A large-scale field experiment was conducted by the UK-based Centre for Ecology and Hydrology (CEH) and commissioned by Bayer and Syngenta, to evaluate the impact on bees of neonicotinoid seed treatments applied to oilseed rape in Germany, Hungary and the UK. A closer look at this research shows that the conclusion in the published report is radically different from what the data actually reveal, illustrating the importance of avoiding bias or sensationalism – and letting the science speak for itself.
// A field trial in three countries was conducted by the Centre of Ecology and Hydrology (CEH) to study the potential impact of neonicotinoid seed treatments on foraging bees under field-realistic conditions.
// While 94 percent of 254 measurements found no significant effects, the press conference and news headlines sensationally claimed “neonicotinoid pesticides harm honey bees”.
// The claims suggest neonicotinoids cause reproductive harm to wild bees, yet a country-by-country review of the data show that neither bumble bee queens nor solitary bee reproductive cell production were directly affected by exposure to the neonicotinoid seed treatments.
// The honey bee study reported no consistent effects of seed treatments across all three countries on any of the primary or secondary assessment endpoints, but claimed evidence of “country-specific effects” on some primary assessment endpoints.
// A subsequent meta-analysis found that the few instances of alleged negative effects reported in the CEH study on honey bee colonies completely vanish once a correction for the initial strength of the colony is considered.
// This study raises important questions about how the need for sensationalism is challenging the traditional, deliberative process that has governed scientific research for many decades.
The neonicotinoids are among the world’s most widely-used insecticides, as a result of their many benefits:
// Excellent performance against destructive insect pests
// Higher worker safety due to lower human toxicity when compared to many alternative insecticides
// A key component of many modern integrated pest management programs
// The environmental exposure of neonicotinoid seed treatments are < 1 percent of that for spray application
// Systemic activity targets the pest, while avoiding exposures to many beneficial species
Since their introduction more than 20 years ago, neonicotinoid insecticides have become an integral part of modern pest management programs all over the world. Neonicotinoids are widely used in agricultural and residential settings because of their performance against destructive pests and their favorable human and environmental safety profile, when compared to many of the older products they replaced (INFOBOX 1, see below). Despite these significant benefits, neonicotinoids have been criticized by some who claim these products are harming honey bees and other pollinators. This has prompted extensive new research, as well as changes in the type of bee safety studies required by government regulators.
All pesticides are tested to assess their safety to bees and no pesticide class has been tested as extensively as the neonicotinoids. Much of this research involves studying individual bees in the laboratory or other artificial environments. While these tests are important indicators of a product’s potential toxicity or hazard, they do not simulate field conditions nor do they provide realistic data on the exposure that might occur during a typical agricultural application. Instead, higher-tiered field studies are needed, to represent realistic potential pollinator/pesticide interactions at the honey bee colony level. Higher-tiered tests require intricate planning (and a little luck) to account for the influence of numerous naturally-occurring variables, such as landscape and weather. In conducting comprehensive field tests to ensure a product’s safety, Bayer often works with independent scientists. Regardless of who conducts the study, care must be taken to ensure the study design properly segregates and replicates experimental treatments to minimize the potential for research bias.
Honey bee colonies at the Thames Valley UK site.
In 2014, Bayer and Syngenta commissioned the Centre for Ecology and Hydrology (CEH) to conduct large-scale field studies in Germany, Hungary and the UK to examine the impact of two neonicotinoids (clothianidin and thiamethoxam) on honey bees when applied as a seed dressing to oilseed rape. Although the basic design was reviewed and agreed upon with Bayer and Syngenta, the study’s implementation, execution and analysis were the primary responsibility of CEH. In addition to the company-commissioned research, both companies approved CEH’s request to conduct a separate wild bee study to evaluate potential effects on a solitary bee (Osmia bicornis) and a bumble bee (Bombus terrestris) species using the same test locations.
“This was one of the largest, most comprehensive studies that we’ve ever envisioned, so we had to have excellent communications to ensure its successful completion.”
