Resolution of the Mexican transgene detection controversy: error sources and scientific practice in commercial and ecological contexts

Abstract

Bernd Schoel and John Fagan (Vice-President and Founder/CEO, respectively, of Genetic ID, henceforth BS&JF) criticize and dismiss our recent publication in Molecular Ecology by focusing on our use of the Polymerase Chain Reaction (PCR) to detect specific DNA sequences. They raise important questions about the standards required to use PCR in various environmental conditions, pointing to the well-known fact that this delicate method may lead an unskilled operator to false results. They further suggest that our observations of transgenic DNA sequences in Mexican landrace maize should be attributed to false positives, i.e. a type I error. After considering their challenge and reviewing the evidence, we find their arguments seriously lacking in substance, and their practice permissive of false negatives, a type II error. We seem to have attracted BS&JF’s attention because, in an effort to corroborate our own results, we utilized the services of Genetic ID as full-paying customers. We established that Genetic ID failed on occasion to detect positive blind samples, which should not be surprising given the known vagaries of the PCR method. Yet for BS&JF this detection failure is not a factual possibility; instead, to explain our observations they would have us both (i) contaminating our samples and (ii) lying about the origin and nature of our materials. Specifically, BS&JF state: 1‘We contend that results such as these are incorrectly interpreted as positive and are more likely to be indicative of contamination in the laboratory.’ and 2‘We would argue that the leaf sample provided by the authors did not contain the claimed NK603 event and, furthermore, does not contain material from any commercialized transgenic single plant.’ Other charges include an implication that we used false evidence and/or withheld inconvenient data (BS&JF, p.5, lines 5–11) to reach our conclusions. All of these are indeed very serious challenges to our technical capacity and expertise, as well as our professional and personal integrity. BS&JF declare their suspicion that all of our PCR positive results arose from systematic contamination. They note the presence of bands in the PCR gels that are weaker than they would expect for a ‘100% (homozygous) or 50% (heterozygous) GMO level’, the only evidence that they would take as a positive result. Such a view is based on the unwarranted expectation that an end-point PCR could be used as a quantitative method. In our experience and that of other independent laboratories, the PCR amplification of transgenic sequences in landrace maize backgrounds tends to produce relatively faint bands of variable intensity in end-point reactions visualized on agarose gels, which so far has been the standard approach in the field (Quist & Chapela 2001; Alvarez-Morales 2002; Piñeyro-Nelson et al. 2009). Genetic ID’s own gels (their standard to screen-out ‘negatives’) show this kind of variability, even for repeats of a single sample in a single assay, or for different assays performed for the same sample at different times [see Fig. S1 (Supporting information)]. Such results should not surprise anyone versed in the PCR method. Although early cycles in the PCR assay may reflect stoichiometric molecular relationships, end-point PCR should be expected to be variable for many reasons including the gradual inactivation of Taq polymerase, competition against primers through amplicon reannealing, and exponential increase in product to be amplified (Butler 2005). Because these are all stochastic processes that become exponentially amplified as the PCR cycling proceeds, variable plateau concentrations of amplicon (i.e. band intensity) should be expected even under the most strictly controlled conditions (Heid et al. 1996). Nevertheless, BS&JF expect homogenous and strong bands in every context, and it is apparently their view that any sample with somewhat questionable intensity in gel bands should be declared a ‘negative’ and should not be studied further. We note that Genetic ID’s procedures recognize but dismiss the existence of variability in samples. For example, a ‘negative’ sample for which there may be a weak but visible band in a duplicate PCR gel is reported as ‘above detection level or detected’, and in other cases where clear bands are present in PCR gels, results were at times reported as ‘negative at the operational limit of quantitation of 0.1%’. BS&JF now choose to assign all the above-mentioned results as ‘negatives’, when in our experience such samples often proved clearly positive upon further molecular analyses. Thus Genetic ID has established standards which are permissive of, and indeed prone to, false negatives (type II error). There is evidence arguing against BS&JF’s PCR contamination hypothesis. We observe, for example, that the presence of positive bands in our samples is neither randomly nor homogeneously distributed as would be required by such a hypothesis. Specifically, at the inception of our study, maize ears were collected, seeds were subdivided from each ear and distributed independently to our separate laboratories (RR in Irapuato and EAB in Mexico City) by an outside