Ecography 32: 504516, 2009 doi: 10.1111/j.1600-0587.2008.05597.x # 2009 The Authors. Journal compilation # 2009 Ecography Subject Editor: Jens-Christian Svenning. Accepted 10 October 2008
Spatio-temporal analysis of an invasive plant pathogen (Phytophthora ramorum) in England and Wales Xiangming Xu, Thomas D. Harwood, Marco Pautasso and Mike J. Jeger X.-M. Xu (
[email protected]), East Malling Research, New Road, East Malling, ME19 6BJ, Kent, UK. T. D. Harwood, Univ. of Reading, School of Biological Sciences, Lyle Tower, Whiteknights, Reading, RG6 6AS, UK. M. Pautasso and M. J. Jeger, Div. of Biology, Imperial College London, Silwood Park Campus, SL5 7PY, UK.
Phytophthora ramorum is a damaging invasive plant pathogen and was first discovered in the UK in 2002. Spatial point analyses were applied to the occurrence of this disease in England and Wales during the period of 20032006 in order to assess its spatio-temporal spread. Out of the 4301 garden centres and nurseries (GCN) surveyed, there were 164, 105, 123 and 41 sites with P. ramorum in 2003, 2004, 2005 and 2006, respectively. Spatial analysis of the observed point patterns of GCN outbreaks suggested that these sites were significantly clumped within a radius of ca 60 km in 2003, but not in later years. Further analyses were conducted to determine the relationship of GCN outbreak sites over two consecutive years and thus to infer possible disease spread over time. This analysis suggested that disease spread among GCN sites was most likely to have occurred within a distance of 60 km for 20032004, but not for the later years. There were 35, 63, 81 and 58 sites with P. ramorum in the semi-natural environment (SNE). Analyses were carried out to assess whether infected GCN sites could act as an inoculum source of infected SNE plants or vice versa. In all years, there was a significant spatial closeness among GCN and SNE outbreak sites within a distance of 1 km. But a significant relationship over a longer distance (within 60 km) was only observed between cases in 2003 and 2004. These analyses suggest that statutory actions taken so far appear to have reduced the extent of long-distance spread of P. ramorum among garden centres and nurseries, but not the disease spread at a shorter distance between GCN and SNE sites.
The increased global trading and tourism has increased the frequency of biological invasions, which is one important driver of ecosystem change (Crowl et al. 2008). Invasive species are the second leading cause, after human population growth and associated activities, of species extinction and endangerment in the US (Pimentel 2002). Emerging and re-emerging infectious diseases are a major consequence of human expansion and ever-increasing global trade (Mayer 2000, Ricciardi 2007). For example, some invasive diseases have caused devastating population declines of their hosts, resulting in substantial ecological and economic impacts, e.g. chestnut blight Cryphonectria parasitica and jarrah dieback Phytophthora cinnamomi (Anagnostakis 1987, Weste and Marks 1987). Thus, identifying and curtailing the spread of invasive pathogens presents a considerable ecological and societal challenge. An important aim of ecology is to reach an understanding of the processes generating the observed species distribution patterns (Holdenrieder et al. 2004), for example, the dispersal mechanisms and distances of invasive species. Such understanding may enable reliable prediction of the spread and hence adoption of appropriate measures
504
to eradicate or contain the species. One approach for describing spatial patterns and inferring the underlying processes is to contrast measured patterns with null models derived from specific assumptions about the underlying processes (Wiegand and Moloney 2004). A useful framework for this approach is provided by spatial point pattern analysis. This provides a set of tools for analysing the spatial distribution of discrete points (Ripley 1981, Diggle 2003, Gosme et al. 2007). Notably, second-order statistics describing the spatial correlation structure of two point patterns can be used to infer the spatio-temporal scale at which these patterns are related. Phytophthora ramorum, the causal pathogen of sudden oak death, was first described as a new Phytophythora species on Rhododendron and Viburnum in Germany and the Netherlands (Werres et al. 2001). The sudden oak death syndrome was first observed in 1994 and associated with widespread mortality of oaks (Quercus spp., Lithocarpus densiflorus) in California (Rizzo et al. 2002, McPherson et al. 2005). The recent emergence of P. ramorum suggest that this pathogen is an introduced species (Werres et al. 2001, Rizzo and Garbelotto 2003, Ivors et al. 2006). In
October 2003, P. ramorum was found causing bleeding cankers for the first time in the United Kingdom (UK), on mature trees of the introduced Quercus falcata (Brasier et al. 2004). This disease affects a wide range of plant species, many of which experience non-fatal foliar symptoms (Rizzo and Garbelotto 2003, McPherson et al. 2005). These plant species, particularly Umbellularia californica in the USA and Rhododendron ponticum (rhododendron) in the UK, play an important role in pathogen dispersal: sporulation on infected leaves of these hosts serve as reservoirs of inoculum (Davidson et al. 2005, Denman et al. 2005). Splash dispersal is likely to be the main mechanism for the short range dispersal of the inoculum (Davidson et al. 2005). Transmission of P. ramorum from infested potting media to stems via infected, symptomless root tissue was also demonstrated to take place (Parke and Lewis 2007). Phytophthora ramorum