Prediction of the combined effect of various GM contamination sources of seed: A case study of oilseed rape under Danish conditions

Abstract

The potential GM contamination at each step in seed production can be predicted for conditions that correspond to a worst-case scenario. When the combined effect of the various GM contamination stages is calculated, the worst-case predictions of the different stages are normally added. This additive procedure to estimate the combined effect of GM contamination may be relevant to estimate the combined effect where a single farm is concerned, but may result in estimation errors when harvested seeds from a large number of farms are mixed. In this study, the consequences of treating the different stages of GM contamination as independent stochastic effects on the combined GM contamination have been estimated. A case study of GM contamination in organic and conventional oilseed rape crops under Danish conditions indicated that the greater part of GM contamination occurred at farm level and much less during transport and storage. Generally, it may be concluded that it is important to consider this uncertainty in the estimates of adventitious presence at each seed production step and also whether the different steps are independent.

Keywords:

• Coexistence
• GMO
• oilseed rape

Introduction

With the introduction of genetically modified (GM) crops for cultivation in the EU countries, regulation, control and management measures become essential in order to secure coexistence between GM, conventional and organic farming (Tolstrup et al., 2003). A central issue is the analysis of the potential GM contamination of seeds at the different stages in the production line. These include seed lots delivered to the farmer, pollen dispersal to the field, GM contamination from the seed bank, seed from volunteers and weeds, seed dispersal by machinery, and transport during harvest and accidental admixture of seed at storage facilities.

In order to control the GM content of different products, generally acceptable tolerance limits for GM contamination also have to be established for end-products. Currently, common EU standards for different uses are being established for seed purity of agricultural crops without GMO. Plant products used for food may contain up to 0.9% GMO (threshold level) without being labelled (EC, 2003a; EC, 2003b). If the GM content is higher, the product must be labelled as containing GM material. However, if the product has been favourably assessed by the scientific committee but not been finally approved, the threshold limit until 2007 is 0.5% GMO (EC, 2004).

Different measures exist to control the amount of adventitious presence of GM content in harvested seeds, and plant products (e.g., for animal feed or use in the food industry). These include: separation distances between fields with similar crops, buffer zones around potential receiver fields, extended cropping intervals, cleaning of machinery and transport vehicles for seed remnants and control of volunteers and wild relatives (Tolstrup et al., 2003). The Danish Ministry of Food, Agriculture and Fisheries has recently employed such measures in mandatory regulations (Plantedirektoratet, 2005).

Currently, when the level of possible GM contamination in different agricultural systems and crops is evaluated in order to generate advisory opinions, the amount of GM contamination at each stage (Figure 1) is predicted or estimated under conditions which may be compared to a worst-case scenario. Afterwards, when the combined effect of the various GM contaminations at the different stages is calculated, the worst-case predictions of the different stages are added.

Figure 1.  Routes for contamination of OSR by GMO during production.

The procedure of adding the worst-case predictions at the different stages in order to estimate the combined effect of GM contamination implicitly makes some important assumptions. It is indirectly assumed that if a worst-case scenario event occurs in one stage then worst-case scenario events will also occur at all the other stages. It is a claim of this article that the adding of worst-case predictions may be a relevant approach when a single farm is considered, e.g., in the matter of legal issues concerning a single farmer. However, if the tolerance limits are at the level of the manufactured products (both food and feed) and the harvested seeds from a large number of farmers are mixed at a processing plant, then the relevant measure of GM contamination is not at the level of the random farmer (who may not meet regulations), but rather at the level of the average harvested seeds. In such a case it is more relevant to treat the predictions of GM contamination as probability distributions, and the possible events at the different stages as stochastic events that may or may not depend on each other in a statistical sense.

Here, the consequences of treating the different stages of GM contamination as a series of independent stochastic effects on GM contamination have been studied in a case study of GM contamination in organic and conventional oilseed rape crops.

We have chosen oilseed rape (OSR) as a case species, because it is a common seed crop in most of northern Europe; the genes are easily dispersed by wind or insect pollination to other fields or by hybrids with wild relatives (e.g., Darmency et al., 1998; Hansen et al., 2001). Furthermore, the seeds may persist viably for long periods in the soil (5– > 20 years) (Thompson et al., 1997) and seed spillage during cultivation, usually 5 to 10% but up to more than 20% (Lutman, 1993; Pekrun et al., 1998) can result in problems with volunteer populations in later OSR crops.

Model

Seeds of sown crops may either contain one or more copies of different transgenes, originally coming from a genetically modified plant (denoted GM seed), or it may be free from any such transgene or copies of it (denoted non-GM seed). At different stages from seed sowing to harvest and sale of the seed crop, another seed may, in principle, be sampled and again this seed may either be a GM or a non-GM seed.

