Non-Specific dsRNA-Mediated Antiviral Response in the Honey Bee

Honey bees are essential pollinators of numerous agricultural crops. Since 2006, honey bee populations have suffered considerable annual losses that are partially attributed to Colony Collapse Disorder (CCD). CCD is an unexplained phenomenon that correlates with elevated incidence of pathogens, including RNA viruses. Honey bees are eusocial insects that live in colonies of genetically related individuals that work in concert to gather and store nutrients. Their social organization provides numerous benefits, but also facilitates pathogen transmission between individuals. To investigate honey bee antiviral defense mechanisms, we developed an RNA virus infection model and discovered that administration of dsRNA, regardless of sequence, reduced virus infection. Our results suggest that dsRNA, a viral pathogen associated molecular pattern (PAMP), triggers an antiviral response that controls virus infection in honey bees.

Honey bees are eusocial insects that live in colonies of up to 50,000 individuals [7,[15][16][17][18][19]. High population density within bee hives and behaviors such as mouth-to-mouth food exchange (trophallaxis) facilitate pathogen transmission [8,17,20,21]. Unlike solitary insects, bees benefit from social immune behaviors (i.e., grooming and behavioral fever) that may reduce colony pathogen burden (reviewed in 9,22,23). A critical number of bees are required to maintain colony temperature, rear brood (young bees), and carry out food gathering and storage. Thus an effective immune system is important at both the individual bee and entire colony level.
In this study, we report the development of an experimental model of honey bee virus infection, Sindbis virus expressing enhanced green fluorescent protein (SINV-GFP) [15,77]. We examined virus infection in the presence and absence of virussequence specific and non-specific dsRNAs. Based on evidence from honey bees [50][51][52] other insects (i.e., fruit flies and mosquitoes) [50,73], we expected that sequence specific RNAi would be the major honey bee antiviral defense mechanism. Instead, we determined that co-injection of nonsequence-specific dsRNA reduced virus copy number. This reduction suggests that dsRNA, a viral PAMP, triggers additional immune responses in honey bees. Importantly, a non-specific dsRNA-triggered antiviral response has not been observed in other adult insects. Transcriptional level response to virus and dsRNA, evaluated using honey bee gene expression microarrays, indicated that the majority of genes with previously characterized roles in insect immunity were not appreciably regulated by virus or dsRNA, suggesting that dsRNA-mediated antiviral defense in honey bees may involve unique genes and signal transduction cascades, alter or suppress known immune pathways, or may be predominantly post-transcriptionally regulated.

Honey Bee Virus Infection Model
The majority of honey bee viruses are positive sense singlestranded RNA (+ssRNA) viruses that replicate via a dsRNA intermediate. Although the genomes of numerous honey bee viruses have been sequenced, there are no infectious clones available [78]. Since many insect viruses have broad host range we tested the ability of Sindbis virus (SINV) to infect honey bees. To facilitate monitoring virus infection we employed a recombinant virus, SINV-GFP, that expresses green fluorescent protein (GFP) [15,77]. Bees were inoculated with SINV-GFP via intra-thoracic injection and monitored daily. GFP fluorescence increased over time and spread throughout the bee (Figure 1 and S1). SINV-GFP is most readily visualized in the abdomen due to the relatively thin exoskeleton in this region ( Figure 1A-B). The accumulation of virus in protein lysates from both the injection site (thorax) and distal regions (head, abdomen) demonstrated that SINV-GFP infection was systemic ( Figure 1C and S1) and the ability to infect naïve bees with lysates from SINV-GFP infected bees demonstrated its infectivity ( Figure S1). Relative abundance of SINV-GFP in individual, dissected bees was assessed by RT-qPCR and demonstrated productive infection and virus spread from the injection site (thorax) to distal regions (head and abdomen) ( Figure S2). Together these data established that SINV-GFP productively infects honey bees and thus may function as a model of virus