FOUND-AND-LOST: TRANSGENIC MAIZE IN OAXACA, MEXICO
Kelly M. Paulson
November, 2005

Since the advent of transgenic maize (Zea mays L.) in the United States and Canada in 1996, gene flow to other, non-transgenic varieties of corn has been of concern to scientists, farmers, and the public. However, the concern about gene flow to landraces of maize in Mexico intersects with a suite of cultural, historical, and spiritual connotations.¹ The extraordinary significance of maize in Mexico is at least part of the reason that the country enacted a de facto moratorium on the commercial planting of genetically engineered corn varieties in 1998,² in spite of permitting transgenic varieties of other crops like soya and cotton. Perhaps this is why Quist and Chapela’s November 2001² revelation that gene flow from transgenic corn had occurred in Oaxaca, Mexico, came as such a surprise. The controversy that was sparked by this paper continues to simmer, but most scientists agree¹ with the major conclusion of Quist and Chapela’s research^2,3: transgenes were present in native landraces of maize in Oaxaca. However, the Quist and Chapela study was not designed to detect frequency, only presence. The most recent contribution to this story, by Ortiz-García et al., ⁴ represents the first published attempt to quantify the frequency of transgenes in the same region (personal communication, Dr. Allison A. Snow (AAS), 07 October 2005).

Ortiz-García et al.⁴ collected corn cobs and seeds from eighteen different villages ("locations") during November-December 2003 and 2004 (Figure 1).

At each location, from one to eight different fields were sampled. In 2003, four to five maternal plants were selected haphazardly within each field meaning that they were distributed throughout the field and not growing close to one another (pers. comm., AAS, 07 October 2005). In 2004, five "normal" and five "stressed" maternal plants were selected from each field, on the basis of the assumption that early hybrids between transgenic and locally adapted maize plants might appear maladapted to the local field conditions. In both years, one cob was collected from each maternal plant, and kernels were taken from each cob for analysis (a range of 104-503 seeds per cob in 2003 (pers. comm., AAS, 10 October 2005)).

Seeds from 2003 were analyzed by Genetic ID (www.genetic-id.com) and seeds from 2004 were divided evenly between Genetic ID, GeneScan (www.gmotesting.com), and an archive in Mexico. Both of the labs are capable of detecting transgenes at a frequency of 0.0001 (i.e., one transgenic seed in a homogenized sample of 10,000 seeds) with nearly 100% accuracy. In 2003 the researchers chose the conservative strategy of delivering ground samples representing ≤503^a seeds each, while in 2004 the sample size was 810 to 5,630 seeds per ground sample. Both laboratories used two markers to probe for transgenic DNA in the samples: the CaMV 35S (cauliflower mosaic virus) promoter and the NOS (nopaline synthase, from Agrobacterium tumefaciens) terminator sequence. The CaMV 35S sequence is present in all varieties of commercialized transgenic corn, with the exception of the GA21 Roundup-Ready^® event. The NOS terminator sequence, however, is present in the GA21 corn and in several other varieties of transgenic corn. In addition, the adh1 gene, which is native to maize, was amplified as a positive control. Using a combination of quantitative and qualitative PCR techniques, all samples were scored negative for both transgenic markers.⁴ The possibility that transgenic seeds were sampled but were undetected by the PCR analysis is unlikely because both companies use proper controls designed to avoid false negatives as well as false positives, in compliance with international seed-testing standards.⁴

Given the previous reports of the discovery of transgenes in this same region of Mexico (e.g.,^{2, 1}), one might wonder how likely it is that these researchers simply "missed" the transgenic kernels. Handily, Ortiz-García et al.⁴ address this very question in two different ways: using the kernel, then the cob, as the unit of observation. Understanding the reasoning for these two alternative analyses requires a brief lesson on corn reproduction: a single cob on a maternal maize plant can contain several hundred seeds. Theoretically, each kernel on a plant could have a different paternity, subject to the diversity of pollen donors growing nearby.⁵ While corn is able to self-pollinate, most of the kernels on a cob are the result of outcrossing.⁶ In other words, it is unlikely that all the kernels on a cob have the same paternity, but it is also unlikely that no two kernels are full siblings, either because they are derived from self-pollination or from the same paternal plant’s pollen. The proportion of kernels on a maternal plant in a field situation that are full sibs does not seem to have been investigated.

First, the authors calculated the binomial probability that they missed the transgenic elements, assuming that transgenes did exist at a frequency q of 0.0001. The use of such a low frequency for these analyses was a conservative strategy, especially considering the 2003 genetic analysis was designed for the possibility that transgenes were in excess of 5% in some of the fields.⁴ Using the hypothetical underlying frequency, the joint probability of missing all transgenic seeds in the sample from all locations in a given year can be calculated as P_overall (0 inclusions|q = 0.0001). When each kernel was considered as an independent observation, P_overall is equal to 0.00003 in 2004. If the maternal cob is considered as the unit of observation, however, the binomial probability of detecting no transgenes in 2004 increases to 0.932.^c In the latter case, failing to detect any transgenic seeds, if they were actually present at the underlying frequency of 0.0001, becomes unsurprising. The range between these two estimates, from 0.003% to 93.2%, certainly makes interpretation difficult. A second calculation estimated the transgene frequency at which at least one seed ought to have been sampled with 95% certainty (q_0.95) across all locations. Ortiz-García et al.⁴ estimated that in 2004, if kernels are the unit of observation, they could be 95% certain that transgenes were present at <0.003%. However, when the same analysis was based on cobs, they could only be 95% sure that transgenes were present at less than 0.43%. Because the real sample size is somewhere between the kernel and the cob as experimental unit, the authors conclude that transgenes are "absent or extremely rare" in the sampled fields, and that 0.01% might be a realistic mid-point estimate based on the second set of analyses, and considering data from both years.⁴

Two questions relevant to biosafety might be raised by this study. First, could one have predicted these results? Second, how could the Ortiz-García et al.⁴ methodology inform future experiments to monitor for transgene escape?

