Reference Wilson, Bretsky, Cooper, Egbertson, VanKruiningen and Cartter1997)).

Although human-mediated long-distance movement was found to be a significant feature of raccoon rabies spread (Smith et al. Reference Smith, Lucey, Waller, Childs and Real2002), it was still rare compared to natural dispersal, and stochastic in its occurrence. In contrast, human-mediated dispersal of dog rabies appears more widespread and potentially predictable due to possible links with human activities and migration patterns. The European colonisation of Africa is thought to have led to the introduction and subsequent expansion of canine rabies across the continent (Bourhy et al. Reference Bourhy, Reynes, Dunham, Dacheux, Larrous, Huong, Xu, Yan, Miranda and Holmes2008; Talbi et al. Reference Talbi, Holmes, de Benedictis, Faye, Nakouné, Gamatié, Diarra, Elmamy, Sow, Adjogoua, Sangare, Dundon, Capua, Sall and Bourhy2009; Lemey et al. Reference Lemey, Rambaut, Welch and Suchard2010); increasingly frequent records of re-introductions into countries where rabies has been eliminated have been recorded (Zanoni and Breitenmoser, Reference Zanoni and Breitenmoser2003; Weiss et al. Reference Weiss, Hoffmann, Freuling, Müller, Fesseler and Renner2009; Gautret et al. Reference Gautret, Ribadeau-Dumas, Parola, Brouqui and Bourhy2011;), and canine rabies has recently emerged on several previously rabies-free islands in Indonesia via fishermen importing incubating dogs (Windiyaningsih et al. Reference Windiyaningsih, Wilde, Meslin, Suroso and Widarson2004; Susilawathi et al. Reference Susilawathi, Darwinata, Dwija, Budayanti, Wirasandhi, Subrata, Susialarini, Sudewi, Wignall and Mahardika2012). Even on smaller scales, phylogeographic patterns frequently imply translocation events that increase the level of mixing between lineages and may obscure patterns of lineage clustering (David et al. Reference David, Yakobson, Gershkovich and Gayer2004; Talbi et al. Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010; Hayman et al. Reference Hayman, Johnson, Horton, Hedge, Wakeley, Banyard, Zhang, Alhassan and Fooks2011).

Although often inferred, it is difficult to quantify the influence of human-mediated movements on the diffusion of RABV. Though translocations may be frequent, the likelihood of establishment and spread is much lower (Smith et al. Reference Smith, Waller, Russell, Childs and Real2005). But, unlike for wild animals, where translocation is risky and often unsuccessful, human-mediated displacement from a dog's original home range confers fewer risks. Dog translocations are therefore likely to be more successful in initiating new disease foci, and with potentially higher consequences, introducing disease into previously disease-free areas. Given that long-distance movements instigated by humans are often in response to social drivers, the genetic structure of dog rabies is expected to reflect variation in cultures and socioeconomic factors. For example, rural workers in Thailand often relocate to urban areas where work is more readily available in the off-growing season, taking their dogs, and sometimes rabies, with them (Denduangboripant et al. Reference Denduangboripant, Wacharapluesadee, Lumlertdacha, Ruankaew, Hoonsuwan, Puanghat and Hemachudha2005). Similarly, the resurgence of canine rabies in KwaZulu-Natal in the 1970s was linked to refugee movements from Mozambique (Swanepoel et al. Reference Swanepoel, Barnard, Meredith, Bishop, Bruckner, Foggin and Hübschle1993; Cleaveland, Reference Cleaveland1998) and phylogenetic patterns of RABV in parts of northern Tanzania can be tentatively explained by the movement of nomadic Maasai pastoralists with their dogs (Lembo et al. Reference Lembo, Haydon, Velasco-Villa, Rupprecht, Packer, Brandao, Kuzmin, Fooks, Barrat and Cleaveland2007). Counterproductively, dog-owners have been reported to move their dogs to avoid the threat of culling during attempts to control rabies, which may explain more rapid spread of rabies than would be predicted by dog movement alone. Importantly, many of these aspects of human geography are quantifiable, making it possible in principle to generate testable hypotheses about their role in determining rabies phylogeographic structure.

