RA&RM of LMOs with Stacked Genes or Traits

The draft outline “stacked events” of 2009-06-09 needs revision in order to be in line as well with other widely accepted guidance and practices as with science based evidence:

• The use of terms should be consistent throughout the document e.g. “stacked (transformation) events.

• The term stacked (transformation) events should be defined in accordance with:
“Revised 2006: OECD Guidance for the Designation of a Unique Identifier for transgenic plants”.

• The CPB uses the terms “living modified organisms” and “modern biotechnology”. The term “transgenic” should be replaced accordingly.
(Comment: What is a “transgene? How does the term relate to the genetic cis-trans-test-method? How would such term cover the genetical engineering of deletions?)

• The inclusion of „multiple, but not necessarily different transgenes“ in the definition would cover also transformants with two insertion sites resulting from a single transformation and transformants with a repeat of the transforming DNA at a single insertion site.
Correction required.

• “The combination of transgenic traits via cross breeding may change the molecular organisation of the inserted genes…”.
If relevant for plants please give examples.

• “Conformation of the presences of the transgenic characteristics in the GM stacked event”.
Not relevant in the majority of case, because  F1-hybrids will be grown for one season only.
• “Having the same of similar promotors may lead to gene silencing…”.
In a number of gm plants the same promoter is present twice, in single events as well as in stacked events. But no silencing observed due to the promotors.
Please give example.

• “The genetic background … may influence not only the expression of the transgenes itself but also the metabolism of the transgenic plant”.
What is meant, please give example.

• “The combination of transgenic traits via cross breeding may change the molecular organisation of the inserted genes…
What is meant, please give example

• “Points to consider”
Why are the first three points mentioned specific for stacked events?
      Please give examples.

Thank you for developing the RA guidance for stacked LMOs and appreciate very much the suggested improvements for its structure and language in earlier postings.

RA for LMOs with stacked traits should focus on gene products and plausible interactions that would affect the risk assessments of the individual trait products.

Many traits will be combined using traditional breeding techniques. This process will have no affect on the safety of the derived product, and the safety assessment previously undertaken for the individual single trait products should be applicable for the combined trait product.  Redundant regulatory review of multiple dossiers for such combined trait products will not contribute to enhance safety for human or animal health, or for environment and therefore such products should not be subject to specific regulatory oversight.

When the traits have the potential to interact, further analysis may be required on a case-by-case basis to determine the impact of any additive, synergistic or antagonistic effects.

For LMOs with stacked traits produced by introducing at the same time in a single transformation event or by re-transforming a plant that contains an existing biotechnology-derived trait, these LMOs would be subject to the same regulatory data requirements as other transformation events.  Risk assessment for these stack events should be addressed in the Roadmap and not in the special guidance.

Any risk assessment to be done should be proportionate to potential safety concern arising from such LMOs and should not result to prolonged and costly regulatory procedures.

Sonny Tababa
CropLife Asia

I also believe that the assessment for stacked traits should focus on gene products for all risk pathways that would derive from products.

How do we recognise when or if a product of modern biotechnology will produce a gene product with an unintended or unanticipated property? For the most part, we extrapolate from information about known genes in donor organisms and how those products behave in that context. Then, we either assume that the new context will preserve those functions or we test to see if the same properties have been extended to the LMO. Since such information is extremely limited (because we do not have extensive functional studies on all gene products in many different genetic backgrounds), assumption-based assessments are of low reliability.

This is why at least in the context of food safety testing Codex says that comparators should not be other products of modern biotechnology. To use them as a baseline is to give them the status of having a history of safe use. To extend an assessment of safety to the progeny of two single event LMOs is to elevate the parents to organisms with a history of safe use. For the foreseeable future, I do not believe that a pre-assessment based on expectations of gene x gene or gene + gene x environment interactions in stacked products of modern biotechnology should replace actual testing.

Such pre-assessments rely on the intuition of the regulator or developer to make a case for the "plausibility" or "potential" of interaction, an intuition that I believe is unnecessary while scientific tests are available. If we can accept that LMOs "stacked" by virtue of "introducing at the same time in a single transformation event or by re-transforming a plant that contains an existing biotechnology-derived trait" would be subject to a full safety evaluation that would in theory detect unanticipated interactions, it does not seem necessary to exclude such an evaluation for only those that are combined by breeding.

