Conflicts of Interest in GM Bt Crop Efficacy and Durability Studies

Public confidence in genetically modified (GM) crop studies is tenuous at best in many countries, including those of the European Union in particular. A lack of information about the effects of ties between academic research and industry might stretch this confidence to the breaking point. We therefore performed an analysis on a large set of research articles (n = 672) focusing on the efficacy or durability of GM Bt crops and ties between the researchers carrying out these studies and the GM crop industry. We found that ties between researchers and the GM crop industry were common, with 40% of the articles considered displaying conflicts of interest (COI). In particular, we found that, compared to the absence of COI, the presence of a COI was associated with a 50% higher frequency of outcomes favorable to the interests of the GM crop company. Using our large dataset, we were able to propose possible direct and indirect mechanisms behind this statistical association. They might notably include changes of authorship or funding statements after the results of a study have been obtained and a choice in the topics studied driven by industrial priorities.


Introduction
Despite the extraordinary financial, political and ideological stakes relating to genetically modified (GM) crops, we were able to identify only two research studies [1,2] focusing on conflicts of interest (COIs)-defined as "a set of circumstances that creates a risk that professional judgment or actions regarding a primary interest will be unduly influenced by a secondary interest" [3]-in this field of research. These two studies reported similar frequencies of COIs in research articles on GM crops, but differed markedly in their conclusions. Sanchez [1] concluded that COIs were not frequent enough to be a serious issue in GM crop studies, whereas Diels et al. [2] warned that COIs might be a problem because they were significantly linked to study outcomes. Unfortunately, one of these studies was potentially subject to COI itself [1] (the author's affiliation was ChileBio, http://www.chilebio.cl, which is funded by companies that develop GM crops) and the other included a limited number of articles [2] (n = 94).
We investigated the occurrence of COIs in GM crop studies and their association with the outcome of those studies further, by analyzing 672 full-text articles focusing on two topics: the efficacy and durability of GM crops producing Bacillus thuringiensis (Bt) toxins-often referred to as Bt crops-against target pest insects (see the flowchart of article selection in Fig 1 and S1 Table for a list of all the articles, with the associated classifications).
The efficacy and durability of Bt crops are defined here as the level of pest control provided by the Bt toxin they produced or by the crop themselves and the sustainability of this pest control over time, respectively. We used the COIs created by ties with GM crop companies as a model of COIs for this study because the best interests of these companies was easy to characterize (efficacy and durability of the crops) and COIs were easy to detect (affiliation of an author to a GM crop company or part of the funding provided by a GM crop company).

Goal of this study
We explored the occurrence of COIs and the association between COIs and outcomes in studies of genetically modified (GM) crops. We focused our analysis on articles evaluating the efficacy or durability of GM crops producing toxins (Cry and Vip proteins) from Bacillus thuringiensis (Bt) against targeted pests. These articles included studies on the main Bt crops developed by Monsanto (e.g. maize and cotton varieties expressing the MON810 and MON1076 events, respectively), Syngenta (e.g. maize varieties expressing the Bt11 event), Dow AgroSciences and DuPont Pioneer (e.g. maize varieties expressing the 59122 event).
We used the COI created by the professional affiliation of authors to GM crop companies or the provision of funding for the study by these companies as the model of COI for this study. We considered only COIs linked to GM crop companies because the best interests of these companies were easy to characterize (efficacy and durability of crops) and the presence of such COIs was easy to detect (an author was affiliated to a GM crop company or some of the funding came from a GM crop company). The COIs investigated here correspond therefore to financial COIs. Non-financial COIs [4], also known as intrinsic COIs or intellectual COIs, were not investigated.
The goal of this study was 1) to measure the extent of COIs in this field of research, 2) to test the hypothesis that study outcomes for the efficacy or durability of Bt crops are more frequently favorable to the interests of GM crop companies in the presence than in the absence of COIs and 3) to investigate the mechanisms underlying the association between COIs and study outcomes.

Article selection-first step
The flowchart describing the literature included/excluded is shown in Fig 1. We initially selected articles by searching the Web of Science (WoS, ISI Web of Knowledge, Thomson Scientific) bibliographic database. We used a Boolean search term including (1) most crops (e.g. cotton, maize, potato) for which GM varieties producing Bt toxins have been commercialized, and (2) most of the pests targeted by Bt crops (e.g. Diabrotica virgifera virgifera, Ostrinia nubilalis, Pectinophora gossypiella).
Two of us (DB and TG) then independently screened the titles and abstracts of these 1,227 articles, to eliminate those beyond the scope of our survey. Articles that were classified either with uncertainty by at least one of us (4.6%) or differently (7.9%) were jointly discussed to reach a consensus. We excluded 307 articles which dealt with various subjects, including assessments of the effects of Bt crops on non-target organisms, GM crops producing toxins other than those from Bt (e.g., toxins from spiders, lectins) and Bt crops developed by academics rather than by GM crop companies.