Talbi et al. (Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010) used spatial simulations to show that the observed patterns of spread in North Africa could only be explained by the inclusion of long-distance translocation events as opposed to natural dog movements alone. The wave-like patterns that are a signature of wildlife rabies and determined by natural host movements are less evident in domestic dog rabies, and it is unknown to what extent natural dog movements vs. human-mediated movements determine the observed phylogeographic patterns. A recent study in a Kenyan rangeland found that healthy domestic dogs rarely moved more than 50 metres from their home bases, and the maximum distance recorded was 3·2 km (Woodroffe and Donnelly, Reference Woodroffe and Donnelly2011). In contrast, dispersal distances in wildlife hosts tend to be at least an order of magnitude higher (Cullingham et al. Reference Cullingham, Pond, Kyle, Rees, Rosatte and White2008). However, rabies infection leads to behavioural changes, which may alter movement patterns; in northern Tanzania, most contacts with rabid dogs occurred within a kilometre of an animal's homestead (mean ∼0·88 km), but a small proportion of rabid dogs ran over 15 km while infectious (Hampson et al. Reference Hampson, Dushoff, Cleaveland, Haydon, Kaare, Packer and Dobson2009). Dog movements also vary depending on the societal context and livelihoods of the communities to which they belong (Woodroffe and Donnelly, Reference Woodroffe and Donnelly2011). For their dispersal simulations, Talbi et al. (Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010) used measures of natural dog movements based on a specific locality in Tanzania (Hampson et al. Reference Hampson, Dushoff, Cleaveland, Haydon, Kaare, Packer and Dobson2009) but it remains unclear whether such data are transferrable among different geographic localities. Research on domestic dog ecology and movement is lacking and detailed, location-specific data are required in order to reliably test hypotheses for alternative drivers of viral diffusion.

Fig. 1. Varying spatial complexity in areas with endemic dog rabies as a result of increasing dog population density, A) low density: Ngorongoro District, Tanzania; B) medium density: Serengeti District, Tanzania, C) high density: Hermosillo, Mexico. Red circles highlight settlements in rural areas. Maps obtained using Google Earth (http://earth.google.com).

Landscape attributes influencing rabies spread

As a parasite transmitted through direct host-to-host contact, spatially-defined genetic discontinuities in RABV populations may indicate a barrier to host contact (Biek and Real, Reference Biek and Real2010). We consider a ‘dispersal barrier’ to be any spatial feature of the landscape that impedes the gene flow of a pathogen. Spatially explicit models incorporating landscape features can help uncover barriers to gene flow based on slower progression than expected over a homogeneous area. However, the interpretation of genetic discontinuities requires caution as phylogeographic patterns may arise as a result of historical colonisation events and irrespective of physical barriers (Real and Biek, Reference Real and Biek2007; Talbi et al. Reference Talbi, Holmes, de Benedictis, Faye, Nakouné, Gamatié, Diarra, Elmamy, Sow, Adjogoua, Sangare, Dundon, Capua, Sall and Bourhy2009).