In January 2004, Information Systems for Biotechnology sponsored a workshop to discuss the application of the net fitness model⁷ to gene flow from crop plants.⁸ The net fitness model was originally developed by Muir and Howard to predict gene flow from a group of transgenic fish into a population of wild-type conspecifics. To predict population size and transgene frequency over a number of generations, the model uses six quantitative life history measurements or "net fitness traits": juvenile viability, age to sexual maturity, mating success, female fecundity, male fertility, and adult viability. Workshop participants discussed the potential for making these measurements suitable to the life history of plant species. While Brassica spp. and cotton were suggested as useful test cases for the model’s application to crop species,⁸ thinking about transgenic corn varieties through the lens of the model may provide a useful indicator of what data is lacking. For example, Ortiz-García et al.⁴ assumed that transgenic-landrace hybrids would appear "stressed" in the Mexican fields they sampled. While this assumption was based on sound reasoning, a study to characterize relevant performance traits of landraces and commercial cultivars in the environment of interest would be useful for future studies such as this one. Further, more information on the fitness of F₁ and advanced-generation hybrids between transgenic and local varieties could give scientists a better understanding of the likelihood that transgenes would persist, or spread, in the years following a hybridization event. If there were particular agronomic traits that distinguished hybrid from landrace plants, small-scale farmers could select or deselect them from their fields (pers. comm., AAS, 07 October 2005). If the challenge posed by measuring life history traits on corn plants is not enough, imagine modeling human behavior as a component of the potential for transgene flow and dispersal.

In any large-scale environmental release of a transgenic plant or animal, it is important to have a monitoring plan. This study raises several issues relevant to the design of monitoring experiments: (1) site selection; (2) sample size determination; (3) sampling methodology; and (4) detection of transgenic elements. What decisions to make regarding these issues will vary depending on the investigator’s specific objectives. First, while a few sites in this study were selected in the same villages as the original discovery by Quist and Chapela,² it is likely that different farmers’ fields were sampled (pers. comm., Dr. Sol Ortiz-García, 10 October 2005). In the case of corn in which most pollen settles within 100 m of the source plant,⁵ it may be important to design studies where site selection is informed by the location of previous discoveries. However, crop rotation regimes and seed exchange will make site selection difficult in managed cropping systems. Next, Ortiz-García et al.⁴ provided two analyses using the cob and the kernel as the unit of observation. Given the current understanding of relatedness of kernels on a single cob, their strategy of providing an upper and lower bound for n is warranted. However, using their 2004 data for illustration, we know that neither n = 706 (cobs) nor n = 103,620 (kernels) is true. Pending further studies of the paternity of kernels on a cob, one sampling strategy is to collect fewer kernels per cob to minimize the chance that two kernels are full siblings (pers. comm., AAS, 07 October 2005) and to collect samples from a larger number of maternal plants. Of course, this entails quite a bit more fieldwork and cooperation from farmers. Related to site selection, one can also decide how to sample within the locations and fields. In this study, plants were chosen haphazardly in that they were scattered throughout the field, and some plants that appeared "stressed" were also chosen on the assumption that they would be more likely to carry a transgene. One might choose instead to select plants purely randomly, or to select individual cobs or kernels at random after they have been harvested to remove any possibility of sampling bias. On the contrary, one could intentionally sample plants on the edges of fields; for example, fields or ditches bordering roads where corn kernels in transit could have bounced free of a truck and appeared as volunteer plants. Finally, there is a need for independent, empirical testing of the limits of transgene detection at the commercial labs that perform such services (pers. comm., AAS, 07 October 2005).

Clearly, developing monitoring strategies that are scientifically robust and cost-effective will continue to challenge biosafety science as more transgenic products are released around the world. We have an opportunity to meet this challenge by building on previous research, identifying gaps in existing science, and refining methodologies in an iterative process.

1. Commission for Environmental Cooperation of North America (2004) Article 13 Report, available online: http://www.cec.org/files/PDF//Maize-and-Biodiversity_en.pdf

2. Quist D & Chapela I (2001) Nature 414, 541-543

3. Quist D & Chapela I (2002) Nature 416, 602-603

4. Ortiz-García S, Ezcurra E, Schoel B, Acevedo F, Soberón J & Snow AA (2005) PNAS 102, 12338-12343

5. Luna VS, Figueroa MJ, Baltazar MB, Gomez LR, Townsend R & Schoper JB (2001) Crop Sci 41, 1551-1557

6. Hoeft RG, Scott WO, & Aldrich SR (2000) Modern corn and soybean production. MCSP Publications, Champaign

7. Muir WM & Howard RD (2001) Am Nat 158: 1-16

8. Information Systems for Biotechnology (2004) ISB News Report January 2004, available online: http://www.isb.vt.edu/news/2004/news04.Jan.html

Kelly M. Paulson
Conservation Biology Graduate Program
University of Minnesota – Twin Cities
Saint Paul, MN, USA
kmp@umn.edu

(Endnotes)
^a This appeared in the original paper as 300; this correction is in press.