Classification by the type of conclusion
The outcomes of the 920 articles retained for the analysis were classified as 'favorable', 'neutral' or 'unfavorable' to the financial interests of at least one GM crop company. This classification was done independently by two of us (DB and TG) and was based exclusively on the titles and abstracts of the articles, with no knowledge of the authors' names and affiliations or of funding sources. This classification was also preformed independently of the classification on the basis of study topic. Articles that were classified either with uncertainty by at least one of us (11.2%) or differently (19.6%) were jointly discussed to reach a consensus.
Articles were classified on the basis of the general tone of the title and abstract and the formal conclusions given in the abstract, on the basis of the following criteria: 1. Favorable if the article reported that: a. One or several of the Bt crops tested provided a satisfactory level of control of the targeted pests; b. The susceptibility of the targeted pest populations to Bt toxins was high and/or remained unchanged over time (i.e. indicating an absence of evidence for the selection of mechanisms of resistance to the Bt toxins produced by Bt crops); c. Laboratory selection did not yield individuals sufficiently resistant to develop on Bt crops; d. Biotic and abiotic factors (relating to the pests and/or the plants) provided favorable conditions for the management of Bt resistance in targeted pests (e.g. Bt crops produced high doses of toxins, Bt resistance was recessive and/or associated with fitness costs, non-Bt crops (called refuges) were planted in sufficient quantities and sufficiently close to Bt crops); e. Pest populations resistant to a particular Bt toxin were controlled by another toxin or by other methods for which there was no cross resistance; f. Several Bt toxins acting via different receptors (i.e. decreasing the risk of cross-resistance) could be successfully combined in Bt crops through pyramiding. d. Biotic and abiotic factors (relating to the pest and/or the plant) created unfavorable conditions for the management of Bt resistance in the targeted pests (e.g. Bt crops did not provide high doses of toxins, Bt resistance was not fully recessive and/or was not associated with fitness costs, non-Bt crops (i.e. refuges) were planted in insufficient quantities and/or too far away from Bt crops).

Article selection-second step
We were able to retrieve PDF files containing the whole content of the article for 893 of the 920 articles classified by the type of conclusion. These 893 PDF files were screened independently by two of us (DB and TG) to check again the relevance of the papers for this study. Articles that were classified either with uncertainty by at least one of us (5.4%) or differently (5.7%) were jointly discussed to reach a consensus. We identified 672 of these articles as relevant (S1 Table).

Classification by topic
The 672 articles retained were classified into two categories on the basis of the topic considered: the 'Efficacy' or 'Durability' of the GM crops studied. This classification was done independently by two of us (DB and TG) and was based on titles and abstracts only, blind to the names and affiliations of the authors and to sources of funding (see below). Articles that were classified either with uncertainty by at least one of us (25.1%) or differently (4.3%) were jointly discussed to reach a consensus. Papers assigned to the Efficacy category reported data on:

Classification by funding source and author affiliation
We also analyzed the authors' professional affiliations and the funding sources (see below) of the 672 articles retained for the analysis. This analysis was again performed independently by two of us (DB and TG) blind to the classification of the article for topic and study outcome, as described above. Articles that were classified either with uncertainty by at least one of us (2.1% and 13.7% for authors' affiliations and the funding sources, respectively) or differently (5.5% and 10.1% for authors' affiliations and the funding sources, respectively) were jointly discussed to reach a consensus. For each paper, we recorded: 1. The presence or absence of authors professionally affiliated to a GM crop company. This was established by screening the address of each author appearing in the authors' affiliation section of the article; 2. For publications co-authored by academics and one or more authors professionally affiliated to a GM crop company, whether the senior author worked for a GM crop company or was an academic researcher. The senior author was identified as the corresponding author for the paper; 3. Whether the sources of funding were declared. In cases in which the funding sources were declared, we distinguished between funding from GM crop companies (partial or complete) and funding from other sources (governmental agencies, non-profit organizations, foundations, universities, etc.).
Using this information, we determined, for each article, whether there was: 1. An affiliation COI, defined as the presence of at least one author from a GM crop company on the paper; 2. A funding COI, defined as at least partial funding for the study from a GM crop company.
Overall, a paper was considered to display COI if it displayed either an affiliation or a funding COI.