Smith et al. (Reference Smith, Lucey, Waller, Childs and Real2002) demonstrated a seven-fold reduction in rates of raccoon rabies spread in North America due to the presence of rivers and forest cover. Similarly, the Vistula River separated distinct clusters of a red fox variant and raccoon dog/fox variant in Europe (Bourhy et al. Reference Bourhy, Kissi, Audry, Smreczak, Sadkowska-Todys, Kulonen, Tordo, Zmudzinski and Holmes1999). Clearly, large water bodies impede host movement, but have differential success as barriers according to additional factors such as habitat suitability (Smith et al. Reference Smith, Waller, Russell, Childs and Real2005; Cullingham et al. Reference Cullingham, Kyle, Pond, Rees and White2009), physical geography (Rees et al. Reference Rees, Pond, Cullingham, Tinline, Ball, Kyle and White2009), and width and flow rates (Bourhy et al. Reference Bourhy, Kissi, Audry, Smreczak, Sadkowska-Todys, Kulonen, Tordo, Zmudzinski and Holmes1999). In an attempt to assess the differential permeability of barriers, Rees et al. (Reference Rees, Pond, Cullingham, Tinline, Ball, Kyle and White2008) used genetic simulation modelling to assess the effect of the Niagara River on rabies spread from New York State to Ontario. Comparing genetic population structure derived from field data, with simulated population expansion scenarios they ascertained a 50% barrier effect. In contrast, homogeneous landscapes that lack environmental barriers are particularly vulnerable to the rapid expansion of an introduced pathogen, as demonstrated by the increased speed with which raccoon rabies is predicted to cross central Ohio, which lacks major natural barriers, compared to neighbouring states (Russell et al. Reference Russell, Smith, Childs and Real2005). With the exception of major landscape features inferred as barriers to dog rabies on a global scale (Bourhy et al. Reference Bourhy, Reynes, Dunham, Dacheux, Larrous, Huong, Xu, Yan, Miranda and Holmes2008), we know relatively little about barrier effects at smaller scales. Determining whether natural barriers prevent the dissemination of dog RABV or if ties to human ecology negate their effect is an important research question with significant implications for control.

Political borders are another potential form of barrier that may play a role in the containment of dog rabies, but have no evident impact on wildlife rabies dispersal. For example, distinct and almost monophyletic RABV groups associated with North African countries indicate restricted movement across geopolitical boundaries (Talbi et al. Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010). This well-defined population structure at a regional scale contrasts with a relatively fluid dissemination within countries. On the one hand, this suggests that country level vaccination programmes should have a good chance of eliminating dog rabies even in contiguous landscapes. However, these findings are not generalisable, with epidemiological analyses of RABV in Ghana highlighting frequent cross-border incursions (Hayman et al. Reference Hayman, Johnson, Horton, Hedge, Wakeley, Banyard, Zhang, Alhassan and Fooks2011). In addition, time-series analyses indicate large-scale synchronous dynamics of rabies across multiple countries in eastern and southern Africa, possibly due to a combination of human or wildlife-mediated long-distance dispersal and a lack of sustained control programmes (Hampson et al. Reference Hampson, Dushoff, Bingham, Bruckner, Ali and Dobson2007). Understanding the circumstances whereby political boundaries act as dispersal barriers should provide guidance for whether control programmes require regional co-operation or can be sustained at a national level with appropriate border controls. As a first step, it would be useful to compare phylogeographic structure of dog RABV among different parts of the world and across hierarchical spatial scales. Objective measures for such a comparison can be obtained by quantifying the degree to which sequences cluster on a phylogeny according to their geographic location. Several statistics are available for this (Parker et al. Reference Parker, Rambaut and Pybus2008), of which the association index (AI) (Wang et al. Reference Wang, Donaldson, Brettle, Bell and Simmonds2001) has been found to be of particular utility. This measure of phylogeny-trait correlation may provide an initial descriptive analysis of consistency or variability in the hierarchical genetic structure of dog rabies between continents or countries that could determine the generalisation of results across different systems. For example, we might hypothesise that regions with large population densities and a high level of transport infrastructure, e.g. parts of Asia, show less distinctive phylogeographic patterns (a low AI value) compared to more sparsely populated and less developed landscapes like those in sub-Saharan Africa (high AI value).