In addition, risk pathways that might involve DNA level effects will require genomic information to assess. For example, if the risk pathway is dependent upon frequencies of homologous recombination, then surrounding sequence information may be relevant. If the risk pathway involves the production of unanticipated RNA products (through splicing or other processing pathways), then DNA level information can help to predict those. The transcriptome should also be interrogated for such possibilities.

It may not always be possible to assess harm pathways from a description of the gene products (e.g. proteins and translated or untranslated RNA) or the traits themselves. The two papers on testing dsRNA species as insecticides published back to back in Nature Biotechnology (Mao et al and Baum et al) describe a "trait" that would not be revealed by a phenotypic description of the LMO. In this case, the LMOs were created using a piece of rDNA that would be transcribed to produce a small RNA product. This product had the potential to be processed by either the LMO's endogenous siRNA pathways or the same pathways in insects that eat the LMO. They demonstrated very well that the RNA survives in the LMO tissues long enough to be ingested by the insect, then survives digestion by the insect gut to transfer to insect cells and there to be amplified independently of the rDNA, ultimately killing the insect. The actual "toxic" RNA product was not created in the plant. The large precursor RNA was, but the form necessary for activity in the insect was created after the RNA transferred to insect cells through food. As Baum et al. (p.1323 Baum, J.A. et al., 2007) noted, the “small amounts of dsRNA required for gene silencing and larval mortality suggest an amplification pathway in which ingested dsRNAs are processed to siRNAs, presumably within insect gut epithelial cells, which may prime the synthesis of more abundant secondary siRNAs”. What range of genes could these dsRNA molecules prime a silencing response to? Do they transfer infectiously to animals that might eat the insect? Are there other non-target effects?

In this example, the authors attributed the quantity of precursor RNA as the reason why they could manage to deliver an efficacious dose of RNA to the insect. If they hadn't been able to make so much precursor RNA in the original LMO, then the LMO would have processed the precursor into a potentially ineffective form. By saturating the endogenous processing pathways, precursor RNA survived to be processed in the insect gut cells.

This has two points of relevance for our discussion. One is that stacking even through breeding may result in quantitative and not just qualitative differences. The other is that certain traits would be invisible to us if we restrict our analysis to the phenotype of the LMO. A description of the transcripts from the transgenes or arising across the borders of insertion events might help to predict whether such novel dsRNAs might form and in what organisms and inform an assessment of potential harm that would be revealed in the phenotype of something else, human or otherwise.



Mao, Y.-B., Cai, W.-J., Wang, J.-W., Hong, G.-J., Tao, X.-Y., Wang, L.-J., Huang, Y.-P. & Chen, X.-. (2007). Silencing a cotton bollworm P450 monooxygenase gene by plant-mediated RNAi impairs larval tolerance of gossypol. Nat. Biotechnol. 25, 1307-1313.
Baum, J.A., Bogaert, T., Clinton, W., Heck, G.R., Feldmann, P., Ilagan, O., Johnson, S., Plaetinck, G., Munyikwa, T., Pleau, M., Vaughn, T. & Roberts, J. (2007). Control of coleopteran insect pests through RNA interference. Nat. Biotechnol. 25, 1322-1326.

Jack’s contribution [#1261] offers us much food for thought. Because there are disparate subjects to address, I've divided this into separate posting to help with clarity of the message.

First, his reference to Codex is a good reminder to this AHTEG and the Stacked Event SWG that we are not reinventing the wheel. Codex provides us with guidance and a precedent to addressing some of Jack’s questions in this posting. In particular, “How do we recognize when or if products of modern biotechnology will produce a gene product with an unintended or unanticipated property?” Codex addressed this clearly in Section 3, paragraphs 14-17 (Unintended Effects) of Guideline for the Conduct of Food Safety Assessment of Foods Derived from Recombinant-DNA Plants. http://www.codexalimentarius.net/download/standards/10021/CXG_045e.pdf

The main points made by Codex were:

(1)    Unintended effects are not restricted to products of modern biotechnology; they are “an inherent and general phenomenon that can also occur in conventional breeding”.
(2)    They have many sources at a DNA level that can ultimately affect levels of metabolites, formation of new metabolites, expression of existing genes, etc.
(3)    Unintended effects due to GM may be predictable or unexpected, and with “expanding information on plant genome and the increased specificity in terms of genetic materials introduced through recombinant-DNA techniques compared with other forms of plant breeding, it may become easier to predict unintended effects of a particular modification.” (emphasis added)
(4)    A variety of data creating a weight of evidence is “necessary to assess unintended effects”. “The assessment for unintended effects takes into account the agronomic/phenotypic characteristics of the plant that are typically observed by breeders in selecting new varieties for commercialization.”

