Genetically modified plants in test-tubes 

EFSA dismisses off-target effects that concern scientists all over the world as "of very limited value for risk assessment". Report by Claire Robinson; technical advice by Dr Michael Antoniou

A new opinion from the European Food Safety Authority (EFSA) dangerously underestimates the risks posed by gene-edited crops and paves the way for their de-regulation – envisioning a future in which few or no safety checks would be carried out.

Below is our analysis of this opinion, divided into two parts: Part I is an overview for those who want the broad picture, while Part II is a detailed scientific analysis (with technical terms explained).


The EFSA opinion focuses on plants produced with three gene-editing procedures: Site-directed nuclease-1 (SDN-1), site-directed nuclease-2 (SDN-2), and oligonucleotide-directed mutagenesis (ODM).

In its opinion, EFSA’s GMO panel said it “did not identify new hazards” specifically linked to gene-editing technology, compared with either conventional breeding or genetic modification techniques giving rise to genetically modified organisms (GMOs). The current risk assessment guidelines for genetically modified (GM) plants were “sufficient”, it said.

Most worryingly, EFSA dismissed the importance of off-target effects resulting from gene editing, saying, "The analysis of potential off-targets would be of very limited value for the risk assessment".

EFSA also noted that less data for the risk assessments of gene-edited plants might be needed compared with older-style transgenic GM plants because of the absence of foreign DNA. This is a step towards actively weakening regulatory processes, seemingly in order to speed risky new gene-edited products to market.

Misleading comparison

EFSA justifies its de-regulatory approach to gene editing via a misleading comparison. It claims that gene editing causes fewer mutations than conventional breeding. But EFSA is not comparing gene editing with standard conventional breeding as it is most commonly practiced, in the form of cross-breeding and/or seed saving. Instead EFSA chooses to compare gene editing with a minority set of techniques that many regard as risky, even if not as risky as genetic engineering. That set of techniques is mutation breeding, which has been used alongside conventional breeding for several decades, but which is not in itself conventional breeding.

In mutation breeding, plant seeds are exposed to mutagenic agents such as radiation or chemicals to induce mutations (DNA damage) in the hope that one of the mutations induced may result in a useful trait (for a detailed analysis and references, see the book GMO Myths and Truths).

Most often, the induced mutations are damaging or silent (have no observable effect). Very occasionally, a mutation can produce an effect (such as the semi-dwarfing trait in barley) that is deemed desirable by the breeder and the mutant plant can be bred on to develop new varieties.
By comparing gene editing with chemical- and radiation-induced mutagenesis, EFSA is able to conclude that the former causes fewer mutations than “conventional breeding”. But through this apparent sleight-of-hand, EFSA is using one problematic procedure to justify another. It’s equivalent to allowing a child to be exposed to a toxic substance on the grounds that she will also be exposed to other known or suspected toxins. EFSA should instead use the most common method of conventional breeding – cross-breeding – as the comparator.

EFSA's approach threatens public health

EFSA's opinion is diametrically opposed to the approach to gene editing recommended by GMWatch and many other NGOs. In our collective view, the best way to protect public health and the environment is to keep the GMO regulations as they are but to issue stringent new risk assessment guidelines for gene-edited crops that recognise the risks and uncertainties posed by these products.

Such guidelines would include a detailed analysis of the consequences of off-target DNA damage (mutations) of gene editing and unintended on-target mutational effects (at the intended gene editing site in the genome) and what they might mean for health and environmental safety.

Our view is backed by a group of scientists who published a paper this year on risk assessment of gene-edited organisms. They wrote, "Many of the concerns associated with first-generation GMOs also apply to organisms developed through new genetic engineering techniques. However, genome editing techniques can cause additional, specific unintended genomic irregularities in many cases," meaning that risk assessment guidelines developed for first-generation GMOs will require revision and expansion to ensure they capture all hazards associated with genome-edited organisms.

These hazards, they warn, could include "toxicity and allergenicity".

But perversely, EFSA wants to keep the guidelines as they are, avoid looking at unintended effects, and ask for less safety data than is usual for GMO risk assessment. This places public health and the environment at risk.


