Risk assessment of honey bee stressors based on in silico analysis of molecular interactions

Abstract

A global decline of the honey bee Apis mellifera has been observed in the last decades. This pollinator plays a fundamental role in food production and the economy in Europe. The decline of honey bee colonies is linked to several stressors, including pesticides. The current pesticide risk assessment of honey bees in Europe focuses on lethal effects and lacks reflection on sublethal effects. A better understanding of the consequences that exposure to these chemicals has on honey bees is still needed. In this context, the aim of this European Food Risk Assessment Fellowship Programme fellowship project has been to use in silico methodologies, such as virtual screening, as a first step to identify possible interactions at the molecular level between A. mellifera proteins and pesticide ligands. For this purpose, a docking study of the proteins from A. mellifera and pesticide ligands extracted from online databases has been performed by using the software Autodock Vina. The results obtained were a ranking based on the predicted affinity of the pesticides for specific and non‐specific binding sites on bee macromolecules. These results were compared with data obtained from the literature and linked to potential sublethal effects. Finally, a risk assessment analysis of the identified molecular stressors of honey bees was performed. The results of this study are considered a starting point to identify new sources of possible stress for honey bees and thereby contribute to the overall understanding of the honey bee decline.

1. Introduction

1.1. European Food Risk Assessment Fellowship Programme

The present work is part of the European Food Risk Assessment Fellowship Programme (EU‐FORA). This programme is an initiative of the European Food Safety Authority (EFSA) created to provide practical training to scientists, increasing their knowledge in food safety risk assessment and knowledge community. The fellow was hosted by the Institute of Biochemistry and Cell Biology at the Italian National Research Council (CNR). The project focused on the use of the in silico approach using docking analyses to identify possible adverse interactions between Apis mellifera proteins and pesticide ligands. This approach could enhance the protection of honey bees in Europe.

1.2. Virtual docking in risk assessment of Apis mellifera

There is an observed global decline of the honey bee A. mellifera (Di Noi et al., 2021). This pollinator plays a fundamental socio‐economic role in food production because of its pollination services and bee products such as honey (Gallai et al., 2009). The use of pesticides is one of the main stressors linked to the observed decline of this species (Sanchez‐Bayo and Goka, 2014; Chmiel et al., 2020). In the European Union (EU), there are currently 1,461 active substances registered, of which 454 are approved (EU Pesticides Database, 2021). In addition, non‐approved active substances, such as the neonicotinoid imidacloprid, are still used under exceptional circumstances laid in Article 53 Emergency authorisation of Regulation (EC) No 1107/2009.

The current bee risk assessment of pesticides in Europe for approval of pesticides focuses on lethal effects (Sgolastra et al., 2020). However, sublethal endpoints such as effects in immunity, behaviour, or sensor ability can affect honey bee performance and reduce populations. Studies aiming to describe the non‐lethal effects of pesticides in A. mellifera are available in the literature (Chmiel et al., 2020; Di Noi et al., 2021). Nevertheless, most of these studies are performed with insecticides, mainly from the neonicotinoid family. Other pesticide classes, such as fungicides and herbicides, are not well represented and evidence indicates that they could also affect the honey bee's health. In addition, certain types of effects particularly within enzymatic and molecular responses are not well characterised (Di Noi et al., 2021). Thus, a better understanding of the consequences that exposure to pesticides has on honey bees is still needed, but this is a challenging task considering the number of available pesticides and possible endpoints. A quick and cost‐effective screening assay that identifies chemicals and pathways of concern could be beneficial.

Virtual screening is commonly used in drug discovery to predict the binding free energy of libraries of small molecules to a target protein structure and the bound conformation (Forli et al., 2016; Naqvi et al., 2018). This technique could also be a promising tool for chemical toxicity screening at a molecular level (Goldsmith et al., 2014). Proteins are essential components of living organisms with different functions such as enzymes, signalling components and transport/storage constituents. They usually bind to other molecules (ligands) with high specificity and affinity to perform their respective task. The binding is determined by the physicochemical properties of the amino acid residues and the shape of the binding pocket. Alteration in the protein–ligand interaction can affect protein functions and molecular pathways which could ultimately be linked to sublethal effects such as immunity dysfunctions or communication (Eder et al., 2007; Li, 2007). Currently, dozens of 3D structures of proteins from A. mellifera are present in online databases including odorant‐binding proteins, pheromone binding proteins and proteins from the immune system. The objective of this fellowship was to verify if these proteins could be affected by toxic compounds, such as pesticides, using virtual screening and docking approaches. Furthermore, the in silico results were compared with literature data.