The probability of sampling a GM seed, P ^tot _GM may be calculated in a stochastic model where the uncertainties at the different stages of seed production are included in the model. If the events at the different stages of seed production are assumed to be independent, then the probability of not sampling a GM seed after n different stages are:

1

where P ⁱ _GM is the probability that a seed sampled at random at stage i is a GM seed. For example, during the stage of pollination, P ^pollination _GM is the probability that an ovule without a transgene is pollinated by a pollen grain with a transgene or, during harvest, if 1000 GM seeds are accidentally mixed into a seed lot of 10,000,000 non-GM seeds, then this corresponds to a P ^harvest _GM of 0.01%. EU legislation does not distinguish between heterozygous and homozygous transgenic seed.

Treating the events at the different stages as probabilities allows the uncertainties of the specific processes at a particular stage to be integrated in the risk analysis. For example, a P ^harvest _GM of 0.01% may be known to be a rare situation where a particular farmer does not act according to regulations and good farming practice. Thus, rather than using a rare case as a point estimate, we argue that the likelihood of sampling a GM seed is better described by a probability distribution. In such a probability distribution, knowledge of the likelihood that farmers do not act according to good farming practice, or fulfil only some of the requirements, may be included in the calculation. Some of the information underlying the probability distribution may be rather vague due to e.g., limited data, but this is no argument for ignoring the little information that is available. Furthermore, in some cases the probability distribution may be estimated directly from data, e.g., the probability of GM pollination using a Bayesian approach (Damgaard & Kjellsson, 2005), and then it will be meaningless to summarize the estimated posterior distribution by a point estimate.

Calculations involving probability distributions used to be a job for specialists, but new software has simplified the calculations considerably. The calculations in this article were made by Monte Carlo simulation of the specified probability distributions using the Excel-add-in ‘@RISK’ (Palisade, 2006) until there was a convergence within 0.01%.

Results

The likelihood of a worst-case scenario of GM contamination in each of the different stages in production of oilseed rape seed has been predicted (EC, 2001) and the combined effect of the different production stages has been calculated by adding the worst case predictions (Tables I and II). It should be noted that, since the predicted probabilities are small, approximately the same result is obtained by treating the worst-case estimates as point estimates of the probability and calculating the combined probability by using equation (1).

Table I. Stages in production of oilseed rape seed and the expected probability of contamination (see text). ‘≈95%’ denotes the approximate 95% percentile level of the predicted distribution.

Stage	SCP^a	SCP as uniform dist. distribution	Organic farming distribution	≈95%	Conventional farming distribution	≈95%
Certified seeds	0.003	∼uniform[0, 0.003]	∼trunc. normal[0.0001, 0.0005]	0.001	∼trunc. normal[0.0003, 0.0015]	0.003
Seed bank (volunteers)	0.002	∼uniform[0, 0.002]	97%∼trunc. normal[0.0002, 0.001]	0.002	97%∼trunc. normal[0.001, 0.005]	0.01
			+3%∼trunc. normal[0.001, 0.01]	0.01	+3%∼trunc. normal[0.002, 0.01]	0.02
Pollination	0.002	∼uniform[0, 0.002]	∼gamma[0.0889, 0.000077]	0.00004	∼gamma[0.0933, 0.000350]	0.0002
Harvest	0.0001	∼uniform[0, 0.0001]	∼trunc. normal[0.0001, 0.0005]	0.001	∼trunc. normal[0.0001, 0.0005]	0.001
Transport	0.0005	∼uniform[0, 0.0005]	∼trunc. normal[0.00001, 0.00005] + 1/10.000	0.0001	∼trunc. normal[0.00001, 0.00005] + 5/10.000	0.0001
Storage	0.0005	∼uniform[0, 0.0005]	∼trunc. normal[0.00001, 0.00005]	0.0001	∼trunc. normal[0.00001, 0.00005]	0.0001
^aPredictions of the adventitious presence of GM seeds from the Scientific Committee on Plants (EC, 2001).

Table II. The probability distribution of sampling a GM seed after the different stages (in %), reported by the mean and the {5, 50, and 95 percentile}.

Stage	SCP^a	SCP as uniform dist.	Organic farming	Conventional farming
P ^farm GM	0.71^b	0.35 {0.16, 0.36, 0.55}	0.19 {0.08, 0.18, 0.33}	0.63 {0.19, 0.58, 1.25}
P ^tot GM	0.81^b	0.40 {0.20, 0.40, 0.60}	0.20 {0.09, 0.20, 0.35}	0.68 {0.24, 0.64, 1.30}
^aPredictions of the adventitious presence of GM seeds from the Scientific Committee on Plants (EC, 2001).
^bFound by adding the probability estimates. Note that since the probabilities are small approximately the same result is obtained by using equation (1).