infection. Infection with a model virus has several advantages including the ability to precisely control the titer of virus inoculum and track the virus during infection. Intrathoracic injection, as opposed to an oral infection route, ensured that each bee received the same infectious dose, and mimics mite (Varroa destructor) vectored honey bee virus infection [79][80][81][82]. The use of a model virus also ensured that honey bees, obtained from managed colonies housed outside of the laboratory, were not previously infected with SINV. This is particularly important to honey bee host-pathogen interaction research, since bees are readily infected with pathogens and their pathogen status has been show to vary on a weekly basis [3]; therefore, the pathogen status of individual bees at the onset of an experiment is unknown. Use of a model virus inoculum, rather than a semi-purified bee virus inoculum, ensures that all experimental subjects have the same initial dose at the onset of an experiment. While there are advantages and disadvantages to model and natural virus infection systems, both are required to furthering our understanding of honey bee antiviral immune responses. mediated antiviral immunity in plants and insects. To investigate the role of dsRNA in honey bee antiviral immunity, bees were injected with virus (SINV-GFP) alone or co-injected with multiple species and lengths of dsRNA including virusspecific dsRNA (SINV dsRNA, 928 bp), non-specific dsRNA corresponding to Drosophila C virus sequence (DCV dsRNA, 1017 bp and 448 bp), luciferase sequence (LUC dsRNA, 355 bp) and polyinosinic-polycytidylic acid (poly(I:C)), a synthetic mimic of dsRNA, the building blocks of dsRNA nucleotidetriphosphates (NTPs), or dsDNA (SINV dsDNA 928 bp) ( Figure  2A-B). Three-days post-injection virus abundance was assessed by microscopy, Western-blot analysis, and qPCR. In Drosophila, sequence-specific dsRNA triggers an RNAimediated mechanism of antiviral immunity, whereas nonspecific dsRNA has no effect [15]. Surprisingly, co-injection of non-specific dsRNA dramatically reduced virus infection in honey bees ( Figure 2). Fluorescence imaging provided qualitative evidence of this reduction ( Figure 2C) and quantitative data was obtained at the protein (or viral capsid level) from Western blot and fluorimetry analyses of bee lysates, and at the genomic level by quantitative PCR (qPCR) (Figure 2 D-F and Figure S3). Analysis of virally produced GFP-protein in pooled lysates (representing 10 bees each) demonstrated that co-injection of multiple species and lengths of dsRNA (SINV dsRNA, DCV dsRNA, LUC dsRNA, and poly(I:C)) decreased Sindbis virus production as compared to bees injected with virus only ( Figure 2C). In order to rule out the possibility that dsRNA (a long poly-anion) was simply interfering with viral infection, dsDNA of the same length and sequence (SINV dsDNA) was co-injected with virus and no reduction in infection was observed (Figure 2 C-D, Figure S3). In addition, co-injection of nucleotide-triphosphates (NTPs), the building blocks of RNA present in excess in dsRNA production reactions, did not reduce SINV-GFP infection ( Figure 2C-D and Figure S3). Fluorescence measurement of individual bee lysates validated differences between dsRNA-treated bees as compared to those co-injected with NTPs or dsDNA ( Figure  2E). Likewise, Western blot analysis of individual bees supported data from pooled samples demonstrating dsRNAmediated reduction in virus load, but also identified experimental outliers ( Figure S3). These individuals were not further analyzed in this study and may be attributed to genetic differences between individual members of this out-bred honey bee colony. The average GFP to actin ratio in Western blots of individual bee lysates was 3.0 for virus-infected bees, whereas the ratio in bees that were co-injected with virus-specific dsRNA (SINV dsRNA), non-specific dsRNA (DCV dsRNA), or poly(I:C) was reduced to 1.2, 0.9, and 1.2 respectively ( Figure   Figure 1. Honey bees are productively infected with a model virus, SINV-GFP. (A) Honey bees were infected with SINV-GFP and fluorescence microscopy revealed increased amounts of GFP each day post-injection (d.p.i.) (GFP pixel counts quantified using ImageJ in lower right corner of each image) and represented graphically in (B). (C) SINV-GFP was detected in honey bee protein