Jack offers this SWG the same advice as Codex that molecular information may be helpful. However, Codex offers that solid guidance to rely on the history of safety associated with breeding and assessing for unintended effects using a weight of evidence approach. This recommendation likely applies to all SWGs currently operating under the AHTEG.

An important point for our work on stacked events is that conventional breeding is not necessarily safe, but it has a history of safety developed over time. In addition, we can use our knowledge of breeding to produce stacked events that meet criteria of acceptable safety (as safe as).

We will need to be cautious with regard to making any recommendations about requiring a course of molecular experimentation as part of a risk assessment when there is no clear relationship between the experimental outcome and the potential adverse effect. We should also keep open to the possibility that in the future some areas currently described as experimental science will become relevant to risk assessment. In this case, the risk assessor will have to have sufficient flexibility.

Reply to Tom Nickson

While Codex invokes phenotypes as part of a search for potential unintended and, most importantly, unanticipated hazards, it does not limit one to this methodology. For example, if you continue with Codex text after that which you quoted, it says “Molecular biological and biochemical techniques can also be used to analyse changes that occur at the level of transcription and translation that could lead to unintended effects.” That sentiment was reaffirmed in its 2008 document on GM animals, indicating that Codex has not softened on this point.

The results of molecular studies should be considered on a case-by-case basis, not whether or not molecular studies should be done. With other ambiguous standards (e.g. sound science, substantial equivalence), weight of evidence is highly dependent on judgment, or what I call intuition (and not disparagingly so, since I think intuition is a powerful source of creativity and hypothesis formation), and thus will vary from commentator to commentator and regulator to regulator. Weight of evidence conclusions are not themselves tested by science, but by the imagination of scientists or regulators.

I disagree with what I think is meant by the statement that “We will need to be cautious with regard to making any recommendations about requiring a course of molecular experimentation as part of a risk assessment when there is no clear relationship between the experimental outcome and the potential adverse effect.” We need instead to be cautious about how much we assume we know about the GMO and the environments it may plausibly be in.

I believe that strong hypotheses in safety testing are warranted and necessary for precaution, but to instead impose the standard of 'expectation of interaction that could lead to hazard' is simply to impose an assumption of safety based on the intuition of the assessor. Intuition at that end of the safety equation is based on how much one knows of the biochemistry of the transgenic organism and the environment into which it might placed, and on the pace of relevant science instead of the use of actual scientific method.

There are several things in posting #1261 that require clarification, correction and more discussion for them to be of utility to the SWG on Stacked Events. 

First, we need clarification in paragraph #4 on what is meant by “Such pre-assessments rely on the intuition of the regulator or the developer to make a cased for the “plausibility” or “potential of interaction”.  The process of developing hypotheses for risk assessment as practiced currently cannot be accurately described as intuition.  The current practice of risk assessment by developers and regulators involves careful consideration of all available scientific information (literature, experimental, and expert advice).  The decision-maker will evaluate its applicability and validity to the risk assessment using sound scientific principles.  In some countries this is done in accordance with Annex III.  My hope is that others will better see the importance of using scientific information from broad sources in postulating plausible hypotheses.  This is not accurately described as merely intuition.  The challenge before this SWG is to provide additional guidance beyond the Roadmap that will aid others in formulating relevant and testable risk hypotheses for Stacked Events.  If we can agree that in fact there are “scientific tests that are available” and come to a consensus that they, in addition to being available, are relevant to the task of meaningfully informing the risk assessment so much the better.

Second, posting #1261 needs to be corrected for a misstatement where it is stated that “The actual “toxic” RNA product was not created in the plant”.  This is incorrect.  DsRNAs produced by the plant cell are ultimately toxic to corn rootworm larvae, therefore the actual toxic RNA product was created in the plant.  This potential product works much as an insect protected crop (Bt) wherein the plant gene encodes a protoxin which is processed in the insect.  More importantly for this SWG is the inaccurate conclusion that “certain traits would be invisible to us if we restricted our analysis to the phenotype”.  In this case (Baum et al, 2007), the phenotype is insect protection.  It would thus be a focus of the risk assessment and reasonably plausible hypotheses would be formulated.

If posting #1261 is asserting that the risk assessment might miss unintended RNA effects, more discussion is needed.  What is the hypothesis that RNA would have potential adverse effects?  Work by Baum and Mao for RNA-based insect control taps into pre-existing processes present in natural systems.  Is there something about the transgenic stack that creates a hypothesis that one would not formulate based on our knowledge of conventional plants?

