Inductive hazard analysis for GMOs
Keith R. Hayes
CSIRO Division of Marine Research, 2004
Hazard identification is arguably the most important component of any risk assessment. Hazards that are not identified in the early stages of a risk assessment are not carried through the assessment, leading ultimately to underestimates of risk. Hazard identification for all new technologies initially must be inductive. As operating experience grows, and adverse events are recorded, the analysis can also adopt deductive approaches. The most common (deductive) approaches are checklists and unstructured brainstorming. Checklists may be lengthy and well developed or quite short, and are clearly the "status quo" in the majority of risk assessment frameworks for GMOs. This is surprising for such a new technology and worrying-checklists do not ask "what can go wrong" with the system in question, and do not confirm that all components of the system have been questioned. Indeed they tend to mislead the analyst into believing that all aspects of the system have been questioned without confirming this to be true.
The objective of this report is to apply two inductive techniques - Hierarchical Holographic Modelling (HHM) and Fault Tree Analysis-to identify the potential ecological hazards associated with the unconfined release of Herbicide Tolerant (HT) canola, Brassica napus. The aim of the analysis is to demonstrate the potential value of inductive hazard identification techniques as applied to GMOs. It does not aim to identify hazards specific to a particular product. The demonstration does not therefore apply to a particular type of HT canola nor identify particular herbicides or release conditions in a particular environment. It does, however, identify the general types of ecological hazards that may be associated with HT canola. In a real analysis of a GMO intended for release, the identification of hazards would be supported by product-specific and geography-specific information that is not presented here.
Hierarchical Holographic Modelling (HHM) captures the complexity of a large system by identifying the components and processes of all sub-systems and analysing how they interact with each other. The technique decomposes the system by looking at it from many different perspectives including, for example, the functions, activities, geo-political boundaries, or structures of the system. The analyst constructs an HHM by first identifying the most appropriate perspectives for the problem in hand. These are used to define the sub-systems which in turn are further decomposed into components, processes, functions or activities, which may or may not overlap with other sub-systems. The analyst(s) identifies hazards by comparing potential interactions between the sub-systems in a qualitative fashion. This is best achieved by a team expert in one or more of the chosen perspectives.
The Hierarchical Holographic Model developed here identified a total of 153 potential hazards, 13 potential benefits and 30 event scenarios that may present a benefit or a hazard depending on the specific environmental and agricultural conditions. It is important to note that the analysis did not actively seek to identify potential benefits-the ratio of hazards to benefits in this study is not in any way indicative of the cost-benefit ratio that might result from the introduction of HT canola to any given area. Approximately 43% of hazards were identified only once. A further 42% of were identified between 2 and four times, whilst 2 hazards (1%) were identified over 15 times. All hazards were grouped into broad categories and scored by degree of concern and confidence. The final hazard score does not represent a formal assessment of risk and uncertainty-it is simply a way to prioritise each of the hazards for further analysis. In particular, some hazards, which are probably quite unlikely, might have received a disproportionately high hazard score because one team member, perhaps unfamiliar with that particular biological process or group of organisms, over-rated the likelihood or the severity of the consequences.
The incidence of HT volunteers (and HT resistant weeds) on the farm has the highest average score of all the hazard categories. HT volunteers on farm occur due to the significant seed loss during harvest and via a variety of natural process (e.g. ants and earthworms) that encourage seed burial and re-emergence. Dispersal of the HT gene beyond the farm (off-site) has the second highest average score of all the hazard categories identified here. There are a large number of ways in which the HT gene might disperse beyond the farm, either as HT canola pollen and seed, or as HT pollen and seed of a weedy relative following gene flow.
Adverse changes to weed spectra were the most frequently identified hazard of HT canola, although this ranked relatively low. Farming practice associated with HT canola, particularly the expected increase in post-emergent herbicide application and subsequent selection of herbicide resistant hybrids or volunteers, may increase the resources available to brassicaceous pests whilst at the same time reducing the resources available to beneficial insects and invertebrates. The largest source of potential ecological hazards arises through potential changes to farming practice that may follow widespread use of HT canola. Most of these hazards are associated in one way or another with either the way in which HT canola is grown-such as closer crop rotations, minimum tillage and post-emergent application of herbicides-or with the farmer's more attentive behaviour to a high value/high return crop-such as increasing acreage into marginal or remnant land areas and altered spray strategies.