lysates from the injection site, thorax (Th), and distal sites (D; head and abdomen) each d.p.i. by Western blot analysis using α-GFP (green lower arrow) and a-actin (top arrow) antibodies. Western blot bands were quantified using ImageJ and the GFP:actin ratio was normalized to 1* at day 2 p.i.. doi: 10.1371/journal.pone.0077263.g001 S3). In contrast, the GFP to actin ratio was not as dramatically reduced in bees co-injected with NTPs, 2.3, or SINV dsDNA 2.4 ( Figure S3). Overall the majority of bees that were individually assayed by Western-blot exhibited strong (> 50% less virus-GFP) dsRNA-mediated reduction in virus production. Specifically, 58% of bees co-injected with SINV dsRNA (n=66) and 69% of bees co-injected with DCV dsRNA (n=88) exhibited more than 50% reduction in virus-GFP. In contrast, only 3% of bees co-injected with NTPs (n=37) and 10% of bees coinjected with SINV dsDNA (n=20) had low levels of virus-GFP, which were likely a consequence of ineffective injections since these percentages were similar to the percentage of bees that were not infected after intra-thoracic infection of virus (8%, n=87). Western blot analysis was used to routinely assess dsRNA-mediated knock down of virus in bees and quantitative PCR (qPCR) was used to determine the relative abundance of SINV-GFP genome copies in a subset of differentially treated individuals ( Figure 2F). Bees injected with only virus and buffer had the highest SINV genome copy number (1.65x10 6 + 4,100 copies) 3 days p.i. relative to those treated with either virus specific dsRNA (SINV dsRNA) or non-specific-dsRNA (DCV dsRNA) in conjunction with virus, 1.66x10 5 +1,050 and 2.32x10 5 + 1,070 average genome copies, respectively. We conclude that co-injection of non-specific dsRNA effectively limited Sindbis virus infection in honey bees.

Transcriptional level evaluation of virus and dsRNA induced immune responses
We hypothesized that some of the honey bee genes involved in antiviral defense would be transcriptionally regulated and that a subset of these genes would also be regulated in response to dsRNA, a viral PAMP. In order to both examine the role of previously characterized insect immune genes and potentially identify additional honey bee host factors responsible for antiviral defense, we utilized custom oligonucleotide microarrays to evaluate the transcriptional profiles of virus-infected or dsRNA-treated bees [27,[31][32][33]83].
Long-term laboratory rearing of honey bees is not feasible, therefore we obtained frames of brood from colonies outside the laboratory. To minimize genetic variability between samples, all microarrayed bees were obtained from a single brood comb from a naturally-mated queen and were therefore age-matched half-sisters. Bees that co-emerged in the lab were grouped into samples of 10 and then intrathoracically injected with SINV-GFP virus, dsRNA (dsRNA SINV 928 bp; 1 μg), or buffer (mock-infected). After 3 days, honey bee samples were individually analyzed by Western-blot, fluorimetry, and qPCR as described above. In addition, since bees are readily infected with pathogens we utilized the arthropod pathogen microarray (APM), which is capable of detecting all common honey bee pathogens (at levels greater than 10,000 genome copies) and over 200 additional arthropod pathogens [3], to screen for pre-existing conditions prior to transcriptome analysis. APM profiling ensured that the predominant infection and prevailing immune response in the bees analyzed by microarray analysis was due to the experimentally introduced SINV-GFP or dsRNA and not a consequence of pre-existing infections (Table S1). All microarrayed samples were also tested for deformed wing virus (DWV), a common honey bee pathogen detected in APM analysis of two RNA-treated bees (Table S1). DWV qPCR determined that all the samples used for transcriptome analysis had DWV levels less than 10 4 copies of virus per 500 ng RNA (< 10 5 copies per bee) (Table S1), whereas SINV-GFP levels were approximately 10 6 copies per 500 ng RNA (~ 10 8 copies per bee) ( Figure 2, Table S1). Levels of common bee viruses are typically more than 10 6 copies per 100 ng RNA (> 5x10 8 copies per bee) [3,[84][85][86]. In general, bees that emerged in laboratory conditions had less infections than older bees obtained from colonies housed outside of the laboratory [3] (Table S1, positive controls).
