RNA interference (RNAi) technologies (RNAi) are important tools for manipulating transcript levels and exploring gene function in a wide range of species. RNAi is a particularly critical tool for functional genomic studies in species where other genetic manipulations, such as the development of transgenics or mutant strains, are not feasible [1]. Honey bees (Apis mellifera), for example, are an excellent model system for sophisticated studies of cognition, neurobiology, plasticity and complex social behavior, but thus far, the development and maintenance of knockout and transgenic lineages in this species have not been possible [2]. Over the last decade, RNAi has been successfully adopted as the major genetic tool for gene function analysis in honey bees [3,4,5]. However, there can be considerable variability in the levels and duration of transcript knockdowns between tissues, individuals and experiments. In order to take full advantage of this powerful technology to study gene function in honey bees and other insects, it is necessary to develop a complete understanding of the mechanisms by which RNAi decreases transcript abundance and, in particular, to examine and characterize any off-target effects.

Off-target effects are non-specific and caused by undesired base-pairing of non-target genes with small interfering RNA (siRNA) derived from double-stranded RNA (dsRNA) [6]. Off-target effects can be widespread and can alter expression of large numbers of genes, as previously reported in RNAi experiments involving plants, invertebrates, vertebrates [7,8,9], as well as honey bees [10].

A green fluorescent protein (GFP)-derived dsRNA (dsRNA-GFP) has been used as an exogenous control for RNAi assays in several arthropod species, including Marsupenaeus japonicus [11], Pacifastacus leniusculus [12], Spodoptera exigua [13,14], Acyrthosiphon pisum [15], Aedes aegypti [16], Antheraea sp. [17], Locusta migratoria [18], Schistocerca gregaria [19], Bactericerca cockerelli [20] and Apis mellifera [5,21,22,23,24,25,26,27]. Its gene sequence is not found in the honey bee genome. Although dsRNA-GFP is not expected to trigger an RNAi response in treated bees, undesirable effects on gene expression, pupal pigmentation or developmental timing have been routinely observed. To better understand the molecular and phenotypic effects of dsRNA-GFP in honey bees and to evaluate its use as a control for RNAi studies, we examined the impact of dsRNA-GFP on global gene expression patterns in developing workers. The dsRNA-GFP was introduced using a non-invasive feeding protocol [23]. We found that dsRNA-GFP causes large-scale changes in gene expression associated with multiple biological processes. Furthermore, dsRNA-GFP exposure tended to preferentially decrease, rather than increase, expression of genes compared to controls.

The double-stranded RNA for green fluorescent protein (dsRNA-GFP) is widely used as an exogenous control in honey bee RNAi studies [5,21,22,23,24,25,26,27]. In our studies, changes in gene expression, pupal pigmentation or developmental timing are the most frequent undesired effects in RNAi screens that use dsRNA-GFP as control. These effects have been also noted by other bee researchers during discussions related to RNAi approaches at the IUSSI Congress 2010 (Copenhagen, Denmark) and Workshop on Honey Bee Genomics & Biology 2011 (Cold Spring Harbor, USA). Therefore, those observable side effects are recurring and not yet reported in the literature. Based on this, we decided to perform a large-scale gene expression analysis to compare untreated and dsRNA-GFP treated developing workers.

Three technically distinct experiments were conducted in two different laboratories. Similar numbers of up- and downregulated genes were found for experiments 1 and 2. In addition, results from both experiments showed a greater number of down- than upregulated genes (Table 1). On the other hand, expression of only a few genes was altered by dsRNA-GFP in experiment 3, which examined adult workers.