was primarily recovered from the top (05 cm) soil in ornamental nurseries (Dart et al. 2007) and can survive in forest soils over summer (Fichtner et al. 2007). Long-range dispersal mechanisms include movement of infected plants in the nursery trade (Osterbauer et al. 2004, Stokstad 2004), aerial dispersal (Rizzo et al. 2005), movement of inoculum in water (Davidson et al. 2005), and human recreational activities (Tjosvold et al. 2002). Longdistance trade links among nurseries have the potential to facilitate the regional establishment of invasive pathogens (May 2006, Jeger et al. 2007). As the impact of P. ramorum on ecological communities increases (Monahan and Koenig 2006), so has the research on how landscape pattern, climatic conditions and human activities affect its establishment and spread (Meentemeyer et al. 2004, Condeso and Meentemeyer 2007, Kelly et al. 2007, Kluza et al. 2007). The invasion of this pathogen is influenced by landscape configuration and local characteristics of host habitats (Kluza et al. 2007), including human activities (Cushman and Meentemeyer 2008). The effect of landscape composition on the pathogen spread is scaledependent (Condeso and Meentemeyer 2007). Most investigations of the spatial patterns of the disease have focussed on forests and surrounding landscapes (Kelly and Meentemeyer 2002, Kelly et al. 2008). By contrast, the characteristics of P. ramorum spread among commercial nurseries have not received significant attention despite the risk of long distance spread due to commercial trading. Since 2002 when P. ramorum was first discovered in a nursery in the UK (Lane et al. 2003), extensive surveys have been carried out to check for the presence of this disease in nurseries, garden centres, and the semi-natural environment. Statutory actions are being taken whenever the pathogen is reported, including destruction of affected plants and sometimes tracing of related stocks (Appiah et al. 2004, European Commission Decisions 2002/757/EC and 2004/426/EC). We applied spatial point analysis to these outbreak cases during the period of 20032006, focusing on the spatial relationships of P. ramorum occurrences in nurseries, garden centres and the semi-natural environment in England and Wales. The main objective was to determine whether there was evidence for disease spread between commercial nurseries and garden centres, and the semi-natural environment, and, if so, at what spatial scale.
Material and methods Disease survey Since P. ramorum is a quarantine pathogen, all suspect plant materials with symptoms similar to those caused by this pathogen are required to be sent to Central Science Laboratory (CSL) in York, UK, for confirmation of the presence of P. ramorum using molecular detection methods (Lane et al. 2006, Tomlinson et al. 2007). Thus, most occurrences of this disease are expected to be recorded, particularly those in garden centres and nurseries. Nurseries and garden centres
The Plant Health and Seeds Inspectorate (PHSI) of the UK government Dept of Environment, Food and Rural Affairs (Defra) annually surveyed all major garden centres and nurseries (GCN) in England and Wales during the period of 20032006 for the presence of P. ramorum. The inspection policy is to visit all nurseries twice a year. All suspect materials were sent to CSL for molecular diagnosis. Once confirmed, stringent eradication and containment actions were immediately taken under official notice, including 1) destruction of infected plants, all susceptible plants within 2 m of an infected plant and any associated crop debris by burning or deep burial, 2) disinfection of surfaces, and 3) prohibition of movement of susceptible plants within a 10 m radius of infected plants for 3 months. Around 2500 garden centres were checked annually 2003 2005, whereas in 2006 around 2000 were surveyed. In total, there were 4301 different GCN sites monitored over the four-year period (Fig. 1), which represented nearly all GCN sites in England and Wales. Semi-natural environment
Between December 2003 and April 2004, P. ramorum in woodland was also surveyed by the Forest Commission, focusing on locations with rhododendron (n 1217 sites). In addition to systematic surveys, all suspect plant materials were reported and analysed for P. ramorum infection using molecular techniques, irrespective of their location. Once confirmed, similar eradication and containment actions to those in GCN were then taken. In total, 2798 sites were surveyed in the semi-natural environment (SNE: woodlands, private gardens, and historic gardens) during 2003 2006 (Fig. 1). Spatial locations of all surveyed sites (regardless of whether the pathogen was confirmed or not) were recorded, along with other relevant details including the genus and species of infected plants. However, disease incidence and severity at each outbreak site were not determined. Statistical analysis The second-order statistics, the pair-correlation function (g) (Stoyan and Stoyan 1994, Stoyan and Penttinen 2000), the related O-ring statistics (Wiegand and Moloney 2004) and Ripley’s K (Ripley 1981), were used to characterise a single type of outbreak sites (GCN or SNE) and to determine relationships between GCN and SNE outbreak sites. 505
Figure 1. Sites surveyed for the presence of P. ramorum during the period 20032006 in England and Wales. Two types of sites were surveyed: garden centres/nurseries and those in the semi-natural environment, including woodland, estates, parks and private gardens.
sffiffiffiffiffiffiffiffiffiffiffiffi K12 (r)
Statistical testing was done by comparing the observed patterns to Monte Carlo simulations under a specific null model.