researcher (S. Ortíz-García, a co-author with B.S. in Ortíz-García et al. 2005). Maintaining each laboratory in complete isolation from the other through double-blind coding, seeds from these subsamples were germinated, emergent leaf tissue lyophilized in facilities free of cloning or PCR products, extracted and PCR-amplified, after which we compared all results for congruence. As explicitly described in our original study, we took a highly conservative position before we would call a positive sample: samples were never scored as positive unless we had at least two repeated confirmatory results in each separate laboratory based on independent DNA extractions and amplifications. Under these circumstances, the laboratory contamination implied by BS&JF should be expected to either appear in all samples or to be randomly distributed among families within laboratories, with a possible differentiation between the two laboratories reflecting their differing patterns of contamination. None of these scenarios occurred. Families and/or localities consistently appeared with positive individuals in both laboratories while others consistently failed to show positives. We have now subjected all our results to a statistical analysis showing that the distribution pattern of positive samples among seed families or localities is indistinguishable between the two independent laboratories; i.e., overall, the frequency of positives among families matches across laboratories. The probability of this pattern emerging from a contamination source is <0.001. Naturally, there were instances where we suspected false positives and false negatives based on the performance of appropriate controls. On occasion, we also saw lack of reproducibility of results within or across laboratories for individual samples that were further analysed. Such results are to be expected in any normal laboratory procedure, and certainly in a PCR-based process, but we designed protocols and controls to deal with this expectation and never allowed questionable results to be included in our publications. Thus we remain confident of our statements even if considered only from the partial perspective of PCR amplification. However, other evidence supports our claims. We believe that while Genetic ID’s methods may well be suited for commercial use on commercial crop varieties in the U.S., they seem inadequate for research-oriented and environmental applications. BS&JF’s vague assertion that Genetic ID has ‘standardized validation procedures using all transgenic maize events and 14 species commercialized to date’ (Schoel and Fagan, p. 4) covers at best their proficiency in working with the limited genetics of commercial hybrid maize. Most commercial GMO assays (including Genetic ID’s) have been optimized using maize varieties widely grown in the United States including commercial transgenic lines, all of them built upon a very homogenous genetic background, a cross of Northern Flint x Southern Dent. Both inbreds, and Northern Flint in particular, are quite divergent genetically from Mexican maize land races (Doebley et al. 1986, 1988). There are good reasons to believe that such limited focus may place Genetic ID’s methods at a relative disadvantage for detecting transgenic DNA sequences in landrace maize. Using real-time PCR, we found that there are significant differences when comparing a hybrid transgenic commercial line against a landrace sample in the relative amplification of an internal control, a zein gene, included in the TaqMan® kit for the quantification of the 35S CaMV promoter sequence (see Fig. 1). This work was completed in 2005 using materials relevant to our publication (Piñeyro-Nelson et al. 2009). We also observed unexpected results such as high background levels when using ELISA kits for the Cry9c protein on landrace maize, but in this case open discussions with the manufacturer identified the problem and allowed subsequent kits to perform successfully in testing local landrace samples. Amplification curves in qPCR assays using the TaqMan® GMO Maize 35S Detection Kit (Applied Biosystems). A. 35S CaMV amplification and quantification was efficient for both the positive control, a commercial transgenic line (1) and a transgenic landrace variety (3); whereas the negative control, a commercial non-transgenic maize line (2) did not produce a proper amplification signal. B. Amplification and quantification of an endogenous gene (believed to be a zein gene) was efficient in both commercial transgenic and non-transgenic maize lines (1, 2), whereas the quantification in the landrace variety (3) was not possible (even though a PCR product was visible in the gel) perhaps, according to the manufacturer, as a result of a faulty interaction of the PCR product and TaqMan® probe. The challenges faced by a commercial approach to study landrace materials relate to: (i) the diversity of genetic backgrounds leading to expected inefficiencies and higher degree of variability in the PCR; (ii) uncertainty related to internal-standard variability; and (iii) sequence diversity even for expected target sequences. 