As a first rough comparison of including uncertainties in the calculations, the probability of GM contamination at the different stages is assumed to be independent stochastic events, which are uniformly distributed between 0 and the worst-case estimates predicted by SCP (EC, 2001). The combined probability of GM contamination at the farm level P ^farm _GM , and the effect of transport and storage P ^tot _GM are both probability distributions and are reported here by their 5%, 50%, 95% percentiles (Table II). As expected, the assumption of uniformly distributed probability distributions considerably reduced the most likely combined probability of GM contamination.

Predictions based on Danish conditions

We also tried to predict the probability distribution of GM contamination in each of the different stages for seed production of oilseed rape grown organically or conventionally under Danish conditions (see below). The predictions were primarily based on experience with practical oilseed rape farming in Denmark (see below) and, in the case of pollination, a meta-analysis of available data of pollination between oilseed rape fields (Damgaard & Kjellsson, 2005). In most cases the predicted probability distribution of GM contamination in each of the different stages was assumed to come from a normal distribution that was truncated outside the interval [0, 100%]. The mode of the truncated normal distribution was furthermore assumed to be one-tenth of the predicted 95% percentile, so that, given a rather small predicted 95% percentile value, the predicted probability distribution is skewed with a right tail (Figure 2). The above truncated normal distribution was chosen as a quite flexible candidate distribution for quantifying our degree of uncertainty, mainly because it has a sizeable right tail, which leads to relatively conservative predictions. However, the choice of distribution is somewhat arbitrary and other distributions may be used instead when more knowledge of the different processes has been accumulated.

Figure 2.  For most of the expected distribution the truncated normal distribution was selected. The distribution is truncated outside the interval [0, 100%]. The mode (µ) of the shown truncated normal distribution is 0.001 and σ is 0.005, with a 95% percentile of about 0.01 (=10 µ).

Certified seeds for sowing

In connection with the production of certified seeds to the farmer for sowing, a number of precautionary measures are expected to be enforced. The early generations (pre-basic and basic seed) will probably be controlled and only batches with no or very little GM content will be used for further multiplication into certified seed for sowing (C1 and C2). However, it will not be possible to avoid unintended admixture entirely. Therefore, it is predicted that the majority of the sown seed will have no or a very little content of GM, but that there will be seed batches with a somewhat larger GM content. Consequently, we expect that the predicted 95% percentile value will be approximately 0.3% GM admixture in seeds sown for conventional farming and approximately 0.1% GM admixture in seeds sown for organic farming. At a seed quantity of 4 kg per hectare, an admixture of 0.1% corresponds to 4 g sown GM seeds per hectare. We do not expect any significant GM admixture during drilling, since it is possible to clean the drilling machines almost completely.

Seedbank and volunteers

Rapeseed can survive several years in the soil, and during harvest a relatively large number of seeds is spilled on the ground and incorporated into the soil. The amount of spilled seeds, the degree of immediate and later germination and the mortality (i.e., the depletion rate) affect the size of the soil seed bank. The amount of GM seed in the seed bank depends on whether GM rape has previously been grown on a given area. In areas where non-GM rape is grown and GM rape has not been grown before (in the calculations assumed to be 97% of the total area), the admixture of GM rape is foreseen to be low. On such areas the occurrence of GM rape in non-GM rape will primarily be due to spread of pollen and/or the spread of seeds with machines. Consequently, we expect that the predicted 95% percentile value will be approximately 1% GM admixture for conventional farms and approximately 0.2% GM admixture for organic farms. On the remaining area, assumed to be 3%, where GM rape has previously been grown, there will be a considerably higher content of GM rapeseed in the seed bank. If we assume that conventional and organic farming of non-GM rape seed can only take place after a cropping period of a minimum of five and eight years for conventional and organic farms, respectively, we expect that the predicted 95% percentile value will be approximately 2% GM admixture for conventional farms and approximately 1% GM admixture for organic farms.

Pollination

The probability of pollination resulting in harvested GM seeds was predicted from a meta-analysis of available gene-flow data in oilseed rape (Damgaard & Kjellsson, 2005). The predicted probability of GM pollination is calculated at the 95% credibility level when the width of a pollen-receiving field, the distance to a GM-oilseed rape field, and the width of a non-harvested border zone are known. The predicted values assumed that the width of both organic and conventional fields is uniformly distributed between 100 m and 500 m. Furthermore, it was assumed that the distances to a GM-oilseed rape field are uniformly distributed between 500 m (the proposed minimum isolation distance in Denmark) and 2000 m for organic OSR crops, and between 150 m (proposed minimum distance in Denmark) and 2000 m in conventional OSR crops. Finally, it was assumed that the border zone is included in the harvest. Simulations using the results of the meta-analysis and the above-mentioned assumption gave a predicted probability distribution of GM pollination which, by visual judgement, was fitted to a two-parameter gamma distribution.