Risks not confined to foreign genes or DNA

Those parts of the risk assessment guidelines that EFSA deems dispensable for gene-edited crops and foods relate to the risks of introducing foreign DNA or genes. Because SDN-1 and -2 don't intentionally do this, EFSA believes that these guidelines can be jettisoned.

But the idea that the risks of genetic engineering are restricted to inserting foreign DNA or genes – or that those risks can be significantly reduced by avoiding the deliberate introduction of foreign DNA or genes – is scientifically baseless. Numerous studies show that SDN-1 and -2 gene editing techniques – which do not deliberately introduce foreign DNA or genes – can still cause major genetic errors. Such errors could lead to changes in multiple gene functions, which in turn could lead to altered biochemistry in gene-edited food plants, with consequent health risks (toxicity and allergenicity).

The early warnings are already with us. In one study, the researchers tried to knock-out the function of a single gene using CRISPR gene editing in an SDN-1 procedure to improve the yield of rice varieties. Unexpectedly, they found that the editing process caused large insertions, deletions, and rearrangements of DNA. They also found that the majority of certain edited rice lines contained foreign DNA, as well as the Cas9 gene-editing tool. This foreign material had been introduced via the Agrobacterium infection method of introducing the gene-editing tool.

Unlike EFSA, the researchers did not dismiss their results, but learned from them. They warned, "Early and accurate molecular characterization and screening must be carried out for generations before transitioning of CRISPR/Cas9 system from lab to field". They added, “Understanding of uncertainties and risks regarding genome editing is necessary and critical before a new global policy for the new biotechnology is established".

EFSA failed to cite this paper in its opinion. Far from understanding the uncertainties and risks of gene editing, they have no intention of even looking at unintended effects because supposedly they are of "very limited" value to the risk assessment.

EFSA backs industry’s wish for product-based regulation

EFSA says, "The amount of experimental data needed for the risk assessment will mainly depend on the modified trait introduced".

EFSA is thus lending its support to trait-based or product-based regulation, which the agbiotech GMO industry has always wanted. It is based on only looking at the intended change in the GM plant but would miss the many unintended changes caused by the gene-editing process.

The EU’s GMO regulatory system is currently process-based as well as product-based – that is, it looks at the way that the GMO was developed. This enables regulators to know what to look for in terms of unintended effects and thus is a vital aspect of applying the precautionary principle.

Under product-based regulation, regulators would not necessarily be told which methods were used to develop the crop in question, so they wouldn't know what to look for in terms of unintended effects that result from the genetic manipulation process.

EFSA ignores studies showing unintended effects

The scientific literature is stuffed with an ever-growing pile of papers that discuss the off-target mutational effects of gene editing, as well as unintended outcomes at the intended gene-editing site ("on-target effects"). Clearly, the scientists who published these papers are convinced that unintended effects are important. The papers testify to the huge scientific resources that are being poured into trying to reduce such effects.

But EFSA isn't interested in them because it has already decided that they will be of "very limited" value to the risk assessment.

In its opinion, EFSA sets the tone for this "head in the sand" approach by simply ignoring the vast majority of scientific studies showing off-target and unintended on-target DNA-damaging effects of gene editing. This was not a problem of ignorance, as GMWatch and other groups submitted comments to the EFSA public consultation, detailing many key studies.

The inevitable conclusion is that these omissions were deliberate, allowing EFSA to reach a politically motivated decision to weaken the regulatory requirements around gene-edited plants.

EFSA airbrushes out unintended on-target effects

EFSA's handling of unintended on-target effects of gene editing (errors at the intended edit site) was especially inadequate. These types of effects pose a particular challenge to advocates of de-regulation of gene editing because they cannot be reduced by improving the precision of the targeting of the initial double strand cut made in the DNA by the gene editing tool. This is because unintended on-target mutations occur after the gene-editing tool has completed its task of producing a double-strand DNA break and results from the subsequent DNA repair activity, over which the gene editor has little or no control. Thus no matter how precise the initial DNA cut made by the gene-editing tool, genetic errors still happen as a result of the imperfect DNA repair processes of the cell.