Quantifying barrier effects, to predict the likelihood of incursions or to exploit them for control programmes, is an important area yet to be tackled for dog rabies. The impact of anthropogenic landscape features on spread, at least in a qualitative sense, has been noted in several countries. For example, Tenzin et al. (Reference Sharma, Dhand, Timsina and Ward2010) mapped the spread of rabies in Bhutan showing a strong visual pattern alongside road networks and towns with high dog-to-human ratios; and phylogeographic patterns in north Mexico and Thailand match the distribution of major migration routes (de Mattos et al. Reference de Mattos, de Mattos, Loza-Rubio, Aguilar-Setien, Orciari and Smith1999; Denduangboripant et al. Reference Denduangboripant, Wacharapluesadee, Lumlertdacha, Ruankaew, Hoonsuwan, Puanghat and Hemachudha2005). It can be hypothesised that features important to wildlife rabies become less significant to dog rabies as anthropogenic features take over the landscape and mediate the effect of barriers, i.e. bridges, roads and transportation make it possible for dogs to circumvent natural barriers and support a relatively fluid dissemination of rabies across naturally heterogeneous landscapes. These hypotheses remain largely untested and thus present a fruitful area for landscape genetics investigations.

Uncovering those features that facilitate viral spread is an equally important aspect of assessing the landscape. As previously discussed, anthropogenic effects may facilitate the transmission of rabies across a larger scale than natural movements alone allow. The impact of human-mediated dog transport may by explored by phylogenetic analysis, with translocations implicated by the presence of a cluster-specific variant in a distant locality (see David et al. (Reference David, Yakobson, Gershkovich and Gayer2004) and Cohen et al. (Reference Cohen, Sartorius, Sabeta, Zulu, Paweska, Mogoswane, Sutton, Nel, Swanepoel, Leman, Grobbelaar, Dyason and Blumberg2007), Table 1), essentially an assignment approach (Paetkau et al. Reference Paetkau, Calvert, Stirling and Strobeck1995). While this method is not a definitive measure of a translocation, it can identify the most likely dispersal scenario. Talbi et al. (Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010) used Bayesian phylogeographic diffusion models (Lemey et al. Reference Lemey, Rambaut, Drummond and Suchard2009) in an attempt to quantify the importance of various anthropogenic predictors on the observed spread of rabies cases in North Africa (see Table 1). Model fitting indicated that dispersal patterns among towns were best explained by road distance, consistent with the anticipated role of human movement. Interestingly, further refinements to calculating distance matrices, such as an accessibility index based on road type and travel time, received only limited model support and road distances were only a marginally better predictor than great-circle distances. This may indicate that RABV dispersal follows a rather homogeneous spatial diffusion process, without any particular effect of landscape heterogeneity. More likely however, it means that more useful geographic predictors of human-mediated dispersal have yet to be found. While this was not possible in the Talbi et al. (Reference Talbi, Lemey, Suchard, Abdelatif, Elharrak, Nourlil, Faouzi, Echevarria, Vazquez Moron, Rambaut, Campiz, Tatem, Holmes and Bourhy2010) study, their method can accommodate a wide range of geographical and environmental predictors and thus provides a promising general framework for examining landscape effects in future data sets.

Population level effects and metapopulation dynamics

Infectious disease dynamics are often described in terms of a metapopulation, with host populations divided into smaller, spatially structured sub-populations that exist with different inter- and intra-patch dynamics (Grenfell and Harwood, Reference Grenfell and Harwood1997). Heterogeneities in the spatial configuration of host populations are critical to understanding the persistence of endemic pathogens (Hagenaars et al. Reference Hagenaars, Donnelly and Ferguson2004) and incorporating this social/spatial structure is particularly important for developing mechanistic models of acute-acting infections like rabies (Cross et al. Reference Cross, Lloyd-Smith, Johnson and Getz2005). In the case of dog rabies, human settlements (village/town/city) can be considered habitat patches that vary in host density and connectivity (Fig. 1). For example, rural areas typically exist as an array of villages connected to larger towns and cities by major roads. Fragmentation and low connectivity, i.e. long distances between settlements, natural barriers and lack of transportation networks, likely restrict the ability of a dog to move across a landscape and hence limit rabies spread. Hypothetically, limited contact between dog populations and correspondingly strong spatial structure of rabies would be expected in the least developed regions where travel is most limited.