Dr. Mark Miles, Senior Pollinator Safety Expert, Bayer Global Ecotoxicology Department
The CEH research examined the potential effects of exposing honey bees, bumble bees and wild bees to commercially grown winter oilseed rape treated with one of the two neonicotinoid seed treatments, by comparing them to non-neonicotinoid treated (control) oilseed rape. Observations were made during the 2015 growing season and were extended to include assessments of overwintering survival of honey bee colonies in the spring of 2016.
Close up of bumble bee colonies showing mesh cage to protect the bees from predators.
Ironically, despite its size and complexity, the CEH study does not come close to satisfying the 2013 Bee Guidance Document proposed by the European Food Safety Authority (EFSA). The principles of this document served as the basis to the use restrictions of three neonicotinoid compounds in the European Union. The implications of this were discussed in a previous edition of BEEINFOrmed1. The European Union Member States have still not ratified this guidance document due to concerns about its practicability and the agricultural industry has continued to argue that EFSA’s guidelines are inadequately defined and unfeasible from a practical perspective.
Nevertheless, it was hoped that this large-scale study would come as close as is realistically possible towards satisfying the bee guidance requirements and provide new information on the potential impact of neonicotinoids to bees, when used under field-realistic conditions.
Prior to the study’s implementation, the companies and the researchers worked together, to ensure the development of a comprehensive study plan that would achieve the research objectives. Prior to its implementation, company representatives overseeing the CEH research believed things were well in hand. Regular contact between the sponsors and the research team suggested that the final study would provide a definitive assessment of how pollinators respond to neonicotinoid seed treatments in oilseed rape under realistic field conditions.
A wild solitary bee on oilseed rape flowers; there was a thriving natural population at the sites.
Wild bumble bee (Bombus terrestris) foraging on off-field vegetation.
Dr. Mark Miles, senior pollinator safety expert at Bayer’s Global Ecotoxicology Department, was well aware of the challenges the CEH research team faced in implementing the study. “Conducting a large-scale field study in one country is difficult enough, but working across three countries is extraordinarily ambitious,” he notes. “This was one of the largest, most comprehensive studies that we’ve ever envisioned, so we had to have excellent communications to ensure its successful completion.” Although frequent discussions between sponsors and scientists are normal in commissioned research, outside interest groups questioned industry’s role in the process and aggressively criticized the study and its lead researcher. “Once these groups became publicly involved, our correspondence with the CEH research team began to suffer,” noted Miles.
1 BEEINFORMED N° 4_2017
Title: ‘Neonicotinoid Restrictions in EU miss Target of protecting Bees’
Although Bayer and Syngenta were not directly involved with the CEH research during the implementation phase of the study, they were provided with an interim study report, which showed no discernable effects on bees exposed to the neonicotinoid treatments. Despite these encouraging findings, there were signs from the earliest stages that at least part of the research may have been compromised. A fundamental concern centered on the small colony size in the UK, which failed to meet the minimum EFSA requirement of 4,000 honey bees per colony. “We thought we had an agreement that the UK component of the study should not be continued due to the poor health of the colonies,” said Miles. “But by this time, our communication had deteriorated and the research continued nonetheless.”
Entrance to a bumble bee nest with two bees, one leaving and another guarding.
Before the results of the CEH study were published in Science Magazine and announced at a highly-publicized press conference, the company sponsors received various draft versions of the manuscript but had relatively little discussion with the study authors. By the time that a copy of the final report was provided, the sponsors were surprised by the study’s overall conclusion. While there was no consistent evidence of seed treatment effects in any of the three countries regarding the primary endpoints (e.g. colony strength, forager mortality, and overwintering success), or the secondary endpoints (e.g. honey bee behavior and disease susceptibility), the CEH report concluded that there was evidence of “country specific effects” on these endpoints.
In addition to their remarks regarding the honey bee study, the CEH scientists also commented on their parallel research project involving wild bees, in which lower reproductive success was reported in the solitary and bumble bee species examined. Summing up both studies, the lead author, Dr. Ben Woodcock, concluded “The neonicotinoids investigated caused a reduced capacity for all three bee species to establish new populations in the following year, at least in the UK and Hungary.”
Based on the interim results that were provided to the sponsors, the final CEH conclusion on honey bees was certainly a surprise to the Bayer team that had been following the study’s progress. And while Bayer was not privy to the research details regarding the wild bee study, the conclusion that these species were affected by the neonicotinoid treatments was unexpected too, given the results seen in previous field trials. Clearly, a closer look at the CEH study data was needed.