1 We have already pointed out the expectation of much higher levels of molecular diversity in Mexican landrace samples with a diverse genetic background compared with hybrid, commercial varieties (Piñeyro-Nelson et al.2009). Significant genome size variation among landraces has been reported (1700 to 3300 megabases; see comment in Walbot 2008), while lack of genetic colinearity and pervasive gene duplication have been described (Fu & Dooner 2002; Wang & Dooner 2006). We stand by our expectation that such diversity could cause inefficiencies and variability in PCR results stemming from direct or indirect molecular effects on any of the components and conditions of PCR assays. In these conditions, a protocol with no flexibility for careful observation and follow-up of bands that are less than optimal would create ample opportunity for false negatives. 2 BS&JF dismiss any discussion of PCR inconsistencies by vaguely invoking an undefined and unaccountable protocol, thus: ‘[Genetic ID] includes at several points in its analytical procedures controls that would detect the kinds of problems cited by the authors and therefore ensure accurate reporting of results. For example, PCR inhibition tests are routinely conducted to rule out the presence of compounds (‘metabolites’) that could interfere with PCR amplification.’ Of particular interest is their claim of a standard, routine test for inhibition of the PCR assay, which should stand for any and all sources of inhibition possible from commercial and landrace materials that have a wide range of, for example, phenolic compounds in their constitution (Arnason et al. 1994); no details are given about the specific sequences used in such tests, gene dosage or specific genetic behaviour. Our own experimental routine shows this facile dismissal of the inhibition problem to be fallacious. Specifically, we showed as part of our careful method evaluation that there are indeed differences between commercial, hybrid maize varieties and landrace materials as far as their PCR performance is concerned. 3 In their critique, BS&JF deride our expectation of sequence diversity in our target sequences by claiming that such an expectation violates ‘the known and accepted norms of genetics’ (Schoel & Fagan, p. 3). BS&JF’s sole source of support is a general evaluation of the average rate of spontaneous mutation across broad taxonomic groups (Drake et al. 1998). This approach fails to recognize site-specific differences in mutation rates, especially well known in transgenic constructs where, for example, the borders of the transgenic construct are prone to sequence variation (Matsuoka et al. 2002). Maize itself has highly variable mutation rates at different loci, ranging from <0.1 × 10−5 to 49.2 × 10−5 mutations per gamete, while the equivalent rate for retrotransposons is a much higher value of 0.1 per genome per replication (Drake et al. 1998). Our sequencing data do confirm that there was indeed sequence diversity among the events of transgenic transformation that we were able to detect (Piñeyro-Nelson et al. 2009), which is also further evidence against the possibility of DNA contamination suggested by BS&JF. The fact stands that Genetic ID was unable to detect clear positive samples in various forms. Their standard methods failed to detect positives not only in landrace materials, but also in a non-commercial hybrid variety produced by CIMMYT and even in the well-documented Monsanto transgenic event NK603. Our own positive detection of transgenic sequences in the same samples through a variety of methods shows that they were present, so we must assume that some or all of the possibilities discussed above may be responsible for Genetic ID’s false negatives. As stated in our previous manuscript (Piñeyro-Nelson et al. 2009, p.758), specified on Genetic ID’s webpage (Verified on July 16th, 2009; http://www.genetic-id.com/About-Us.aspx), and commented to one of us by B.S. himself (personal communication to APN), the CIMMYT event was likely not detected by GID as a result of variation in the 35S CaMV promoter sequence; the very suggestion BS&JF now dismiss. Finally, it is clearly strange that BS&JF would decide to question the positive nature of our sample #5, which by their own standards showed specific PCR amplification of the 35S CaMV promoter with 100% or close to 100% in duplicate assays from independent DNA extractions. In the latter case, BS&JF appropriately call for non-PCR methods to strengthen the hypothesis that such a sample is positive, a surprising call as this standard does not seem to carry value anywhere else in their discussion of our study. While BS&JF focus on a critique of our PCR methods to entirely dismiss our study, we are convinced that the only way to resolve the basic quandary of whether transgenic sequences are found in Mexican landraces of maize or not, is not hypothetical argumentation, but independent evidence. We have produced non-PCR evidence which makes us feel strongly that our conclusions are fully warranted: (i) the sequencing of PCR products; (ii) DNA–DNA hybridization (Southern Blot); (iii) immunological detection of proteins derived from a transgenic sequence; and (iv) physiological tests of transgenic sequence expression, such as herbicide resistance tests. We have systematically confirmed the presence of 35S CaMV sequences in our samples, a strong sign of transgenic DNA presence. From sequence data we conclude again that contamination with a laboratory DNA sequence