For example, the study on CRISPR gene editing in rice varieties that found large insertions, deletions, and rearrangements of DNA located these unintended effects both at off-target and on-target locations in the genome. These types of mutations could cause changes in the plant's patterns of gene function, leading to an altered biochemistry that could result in the production of toxins or allergens.

The authors of the rice paper warned, "Understanding of uncertainties and risks regarding genome editing is necessary and critical before a new global policy for the new biotechnology is established".

But it seems that EFSA prefers not to examine these risks and uncertainties but to follow the philosophy of "Don't look, don't find." It appears to be blind to the issue of unintended on-target effects.

EFSA’s claim that gene editing causes fewer mutations than conventional breeding is baseless

EFSA believes that “The number of off-target mutations generated by SDN-based methods is lower than the number of mutations observed in conventional breeding due to spontaneous or induced mutations”.

Huge and unjustified assumptions underlie this statement (for a particularly misleading assumption see “Misleading comparison” in the OVERVIEW section above), including:

1) That the balance of mutations does break down in this way. There is very little research systematically comparing the numbers of mutations acquired in the same types of plant via gene editing versus mutation breeding. Not enough types of plant have been examined, and mutations have not been searched for with adequate screening methods.

2) That the randomness of mutations in mutation breeding equates to higher risk and that the claimed precision of gene editing (a precision that is illusory) equates to lower risk. As we note below (“Gene-editing changes not of same type as conventional breeding”), the reverse may be true. The apparent randomness of chemical- and radiation-induced mutagenesis may in fact be saving the plants from more severe genetic damage if the mutations were targeted to active gene regions.

Off-target mutations of gene editing will not be random but will be located in other gene regions of similar sequence to the gene being edited. Thus gene editing may have a higher risk of unintentionally altering gene function than random mutagenesis.

3) That the number of mutations is a determining factor when it comes to safety. It may be that the types of mutations caused by different techniques are what determines safety. In other words, it’s not so much a matter of how many mutations there are, but where they are located and what they do. We are not aware of any research in the plant biotechnology field that addresses this question. And EFSA presents no evidence that the mutations caused by gene editing are of the same type as those from conventional breeding and thus no more risky. In order to address this question, long-range PCR and long-read DNA sequencing would have to be carried out, as well as in-depth molecular analyses (via “omics” techniques) of gene-edited plants and long-term animal feeding studies to ensure that any biological effects are given the chance to show up.

EFSA’s evidence doesn’t stand up

EFSA cites three studies to back up its claim that the number of off-target mutations generated by gene editing is lower than the number of mutations found in mutagenesis breeding: Tang et al., 2018; Lee et al., 2019; and Li et al., 2019.

Is this really true? Do these three studies suggest that mutations caused by gene editing are fewer in number than in conventional breeding and thus unimportant for risk assessment? Let’s take a closer look.

Tang et al., 2018

This study analysed gene-edited rice plants, engineered with two different CRISPR editing tools, for off-target and on-target effects. While the study found that the CRISPR tools in themselves did not introduce many off-target mutations in the plants, it found many off-target mutations caused by other aspects of the CRISPR genetic manipulation process – namely tissue culture (200 per plant) and Agrobacterium infection. The Agrobacterium infection amplified the number of mutations over and above the number caused by tissue culture.

In comparison, saving seed from wild type non-GM rice plants (a widely practiced method of conventional breeding) resulted in relatively few (30 to 50) spontaneous mutations per plant.

Tissue culture is sometimes used in conventional breeding but when making gene-edited plants, it is virtually always used. And Agrobacterium infection is often used to deliver the gene-editing tool. Both tissue culture and Agrobacterium infection are known to be highly mutagenic, creating many genetic errors. Mutations resulting from the Agrobacterium infection – potentially including insertion of a foreign gene – will be added to those resulting from the tissue culture and the use of the CRISPR editing tool, meaning that the cumulative effect of all these procedures will be highly mutagenic, with uncertain outcomes.

Thus the inevitable conclusion of Tang et al.'s study is that the CRISPR process, taken as a whole, causes large numbers of off-target mutations – far more than conventional breeding. By only looking at one part of the process – the actions of the CRISPR editing tools – EFSA is following a narrow and reductionist approach that enables it to reach a reassuring – but baseless – position on the risks of gene editing.