L12 (r)
Definition and statistical significance testing of Ripley’s K function and O-ring statistics
is usually used. The bivariate pair-correlation function g12(r) is analogous to Ripley’s K12(r): K12(r) measures the cumulative amount of clustering/regularity up to distance r and g12(r) measures the instantaneous amount of clustering at distance r,
Ripley’s K function (Ripley 1981) is the accumulative version of the pair correlation function g(r) (Stoyan and Stoyan 1994), i.e. r
K(r) 2p
g g(t)tdt: t 0
Both K and g use information on all inter-point distances and provide more information on the scale of the pattern than do statistics based on nearest neighbour distances only (Diggle 2003). They describe the characteristics of the point pattern over a range of distance scales. The O-ring statistics O(r) can be defined as the intensity of points of a given point pattern within a ring of radius r and width dr around a representative (or arbitrary chosen) point of the pattern and is related to the normalized pair correlation function g(r) via O(r) lg(r) where l is the intensity of the pattern. Depending on the biological question, g- or O-ring functions may be preferred. If an interpretation of neighbourhood density is required O-ring statistics should be used, whereas g functions are preferred if the relative degree of clustering of two patterns needs to be investigated. The non-accumulative statistics have the advantage of easyyet-direct interpretation over the K function, since the cumulative nature of the K function confounds effects at larger distances with effects at shorter distances (Wiegand and Moloney 2004). Because the K-function uses more points, it is more powerful to reject a given null model. The bivariate K-function K12(r) calculated from patterns of two types of point has the interpretation of the expected number of points of type 2 within a given distance r of an arbitrary point of type 1, divided by the density (l2) of pattern 2. To stabilise the variance and remove the scaledependence, a square root transformation of K(r), called the L-function (Besag 1977), 506
g12 (r)
p
r;
1 dK12 (r) 2pr
dr
:
A transformation, O12(r) l2g12(r), gives the expected density of points of pattern 2 at distance r of an arbitrary point of pattern 1, i.e. local neighbourhood density. These statistics were calculated using the Programita software (Wiegand and Moloney 2004). Monte Carlo simulations under a specific null model were used to construct simulation envelopes for the L- function, g or O-ring statistics of the observed pattern. Each simulation generates a g(r), O12(r), or L-function. At a given distance (r), approximately n/(n1) 100% simulation envelopes were calculated individually from the highest and lowest values of the n test statistics calculated from n simulations of the null model (Stoyan and Stoyan 1994). In the present study, 199 simulations were conducted, i.e. n 199, corresponding to 99% simulation envelopes. Thus, if a statistic is outside the simulation envelope at a given distance, it is judged as an indication for a departure from the null model at that spatial scale. Simulation envelopes cannot be interpreted as confidence intervals for formal hypothesis testing because type I error inflation may occur due to simultaneous inference (i.e. tests at many spatial scales) (Diggle 2003). For each null hypothesis on spatial relationships of outbreak sites, two analyses were carried out, one analysis using the L function to test for possible departure from the null model and one analysis with a non-accumulative statistic to explore the nature of the departure in more detail. Underestimation of type I error is less an issue when using non-accumulative statistics such as g(r) and O-ring statistics, which, unlike the L function, for each distance r use a different set of pointpoint pairs.
Table 1. Null models, questions and methods used to investigate the spatial structure of garden centres/nurseries (GCN) and semi-natural environment (SNE) sites where P. ramorum had infected plants. Analysis
Pattern
Figure
1 Univariate Only garden centre and nurseries cases
Trivariate Only garden centre and nurseries cases
Semi-natural environment versus garden centres and nurseries Garden centres and nurseries versus semi-natural environnent
Null model
Questions asked
2
2003 2004 2005 2006
5a 5b 5c 5d
Random labelling (pattern 1 random on all garden centres and nurseries).
Are infected sites clumped given the locations of all garden centres and nurseries?
2003 2004 2005
2004 2005 2006
6a 6b 6c
Random labelling under antecedent conditions (pattern 1 fixed and pattern 2 random).
SNE 2003 SNE 2003 and 2004 SNE 2004 and 2005 SNE 2005 and 2006 GCN 2003 and 2004 GCN 2004 and 2005 GCN 2005 and 2006
GCN 2003 GCN 2004 GCN 2005 GCN 2006 SNE 2004 SNE 2005 SNE 2006
7a, 8a 7b, 8b 7c, 8c 7d, 8d 7e, 8e 7f, 8f 7g, 8g
Are later infected sites (pattern 2) clumped around the earlier infected sites (pattern 1) given the locations of all garden centres and nurseries?
Characterising a single pattern of outbreak sites
L(r) and g(r) statistics were calculated to determine whether GCN sites with P. ramorum are closer to each other than expected given the spatial locations of all GCN sites. Random labelling was used as a null hypothesis to test the significance of aggregation/repulsion among GCN sites with P. ramorum detected (Table 1). The assumption of random labelling is that there were two independent processes acting in sequence: the first process created the spatial pattern of GCN sites and a subsequent process acted over those locations deciding whether or not a site was a P. ramorum outbreak site (Goreaud and Pe´lissier 2003). Under the hypothesis of the random labelling, the second process which distributed the label ‘‘outbreak’’ and ‘‘nonoutbreak’’ over the GCN sites is considered a random process. Therefore, the n1 cases (outbreak sites) represent a random sub-sample of the combined pattern of the n2 nonoutbreak sites and the n1 case sites. This analysis was not extended to SNE sites with P. ramorum. The lack of spatial locations for all possible SNE sites with susceptible hosts made it impossible to distinguish true aggregation/repulsion among outbreak sites due to inoculum dispersal from clumped/over-dispersed patterns of susceptible hosts. Determining relationships between two patterns of outbreak sites
A special type of L12(r) and O12(r) analysis was used to test the hypothesis that P. ramorum spread from infected GCN sites to neighbouring ones, and from GCN sites to SNE sites or vice versa (Table 1), i.e. earlier infected sites acted as inoculum sources for subsequent new infections at other
Random labelling under antecedent conditions (pattern 2 fixed and pattern 1 random due to the lack of SNE locations for all susceptible hosts).