Leading authors
For each article, the leading author was identified as the author present on the largest number of papers from among the 672 papers considered in this analysis. If several authors of a given paper had been authors on the same number of articles in our list, the leading author was identified as the first in alphabetic order. We identified 193 leading authors in total (mean number of papers per leading author = 3.48, median = 14.4, minimum = 1, maximum = 53, 5th percentile = 1, 95th percentile = 15)

Statistical analysis
The associations between (i) the presence/absence of COI and study outcome, (ii) the presence/absence of COI and the topic of the article, and (iii) the topic of the article and its outcome, were assessed in Fisher's exact tests, with the function fisher.test of R [5].
For a given article, the effect of its topic (i.e. efficacy or durability) on its number of citations (TC), as recorded on September 11, 2015 in the WoS, was evaluated with a mixed linear model, using the lmer function of R. The data were normalized by log(TC+1) transformation. The number of authors was included as a fixed effect and the identity of the leading author of each article was treated as a random effect. The age of the paper (i.e. 2015 minus the year of publication) was also considered as a fixed effect.
For a given author and a given topic, we evaluated the effect of COI on study outcome, by a generalized linear mixed model approach, with the glmer function of R (library lme4). The topic of the article and COI were considered as fixed effects and the identity of the leading author was treated as a random effect. The outcome of the study was considered as a binary variable with levels 'favorable' and 'other' (containing both 'neutral' and 'unfavorable' categories). We used the Akaike information criterion (AIC) [6] to select the best model.

Direct and indirect COI links
The global link between COI and the probability of obtaining a positive outcome can be written as follows:

Results and Discussion
A widespread occurrence of COI Like Sanchez [1] and Diels et al. [2] we found that COIs were widespread in the articles considered (Fig 2). Only about 7% (n = 46) of the articles contained a declaration of COI, but about one fifth (n = 141) of the 672 articles had at least one author from a GM crop company. Details about funding were provided by 524 articles, 29% (n = 153) of which acknowledged financial support from a GM crop company. By combining information about authors' affiliations and funding statements, we were able to assess the presence or absence of COI for a total of 579 articles ( Table 1)

A significant association between COI and study outcome
For the 579 articles we were able to classify, we found that COIs were associated with a 49% higher frequency of outcomes favorable to the interests of the GM crop company (p = 4x10 -5 , Fig 3A). Such an association between COI and study outcome was also reported by Diels et al. [2] and has been identified in other fields, including those of tobacco [7], biomedical [8], nuclear [9], sugar [10] and pharmaceutical [11] research. However, the mechanisms underlying this association have been little investigated. We studied these mechanisms in more detail.
The best model describing the outcomes of the studies was that with all fixed (Topic and COI) and random (leading author) effects but no term for interaction between Topic and COI effects. The main factor was Topic, with efficacy leading to a favorable outcome in 61% of cases, whereas only 30% of articles on durability resulted in a favorable outcome (p = 8x10 -16 ). The second most important factor was the presence of a COI, which increased the probability of a favorable outcome by +26% for articles on efficacy and +50% for articles on durability, respectively (p = 2.2x10 -3 ). The leading author's identity had an effect of only marginal significance (p = 0.07). This model suggests the existence of direct and indirect links between outcome and COI.

Direct and indirect links between outcome and COI
First, we detected an indirect link between outcome and COI through the topics studied. Studies with COIs investigated efficacy 70% more frequently than studies without COI (p = 3.8x10 -7 , Fig  3B). Moreover, outcomes were 104% more likely to be favorable for articles focusing on efficacy than for those exploring durability (Fig 3C, p = 8x10 -16 ). This would naturally lead to, and, therefore, partly explain, the association between COI and favorable outcome.
Second, we found a direct link between outcome and COI within each topic. For a given leading author and a given topic, study outcomes were more likely to be favorable to the GM crop company in the presence than in the absence of COI (+26%, p = 0.05 for efficacy, +50%, p = 0.02  for durability, and p = 2.3x10 -3 overall, Fig 3D). The statistical effect of COI on outcome (i.e. the difference of probability of obtaining a positive outcome with and without conflict of interest) via this direct link (Δ direct link = 0.12) was twice that via the indirect link (Δ indirect link = 0.07). Numerical computation of the whole effect, gives Δ = 0.18. Note that Δ indirect link + Δ direct link > Δ simply because of an interaction term.
A cause or a consequence of COI?
Part of the observed COIs may be due to the favorable outcomes. For instance, authorship or funding statements may be modified after the results of a study have been obtained. GM crop companies may prefer not to be listed as having provided support for studies leading to unfavorable outcomes. Alternatively, academics might be more inclined to propose researchers from GM crop companies as co-authors on publications presenting results favorable to the interests of the company. Conversely, COIs may be partly responsible for the favorable outcomes, for at least two reasons: (i) GM crop companies may promote research on efficacy rather than durability, thereby indirectly increasing the frequency of favorable outcomes for papers with COI. The interests of the scientific community at large seem to be different, as studies without COI focused mostly on durability, and on average, peer scientists cited durability papers more frequently than efficacy papers (between +23% and +51%, p = 9.7x10 -3 , depending on the number of authors, the leading author and the year of publication, Fig 4). (ii) Data against the interests of GM crop companies may be less likely to be published in the presence than in the absence of COI. This scientific bias might underlie the direct link between favorable outcome and COI found in our study. However, as previously reported by Krimsky [12], this direct link is not necessarily the consequence of such a bias. Here, studies displaying COI might have investigated more effective and durable GM crops than those without COI, or they may have used different methodologies (e.g. laboratory vs. field conditions). A closer examination of studies on GM crops is required to determine which of these possibilities actually accounts for the direct link between COI and outcome.