Analyses of the downregulated gene set revealed that only five genes appear in both lists of differentially expressed genes obtained in experiments 1 and 2 (Table S1). Based on Gene Ontology terms, they are classified as enzymes or proteins with binding functions involved in cellular metabolic process, such as RNA processing and transport. In RNAi experiments, long dsRNA molecules are cleaved into small interfering RNA duplexes (siRNA) of ~21 nucleotides by Dicer enzyme [28]. If these siRNAs have sequence similarity to additional, non-target transcripts, they can cause degradation of these transcripts [6]. To determine if the decreased transcript abundance of these five genes is due to off-target effects, we aligned the sequences of each of these transcripts with the dsRNA-GFP sequence. For each gene, alignments revealed multiple perfect base-pairing matches ranging from eight to 11 nucleotides-long, as well as regions of imperfect complementarity (File S2). Previous studies have demonstrated that off-target gene silencing can be mediated by 7 nt perfect matches between a siRNA and targets [29] and even by microRNA-like mechanisms [30,31] targeting both coding and untranslated regions [32,33,34]. Moreover, the occurrence of multiple off-target sites is suggested to enhance the disturbances [29]. It is also likely that siRNA molecules generated from dsRNA-GFP saturate cellular RNAi machinery, so that siRNAs compete with endogenous microRNAs [34]. Thus, the expected microRNA action is perturbed, and their natural targets may be upregulated [35].

Nine genes were found to be upregulated in both experiments 1 and 2 (Table S1). This is not unexpected, since siRNA off-target effects were previously described to cause both down- and upregulation in mammals [36]. In our studies, these upregulated genes are involved in immune responses, oxidoreductase activity, aging, cell homeostasis, morphogenesis, response to external stimulus and response to stress. Among them, GB10133 (superoxide dismutase 1) and GB15855 (thioredoxin 2) are members of an antioxidant system [37] potentially involved in detoxication by modifying the chemical structure of xenobiotics, such as dsRNA-GFP. Also, GB10398 codes for a ninjurin-1-like protein, member of the metazoan conserved Ninjurin family. Ninjurins are cell adhesion molecules, and it is known that their mRNA levels increase after injury, infection or stress [38,39]. Finally, two members of the apidaecin gene family (GB13473 and GB17782) related to humoral immunity [40] were also upregulated. Thus, of these nine genes, five appear to be involved in response to infection or stress. RNAi machinery plays an important role in the insect immune system [41] by triggering antiviral responses in response to exogenous dsRNA viruses [42]. Thus, it is plausible that dsRNA-GFP molecules are recognized as a viral infection, culminating in the activation of immune genes, RNAi systems, siRNA production and consequent off-target effects.

No overlapping genes were found between experiments 1 and 3. Four genes are upregulated in both experiments 2 and 3 (Table S1), including GB10708, a gene encoding for immune responsive protein of 30 kDa (IRP30). Honey bee IRP30 gene was previously reported as a non-canonical immune factor found in social hymenopterans upon bacterial challenge or viral infection [43,44,45].

Little dsRNA-GFP effects were observed in the adult bees. It is important to remember that the dsRNA-GFP was offered only once per experiment in the diet of second instar larvae. Regarding experiments with adult workers, sampling occurred ~ 23 days after treatment. Previous works have shown that dsRNA molecules remain intact for several days in A. mellifera and produce long lasting effects [4,23]. Therefore, it is unlikely that such molecules have been degraded or eliminated in adults. We believe that over ~ 23 days, the honey bee immune system has become adapted to the presence of this molecule, no longer recognizing it as viral particles.

These results suggest that: (i) RNAi treatment effect is different during worker development, as in different life stages of other insects [46]; (ii) few target genes are primarily affected by dsRNA-GFP off-target effects; (iii) undesired consequences on gene expression and phenotypes observed in bees treated with dsRNA-GFP could be triggered by disturbances in those targets, affecting downstream gene networks in association with particularities of physiological context of each developmental stage. Furthermore, the physiological and developmental context can influence the effects of dsRNA-GFP treatment: in some genes, dsRNA-GFP treatment has opposite effects in the two experiments, i.e., increased expression in experiment 1, decreased expression in experiment 2 or vice-versa (Table S2).