nearby locations. To test the hypothesis of disease spread among GCN sites over two consecutive years, a novel trivariate analysis was conducted in which random labelling ‘‘under antecedent conditions’’ was used as the null model. There are three patterns in this analysis. The first pattern is the spatial pattern of GCN sites that were infected by P. ramorum in the first year (i1). The other two patterns are the pattern of sites newly infected at year 2 (i2), i.e. those sites not infected in year 1 but infected in year 2, and the pattern of sites that were not infected in years 1 and 2 (ni2). Testing this hypothesis is essentially assessing whether i2 sites at a given spatial scale r are more frequently around i1 sites than ni2 sites. Conventional random labelling was executed between the two patterns of year 2 (i.e. i2 and ni2). However, in contrast to conventional random labelling, an O-ring statistic that determines the density of i2 sites at distance r from i1 sites was calculated to assess whether i2 sites were more frequent around i1 sites than expected under the null hypothesis of no relationships among i1 and i2 sites by random labelling of the joint i2 and ni2 sites. The O12(r) statistics instead of g12(r) was used because the direct interpretation of a neighbourhood density is needed. Spatial relationships between GCN and SNE outbreak sites in two consecutive years were also assessed similarly under the same null hypothesis. Outbreak sites of either of the two types (GCN or SNE) in both the previous and the current year (pattern i1) were used as possible inoculum sources for new outbreak sites of the other type in the current year (pattern i2) (Table 1). For example, both 2003 and 2004 GCN outbreak sites were used as pattern 1 to determine whether new SNE outbreak sites (pattern 2) in 507
2004 were spatially closer to these GCN sites, thus to infer the likelihood that these new SNE outbreak sites had resulted from inoculum originating from those GCN outbreak sites in the current year or the year before. When testing whether GCN infected sites were possible sources for SNE infections, the infected GCN sites (pattern 1) should in principle be fixed and infected SNE sites (pattern 2) randomised among all possible SNE sites. However, because all possible SNE locations with susceptible hosts were not available, we had to fix the SNE infected sites whilst randomising the infected GCN sites among all GCN locations, assuming that spatial distribution of susceptible plants at SNE sites around all GCN sites is the same or very similar. Since 2002 data were not available, the test for the relationship between GCN and SNE infected sites in 2003 can be interpreted in either direction: GCN or SNE infected sites acted as inoculum sources for the other. Data resolution
The resolution for data analysis is 1 km and thus multioccupancy of a single cell occurred in a number of cases. However, finer resolution would lead to excess demands for computing time. A resolution of 1 km was considered sufficient for the current objectives given that the sampling area is ca 600 600 km. The g and O-ring statistics were always calculated with a ring width of 2 km. When analysing the spatial relationship between SNE and GCN outbreaks sites, a finer scale (cell size of 0.25 km and ring width of 1 km) was used because these two types of sites were often close to each other.
Results Overall disease development For GCN cases, most outbreak cases occurred during MarchMay although this outbreak peak was weaker in 2006. Out of the 4301 GCN sites surveyed, there were 164, 105, 123 and 41 sites with P. ramorum in 2003, 2004, 2005 and 2006, respectively. The disease persisted on many GCN sites over several years: there were 33, 16 and 10 infected sites which persisted from 2003 to 2004, from 2004 to 2005 and from 2005 to 2006, respectively. The numbers of SNE sites with P. ramorum fluctuated without a clear trend (Fig. 2). There were 35, 63, 81 and 58 SNE sites with P. ramorum confirmed in 2003, 2004, 2005 and 2006, respectively. Many of these disease sites persisted over the four years in Cornwall, southwest of England (Fig. 3). Spatial relationships among GCN outbreak sites In 2003 and 2004, GCN outbreak sites appeared to clump around two foci: one in the southeast of England and the other in the northwest of England (Fig. 4a, b). These two foci also coincided with the regions of high GCN densities (Fig. 1). In 2005, the southeast centre was no longer apparent, but there appeared to be more outbreak sites in
508
Figure 2. Monthly number of new sites with P. ramorum detected during the period 20032006 in England and Wales.
the Midlands (Fig. 4c). In 2006 there were no clear foci (Fig. 4d). L(r) values from the analysis of the GCN outbreak sites in 2003 clearly suggested aggregation among outbreak sites for all the scales up to 100 km (Fig. 5a). The pair correlation (g11) showed that outbreak sites were more likely within 60 km of another (Fig. 5a). Of the 60 g(r) values when r560 km, 51 were above the 1% upper simulation envelope; in contrast, only in 1 out of 60 values was g(r) above or below the 1% simulation envelope when 60 Br5120 km. The aggregation was weaker in the other three years (Fig. 5b, c, d). In 2004, aggregation among outbreak sites was indicated by the L(r) values for r6 (Fig. 5b). This appeared to result mainly from the greater than expected pair correlation at the distance of 58 km apart (Fig. 5b) and there are further seven g11 values above the 1% upper simulation envelope at r 20, 37, 44, 55, 56, 58 and 65. For 2005, L(r) values were above 1% upper simulation envelope for r21 (Fig. 5c), due to individual positive correlation at r 15, 19 and 21 and further seven g11 values were above the 1% upper envelope at r 35, 36, 38, 39, 45, 52 and 53. In 2006, L(r) values did not indicate any significant deviation from the null hypothesis (Fig. 5d). For all years, the other two test statistics (g12 g11) and (g12 g22) [(L12 L11) and (L12 L22)] showed nearly the same patterns with r as g11 (L11) with r. Spatio-temporal relationships among GCN outbreak sites L12(r) values from the analyses of the infected GCN sites between two consecutive years showed a significant departure from the null model for the 20032004 data (Fig. 6a) when r5, initially due to greater than expected O12(r) values at r 5, 7, 10 and 11. There were further 18 O12(r) values above the 1% simulation envelope in three regions: 21 BrB26, 35 BrB48 and 52 BrB60 (Fig. 6a), indicating that infected GCN sites in 2004 occurred significantly more frequently within a distance of 60 km from GCN sites infected in 2003. L12(r) values did not indicate any significant deviation from the null hypothesis for 2004 2005 and 20052006 data (Fig. 6b, c).