Differences between treatment and control groups
In the UK, an unusually strong infestation of Varroa mites caused the initial colony strength in both groups to fall below EFSA’s minimum requirement of 4,000 bees per colony.
Along with its publication in Science Magazine, the CEH authors issued a news release with a sensational headline claiming that “neonicotinoid pesticides harm honey bees and wild bees,” which was carried by many media outlets following the London press conference.
Soon after the report’s release, however, a serious discrepancy between what was announced and what was actually observed in the research became apparent: the study data and hundreds of pages of supporting information not included in the original manuscript suggested that the overarching conclusion was wrong.
A closer look at the study data found no consistent differences between the control and treated colonies. Out of 254 different parameters measured to assess honey bee colony performance and health, 238 (94 percent) showed no significant differences between the treated and non-treated groups. Interestingly, nearly three percent of the treated groups performed better than the controls and about three percent of the treated group performed worse. That the study authors drew their primary conclusions from a selective sliver of data is strange, considering that the small percentage of seemingly negative effects could easily be explained by random statistical chance alone (also known as ‘falsenegatives’), just as the ‘positive’ effects are equally likely to be false-positives. A clear-eyed assessment of the data might only conclude that there were no significant differences in the health of honey bees exposed to neonicotinoid seed treatments under field-realistic conditions.
Bumble bee data for the UK and individual bumble bee data for Germany and Hungary.
The Coefficient of Determination (or R2 ) is a statistical term used by scientists to measure the predictability of a relationship – in this case the relationship between neonicotinoid residues in the nest and the number of bumble bee queens.
R2 values range from 0 (no predictive relationship) to 1 (highly predictive relationship).
The very low R2 values in the UK, Germany and Hungary suggests that the health of bumble bee queens is not affected by the presence of neonicotinoid residues found in the nests in any of the test locations.
Solitary bee nests in the ground. Around 80 percent of wild bee species are soil-nesting.
A similar discrepancy between the study conclusions and the actual data was found for the wild bee species investigated. Despite entitling the paper “Country-specific effects…”, the authors chose to aggregate data from across the three countries in this case in order to demonstrate an apparent effect. Contrary to the report’s conclusion of reproductive harm, neither bumble bee queens nor solitary bee breeding cell production were directly affected by exposure to the neonicotinoid seed treatments. Lacking consistent, statistically significant effects from direct head-to-head comparisons, the authors pooled the data across all three countries to bolster their conclusion that neonicotinoid treatments harmed wild bees. When viewed on a country-by-country basis, however, Figure 1 clearly shows that there is no correlation between bee reproduction and the amount of neonicotinoid residues found in the nests of wild bees. This was especially true in the UK bumble bee analysis, where the statistical probability of predicting a cause-and-effect relationship was essentially zero.
An Osmia nesting unit on site.
If there are environmental circumstances that are so important for honey bees, why are they not for wild bees?
Even more troubling are the inconsistent statistical approaches reported in the CEH report to assess the impacts of seed treatments on honey bees and on wild bees. In the case of honey bees, the authors strongly emphasized proposed “country-specific responses” to account for environmental variability between countries. However, in the case of wild bees they used pooled data collected from all of the countries to find statistical differences where none were found within the individual countries. This raises a fundamental question: if there are environmental circumstances that are so important for honey bees, why are they not also important for wild bees? Conversely, if environmental circumstances are not important for wild bees, why are they so important for honey bees?
The glaring contrast between what was reported and what was actually observed in the CEH study suggests that some element of bias may have been introduced. One of the most pervasive forms of bias in research is ‘confirmation bias’, which occurs when researchers form a hypothesis and weigh responses that confirm their hypotheses as relevant and reliable, while minimizing or dismissing evidence that doesn’t support it. Some degree of bias is almost always present in any published study and can occur at any phase of research, including study design or data collection, as well as in the analysis of data and in the process of publication. Bias alone does not negate the benefits or contributions made by a study but it is a reason why all scientific research should be subject to review and reinterpretation.