2. Description of work programme

2.1. Aims

The main aim of this project was to investigate in silico interactions of bee proteins and pesticides to produce a risk assessment analysis that can enhance the protection of honey bees in Europe. To achieve this goal, the fellow prepared data sets of proteins from A. mellifera and pesticides for the docking analysis. Subsequently to the virtual screening analysis, the data were processed to obtain a ranking of affinity and possible interference of pesticides in the protein functions and organism pathways. These results were compared with data obtained from the literature. The last activity of the present work was dedicated to the risk assessment analysis of the effects of pesticide stressors on honey bees.

2.2. Activities/methods

2.2.1. Proteins and ligands selection and preparation

All the different steps performed in this section are summarised in Figure 1. In the first part of the program, the fellow accomplished a search in databases of protein 3D structures (RCSB PDB) from A. mellifera. Once identified, the structure files were retrieved and stored in a pdb format. The pesticide molecules were selected based on the list of approved active substances from the EU pesticides database. Subsequently, the 3D structures were downloaded from PubChem. If the 3D structure was disallowed (e.g. mixtures, salts or too large compounds), the pesticide was not included in the final ligand data set. In addition, ligands were classified by pesticide group and mode of action (MoA). The protein and ligand structures were optimised and converted to pdbqt format using the program MGLTools (http://mgltools.scripps.edu/). This step was essential as the program used for docking, Autodock Vina (http://autodock.scripps.edu/), requires as inputs molecular pdbqt files. The preparation included removing water molecules and any residual non‐standard ligands and adding polar hydrogen atoms and Gasteiger charges. Finally, grid boxes with a spacing of 1 Å were created to scan the protein surface. The boxes could incorporate the entire protein, in the case of blind docking or only the binding pocket in the case of a more targeted docking (see Figure 2).

Figure 1.

Different preparation steps followed to perform the virtual screening. The molecular docking analysis was performed with Autodock vina. The program requires as inputs a pdbqt molecular files of the 3D structure of the protein and the ligand as well as a grid‐box to scan the protein surface. The analysis returns an affinity score (kcal/mol) of the protein‐ligand affinity

Figure 2.

Preparation example of the grid‐box, which defines the space around the protein used to perform the docking analysis. The box can incorporate the entire protein (a) or just the binding pocket (b)

2.2.2. Virtual docking and data processing

The fellow performed the molecular docking analysis using AutoDock Vina (Trott and Olson, 2009). The exhaustiveness was set to 16 and the maximum number of simultaneous threads was set to 9. The analysis was performed with an I7 Intel workstation equipped with 32 GB RAM, 2 SATA HD of 2 TB each for data storage, 1 SSD with the operative system (linux OS debian based), a dual monitor, online connection with the server for remote control. Shell and python scripts were developed to automatise some steps. The best affinity scores of each protein–ligand complex were examined with a hierarchical heatmap. Good affinity scores were considered for interactions ≤ −6 kcal/mol. Visual inspection of the complex and predicted possess was performed with Pymol, Charmm‐GUI and MGLTools.

2.2.3. Literature search

A more in‐depth bibliographic search on the principal bibliographic sources, such as PubMed or Web of Science, was performed to collect as much as possible data on molecular stressors for honey bees. The fellow compared the data obtained from literature with the data set obtained in the in silico analysis to identify a new source of possible stress for honey bees.

2.2.4. Risk assessment analysis

The final part of the work programme concerned the risk assessment analysis of the identified molecular stressors in the investigated area of honey bees' increased mortality. The fellow identified the molecular hazards which could negatively impact relevant honey bees' pathways and ultimately jeopardise their survival. These results were compared with the results obtained from the literature.

3. Conclusions

3.1. Overall conclusion

In the present work, in silico results obtained from the virtual docking analysis were compared with information collected from the literature.