Figure 3. Sites in the semi-natural environment (e.g. woodland, historic gardens, parks and private gardens) with P. ramorum confirmed each year in England and Wales.
Spatio-temporal relationships between GCN and SNE outbreak sites SNE sites as inoculum sources
In 2003, L12(r) values were above the 1% simulation envelope for r2 (Fig. 7a), initially due to greater than expected O12(r) values at r 35 and 79. There were further 10 O12(r) values above the 1% simulation envelope, all for rB45 (Fig. 7a). Again for the 20032004 data L12(r) values were above the 1% simulation envelope for r2 (Fig. 7b), initially due to greater than expected O12(r) values at r 2, 4 and 5. Further 10 O12(r) values were above the 1% simulation envelope, all for rB46 (Fig. 7b). For the 20042005 data, L12(r) values were above the 1% simulation envelop in the region of 57 BrB108 (except r 7577) (Fig. 5c); but only O12(4) was above the 1% simulation envelope. For the 20052006 data, L12(r) values did not indicate any significant deviation from the null hypothesis (Fig. 7d). When analysed at a finer scale (0.25 km), both L12(r) and O12(r) were on or above the 1% simulation for one or more r values when r51 (Fig. 8ad). GCN sites as inoculum sources
For the 20032004 data L12(r) values were above the 1% simulation envelope for r4 (Fig. 7e), initially due to greater than expected O12(r) values at r 4 and 5. Further 11 O12(r) values were above the 1% simulation envelope, all for rB46 (Fig. 7e). For the 20042005 data, L12(r) values
were all above the 1% simulation envelop for r2 (Fig. 7f), initially due to greater than expected O12(r) values at r 4, 5, 7 and 8. Further 12 O12(r) values were above the 1% simulation envelope, all for rB61 (Fig. 7f). For the 2005 2006 data, L12(r) values did not indicate any significant deviation from the null hypothesis (Fig. 7g). When analysed at a finer scale (0.25 km), both L12(r) and O12(r) were above the 1% simulation for one or more r values when r5 1 (Fig. 8e, f, g).
Discussion Analytical methodology The data used represent the results of regular inspections of sites with susceptible host species. There are limitations to this type of data, notably that the initial infection time cannot be determined, and thus the temporal pattern is uncertain. Inspectors may fail to observe the disease at an infected site because the plants are asymptomatic at the time of infection, because symptomatic plants have been pruned or removed, or because of inherent variability among inspectors. In addition, the sheer volume of plants may make detection difficult, especially in dense semi-natural or garden sites. Consequently, whilst it is likely that the presence of the disease can be detected over several inspections, it is not possible to trace the exact timing of 509
Figure 4. Garden centres and nurseries with P. ramorum confirmed each year in England and Wales.
an infection event. Furthermore, there were several different dispersal routes that may lead to varying dispersal distances, including trading through nursery network, short-distance dispersal via rain-splash, aerial dispersal, possible movement in water streams and dispersal through human recreational activities. Testing for dispersal at several scales from the present data set is difficult. Thus, the present data were not suitable for an analysis to derive a dispersal kernel model due to the uncertainty in determining initial infection time and the large variability in dispersal distances caused by different dispersal routes. Instead, we conducted exploratory analysis of the data using spatial point analyses by aggregating observed cases by calendar year to minimize the above-mentioned problems. Disease spread among GCN sites Spatial analysis of point patterns suggests that the spread of P. ramorum among commercial nurseries and garden centres is most likely to occur within the distance of 60 km (Fig. 5, 6). The degree of spatial aggregation of GCN sites with P. ramorum within a given year was greatest in 2003 and then decreased through time. Indeed, there was no significant aggregation detected among diseased sites in 2006. In 20032005, nearly all significant aggregation of diseased sites was detected within the distance of 60 km. 510
This ‘‘maximum’’ dispersal distance was further confirmed by the results of the trivariate analysis of infected GCN sites over two consecutive years, which also suggests a lack of spatial relationships among infected sites in later years. The decreasing spatial relationships among outbreak sites through time are probably due to the stringent measures taken by authorities, including plant passport systems, site eradication and strict port checking, to contain the disease. These measures taken, including limited traceforward and trace-back actions, are likely to have reduced the disease spread from outbreak nurseries and hence constrained rapid increase throughout the UK nursery network. In the semi-natural environment of the UK, P. ramorum was first discovered at several sites in Cornwall and in these sites the disease has persisted over the fouryear period analyzed. Thus, it may be reasonable to assume that these sites were the initial source for subsequent infections of GCN sites. The subsequent rapid spread of P. ramorum among GCN sites in England was most likely due to movement of infected but yet symptomless plants via commercial trading since disease spread appeared to occur within distances up to 60 km, much greater than the distance normally achieved by other dispersal mechanisms. Aerial dispersal of P. ramorum propagules (sporangia and zoospores) in wind-driven rain
Figure 5. Univariate analysis of the spatial pattern of garden centres and nurseries with P. ramorum detected within each year in England and Wales using the pair-correlation function (solid line) with 99% simulation envelopes (dotted lines) for the null model of random labelling. The simulation envelopes were constructed using the highest and lowest g(r) from 199 permutated data sets under the null model. The value g 1 is the theoretical expectation for a random point pattern. The inset graph is the corresponding plot for the L(r) function.
