6 December 2013
Imported Plant Diseases:
An Epidemiological Perspective
Professor Chris Gilligan
This lecture is about infectious plant diseases; how they arrive in new regions, how we detect them and how we use models to predict the spread and to optimise control strategies, often under emergency conditions. Plant diseases are caused by a range of micro-organisms that include fungi, viruses, bacteria and nematodes amongst others. Some of these pathogens are transmitted by direct contact between plants: some are wind-dispersed, others are dispersed by movement in water. Many are transmitted by insect vectors and many are also transmitted by human activity, including movement by trade, travel, personal importation, on machinery and even workers clothing and tools.
The science of epidemiology plays a central role in understanding patterns and rates of spread of disease invasions and the effects these have on plant populations. This in turn enables us to predict whether or not an epidemic is likely to become severe. It also enables us to identify how best to deploy methods of control to prevent, inhibit or slow the rate of spread of the epidemic. Applications range from control and management of plant disease in agricultural, horticultural and forest crops through to disease control in natural and semi-natural environments. The options for control include pesticides, but also genetic-based resistance control, i.e. breeding crop plants and ornamentals for disease resistance, but also cultural methods such as removing infected plants or establishing a cordon sanitaire around an invading epidemic.
The recent spread of Ash Dieback, caused by the fungus Chalara fraxinea, received considerable attention in the media. This is an example of a newly arrived pathogen killing an iconic tree species that could have profound effects on the landscape and the biodiversity that ash trees support. Some will remember the devastating effects of Dutch elm disease on the landscape in the 1960s and 70s. There are many other diseases, however, that are threatening crops and natural ecosystems that go largely unreported in the national and international media. Foremost amongst these are new variants of rust diseases that are threatening the productivity of wheat crops worldwide. Wheat stem rust and wheat yellow rust, caused by two different fungal pathogens, have been successfully controlled for many years by genetic resistance through conventional plant breeding. Arising from the green revolution, this approach to plant breeding has focused on identifying so-called major genes for resistance to pathogens. One consequence of this is that the many different varieties of wheat grown throughout Asia, Africa, Europe, the Americas and Australia all rely on the same genes to resist the rust pathogens. The result of this is a strong selection pressure on the pathogen populations, such that if a mutant strain arrives that can overcome the resistance of the host, i.e. it can infect and reproduce on the host, the mutant strain is likely to spread very rapidly. This is occurring with strains of yellow rust that have spread globally across several continents with movement of spores by wind.
A second major threat is posed by a particular strain of wheat stem rust, known as UG99. This strain was first detected in Uganda and has subsequently spread through North Africa and the Gulf States into Iran. If the pathogen were to spread into the Indian subcontinent it would have a significant effect on one of the major wheat producing areas. Rapid spread by aerial dispersal is not restricted to pathogens of agricultural crops but can also occur in pathogens of natural vegetation. Chalara fraxinea is one example that has spread throughout continental Europe ultimately arriving in the UK. Others include fungal-like organisms that can infect many different species. Two striking examples of the genus Phytophthora are P. ramorum and P. cinnamomi. Phytophthora ramorum causes sudden oak death in California and ramorum dieback of larch and rhododendron in the UK. Phythophthora cinnamomi, has laid waste to large areas of natural vegetation including Jarrah vegetation in Australia.
What then are the key scientific and policy questions associated with these invading pathogens and how can science and epidemiology in particular help? We identify four key questions. Why do some diseases take-off and others do not? Why do some strains and diseases die out, some co-exist and others dominate? How can we translate variability into risk? New infections occur at the small scale, for example at the level of an individual plant but serious economic and epidemiological problems are manifest at the large scale: how can we scale from the individual to the population?
A further question arises in how to use epidemiology to optimise the deployment and durability of disease control methods?