In total, more than 12,000 protein–ligand pairs, corresponding to > 300 pesticides and > 40 protein structures analysed, were evaluated. In Figure 3, a hierarchical heatmap of the affinity scores obtained from a screening search was reported. The highest affinity score was −12.5 kcal/mol and the lowest −0.6. Good affinity scores were considered for interactions ≤ −6 kcal/mol. According to the heatmap, the pesticide structures can be clustered into three main groups. The first group indicated that most of the tested pesticides do not target honey bee protein (binding affinity ≥ −6 kcal/mol) or only interact with a few of them such as pheromone‐binding proteins. The second group revealed the multi‐target behaviour of a large number of pesticides that exhibited good affinity scores (≤ −6 kcal/mol). Interestingly, not only insecticides but also herbicides and fungicides were among the pesticides with the highest affinities. This result suggests that herbicides and fungicides can also affect honey bees. The last cluster belongs to a smaller group of pesticides that did not interact with any of the protein structures.

Figure 3.

Hierarchical heatmap of the molecular docking affinity scores obtained in the virtual docking. The right and upper dendrogram correspond to the pesticide and protein molecules respectively. Apis mellifera protein structures are coloured by functions. The protein–ligand pairs with the strongest binding affinities are represented in blue and the ones with lower affinity in red

The results from the literature search showed that pesticides affect cognition, immunity and reproduction among other sublethal effects. Effects on proteins could be linked to some of these responses, for instance, Li and collaborators suggested that olfactory recognition could be affected as imidacloprid decreases the binding affinity of odorant‐binding proteins to floral volatiles (Li et al., 2015). Moreover, neonicotinoids can affect the Toll and Imd immune pathways, which are essential in triggering the innate immune response in insects (Di Prisco et al., 2013; Chmiel et al., 2019). In accordance with literature data, we found an interaction between some proteins involved in the innate immune response with some pesticides including neonicotinoids.

In conclusion, this research contributes to understanding how certain pesticides could interact with honey bee proteins with a high theoretical affinity. Some studies have linked sublethal exposure to pesticides with adverse effects on honey bees. However, these studies have mostly been performed on insecticides (mainly neonicotinoids). Other types of pesticides are underrepresented. Furthermore, a recent review by Chmiel et al. (2020) also highlighted that knowledge gaps exist in certain responses, particularly within enzymatic and molecular levels, such as those regarding the immune system and genotoxicity. Therefore, the protein–ligand interactions presented in this study could be used as guidance for the experimental testing of pesticides and in this way contribute to the overall protection of honey bees in Europe.

3.2. Additional scientific activities

During the EU‐FORA fellowship programme, the fellow was involved in extracurricular activities. These included the participation in the SETAC Europe 32nd Annual Meeting Conference held 15–19 May 2022 in Copenhagen. The fellow had the opportunity to expose a poster titled ‘Risk assessment of honey bee stressors based on in silico analysis of molecular interactions’ at the SETAC Europe 32nd Annual Meeting Conference 15–19 May 2022 in the Bella Center Copenhagen. Moreover, the fellow attended the ONE – Health, Environment, Society – Conference, 21–24 June 2022. In addition, the fellow attended a meeting organised by the Italian focal point with researchers working on bees and other pollinators. Finally, the fellow collaborated with the university group at the Department of Science and Technology University of Naples Parthenope (Italy). Currently, two students of this university are performing their bachelor thesis on informatics about this project. The fellow had the opportunity to help define the objectives of the thesis and to supervise the students. The writing of two manuscripts is in progress.

3.3. Disclaimer

Detailed results obtained from the method development, results and risk assessment are not included in this report to avoid certain copyright claims, as these results will be subsequently published in other scientific journals.

Abbreviations

CNR

National Research Council

EU‐FORA

European Food Risk Assessment Fellowship Programme

MoA

mode of action

SETAC

Society of Environmental Toxicology and Chemistry

Suggested citation: del Águila Conde M and Febbraio F, 2022. Risk assessment of honey bee stressors based on in silico analysis of molecular interactions. EFSA Journal 2022;20(S2):e200912, 9 pp. 10.2903/j.efsa.2022.e200912