occurred within 10 m from the source (Davidson et al. 2005), though it should be noted that detectable dispersal distance is much less than the realized median dispersal distance. However, for wind-driven dispersal appreciable long-distance dispersal is expected (Shaw et al. 2006). In Oregon, most of infected trees were within 100 m of trees killed the previous year though occasional long-distance (up to 3 km) dispersal was possible (Rizzo et al. 2005). Spatial-pattern analysis of infected trees in California suggested that new dead oak trees tend to be located within up to 300 m of past dead oak trees (Kelly et al. 2008) and that a strong spatial association between oak tree mortality and California bay trees exists within 150 m (Liu et al. 2007). In the present study, the possible disease spread distance of up to 60 km found among GCN sites is probably specific to commercial trading characteristics
during the survey period. Long-range dispersal due to movement of infected plants in the nursery trade was also observed in USA (Osterbauer et al. 2004, Stokstad 2004). The structure of networks formed by plant movements among commercial nurseries is likely to have an important effect on the epidemic threshold (May 2006, Jeger et al. 2007, Pautasso and Jeger 2008). Our study suggests distance decay of disease spread over a certain distance within the nursery network. Further research is needed to elucidate the consequence of varying network characteristics on disease spread, which may enable development of effective intervention measures to reduce the risk of further economic losses due to P. ramorum. Disease spread between production facilities via movement of plants was also previously reported for P. nicotianae and P. drechsleri (Lamour et al. 2003).
511
Figure 6. The O-ring statistics (solid line) from the trivariate analyses using the null model of random labelling under antecedent condition to test for independence in the garden centres and nurseries with P. ramorum outbreak between two consecutive years. The simulation envelopes (dotted lines) for the null model were constructed using the highest and lowest O12(r) from 199 permutated data sets under the null model. The inset graph is the corresponding plot for the L(r) function.
therefore, not possible to attribute aggregation (or lack of aggregation) to the inherent pattern of susceptible hosts or true dependence among outbreak sites. Instead, we carried out analyses to determine whether GCN sites could act as inoculum sources for SNE infections and vice versa. In testing the former hypothesis (i.e. GCN 0SNE), we need to fix the GCN infected sites and randomise infected SNE sites. Since spatial locations of all susceptible hosts at SNE sites are not known, we instead randomised the GCN ‘‘source’’ sites among all GCN sites. To interpret the results, it is thus necessary to assume that the spatial distribution of susceptible plants at SNE sites around all GCN sites is very similar. Bidirectional spread of diseases between GCN and SNE is expected to occur within short distances mediated by rain-splash or aerial dispersal, including wind-blown rain. This short-distance dispersal is supported by the results of trivariate analysis of GCN-SNE outbreak sites, which showed consistent strong positive correlation among GCN and SNE outbreak sites within a distance of 1 km in either direction, i.e. GCN 0SNE or SNE 0GCN. This agrees with other findings of short-distance dispersal associated with aerial and rain-splash dispersal (Davidson et al. 2005, Rizzo et al. 2005, Swiecki and Bernhardt 2005). However, molecular characterisation of pathogen isolates suggested little gene flow of P. ramorum between nursery and forest environments in Oregon (Prospero et al. 2007). It should, however, be noted that the Oregon forest infestation is separated by 100 km from the infested nurseries in Oregon and California. Further work, particularly in molecular characterization of individual fungal isolates or populations, is needed to confirm this interpretation. This trivariate analysis of observed cases also suggested that both (GCN 0SNE, SNE 0GCN) transmission routes are probable over a longer distance (within 50 km). Apart from the several early outbreak sites in Cornwall, it might be reasonable to assume that the sporadic occurrence of P. ramorum at those SNE locations far away from Cornwall is due to inoculum from the nearby infected GCN sites or plants with asymptomatic infection sold from an infected GCN site and planted in a SNE site. This is because longdistance dispersal among SNE sites can only be achieved by movement of affected soil materials through recreation activities (e.g. hiking, biking) on vehicle tyres (Tjosvold et al. 2002) or by occasional very long-distance wind dispersal (Rizzo et al. 2005). Thus, significant association between GCN and SNE outbreak sites that are long distances apart is likely due to the significant aggregation among GCN outbreak sites. Indeed, the relationship among GCN and SNE outbreak sites separated by long distances progressively became weaker in the same way as the aggregation of infected GCN sites did.