is highly unlikely because the sequences obtained over time and through an independent sequencing laboratory were not identical to each other. Southern Blotting is well known for its propensity to deliver false negatives, providing stringent, independent confirmation of PCR-based results. Our very consistent match between PCR and Southern Blot results leaves little reasonable doubt about the validity of our statements. In a recent study, independent samples were collected by researchers (George Dyer and Hugo different from the for Piñeyro-Nelson et al. samples collected by samples were independently by researcher in a different using type of an ELISA at detecting and The results from this study et al. again support our that are present in Mexican maize We further ELISA data for for the maize seed reported in Piñeyro-Nelson et al. out of the were positive for of these positives were distributed among the families scored as PCR positives for the 35S CaMV [see and in Fig. (Supporting information)]. In these positive individuals for were detected in localities in in and in among We also performed resistance assays and found positive individuals for these localities [see in Fig. (Supporting information)]. The latter data had not been included in the study under discussion (Piñeyro-Nelson et al. because in this we on assays at the DNA level that could be performed on the same and ‘negative’ for which we tissue for ELISA for All the evidence above consistently the critique by BS&JF. BS&JF state: would argue that the leaf sample provided by the authors did not contain the claimed NK603 event and, furthermore, does not contain material from any commercialized transgenic single plant.’ (Schoel and Fagan, 2009, p. such a challenge to our and we only that the sample in question obtained from a NK603 maize grown by us but extracted and by Genetic ID and for which positive results for and PCR were in Piñeyro-Nelson et al. BS&JF not only us but also several other by BS&JF that their from both on this is that results obtained to are not to whether of transgenic the Mexican maize has or has not as if there were only two et al. and the one where Bernd Schoel is the et al. they choose to dismiss the work of many and the fact that the of the presence of transgenic DNA sequences in landrace maize varieties in Mexico is and includes by & Chapela et al. Dyer et al. as well as a of and a of the see & of these include samples from the same considered in our study for and and are confirmatory of our results. for the one study for which Bernd Schoel is an and in which Genetic ID is again all of this of evidence BS&JF’s statements. We that is not by of but by and evidence, but BS&JF do not the evidence and of many and this we find as as it is of evidence has from many using a diversity of independent methods and often under the most (even see standards to that transgenic DNA sequences are present in of Mexican maize BS&JF are not to stand against such evidence but the value of their statements to support their is to test BS&JF’s statements using the standard methods of because they have not provided of many details in their methods as because of by their Genetic with limited we have that BS&JF their statements on a at many levels in of type II i.e. a which is permissive of false negatives. BS&JF seem convinced of the of their at Genetic ID, and do not see the to for it However, the value of a not in its but in the that it by independent confirmation or based on the of methods and that be used for commercial Genetic ID may have very good in their we to the is not The detection and of transgenic DNA in commercial and non-commercial is of environmental We from the experience of this discussion that such is under the conditions of high variability of the most delicate environmental In the case of maize, we see that methods that may be in highly homogenous should not be expected to work in the diverse conditions for example, in the of origin and of Most we conclude given the challenges for transgenic DNA in the should be performed by with a clear of good of This was by the Mexican for and and respectively, to and with various laboratory and with would to for technical support with qPCR for was provided by provided by through a Fig. S1 PCR assays conducted in different by Genetic ID on same samples. A. on from in Piñeyro-Nelson et al. 2009). B. in on in and to the same samples as in A. for analysis at Genetic ID leaf tissue from maize landraces of the maize samples were DNA extracted at this significant amplification variability both between of a given sample (see samples and in and and among of a positive (see in all Fig. ELISA tests performed on maize A. ELISA for and The two bands in the are a positive result for In the is the B. ELISA assay for detection of in maize landrace on the shows of the positive shows with are in and both clearly negative samples (see in the to and clearly positive samples (see in and to Fig. herbicide resistance assays. shows a leaf from a from which is shows a from herbicide was by within the in each S1 of ELISA assays for expression, conducted on leaf tissue of from the maize are not responsible for the or of any materials by the than should be to the for the The is not responsible for the or of any by the than should be to the for the