Local within-patch dynamics are potentially a key component contributing to the persistence and maintenance of disease at larger scales. Therefore, it is important to quantify the relative contribution of different patches to overall disease persistence in a system. This might involve characterizing patches based on dog density, turnover and growth rates, levels of ownership and vaccination, alongside measures of connectivity. For example, dog turnover rates can be extremely high e.g. the Machakos region in Kenya has a dog population estimated to grow by 9% per annum, making rabies control difficult in this area (Kitala et al. Reference Kitala, McDermott, Kyule, Gathuma, Perry and Wandeler2001). Areas such as these may possibly be identified as hotspots or source patches where rabies has a high chance of being maintained and spread from.

A potentially useful tool for quantifying how rabies dynamics differ among local patches is the basic reproductive number, R₀. Theory would suggest that for directly transmitted diseases such as rabies, R₀ should increase with host density. However, empirical evidence for density-dependent dynamics is equivocal, with no detectable differences in R₀ among dog populations with varying densities around the world (Hampson et al. Reference Hampson, Dushoff, Cleaveland, Haydon, Kaare, Packer and Dobson2009). While it is unclear why R₀ is so insensitive to differences in dog density, e.g. Mexico (∼100 dogs/km² (Eng et al. Reference Eng, Fishbein, Talamante, Hall, Chavez, Dobbins, Muro, Bustos, Ricardy, Munguia, Carrasco, Robles and Baer1993)) versus rural Africa (∼10 dogs/km² (Lembo et al. Reference Lembo, Hampson, Haydon, Craft, Dobson, Dushoff, Ernest, Hoare, Kaare, Mlengeya, Mentzel and Cleaveland2008)), this finding is more consistent with a frequency-dependent mode of rabies transmission. For both density- and frequency-dependent disease dynamics stochastic extinctions are expected once density drops below a certain level, meaning that low-density patches may act as barriers to spread. Indeed, rabies has been observed to spread but appears less able to persist in low density dog populations, e.g. Ngorongoro in Tanzania (average of 4·2 dogs/km² (Lembo et al. Reference Lembo, Hampson, Haydon, Craft, Dobson, Dushoff, Ernest, Hoare, Kaare, Mlengeya, Mentzel and Cleaveland2008)), suggesting that neighbouring source populations are required for rabies maintenance. Cross et al. (Reference Cross, Johnson, Lloyd-Smith and Getz2007) discuss the incorporation of heterogeneities in population structure by expanding on the utility of R₀ as a measure of disease transmission. R₀ assumes that the population is evenly mixed, but the hierarchical nature of disease invasion requires stochastic models that incorporate both within-patch transmission (R₀) and the factors contributing to persistence, i.e. between patch transmission – namely the recruitment of susceptibles, group size and the infectious period. These questions of potential metapopulation structure, patch variability and source-sink dynamics (Pulliam, Reference Pulliam1988) represent a particularly pertinent and rich area for future studies of RABV which has yet to be explored using genetic approaches and are vital to inform efficient control strategies.

Host movement rates and connectivity between sub-populations are crucial predictors of disease invasion (Cross et al. Reference Cross, Lloyd-Smith, Johnson and Getz2005; Beyer et al. Reference Beyer, Hampson, Lembo, Cleaveland, Kaare and Haydon2011). Epidemiological data available at a localised scale, e.g. the incidence of dog bites, can provide a simple spatio-temporal measure for metapopulation analysis, as shown by Beyer et al. (Reference Beyer, Hampson, Lembo, Cleaveland, Kaare and Haydon2011). Patch-occupancy models identified two metrics of connectivity, the distance between neighbouring villages and the size of villages receiving infection, as significant factors facilitating the transmission of disease. From a population genetics perspective, coalescent analysis of sequence data could provide the means to identify metapopulation dynamics that may not be discernable using epidemiological data alone. Such techniques have been successfully applied, for example, to understand the persistence of influenza A virus, with molecular clock-based estimates of divergence times of most recent common ancestors (MRCA) and demonstrating persistence due to dynamic migration patterns rather than source-sink dynamics (Russell et al. Reference Russell, Jones, Barr, Cox, Garten, Gregory, Gust, Hampson, Hay, Hurt, de Jong, Kelso, Klimov, Kageyama, Komadina, Lapedes, Lin, Mosterin, Obuchi, Odagiri, Osterhaus, Rimmelzwaan, Shaw, Skepner, Stohr, Tashiro, Fouchier and Smith2008; Bedford et al. Reference Bedford, Cobey, Beerli and Pascual2010).