Limitations of this study
This study has several limitations. First, we explored only two characteristics of Bt crops: efficacy and durability. Other characteristics and consequences of these transgenic plants, including all those relating to the environment (e.g. the impact of Bt crops on non-target insects) or health, merit a similar analysis. As a demonstration of the absence of such effects, rather than their presence, is clearly in the GM crop companies' interests, an association between COIs and the outcomes of studies studying environmental or health impacts is also a testable hypothesis. Sanchez [1] and Diels et al. [2] focused on studies on health impacts, in which they found COIs to be very frequent, and Diels et al. [2] found an association between financial or professional COIs and research outcomes.
Second, as we used the addresses of authors to identify their affiliations, only one type of affiliation, that relating to employment, was considered. However, authors may have affiliations to GM crop companies of other types, such as being members of advisory boards, consultants, or co-holders of patents [13], and this could also have a significant impact on the outcomes of studies on GM crops. We did not consider these affiliations as they are not usually reported in articles (COI statements became obligatory in some journals only recently and, as revealed here, they remain very rare). The consideration of other types of affiliation would require a survey that would be difficult to perform given that more than 1,500 authors were considered in this study.
Third, we have considered only the links between authors and GM crop companies. Other stakeholders (e.g. Greenpeace, The Non-GMO Project, The Organic Consumers Association, The Network of European GMO-free Regions) oppose GM crop companies in being openly against the use of GM crops. An inverse relationship might therefore be expected between the outcomes of studies on GM crops and the presence of COIs relating to these stakeholders. We were unable to test this hypothesis because we identified no financial interests connected with anti-GMO stakeholders, in terms of the professional affiliation of the authors or their declared funding sources.
Finally, this study focused exclusively on financial COIs. Non-financial COIs, also known as intrinsic or intellectual COIs -due to personal, political, academic, ideological, or religious interests -might also have a significant impact on the outcomes of research studies [4]. It is difficult to decipher intellectual COIs and, as for the detection of non-professional affiliations with GM crop companies, it would be a major challenge to perform such an analysis given the large number of authors considered.
We think (although this hypothesis has yet to be tested in this field) that adding studies to overcome these limitations would have resulted in the same qualitative conclusions: COIs are frequent, and there is a link between COIs and study outcomes. It remains unclear whether this would increase the power of the analysis, mostly because some of the abovementioned factors (stakeholders other than GM crop companies, non-professional affiliations to GM crop companies and non-financial COIs) might be difficult to classify and would be subject to considerable background noise.

Policy implications
Scientific journals should make it obligatory for authors to publish a COI statement, as such statements remain too rare (Fig 2A). However, COI statements are not a panacea and do not absolve scientists and industrial partners of their obligation to deal with COIs [14]. It is almost impossible to avoid non-financial COIs, but it does seem to be possible to limit the occurrence of professional and funding COIs in studies investigating the performance and impact of GM crops. However, it is desirable for these studies to be at least partly funded by GM crop companies, because these companies are the principal beneficiaries of these products and should take responsibility for the impact of their products on the environment and health. One way of combining the possibility of GM crop company funding with an absence of professional and funding COIs would be a system of indirect financial support for research from the industry: GM crop companies and other stakeholders (governments, NGOs) would make a financial contribution to a common pot managed by an independent agency. This agency would launch and fund calls for proposals developed by researchers and stakeholders, to which researchers could respond. Projects would be accepted or rejected by a scientific committee independent of the stakeholders and as free from COIs as possible. With this system, which has already been proposed for biomedical research [15][16][17], there would be no direct interaction between GM crop companies and researchers: COIs due to professional affiliations and funding sources would be avoided and the continuity of research funding would be independent of the publication of results favorable or unfavorable to the interests of the companies.
Supporting Information S1 Table. List of the 672 articles with outcomes classified as favorable, neutral or unfavorable, topics categorized as efficacy or durability and with affiliation and funding COI characteristics. (XLSX)