In conclusion, the Stacked Event SWG needs to do much more to define its scope to produce the points to consider.  One issue we must resolve is the disparate views on the need for and utility of molecular information in the risk assessment.  This will undoubtedly require more clarification and perhaps some corrections.

I hope that position of my reply will be appropriate.

First of all, I would like to define the “Stacked genes/traits”. Stacked genes/traits are provided by conventional hybridization between LMOs. They are not provided by multiple transgenic genes in single event. Accidentally occurring stacked genes/traits, such as natural outcrossing, will be also noticed for the future.
The risk of stacked genes/traits should be assessed by expressed character (phenotype) not by gene itself. Because stacked genes/traits do not affect on the biodiversity, when they did not appear as the phenotype.
Of course, we cannot ignore the possibility of introgression without appearing the phenotype. However, such phenomenon occur not only in stacked events but also in other LMO, non-LMO or naturally.
Therefore stacked genes/traits should be assessed as the following steps.
1)Is there any interaction in phenotype between stacked genes/traits?
2)If NO, we assess the effect of stacked genes/traits on biodiversity by the effect of each event, including further results of outcrossing/succession (see below on falling out of genes/traits).
3)If YES, we assess the effect in case by case. We can assess logically in some case, and we need to conduct a certain greenhouse/field trials. In the field trial, statistical approach, such as two-way ANOVA, is effective to analyze interaction among stacked genes/traits.
4)And if we cannot ignore the interaction in biodiversity view point, we need to conduct risk assessment in the similar approach for abiotic stress tolerance.

In case of stacked genes/traits, we have to consider to falling out of stacked genes/traits by outcrossing and/or succession with non LMO and the natural species, in case that the interaction exist among genes/traits. Because the probability is much higher than multiple transgenic genes in single event.
And interaction here is as follows.
a)The character which is expressed only when plural genes/traits exist, such as cascade reaction among enzymes
b)The interaction of chemicals, such as synergistic or antagonistic effect

Best regards,

I stand by posting #1261, and think that it is accurate. 

RE: “First, we need clarification in paragraph #4 on what is meant by “Such pre-assessments rely on the intuition of the regulator or the developer to make a cased for the “plausibility” or “potential of interaction”.

I address this point in post #1306 as well. Hypothesis generation is highly intuitive. Different researchers faced with the same range of experimental evidence and literature can invoke alternative models for future tests, and different experimental approaches to test similar models. To suggest otherwise is to deny the long history of contribution of the creative thinker (and not just the tinkerer) in science. I welcome intuition! However, to apply intuition or assumption-based reasoning to argue against experiments that might challenge a pre-existing expectation, which in biology is often based on incomplete knowledge rather than theory, is to use it against precaution in safety research, at least in my view.

I will assert again that in my view proper RA is based on solid and comprehensive hazard identification. At the molecular level, this must come from testing, not assumption. Sufficient flexibility in what regulators can ask for must be maintained. To argue from the viewpoint of mid-2009 what tests we could agree on is premature. If a regulator has a concern, then a test should be forthcoming to resolve it. Failing that, then the regulator must decide whether in his or her opinion the risk is worth bearing in the absence of confirming evidence.

RE: With regard to the statement that the “toxic RNA” was not created in the plant, I think that the evidence is clear.

Many toxic biochemicals are of low or no toxicity until modified in the exposed organism. The authors of the Monsanto study showed that a precursor dsRNA was produced in the plant, but it was its conversion in the cells of the insect that caused it to be toxic to the insect. Moreover, it may not even have been the original plant-synthesised dsRNA that maintained the toxicity; amplification of the RNAi-effect based on secondarily synthesised dsRNA molecules transcribed in the insect and not the plant were most likely the molecules that achieved efficacy. To date I am not aware that these secondary dsRNAs have not been described, but perhaps you could wonder down the hall and ask Baum et al to update us?

Similar lessons in biochemistry are quite common. That is why S9 fractions are used to assess genotoxicity (for discussion, see https://bat.genok.org/bat/?sp=html/practical_assessment/ch4_gmos_and_health/premarket_tox_allergen/genotox.html). The conversion of a pre-dsRNA synthesised in the plant to a toxic form in the insect is analogous to metabolic activation of protein or other toxins.

RE: “More importantly for this SWG is the inaccurate conclusion that “certain traits would be invisible to us if we restricted our analysis to the phenotype”.  In this case (Baum et al, 2007), the phenotype is insect protection.  It would thus be a focus of the risk assessment and reasonably plausible hypotheses would be formulated.”