Little overlap in the sets of affected genes between the three experiments/developmental timepoints is not entirely surprising. Side effects triggered by dsRNA-GFP seem to be multifactorial. From a single dsRNA, the resulting off-target effects are specific for different cell types or developmental stage, because the sets of expressed (and consequently misregulated) genes are different in each biological context [47]. Even if dsRNA may be recognized as a viral challenge and upregulate non-self RNA sensors and expression of the immune genes [48], we need to take into account that the range of defense responses can be either evolutionary divergent across species [49,50], as well as developmentally modulated [43,51,52,53].

Data from all experiments were combined to produce a comprehensive list of potential gene targets whose expression is altered in treatments with dsRNA-GFP (Table S3). In total, 1,461 genes showed significant changes in expression in response to dsRNA-GFP treatment. This raises concerns about using dsRNA-GFP as a control for RNAi experiments, since it suggest that this treatment can cause changes in about 10% of A. mellifera genes, undoubtedly triggering associated development, physiological and behavioral changes. Also, it is important to note that there may be possible additional effects on the transcriptome that could not be examined in the microarray, since it undoubtedly does not cover all possible transcripts. Indeed, several novel coding and non-coding genes were only recently discovered and, thus, are not represented in the microarrays.

To understand how our data are useful and applicable to other honey bee studies, we performed a multiple alignment of GFP partial sequences used by other authors for dsRNA-GFP synthesis (File S1), which showed that sequences are very close to each other, suggesting similar responses.

To gain additional functional insights, Gene Ontology (GO) comparisons were performed using “all downregulated genes” versus “all upregulated genes” as input. Using GO annotation from Drosophila melanogaster, 253 GO terms were recovered for upregulated genes and 508 GO terms for downregulated ones. A number of different metabolic and developmental processes are altered by dsRNA-GFP, and many of the affected genes function as enzymes or binding proteins (Table S4). Although it is difficult to establish trends from the complete dataset (Tables S3 and S4), the most representative GO categories (top 10, according to number of genes) are illustrated in Figure 1. As noted above, more genes and processes were downregulated than upregulated by dsRNA-GFP treatment.

Interestingly, the microarray data indicate a perturbation in the metabolism of the major honey bee hormones by dsRNA-GFP treatment. Juvenile Hormone (JH) and 20-hydroxyecdysone (20E) precisely govern virtually all biological processes during the honey bee life cycle, such as morphogenesis, developmental timing, molt and metamorphosis and adult behavior. Levels of juvenile hormone acid methyltransferase (JHAMT, GB10517) transcripts were upregulated in experiment 1, but downregulated in experiment 2, while farnesyl pyrophosphate synthase (FPPS, GB12859) transcripts were upregulated in experiment 2. Both genes are associated with JH synthesis. Juvenile hormone epoxide hydrolase gene (JHEH, GB10771) controls one step of the JH degradation pathway, and its mRNA levels were downregulated in experiment 1. Transcript levels of Ecdysone-induced protein 75 (E75, GB11364) were upregulated in experiment 2. E75 plays roles in molting cycle, cuticle formation, ecdysis and regulation of the ecdysteroid metabolic process. Expression of two storage protein genes, hexamerin 70a (GB30362) and hexamerin 70c (GB13613), were respectively down- and up-regulated in experiment 1. Hexamerin 70a is susceptible to exogenous stimulus, since bacterial infections in adult workers also disrupt its expression [54]. Moreover, honey bee hexamerins function as an amino acids source for metamorphosis, as well as they are potential JH-binding proteins modulating JH action [55]. Accordingly, perturbations in the endocrine system affect both JH- and ecdysteroids-modulated gene cascades and could account partially for the undesired effects observed.

Our results demonstrate that treatment with dsRNA-GFP can have substantial direct and indirect effects on transcript levels of genes associated with a variety of biological processes in developing honey bee workers. Furthermore, it is clear that the molecular effects of dsRNA-GFP exposure can vary depending on the physiological and developmental context; thus, while our studies identified ~1400 affected transcripts, different transcripts may be impacted in different studies. We recommend the use of a second exogenous control in RNAi studies in honey bees in order to better control for the off-target effects of both the control and experimental genes. Further studies will be necessary to determine the non-sequence specific effects of dsRNA-GFP, the effects of dosages and the duration of treatment.