Disease spread between GCN and SNE sites
Effectiveness of eradication measures taken
It was not possible to determine spatial relationships among P. ramorum SNE outbreak sites because spatial locations of all susceptible hosts at SNE sites were not known. It is,
In England and Wales, P. ramorum is classified as a quarantine disease, thus stringent measures are being taken to eradicate this disease and to limit its spread. The present
512
Figure 7. The O-ring statistics (solid line) from the trivariate analyses using the null model of random labelling under antecedent condition to test for independence in the sites with P. ramorum between garden centres/nurseries and in the semi-natural environment over two consecutive years. The simulation envelopes (dotted lines) for the null model were constructed using the highest and lowest O12(r) from 199 permutated data sets under the null model. The inset graph is the corresponding plot for the L(r) function.
513
Figure 8. The O-ring statistics (solid line) from the trivariate analyses using the null model of random labelling under antecedent condition to test for independence in the sites with P. ramorum between garden centres/nurseries and in the semi-natural environment over two consecutive years on a finer scale (0.25 km). The simulation envelopes (dotted lines) for the null model were constructed using the highest and lowest O12(r) from 199 permutated data sets under the null model. The inset graph is the corresponding plot for the L(r) function.
analysis of P. ramorum occurrence during the period of 20032006 suggests that the current control measures have managed to reduce the disease spread, due to two aspects. First, the overall incidence of disease occurrence has remained stable and indeed appears to have decreased in nurseries and garden centres. Secondly, the control measures appeared to have limited the disease spread among GCN sites as inferred from the decreasing extent of spatial relationships among GCN sites over the survey period. At such a low annual incidence of infected GCN sites (4.1, 2.7, 3.1 and 1.0% in the four years), it is expected that there
514
would be ample opportunities for disease spread among GCN sites given the closeness between the infected and non-infected sites (Fig. 1, 4). Thus, the finding that the degree of spatial relationships among GCN outbreak sites decreased over the years is more likely to be indicative of effectiveness of control measures in limiting disease spread rather than other causes. However, these measures did not effectively reduce the disease occurrence in the seminaturally environment and neither did they have much effect in reducing the potential spread between GCN and SNE sites within a short distance. Similarly, in Oregon,
P. ramorum eradication in nursery environments is effective, but not completely successful at infested forest sites (Prospero et al. 2007), possibly due to difficulties in eradicating inoculum in the forest (Rizzo et al. 2005). Our findings indicate that the containment and eradication policy at those sites where it has been found reduces the rate of long-distance spread, but also that complete eradication of this disease is unlikely. Acknowledgements This project was funded by the UK Dept for Environment, Food and Rural Affairs (Defra). We thank PHSI (Defra) and the Central Science Laboratory (CSL) for providing the data and Michael Shaw of Reading Univ., UK and Thorsten Wiegand of Helmholtz Centre for Environmental Research UFZ, Germany for valuable suggestions.
References Anagnostakis, S. 1987. Chestnut blight: the classical problem of an introduced pathogen. Mycologia 79: 2337. Appiah, A. et al. 2004. Phytophthora ramorum: one pathogen and many diseases, an emerging threat to forest ecosystems and ornamental plant life. Mycologist 18: 145150. Besag, T. 1977. Contribution to the discussion of Dr. Ripley’s paper. J. R. Stat. Soc. B 39: 193195. Brasier, C. M. et al. 2004. First report of ramorum bleeding canker on Quercus falcata, caused by Phytophthora ramorum. Plant Pathol. 53: 804804. Condeso, T. E. and Meentemeyer, R. K. 2007. Effects of landscape heterogeneity on the emerging forest disease sudden oak death. J. Ecol. 95: 364375. Crowl, T. A. et al. 2008. The spread of invasive species and infectious disease as drivers of ecosystem change. Front. Ecol. Environ. 6: 238246. Cushman, J. H. and Meentemeyer, R. K. 2008. Multi-scale patterns of human activity and the incidence of an exotic forest pathogen. J. Ecol. 96: 766776. Dart, N. et al. 2007. Recovery frequency of Phytophthora ramorum and other Phytophthora spp. in the soil profile of ornamental retail nurseries. Plant Dis. 91: 14191422. Davidson, J. M. et al. 2005. Transmission of Phytophthora ramorum in mixed-evergreen forest in California. Phytopathology 95: 587596. Denman, S. et al. 2005. In vitro leaf inoculation studies as an indication of tree foliage susceptibility to Phytophthora ramorum in the UK. Plant Pathol. 54: 512521. Diggle, P. 2003. Statistical analysis of spatial point patterns. Arnold. Fichtner, E. et al. 2007. Detection, distribution, sporulation, and survival of Phytophthora ramorum in a California redwoodtanoak forest soil. Phytopathology 97: 13661375. Goreaud, F. and Pe´lissier, R. 2003. Avoiding misinterpretation of biotic interactions with the intertype K12-function: population independence vs random labelling hypotheses. J. Veg. Sci. 14: 681692. Gosme, M. et al. 2007. Size, shape and intensity of aggregation of take-all disease during natural epidemics in second wheat crops. Plant Pathol. 56: 8796. Holdenrieder, O. et al. 2004. Tree diseases and landscape processes: the challenge of landscape pathology. Trends Ecol. Evol. 19: 446452. Ivors, K. et al. 2006. Microsatellite markers identify three lineages of Phytophthora ramorum in US nurseries, yet single lineages in