More generally, the fine-grained spatial genetic structure evident in wildlife rabies that is lacking for dog rabies may be due to differences in time frame. The best-studied wildlife rabies dynamics come from invasions that originated a few decades ago (Bourhy et al. Reference Bourhy, Kissi, Audry, Smreczak, Sadkowska-Todys, Kulonen, Tordo, Zmudzinski and Holmes1999; Real et al. Reference Real, Henderson, Biek, Snaman, Jack, Childs, Stahl, Waller, Tinline and Nadin-Davis2005a; Biek et al. Reference Biek, Henderson, Waller, Rupprecht and Real2007), compared to endemic foci in domestic dogs, thought to have persisted for centuries (Bourhy et al. Reference Bourhy, Reynes, Dunham, Dacheux, Larrous, Huong, Xu, Yan, Miranda and Holmes2008). Factors influencing persistent endemic cycles are likely to differ from those determining epidemic expansions, with viral population turnover, increased mixture of lineages and the greater role of human-mediated movement resulting in less clearly defined fine-grain viral structure. Again, metapopulation models that capture population connectivity at appropriate spatial and temporal scales may shed light on the genetic structure of endemic foci. Characterizing the landscape connectivity between rabies endemic areas (or individual rabies cases, given the resolution of spatial data) may be facilitated using techniques applied in conservation genetics that have yet to be exploited for disease dynamics. For example, the program Fractionnator, (http://www.unil.ch/biomapper/frictionnator/frictionnator.html), provides a ‘strip statistic’ quantifying the effect of landscape features on the relatedness of all possible pairs of individuals sampled. This entails defining grid cells within a strip across the sampled landscape and assessing the abundance of landscape features within each strip. In addition, advances in the development of spatial analysis software such as python-based customised GIS tools (e.g. Etherington, 2001) that allow the visualisation and measurement of genetic relatedness and landscape connectivity based on least cost path analysis, R packages such as Maptools (Lewin-Koh et al. Reference Lewin-Koh, Bivand, Pebesma, Archer, Baddeley, Bibiko, Callahan, Dray, Forrest, Friendly, Giraudoux, Golicher, Rubio, Hausmann, Hufthammer, Jagger, Luque, MacQueen, Niccolai, Short, Snow, Stabler and Turner2012) and alternative open-source software such as PASSaGE 2 (Rosenberg and Anderson, Reference Rosenberg and Anderson2011) provide utilities for measuring connectivity between patches, quantifying patterns in spatial data and potentially identifying barriers to dispersal.

Elucidating how heterogeneity at the host level contributes to dynamics at the patch level is an important aspect of understanding the mechanisms underlying rabies persistence and spread. Differential within-patch dynamics resulting from individual heterogeneity including variation in biting frequency, host population size, the structure of a settlement, may confer properties that promote the effectiveness of a patch as a source of infection. Tracing transmission pathways may provide a further means of elucidating source populations that initiate chains of infection, particularly at very fine-scale resolution. Lembo et al. (Reference Lembo, Haydon, Velasco-Villa, Rupprecht, Packer, Brandao, Kuzmin, Fooks, Barrat and Cleaveland2007) (Table 1) used parsimonious transmission networks to infer reservoir host dynamics and patterns of interspecific transmission, but this technique could also be used more generally to uncover dispersal patterns, e.g. if a particular patch repeatedly acts as a source of infection. The construction of transmission trees from epidemiological data on the timing and locality of rabies cases is another useful approach for identifying transmission pathways and can be supplemented by contact tracing data (Lembo et al. Reference Lembo, Hampson, Haydon, Craft, Dobson, Dushoff, Ernest, Hoare, Kaare, Mlengeya, Mentzel and Cleaveland2008). Contact tracing of rabies is aided by the memorable nature of rabies bites and therefore presents a unique situation where the collection of contact data is achievable. However, building transmission trees in this way becomes increasingly unreliable as larger numbers of missing links are inferred, and these models may also simplistically predict unlikely transmission events if knowledge of the landscape is not incorporated into algorithms e.g. predictions of transmission across major landscape barriers. Genetic data could be used to quantify mutation rates and build independent time-scaled transmission networks against which epidemiologically constructed transmission trees could be calibrated. Ultimately, transmission trees will be most effectively constructed with the combined use of genetic and epidemiological data, and appropriate datasets are increasingly becoming available for rabies, but further computational and statistical advances are required.