Harvest

During harvest GM rapeseed can be mixed into non-GM rapeseed if the harvester or the trailers are not cleaned. It is expected that the harvest machinery is carefully cleaned so that no or only very little rapeseed is found in the tank, or other places. It is also assumed that the transport trailers are cleaned. We expect that a maximum of 6 kg seeds may be transported from a GM field to non-GM field, and if at least 2 hectares are harvested at a yield of 3000 kg per hectare, this will correspond to a GM admixture of 0.1% of the harvested seeds, based on the predicted 95% percentile value.

Transport

It is assumed that lorries, trailers and conveyor systems are cleaned thoroughly after use for transport of GM rape seeds, and therefore we expect that the predicted 95% percentile value will be approximately 0.01% GM admixture. There is a risk that single loads of GM rape seeds by mistake are mixed into seed lots of non-GM rape. We have predicted the risk that this will happen to be higher for conventionally grown oilseed rape than for organically grow rapeseed. We estimate that the probability of a load of GM rapeseed by mistake being mixed into either conventionally or organically grown non-GM rapeseed is 0.05% and 0.01%, respectively.

Storage

We assume that the storage facilities of GM OSR and non-GM OSR are physically separated in order to minimize the risk of accidental mixing, and we expect that the predicted 95% percentile value will be approximately 0.01% GM admixture.

Combined probability of GM contamination

The predicted probability distribution of GM contamination varies significantly between organic and conventional farming (Table II). This variation is mainly due to the assumed difference in the amount of GM contamination from the seed bank. Generally, most of the GM contamination was found to be due to effects at farm level, whereas the effects of transport and storage were found to be minor (Table I).

Discussion

Generally, it may be concluded that it is important to consider the uncertainty in the estimates on the degree of adventitious presence at each production step and whether or not the different steps are independent of each other. In this study the probabilities that a seed sampled at random at the different stages are assumed to be independent, but this assumption needs to be investigated critically.

The estimates on the degree of adventitious presence calculated in this study (Table II) are higher than those presented for conventional farming in another coexistence study (Bock et al., 2002). In this case maximum contamination rates from 0.22 to 0.42% were expected, but the seed bank was assumed to be without GM seed from the start.

The subjectively defined probabilities need to be refined by experimentation and detailed modelling as in the case of pollination (Damgaard & Kjellsson, 2005). While modelling based on existing data may provide good advice for isolation distances between neighbouring fields, more complex crop and field distributions in the landscape may require the use of more complex landscape models (see e.g., Colbach et al., 2005). Based on model studies and other existing gene flow information, the requirements for the EU threshold of 0.9% GM content have lead to a recommended separation distance between GM and non-GM OSR fields of 200 m for certified seed and lower for restored hybrids (Weekes et al., 2005). For pre-farm seed production, a 400 m buffer zone has been recommended for coexistence in Australia (GTGC, 2003).

Priority attention should be given to effects of seed banks. Gene flow mediated by surviving seeds and volunteers from earlier GM cultivation could make it especially difficult for farms to grow non-GM crops in rotation or in separate fields (Squire, 2005). Hence, the indicated levels of GM admixture presented here may require a high degree of field preparation and long cropping intervals. Coexistence may be seriously affected by long-term survival of GM-OSR seeds in the farming soil, where new studies indicate that it may take from seven to more than nine years for a reduction to 5% surviving seeds (Lutman, 1993).

The accidental mixing of GM and non-GM seed mediated by machinery during harvest and transport is generally expected to be of relatively minor importance (Tolstrup et al., 2003). Cleaning of transport vehicles and processing equipment and control for seed remnants will minimize mixing of different seed lots. Spillage of seed during transport can be a serious dispersal factor in oilseed rape (Crawley & Brown, 2004; Norris & Sweet, 2002) and good covering of vehicles is needed. The risk of more serious mixing of larger quantities of GM seed with non-GM seed at the processing facilities is known to happen and requires particular control and monitoring.

For agricultural production of conventional seed of cross-fertilized species such as oilseed rape and maize, the limit for adventitious presence of GMO has been suggested to be 0.3%, while the limit for many self-fertilized species could be 0.5% (EC, 2001). Considering the estimates in the present study, this will be difficult to achieve without special efforts to minimize GM presence in the certified seeds for sowing and the soil seed bank.

Acknowledgements

We thank Hanne Østergård for valuable discussions when the outline of the article was conceived, as well as the constructive criticisms of an anonymous reviewer.