In summary, the top ten hazards reflect an underlying concern that commercialisation of HT canola, without careful management, will increase the incidence of HT volunteers, both on and off-farm, which facilitate "secondary" seed and pollen-mediated dispersal of the trait over large distances. This coupled with the potential development of herbicide tolerance amongst weeds, may necessitate the use of alternative, and potentially more toxic, weed control strategies across large areas of agricultural and non-agricultural land. The impact of HT canola, and associated farming practice, on soil fauna communities and processes, also figure prominently in the HHM analysis. This appears to be an important, but as yet poorly understood, aspect of the new technology.
The main drawback with the HHM analysis is the time required to complete it, and the need to co-ordinate experts that, as in this case, might be drawn from several different institutions. It is often difficult to maintain continuity and consistency in these groups. Redundancy and duplication within the analysis also tend to reduce its efficiency. It is difficult, however, to determine a priori where duplication is likely to occur. A posteriori analysis of the hazards and their HHM references may help analysts to design more streamlined approaches to the assessment that require less time without threatening the rigour and completeness of the analysis. This is an area for future research.
Fault-trees are a "top-down" hazard-analysis tool-the analyst specifies a failure event (the "top-event") and then, using two logical functions OR and AND, identifies all of the events that cause the specified failure. The causative events are laid out in a tree with the branches connected by "gates" comprising either of these logical functions. A fault-tree is therefore a graphical model of all the parallel and sequential combinations of events that lead to the top event.
In this report, a fault tree analysis is applied to a well-documented hazard-gene flow between herbicide tolerant canola and a weedy relative. It lays out the logical chain of events that must occur for the gene conferring herbicide resistance to become stably integrated in a weed population. Fault tree analysis can augment a Hierarchical Holographic Model by defining the necessary event chain behind the potential hazards suggested by the model.
The principal advantage of the fault tree analysis is its structured and rigorous approach to identifying exposure pathways. The analysis here has helped to identify potential rate limiting steps and speculates on possible hazard scenarios, such as the potential interaction between viral and bacterial pathways, that do not appear to have been addressed in the literature to date.
Another advantage of a fault tree is that it quickly identifies knowledge uncertainty in the system. This analysis highlights how uncertainty increases moving from sexual gene flow pathways to viral, bacterial and then fungal pathways. Particular areas of uncertainty include:
- sexual gene flow: the importance of mass effect in maintaining competitively neutral or inferior hybrids;
- viral: the relative frequency of RNA-RNA, DNA-DNA, RNA-DNA and DNA-RNA recombination;
- bacterial: the mechanisms of homologous recombination in bacteria, the rate at which endophytic bacteria come into contact with host DNA, the possibility of bacterial mediated transformation of plant cells by species other than A. tumefaciens, and the possibility of bacteriophage infection of plant cells; and,
- fungal: the potential role of fungal-mediated gene flow.
The fault tree provides an excellent platform to quantify the potential for gene flow under commercial conditions because it breaks down the event chain to individual elements that can be analysed experimentally, and therefore potentially quantified. Indeed quantitative estimates already exist for some of the basic events in the tree. Furthermore, many of the current appeals to the safety of GM products rely on the incredibly rare likelihood of undesired events such as bacterial gene flow. These appeals, however, are undermined by the large number of potential exposure pathways (demonstrated by the complexity of the fault tree) and extremely high exposure (billions of plants in commercial production). Quantitative fault tree analysis provides a means to explore the overall effect of these two opposing themes. Developing quantitative estimates of the frequency for each of the basic initiating events within the tree, and the undeveloped events, within a case-specific analysis, is therefore an important avenue of future research.
Fault tree analysis is not designed to identify all potential hazards. Unexpected interactions (outside the experience or imagination of the analyst) could result in additional unidentified hazards or hazard inducing mechanisms. By virtue of its holistic approach HHM analysis is more likely to identify, or at least suggest, unexpected interactions. Taken together these tools enable the analyst to postulate certain hazards and then investigate in more detail how they might occur. There is no guarantee, however, that these processes together will identify all hazards. There are no such guarantees in any form of hazard analysis or risk assessment (hence the need to continually compare the predictions of a risk assessment with reality). The logical and rigorous structure of HHM analysis and fault tree analysis, however, helps minimise the probability of missing important casual pathways and it performs much better in this regard than unstructured brainstorming techniques.