Host transmission is typically heterogeneous, often described by the 80/20 paradigm in which individuals differentially carry and transmit pathogens i.e. 80% of infections are carried by 20% of the population (Woolhouse et al. Reference Woolhouse, Dye, Etard, Smith, Charlwood, Garnett, Hagan, Hii, Ndhlovu, Quinnell, Watts, Chandiwana and Anderson1997). In the most extreme case this heterogeneity can exist in the form of superspreaders (Lloyd-Smith et al. Reference Lloyd-Smith, Schreiber, Kopp and Getz2005). There have been no in-depth studies of superspreading in domestic dogs carrying rabies and the contribution of superspreaders to rabies spread and persistence is controversial (e.g. Hampson et al. Reference Hampson, Dushoff, Bingham, Bruckner, Ali and Dobson2007 vs. Talbi et al. Reference Talbi, Holmes, de Benedictis, Faye, Nakouné, Gamatié, Diarra, Elmamy, Sow, Adjogoua, Sangare, Dundon, Capua, Sall and Bourhy2009), but large variations in biting frequency have been observed (Hampson et al. Reference Hampson, Dushoff, Cleaveland, Haydon, Kaare, Packer and Dobson2009). An important distinction must be made between superspreading in a restricted area, e.g. the same village, and a ‘spatial superspreader’ that has dispersed infection over large distances and multiple settlements (see Fig. 2 for visual example). Although the number of secondary cases caused by such individuals may be equivalent, spatial superspreaders are more important from an epidemiological perspective if they are responsible for connecting sub-populations. Tracing resulting infections to their source through genetic sequence data may identify hotspots where certain patch features promote greater dispersal ability i.e. spatial superspreading events (as suggested by Cross et al. Reference Cross, Johnson, Lloyd-Smith and Getz2007). Ideally, this would utilise retrospective genetic tracing of networks of infection, similar to that used for identifying source infections in foot-and-mouth disease (FMD) (Cottam et al. Reference Cottam, Wadsworth, Shaw, Rowlands, Goatley, Maan, Maan, Mertens, Ebert, Li, Ryan, Juleff, Ferris, Wilesmith, Haydon, King, Paton and Knowles2008). Statistically robust methods are yet to be conceived, but accounting for missing links in transmission chains is an important area for future development. As a major advantage compared to FMD, contact that may have lead to transmission is much easier to define for rabies. Contact tracing data, if available, could therefore complement genetic data to offer a more comprehensive view of how individual heterogeneity can impact on dynamics.

Fig. 2. Dispersal of bites from superspreading dogs resulting in rabies transmission in an area of the Serengeti District in Tanzania. Roads and rivers are shown to highlight the potential influence of landscape features on the dispersal of rabies- tentative observations indicate that superspreader progeny appear to cluster alongside roads and movement may be restricted by the presence of rivers (but other landscape features not shown may also be responsible for influencing dispersal patterns). Two potential types of superspreader are also highlighted in the map: A) a spatial superspreader, which transmits over a large spatial area, potentially connecting sub-populations and may be important from an epidemiological perspective; and B) a superspreader with a limited dispersal range that infects a large number of progeny but remains within a small spatial radius. Inset map shows the location of the Serengeti District within Tanzania.