Following the receipt of the final CEH report, Bayer conducted its own review of the data and evaluation methods. Although the overall study design was by mutual agreement prior to its implementation, many of the experimental details were left to the CEH research team to manage. According to Dr. Linus Goerlitz, Head of Computational Life Sciences – Agronometrics at Bayer, “the CEH study was appropriate from a scientific and statistical point of view. But in our opinion the researchers failed to consider some very important study elements when drawing their overall conclusions.”
Goerlitz comes to this conclusion after the Bayer team not only reviewed the CEH study methodology but also conducted a meta-analysis of data collected from in-house historical field trials. “We analyzed 18 Bayer field studies involving clothianidin conducted over the past 15 years, to identify the key environmental influences affecting colony strength,” said Dr. Kathrin Hatz, also from the Computational Life Sciences – Agronometrics group, and part of the Bayer team. “In fact, we examined 53 such studies, but many of them were either short-term acute studies or conducted outside of Bayer and, unfortunately, we were unable to secure permission to access the raw data from the external scientists to use in our meta-analysis. While we would have liked to examine more studies, the studies we had in-hand provided more than enough information to make some very interesting assessments of the CEH methodology.” The team identified two factors in particular – the initial honey bee colony strength and the air temperature at the study sites – which were found to be critically important in properly assessing the honey bee field study treatment effects.
The bee colonies delivered to the UK site (left) were in poor condition; here you see the lack of bees (middle) and multiple dead bees at the bottom of the box (right).
The meta-analysis revealed that when conducting a field trial, the pre-exposure strength of a honey bee colony is one of the most important drivers affecting a colony’s health development. From a practical research perspective, any honey bee field study aiming to differentiate the effects of pesticide exposure on colonies from other environmental influences has to adequately consider this and other factors in the test design and the subsequent analysis. This did not occur in the CEH study.
The importance of the hive size at the start of the experiment can be clearly seen in the CEH study results. Colony strength at study initiation between treatment and control plots showed substantial differences across countries. In the UK, colonies at treatment plots tended to be larger than at control plots; in Hungary the opposite was observed. The largest difference between initial colony strength at control and treatment plots was found for Germany. These differences remained throughout the course of the study, resulting in the initially stronger colonies outperforming the weaker ones up to the end of the assessment period. Furthermore, in the UK, an unusually strong infestation of Varroa mites caused the initial colony strength in both groups to fall below EFSA’s minimum requirement of 4,000 bees per colony and raised concerns that the field study was in jeopardy. “We voiced our concerns that small size and poor health of the hives in the UK made any further evaluations problematic, but the research continued,” noted Miles. “The high mortality in the treatment and control colonies made it impossible to draw any meaningful risk conclusions.”
In reviewing the data from the CEH study, the Bayer team found that even on the few occasions where significant differences in colony health were observed between the control and treatment groups, these differences were probably caused by the differences in initial colony size.
“The importance of initial hive size has not been adequately addressed in the scientific literature, but our research shows it is an area that should be considered before conducting any future field studies,” said Goerlitz. Adding, “Without this understanding, what at first may appear to be a treatment effect may in fact be an artifact of the experimental setup.”
The Bayer team has submitted a manuscript regarding its meta-analysis research and is awaiting approval for publication in a peer-reviewed journal.
Most real estate agents will usually cite ‘location’ as a primary consideration of any new home buyer. Similarly, setting up a field experiment requires knowledge of the landscape in which the trial is to be conducted, to minimize the environmental variability between treatment sites. No field study will ever be perfect because of the inherent difficulty in managing a complex ecosystem, where temperature, location and weather can all exert significant effects. A deeper look at the CEH report shows that conducting a study across three different countries presented its own set of challenges.
The field study was designed using three to four replicates (‘blocks’) of approximately 40 hectares in each country, with each block consisting of a non-treated oilseed rape control, a clothianidin oilseed rape seed treatment and a thiamethoxam oilseed rape seed treatment. This setup, known as a ‘split-block’ design, is based on the assumption that the key environmental factors within a block are more similar than they are between blocks. When properly constructed, a ‘split-block’ design minimizes the natural variation within a block, so that the potential treatment effects can be clearly observed. In practice, the removal of all variability within a large experimental block is very difficult to achieve, however.