Tang et al.'s study also shows that process-based regulation, which takes account of the inherent imprecision of specific techniques used in developing GM plants, is necessary.

Either the EFSA experts failed to understand Tang et al.'s paper, or they deliberately cherry-picked the findings to falsely imply that gene editing is as safe as conventional breeding. Either scenario should give us cause for concern, since experts working in a body charged with protecting public health should both understand the science and respond honestly.

Lee et al., 2019

This study looked at the effects of two CRISPR gene-editing tools in maize plants. Sequencing analysis showed that one of the tools, CRISPR/Cas9, caused no detectable on-target or off-target mutations in the plants that expressed the editing tool as intended. The other editing tool, CRISPR/Cas12a, didn't work as efficiently. The authors concluded, "The CRISPR/Cas9 system used in this study is highly efficient and specific for genome editing in maize, while CRISPR/Cas12a needs further optimization for improved editing efficiency."

Li et al., 2019

This study looked at the effects of CRISPR gene editing in cotton plants. The authors concluded that most variations following Cas9 editing were due either to mutations from tissue culture ("somaclonal variation") or/and pre-existing/inherent variation from maternal plants, but not off-target effects from the gene editing tool. As with Tang et al., 2018, the tissue culture process used to generate gene-edited plants is shown to be highly mutagenic and thus problematic.

Inadequate detection methods used

Each of the above three studies (Tang et al., 2018; Lee et al., 2019; Li et al., 2019) shows the problems and limitations of the gene-editing process taken as a whole, albeit that EFSA chooses not to see them.

Furthermore, all three studies will inevitably grossly underestimate the number of mutations resulting from gene editing. That’s because the authors used inadequate detection methods – short-range PCR and short-read DNA sequencing – to look for mutations.

As Kosicki and colleagues found in a study on human cells, short-range PCR and short-read DNA sequencing can miss major genetic errors, such as large deletions and insertions and complex rearrangements of DNA. This is because they only look at short stretches of the DNA around the intended editing target and predicted off-target sites.

Kosicki and colleagues concluded that long-range PCR and long-read DNA sequencing are needed to spot the full range of unintended mutational effects. As far as we know, these methods in combination have not been applied in any of the studies that look for unintended effects of gene editing in plants.

Gene-editing changes not of same type as in conventional breeding

In any comparison between gene editing and conventional breeding, it is not just the number of mutations that matter, but the type and the effect they have on the plant. EFSA seems to believe that where unintended changes occur from gene editing, they will be “of the same types as those produced by conventional breeding techniques" – a statement repeated from an opinion it published on a gene-editing technique (zinc finger nuclease 3) back in 2012.

EFSA provided no evidence for this extraordinary claim in 2012 and it provides none now, eight years later. That's probably because none exists.

Commenting, the London-based molecular geneticist Dr Michael Antoniou said, “Gene editing frequently causes large deletions, insertions, and rearrangements in DNA, which can affect the function of multiple genes at off-target and on-target sites.

“I would like EFSA to provide the evidence that compares the frequency of these types of large-scale DNA damage in conventionally bred and gene-edited plants, as I am not aware of any such studies.

“What we do know is that there is clear experimental evidence showing that assumptions that gene editing only causes small insertions and deletions at off-target and on-target sites are false.”

However, Dr Antoniou cautioned, “The size of genetic changes does not necessarily determine risk, since small genetic changes may result in dramatic and novel effects. For example, a small deletion or insertion following a gene-editing event could result in creating a new gene sequence, which can give rise to a novel mutant protein with unknown functional consequences. This is why all of the mutations caused by gene editing must be assessed on the basis of what they do, as well as what type and how numerous they are.”

Dr Antoniou says the mutations induced by gene editing will not be of the same type as those induced by conventional breeding or the older mutagenesis techniques (chemical- and radiation-induced mutagenesis). This is because the majority of the genome is non-coding and non-regulatory, so mutations induced through conventional breeding and chemical- and radiation-induced mutagenesis, being random, will more often occur in areas of the genome that are unlikely to affect gene function.