It is sometimes thought that the sedentary nature of plants might lead to a different epidemiological theory for spread from that for animal, including human disease. There are some differences but there are many similarities. Although plants don’t move, spores, vectors and other propagules are dispersed and many of the concepts from medical epidemiology are applicable with some adaptation to plant disease. One very useful paradigm is to separate the population of plants into susceptible (i.e. healthy plants), infected (those that are capable of spreading infection) and removed or recovered plants (those that are actually removed, for example by rogueing, or those that are effectively removed from taking further part in the epidemic by application of pesticide, or by dying). This leads naturally to the SIR (Susceptible, Infected, Removed) model from which it is possible to derive simple criteria for invasion, for the final epidemic size and for the equivalent of herd immunity. These concepts are useful and can be adapted in a number of ways. For example, the invasion criterion, often designated as Ro, can be calculated from a knowledge of the transmissibility of infection and the infectious period (how long plants remain infectious) such that if an infected individual infects on average more than one individual an epidemic ensues. If an infected individual infects on average less than one individual the epidemic will fade out. The scale of the susceptible and infected units can be adapted to apply not only to an individual plant but to refer to individual fields, farms or other natural regions. The scale of interest can also be reduced to consider leaves, roots or shoots as the units of interest. The equivalent of herd immunity means that it is possible to bring the invasion criterion, Ro, below one without treating all the individuals in a population. Mathematical models provide a means to predict future spread of disease and also to optimise strategies for disease management or control. Two important aspects underpinning epidemiological modelling are addressed in this lecture. These are space and uncertainty.
Much early modelling of disease progress relied on the insights gained from deterministic models. These are models that always give the same answer for a given input. Many natural systems, however, and especially epidemics, are inherently variable. Such variability can be introduced in so-called stochastic models, that don’t give a single value for a given set of inputs but rather give a distribution of possible outcomes for a given input. This is much closer to what we see in nature. It becomes very important in judging not only the success associated with different control scenarios but also the risks. The use of stochastic models to compare different control scenarios may lead to rejection of a potentially high yielding scenario because of greater risks of failure.
We consider now an emerging epidemic for a new pathogen, about which very little is known. One of the major challenges in these circumstances is to predict how rapidly the pathogen will spread. This requires an understanding of the dispersal processes. These can be characterised in a number of ways, for example by distinguishing the principal and likely means of transmission: is it wind-borne, water-borne or transmitted by human activities such as trade? In each of these cases the dispersal process can be described by a dispersal kernel, which defines the fall-off in the probability of successful transmission from an infected to a susceptible host as the distance between the two increases. Historically these dispersal kernels were mapped by experimentally introducing an infected host into completely susceptible population. With emerging epidemics, however, it is likely that the pathogen will have arrived or have been introduced at several sites and begun to spread before it is first detected. Characterisation now requires the use of statistical methods to estimate a dispersal kernel from successive snapshots of where disease has spread to over time. Essentially the methodology involves statistical trial and error to identify the most likely chains of events that have led to the current pattern and from which it is then straightforward to estimate the underlying dispersal kernel and the transmission rate. The dispersal kernel identifies how far infection might occur and which distances are more frequent than others, while the transmission rate is a measure of the likelihood of infection occurring given that the pathogen arrives on a particular host.
There is inevitably a conflict of expediency in this process. The natural reaction upon detecting an invading pathogen is to remove it. That works well if the initial invasion is identified before there has been significant spread. In many cases, however, there will have been cryptic spread ahead of the initial site of infection whereby hosts may be infected but not yet showing visible symptoms. Removal of infected hosts at this early stage may cause a significant problem in planning how best to control the emerging epidemic. Early removal may fail to halt the epidemic when there is cryptic infection. Early removal also prevents the epidemiologist from having enough data to estimate dispersal and transmission rates. Without these it becomes very difficult and sometimes impossible to predict the future spread and to use models to compare different strategies for control. We illustrate this in the lecture using sudden oak death in California and citrus canker in Florida. Mathematical models can also be used to analyse and predict the arrival of new pathogens via different routes. This occurs for Ash Dieback in Great Britain in which there are two sources of introduction, one by trade bringing in infected saplings and the other by spread of airborne spores from the advancing epidemic on the continent. By using meteorological models, originally developed by the Met Office for transmission of particles it is possible to show the continued threat of incursion from the continent for this pathogen. Linking the meteorological models for arrival of spores with epidemiological models for subsequent spread of disease enables prediction of the risks within Great Britain and also to other parts of the UK.