At several sites in the UK, the ‘within-block’ variance in e.g. air temperature was greater than the ‘between-block’ variance. For example, the clothianidin site in Block 2 was actually much closer to the sites in Block 3 than it was to the other sites in Block 2, suggesting that the geographical or environmental variability including air temperature within the block was not uniform. And in Hungary, a clear separation between blocks was altogether missing. As the purpose of randomized blocking is to ensure treatment effects can be segregated from other variables, it appears the goal of the split-block design in the CEH study certainly was not fully achieved even in important influential variables.
Map UK trial sites
Sites 22, 23 and 24 were considered by CEH to be in the same block, however site 24 is actually closer to sites 31, 32 and 33 which are considered to be a totally different block. The confounding effects of using blocking in this way in the statistical analysis does not take into account the actual difference or similarities between the sites.
Map Hungary trial sites
In Hungary there was considerable overlap of study blocks and a distinct lack of separation of sites. For example sites 13, 14, 15 are considered to be one site and 16, 17, 18 are another site whereas it is clear that there is virtually no geographical separation to support this. Site 11 can be seen to be at a large distance from other sites considered to be in the same block (sites 10 and 12) and there is overlap with the sites of another block (sites 19, 20 and 21). The location of the sites cannot be attributed to the blocks in a meaningful statistical analysis as conducted by CEH.
Map Germany trial sites
In Germany the three blocks are well separated which allows for a statistical analysis based on a randomised block design because different conditions between blocks and similar conditions between the sites (i.e. within blocks) are more likely. Overall this layout is the most suitable for the type of statistical analysis performed by CEH.
The meta-analysis shows how certain factors, such as initial hive size and randomized block design can have a profound effect on a honey bee field study. While the CEH study did not fully consider these factors, that does not diminish the quality or value of the research itself. Indeed, it highlights the nature of scientific discourse, in which theories are tested and retested until it brings scientists closer to uncovering the truth. Understanding the importance of hive size, for example, would be a valuable consideration for scientists planning a future honey bee field trial.
But a fundamental question remains unanswered: Why, in the face of substantial and contradictory evidence, did the CEH manuscript ignore 94 percent of the data showing no significant correlations, to make the claim that neonicotinoid seed treatments were harming bees?
Since the number of positive findings was similar to the number of negative findings, the study could have just as easily reported that neonicotinoid seed treatments improve bee health. Claiming that exposure to neonicotinoids helps bees is dubious, especially considering that only three percent of the observations support this conclusion. So why did many in the media readily accept the CEH overall conclusion that neonicotinoids harm bees, which was based on a similarly small percentage?
The pressure to ‘publish or perish’ has long plagued many researchers, especially those who prefer the deliberate and often painfully slow process that defines scientific research. While there are few statistics on this topic, many scientists have complained about the propensity of some journals to accept studies that place more emphasis on sensational results instead of rigorous methods or processes. A few journals, such as PLOS One, are trying to avoid this publication bias by being more accepting of studies in which the results are either statistically inconclusive or that found no observable adverse effects from exposures to treatment.
However, the temptation to publish studies that make good headlines is all too real. Scientists who do so are often rewarded with widespread media attention and public recognition even if the findings are suspect, or inconsistent with other work. Unfortunately, there is little interest among many researchers or publishers in replicating or verifying results because it is a time-consuming process that is perceived to have little ‘impact’ value. According to a recent poll of 1,500 scientists across multiple disciplines, 70 percent of them failed to reproduce at least one other scientist’s experiment.
Clothianidin scored less than 0.5 on the quantitative weight of evidence (QWoE) scale of relevance to adverse effects (0 = weak / 4 = strong), suggesting that normal agricultural uses have no significant effect on honey bee colony health. Dots of different colors stand for different routes of potential exposure of bee colonies to the substance.
1Solomon & Stephenson (2017). Quantitative weight of evidence assessment of risk to honeybee colonies from use of imidacloprid, clothianidin, and thiamethoxam as seed treatments: a postscript. Journal of Toxicology and Environmental Health, Part B.