With gene editing, in contrast, mutations are not random but targeted. They are therefore more likely to happen at locations in the genome that have a similar sequence to the intended targeted gene. So they are more likely to affect important protein-coding gene regions and gene regulatory activity.

Some might interpret this as meaning that chemical- and radiation-induced mutagenesis is not risky, but this isn’t a reliable assumption – for why, see below, “EFSA minimizes risks of gene editing via misleading comparison”. What is more plausible is that the risks of gene editing versus those of chemical- and radiation-induced mutagenesis are different.

Also unimpressed by EFSA's claim that gene-editing errors are the same as those of conventional breeding was the Green MEP Tilly Metz. She accused EFSA of “muddying the waters” by trying to “brush off” inconvenient evidence. “Contrary to what they state, gene editing causes hazards that are new and different from conventional breeding,” she said.

Ms Metz's statement is backed by a scientific review that shows that unlike conventional breeding and older chemical- and radiation-induced mutagenesis techniques, gene editing makes the whole genome accessible for changes because it can bypass natural protections in the cell. EFSA fails to cite this review in its opinion.

EFSA minimizes risks of gene editing via misleading comparison

Throughout its opinion, EFSA compares the risks of gene editing with those of conventional breeding. Its definition of conventional breeding includes “certain mutation breeding techniques” that emerged before the EU’s GMO regulations were enacted. This refers to chemical- and radiation-induced mutagenesis techniques.

As we’ve seen, the risks of these techniques are not the same as those of gene editing and there is suggestive evidence that they may be less risky (see “Gene-editing changes not of same type as conventional breeding”, above).

However, chemical- and radiation-induced mutagenesis techniques are still genetic manipulation techniques and are recognized as such in the GMO regulations. Yet they are exempted from the requirements of the regulations because they are deemed to have a long history of safe use  – which doesn’t apply to gene-edited plants.

While chemical- and radiation-induced mutagenesis are used in conventional breeding, they cannot be equated to conventional breeding.

Also, and importantly, chemical- and radiation-induced mutagenesis is widely recognized as risky, unpredictable, and inefficient at producing plants with beneficial mutations. Some plant cells are killed by exposure to the chemical or radiation, while any resulting plants are often deformed and/or infertile (for references, see the book GMO Myths and Truths).

As we’ve noted above (in “Misleading comparison” in the OVERVIEW section), it is highly misleading to compare the risks of gene editing with those of chemical- and radiation-induced mutagenesis.

ODM has not been “successful”

EFSA claims that the gene-editing technique known as oligonucleotide-directed mutagenesis (ODM) "has been successfully applied to several crops like maize (Zhu et al., 2000), rice (Okuzaki and Toriyama, 2004) and oilseed rape (Gocal et al., 2015)".

It is surprising that EFSA claims success for this technique. The ODM herbicide-tolerant maize and rice crops were announced respectively 20 and 16 years ago, yet as far as we know, they have not been commercialised anywhere in the world, including in countries with few or no regulatory hurdles. This can hardly be defined as “success”.

Cibus's herbicide-tolerant oilseed rape has been commercialized, but the company this year claimed that it did not result from ODM gene editing but was the product of an accident in a petri dish (random mutation in tissue culture). This was a shock turnaround, as Cibus had previously described the canola as gene-edited, including in documents submitted to regulators, over many years.

If Cibus’s latest story is true and the canola is indeed a product of random mutation, the technique for which EFSA claims success is revealed as ineffective, imprecise – and entirely unsuccessful.

If Cibus’s story is false and the canola is in fact gene-edited, this is an argument for transparency on the part of developer companies, including full disclosure of the methods used to generate the crop. Yet EFSA, in favouring product-based regulation for SDN-1 and -2 gene-edited crops, is in effect supporting an absence of transparency.

Backcrossing inadequate to remove mutations

EFSA admits that "information on the off-target mechanism and frequency for ODM is quite limited", but doesn't see this as a reason to adopt a precautionary attitude to regulating ODM-generated plants. Instead it indulges in wishful thinking, claiming that "backcrossing steps which follow the transformation process would likely remove off-target mutations from the genome of the final product".

However, it is clear that the GMO industry has never taken care to do enough backcrossing (breeding the GM plant with non-GM parents or relatives) to remove off-target mutations from its GM plants.