US forest and European nursery populations. Mol. Ecol. 15: 14931505. Jeger, M. et al. 2007. Modelling disease spread and control in networks: implications for plant sciences. New Phytol. 174: 279297. Kelly, M. and Meentemeyer, R. K. 2002. Landscape dynamics of the spread of sudden oak death. Photogr. Eng. Remote Sens. 68: 10011009. Kelly, M. et al. 2007. Modeling the risk for a new invasive forest disease in the United States: an evaluation of five environmental niche models. Comput. Environ. Urban Syst. 31: 689710. Kelly, M. et al. 2008. Spatial pattern dynamics of oak mortality and associated disease symptoms in a California hardwood forest affected by sudden oak death. J. For. Res. 13: 312 319. Kluza, D. A. et al. 2007. Sudden oak death: geographic risk estimates and predictions of origins. Plant Pathol. 56: 580 587. Lamour, K. et al. 2003. Etiology of Phytophthora drechsleri and P. nicotianae ( P. parasitica) diseases affecting floriculture crops. Plant Dis. 87: 854858. Lane, C. R. et al. 2003. First outbreak of Phytophthora ramorum in England, on Viburnum tinus. Plant Pathol. 52: 414. Lane, C. R. et al. 2006. Diagnosis of Phytophthora ramorum evaluation of testing methods. EPPO/OEPP Bull. 36: 389 392. Liu, D. et al. 2007. Characterizing spatial-temporal tree mortality patterns associated with a new forest disease. For. Ecol. Manage. 253: 220231. May, R. 2006. Network structure and the biology of populations. Trends Ecol. Evol. 21: 394399. Mayer, J. D. 2000. Geography, ecology and emerging infectious diseases. Soc. Sci. Med. 50: 937952. McPherson, B. A. et al. 2005. Sudden oak death in California: disease progression in oaks and tanoaks. For. Ecol. Manage. 213: 7189. Meentemeyer, R. et al. 2004. Mapping the risk of establishment and spread of sudden oak death in California. For. Ecol. Manage. 200: 195214. Monahan, W. B. and Koenig, W. D. 2006. Estimating the potential effects of sudden oak death on oak-dependent birds. Biol. Conserv. 127: 146157. Osterbauer, N. et al. 2004. Surveying for and eradicating Phytophthora ramorum in agricultural commodities. Plant Health Progr. (online) doi: 10.1094/PHP-2004-0309-02-RS. Parke, J. and Lewis, C. 2007. Root and stem infection of Rhododendron from potting medium infested with Phytophthora ramorum. Plant Dis. 91: 12651270. Pautasso, M. and Jeger, M. 2008. Epidemic threshold and network structure: the interplay of probability of transmission and of persistence in small-sized directed networks. Ecol. Complex. 5: 18. Pimentel, D. 2002. Introduction: non-native species in the world. In: Pimentel, D. (ed.), Biological invasions: economic and environmental costs of alien plant, animal, and microbe species. CRC Press, pp. 38. Prospero, S. et al. 2007. Population dynamics of the sudden oak death pathogen Phytophthora ramorum in Oregon from 2001 to 2004. Mol. Ecol. 16: 29582973. Ricciardi, A. 2007. Are modern biological invasions an unprecedented form of global change? Conserv. Biol. 21: 329336. Ripley, B. D. 1981. Spatial statistics. Wiley. Rizzo, D. M. and Garbelotto, M. 2003. Sudden oak death: endangering California and Oregon forest ecosystems. Front. Ecol. Environ. 1: 197204.
515
Rizzo, D. et al. 2002. Phytophthora ramorum as the cause of extensive mortality of Quercus spp. and Lithocarpus densiflorus in California. Plant Dis. 86: 205214. Rizzo, D. M. et al. 2005. Phytophthora ramorum: integrative research and management of an emerging pathogen in California and Oregon forests. Annu. Rev. Phytopathol. 43: 309335. Shaw, M. et al. 2006. Assembling spatially explicit landscape models of pollen and spore dispersal by wind for risk assessment. Proc. R. Soc. B 273: 17051713. Stokstad, E. 2004. Plant pathology: nurseries may have shipped sudden oak death pathogen nationwide. Science 303: 1959. Stoyan, D. and Stoyan, H. 1994. Fractals, random shapes and point fields. Methods of geometrical statistics. Wiley. Stoyan, D. and Penttinen, A. 2000. Recent application of point process methods in forest statistics. Stat. Sci. 15: 6178. Swiecki, T. J. and Bernhardt, E. 2005. Disease risk factors and disease progress in coast live oak and tanoak affected by Phytophthora ramorum canker (sudden oak death). Proceed-
516
ings of the sudden oak death second symposium: the state of our knowledge, Albany, CA, pp. 383411. Tjosvold, S. A. et al. 2002. Incidence of Phytophthora ramorum inoculum found in soil collected from a hiking trail and hikers’ shoes in a California park. Sudden Oak Death Science Symposium, 1518 December 2002. Tomlinson, J. et al. 2007. Faster, simpler, more-specific methods for improved molecular detection of Phytophthora ramorum in the field. Appl. Environ. Microbiol. 73: 40404047. Werres, S. et al. 2001. Phytophthora ramorum sp nov., a new pathogen on Rhododendron and Viburnum.. Mycol. Res. 105: 11551165. Weste, G. and Marks, G. C. 1987. The biology of Phytophthora cinnamomi in Australasian forests. Annu. Rev. Phytopathol. 25: 207229. Wiegand, T. and Moloney, K. A. 2004. Rings, circles, and null models for point pattern analysis in ecology. Oikos 104: 209229.