Expo: Exposure Only
Fld: Effects Measured
Some studies measured only the amount of residue exposure (Expo), while others measured the effects on colonies (Fld)
A large-scale field experiment in Germany, Hungary and the UK, to quantify the potential impact of neonicotinoid seed dressings in winter oilseed rape on honey bees and wild bees, was published in the summer of 2017 by the Centre of Ecology and Hydrology. While 94 percent of the data comparisons found no differences between treatment and control, the published report and press release sensationally claimed that neonicotinoid pesticides harm honey bees and wild bees. A closer look at the data, found little to support this claim and highlights the need to let science speak for itself.
Not all scientific studies dealing with a specific topic are of equal strength, quality and relevance when looking to address a particular question. This is also the case in answering the question of whether neonicotinoids harm honey bee colonies under field-realistic conditions.
The best way to go about answering such a question, especially when there is a complex data set with a huge number of seemingly conflicting information, is by collecting the available data which could provide useful information and by evaluating individual studies on their strength, quality and relevance to make conclusions based on a weight of evidence approach. Weighing data according to their strength and relevance involves a review of the experimental methods and assessment endpoints (i.e. the parameters evaluated in the respective study) to determine if the study outcome is reflective of normal agricultural practices or is only related to an artificial experimental design. Comprehensive reviews of research conducted over the past 20 years have largely concluded that risks to honey bee colonies are negligible when neonicotinoids are applied under field conditions.
In a recent series of publications, University of Guelph scientists used a quantitative weight of evidence (QWoE) approach to examine the effect of various neonicotinoids on honey bee colony health. In their words, the scientific community has been without a “defined and transparent process for integrating different and sometimes conflicting sources of information” to reach a definitive conclusion regarding the actual risk of these substances. Using this approach, the authors weighed all the available studies according to their suitability, to predict adverse effects under practical conditions.
The Guelph researchers included all available papers or reports in their QWoE analysis, unless there were insufficient data to allow the quality to be assessed. This inclusive process assumes that even studies of lesser quality or of indirect relevance with regard to the question of effects under realistic exposure conditions can provide important insights on the topic. Since some studies have more predictive power than others, a final quantification step enabled the authors to properly score each study based on its overall quality and relevance.
After using a QWoE approach to examine all available higher-tiered studies involving imidacloprid, clothianidin and thiamethoxam, the researchers concluded that the “use of these insecticides as seed treatments under good agricultural practices did not result in harm to honeybees at the level of the colony”. Contrary to the claims of anti-pesticide groups and many sensational headlines, there is strong evidence that agricultural uses of neonicotinoids have no relevance to honey bee colony health, regardless of the route of exposure. In fact, Figure 2 shows that the evidence to suggest clothianidin will harm bee colonies under realistic field conditions is practically nonexistent. This conclusion is consistent with nearly every other major review of field studies involving neonicotinoids.
There is a growing concern that the trend to popularize research by emphasizing sensational headlines or messages may cause long-term harm to the objective, deliberative process of scientific research.
Take this example: Science Magazine published a report by Mitchell et al. 2017, entitled ‘A worldwide survey of neonicotinoids in honey’, which claims the widespread use of neonicotinoids “may increase harm to pollinators.” The scientists, some of whom have a well-known bias against neonicotinoids, collected honey samples, volunteered by amateur beekeepers, in which they found small traces of neonicotinoid residues. The study perpetuates two common misperceptions the editors of Science Magazine ought to have questioned.
When looking at the study they should have considered that:
// All neonicotinoids are not alike
The study makes no attempt to differentiate between neonicotinoids, even though the highest levels of residues came from acetamiprid and thiacloprid, compounds of such low honey bee toxicity that they can be safely sprayed directly on flowering plants – which may naturally lead to more detectible residues in honey.
// Detecting residues does not imply danger
The detection of trace residues in nectar and pollen in treated crops, in itself, is neither surprising nor alarming. The level of residues found in the study is far below that which could potentially harm honey bee colonies – a result that has been demonstrated in study after study – and which is far below human safety limits, as well.
The findings from Mitchell et al. that neonicotinoids are widely used and that insignificant and harmless traces of residues may be found in honey samples ought to be reassuring to the general public. However, we suspect that such reassurance is not what the authors or Science Magazine intended.