This has led to GM crops with unintended new traits and altered composition being commercialised and widely grown – for example, NK603 maize and MON810 maize. In the case of the latter, the GM maize contained an allergen, zein, that was not present in the parent crop.

Clearly, the extent of backcrossing done by the GMO industry does not ensure removal of off-target mutations and other unintended traits.


EFSA's opinion on the risk assessment of gene-edited plants developed using SDN-1, -2, and ODM procedures grossly underestimates the dangers posed by these products to health and the environment, namely unexpected toxicity or allergenicity. The following shortcomings are apparent in EFSA's opinion:

* EFSA dismisses the importance of unintended effects from the gene-editing process, even though scientists across the world consider them so important that they are dedicating major resources towards trying to minimise them.

* EFSA ignores a large and ever-growing body of evidence of these unintended effects, even though they were alerted to the key studies by GMWatch and other groups during the public consultation.

* EFSA assumes that in cases of gene-edited plants where foreign DNA is not deliberately introduced, the risks will be lower. But the risks of genetic manipulation are not restricted to the use of foreign DNA. Unintended effects arising from gene editing include off-target and unintended on-target mutations that can alter gene function in unexpected ways, leading to altered biochemistry in the plant, potentially resulting in the production of new toxins and allergens. These possibilities need to be investigated in the risk assessment and must not be dismissed, without evidence, as being "of very limited value for risk assessment".

* EFSA cherry-picks the findings of studies investigating the effects of gene-editing techniques, emphasising their reassuring aspects while ignoring problems demonstrated with the gene-editing process taken as a whole (e.g. tissue culture and Agrobacterium infection-mediated introduction of the gene-editing tool).

* EFSA fails to acknowledge evidence showing that the methods generally used to search for on-target and off-target mutations from gene editing are incapable of detecting certain important classes of mutations (large deletions, insertions and rearrangements of DNA).

* EFSA compares the risks of gene editing with those of certain other risky genetic manipulation techniques that are sometimes used in conventional breeding – chemical- and radiation-induced mutagenesis – and implies that gene editing is less risky. This is a misleading comparison, comparing one problem with another in an apparent attempt to draw the conclusion that gene editing is less risky. Instead EFSA should use the most common method of conventional breeding – cross-breeding – as the comparator.

* EFSA falsely claims that the ODM gene-editing technique is "successful", even though it has apparently generated no successfully commercialised plant (if Cibus is to be believed when it claims its herbicide-tolerant canola is not gene-edited).

* It is only due to EFSA's flawed reasoning on all the above points that it is able to force the conclusion that less data is needed for the risk assessment of SDN-1 and -2 plants, compared with the data required for older-style transgenic GMOs. Its minimization of the importance of the unintended effects of gene editing and associated processes (tissue culture, Agrobacterium infection) enables it to conclude that the type of data that is required will depend on the intended modified trait – and ignore the unintended effects of the gene-editing processes used to develop the plant. These conclusions represent a dereliction of duty on the part of EFSA and put public health and the environment at risk.


EFSA’s latest opinion was sought by the European Commission, which tasked EFSA with assessing whether its previous opinion on a certain gene-editing technique (zinc finger nuclease or ZFN-3) applies to three gene-editing procedures: Site-directed nuclease-1 (SDN-1), site-directed nuclease-2 (SDN-2), and oligonucleotide-directed mutagenesis (ODM).

We recommend that the Commission:

* Acknowledges that there are specific risks linked to gene editing that are different from those of conventional breeding (including chemical- and radiation-induced mutagenesis) and ensure that all forms of gene editing (SDN-1, -2, and -3) remain under EU GMO legislation.

* Asks EFSA to draw up specific risk assessment guidance for gene-edited products that takes proper account of all unintended effects of the gene-editing process, even though EFSA says this is unnecessary. The Commission has only asked EFSA so far whether the existing guidance is applicable, not whether specific guidance is needed – which is, perhaps, an example of the dangers of asking the wrong question.

It is crucial that no one relies on EFSA's existing opinion as an objective analysis of the current state of the science. To do so would be to gamble with public health and the environment.