I acknowledge that in this case it is the intended phenotype. Hence, I would expect that these plants would be tested for both target and non-target effects on insects. However, this particular example is illustrative and not exhaustive of the kinds of future products that may be even more clearly in need of hazard identification at the molecular level.

While in some distant way the effect of a plant on an animal could be seen as a plant phenotype, it is not routinely an aspect of phenotype characterising experiments and so the potential to cause an adverse effect would not be routinely detected by experiments that are commonly applied to describing the phenotype of the plant. For example, the observed agronomic qualities of plants will not indicate that it produces a novel allergen when people eat it. In this case, the plant has an agronomic quality that is observed only through its consumption, and possibly other toxic qualities that will vary with who consumes it.

Importantly, the effects will be specific to the consumer and thus only seen if tested in the consumer, not observed in the “phenotype” of the plant.

The plausibility of some part of the dsRNA molecule produced in the plant instigating an RNAi-like response in any consumer is made with the demonstrated effect of the plant-derived dsRNA on the target. That experiment showed that dsRNA is stable enough to persist though digestion in the animal, be taken up by cells and amplify a persistent silencing effect.

At present, the only way we could postulate whether there would be a human-specific toxicity is to either conduct feeding studies on humans or to describe all possible secondary dsRNA molecules that might be produced in human cells, followed by hypothesis formation on whether there would be matches between these and genes of human beings. We know from extensive literature that many dsRNAs cause silencing of many non-target genes. Thus, we may see the generation of multiple novel dsRNAs even in the plant for which there may not be an observed phenotype in the plant but may be in other consumers. These secondary RNAs may not have much in common with the instigating RNAs because of the ways in which dsRNAs may be amplified. See the model by Hutvágner, G. and Zamore, P. D. (2002). RNAi: Nature abhors a double-strand. Curr. Opin. Genet. Develop. 12, 225-232. for more detail.

Hence and again, we should be cautious about limiting safety testing hypotheses and embracing safety-assuring assumptions in their stead.

Thank you for developing the risk assessment guidance for stacked LMOs.
I believe the drafted outline “stacked events” of 2009-06-09 requires some revisions.
First of all the definitions in the outline are not consistent throughout the document.  For example the definition of the term "stacking" in the introduction is not in consistency with the definition of a plant with staked transgenes or traits. In addition to this, the definitions are too broad and encompass the transfer of even two copies a certain gene into one locus as a result of one single event. This also includes many of the transgenic crops already in the market which include at least one agronomically/economically important gene (such as a cry gene) and a selectable marker gene (such as an antibiotic resistance or herbicide tolerance genes).
What if more than one copies of a transgene is inserted in one locus and all of these copies are showing Mendelian segregation? What about a transgene and a selectable marker gene that are inserted in one event in one locus and are co-segregating? I believe these cases should not be considered under the term stacked transgenes.
I believe plant with stacked transgenes or stacked transgenic traits could be defined as plants that contain two or more transgenes, as a result of the consecutive crossing of two or more transgenetic plants with different transgene events, retransformation with a second construct and/or simultaneous transformation with different transgene cassettes (cotransformation).  Mere presence of a selectable marker gene and/or regulatory elements per-se in a plant is not sufficient to consider that plant as a plant with stacked transgene or trait.
In addition, some of the terms such as “transgene” itself, require a definition. Are the regulatory elements considered as transgene? What a bout deletion of genes?
What do we mean by multigene cassettes? Almost all the cassettes are multigene cassettes. The presented definition is too broad. Definition should not include the cassettes that include one gene plus a selectable marker or other genes that are technically required during the process of transformation.

Under the risk assessment also clarification is required. In cases where the parental lines containing an event have passed the regulatory procedures and risk assessment (including the molecular characterization of transgenic plant), “demonstration of the maintenance of the regions flanking the inserts and/or demonstration of the integrity/unchanged organisation of the single transgene loci in the StaEv LMO” are not necessary since the flanking region of any native genes are also subject to re-arrangement during cross breeding and recombination! Repeating   molecular regulatory characterization of multiple genes/traits  for such combined gens/traits will not add to the safety of the LMO in question and therefore such genes/products should not be subject to additional  regulatory obligations.  LMOs with stacked genes/traits should be subject to the same regulatory data requirements as other transformation events.  Risk assessment for these stacked events should not require a special guidance with significant difference compared to the original events.