Quick action to tackle a new invasion is beneficial but, as we have seen, many emerging epidemics are already established by the time the first symptoms are reported. The challenge then is to decide when, where and how to deploy control strategies. Epidemiological models have a place to play in this. There are three components to the epidemiological analysis. The first involves the epidemiological parameters as introduced above. The second requires knowledge of where the susceptible hosts are, how they are dispersed in the landscape, whether they are contiguous or widely spread, with long gaps between clusters of susceptible hosts that might slow down the spread of the epidemic. The third component concerns the effect of meteorological conditions, more simply weather variables such as temperature, humidity and duration of leaf wetness on the epidemic. A simple starting point is to distinguish between those meteorological conditions that favour epidemic spread i.e. the epidemic is ‘switched on’ when those conditions prevail, and the epidemic is effectively ‘switched off’ when they don’t. By combining host landscapes, epidemiological models for spread, and meteorological conditions that favour or inhibit spread it is possible to predict the likely future spread of disease. This is done using stochastic models that take account of uncertainties in knowledge of the biology of the pathogen and the inherent variability of weather, using past data for weather to predict future weather patterns.
The models can be used to produce ‘risk’ and ‘hazard’ maps. A risk map shows where disease is likely to occur, taking account of known occurrences and potential sites for introduction, for example by trade or by aerial dispersal from outside the region of interest. Hazard maps provide a measure of the impact and potential for local spread of disease were the pathogen to be introduced to a particular site. Hazard maps are constructed by introducing the pathogen independently at each and every potential site within the region of interest and modelling the local spread of disease. Risk and hazard maps can be used to inform decisions about where to sample for an emerging epidemic and where best to deploy resources for control. By control in these circumstances we usually mean use of a pesticide or cultural control that involves the removal of infected and susceptible hosts to prevent the continued spread of disease. Risk and hazard maps can also be used to inform the deployment of genetically resistant varieties in agriculture. Here an important trade-off is in growing resistant varieties over a sufficient area to prevent or inhibit the spread of disease, while avoiding overuse of the resistant variety that could lead to a rapid selection in the pathogen population to overcome the genetic resistance of the host. There are analogous problems over use of pesticides leading to pesticide resistance. Cost constraints, as well as societal concerns, may also limit the number of plants that can be removed from the landscape. The challenge then is to match the scale of control with the inherent spatial and temporal scales of the epidemic. These are captured by integrating the natural scales of the host distribution, the transmissibility of infection and the natural scales of weather variability. To do this successfully, it is necessary to allow for cryptic infection: if one detects a symptomatic host (tree, a crop, or whatever is of interest) how far ahead is the infection likely to have spread but not yet been detected or shown visible symptoms? We illustrate the approach in the lecture with some simple simulations and a practical example concerning a disease of sugar beet in the UK.
Logistical and economic constraints have been mentioned. These can be formalised when comparing and optimising different control scenarios by linking epidemiological with economic models. The theory underpinning this approach is now quite well developed and has been widely used in analogous areas such as in fisheries. Routine deployment in epidemiology, however, is still at an early stage. We illustrate the approach, therefore, for a simplified SIS (Susceptible, Infected, Susceptible) epidemic occurring in two regions. We assume there is an epidemic happening in each region with a small amount of transmission between regions and that there is a limited supply of pesticide (or analogously a drug for human diseases). Suppose that the levels of infection are different in each region and we wish to bring the epidemic under control as efficiently as possible but there is insufficient pesticide to treat all infected sites. An obvious question is then which region should be given priority, the region with the higher or lower level of infection? It transpires in this case that preferentially treating the region with the higher level of infection is the worst possible scenario. Instead, preference should be given to the region with the lower level of infection in order to bring the epidemic under control. The intuition supporting this conclusion is that because there are more susceptible hosts to be infected in the lesser infected region, the epidemic is potentially faster in that region so treating it preferentially has a greater effect. Researchers are currently adapting these methods for use in real situations of epidemic spread such as how best to deploy chemical and genetic control for the wheat rusts.
Much still needs to be done. There are many diseases threatening agricultural, horticultural, forest and natural systems worldwide. The UK Government recently convened a taskforce to review tree health and plant bio-security in order to be better prepared in detecting, managing and eradicating invading pests and pathogens. Success depends upon increased understanding of how diseases spread and a linking of these with modern methods for the development of safe, effective pesticides and the development of durably resistant varieties of plants.
© Professor Chris Gilligan 2013