We identified five representative bees from each group (mock-infected, SINV-GFP infected, or dsRNA-treated) that were relatively free of honey bee pathogens and analyzed gene expression at the transcriptional level using honey bee oligonucleotide microarrays. Individual bee transcriptomes at three-days post-injection were examined since at this timepoint the virus was actively replicating and spreading at levels detectable by Western blot ( Figure 1D), no secondary ill-effects of laboratory housing were observed, and this was the timepoint used to assess dsRNA-mediated reduction of virus. Also, we reasoned that the wound-response due to injections (although controlled by mock-infection) would be reduced by three days post-injection, thus facilitating investigation of virus and dsRNA-triggered responses.
Custom honey bee gene expression microarrays were utilized to identify genes that were differentially expressed in virus-infected and dsRNA-treated bees compared to mockinfected (buffer-injected) bees. The honey bee oligonucleotide microarray has features representing 9,867 unique genes [25], 2,729 EST probes, 22 probes specific for honey bee pathogens, and control spots including 11 probes for green fluorescent protein (GFP) (Table S2) [27,[29][30][31][32]. Experiments were designed to meet Minimum Information About a Microarray Experiment (MIAME) standards and the protocols utilized for cDNA synthesis, labeling, hybridization, analysis (Rcode for LIMMA) and the raw microarray data obtained in this study were deposited at ArrayExpress [www.ebi.ac.uk/ arrayexpress (accession no. E-MEXP-3608)]. To facilitate gene expression comparisons we utilized a reference-design strategy in which each Cy5-labeled cDNA experimental sample was hybridized with a standardized Cy3-labeled reference honey bee cDNA sample. Microarrays were scanned and global analysis of the differentially expressed genes (DEGs) was performed using Bioconductor's Linear Models for Microarray Analysis (LIMMA) [35] implemented in the R/ Bioconductor platform [87] and data from duplicate array features was condensed into a single value for statistical analysis [16,19]. The data from the five individual arrays in each treatment group were analyzed in a pairwise comparison to five individual arrays from the mock-infected (buffer-injected) group using a contrast matrix. The data were fit to a linear model using EBAYES and false-discovery rate FDR p-value <0.05; and an adjusted p-value, which accounts for multiple pairwise comparisons, less than 0.05 (adj. p-value < 0.05) was selected to define significant differences between each experimental condition as compared to mock-infection. Using Honey bees were inoculated with SINV-GFP or co-injected with (B) nucleic acid treatments including: virus-specific dsRNA (SINV dsRNA); non-specific dsRNA corresponding to Drosophila C virus (DCV) sequence (DCV dsRNA); poly(I:C); NTPs; non-specific dsRNA corresponding to luciferase sequence (LUC dsRNA); or SINV dsDNA visualized on a 1% agarose gel. (C) Fluorescence microscope images of a mock-infected bee (left panel), a SINV-GFP infected bee (middle panel), or a bee infected with SINV-GFP in the presence of nonspecific dsRNA DCV. (D) Western blot analysis using α-GFP and α-actin antibodies demonstrated reduced SINV-GFP in bee protein lysates pooled from ten bees treated with dsRNA (lanes 1, 2, and 5) and poly(I:C) (lane 3), but not in bees treated with NTPs (lane 4) or dsDNA (lane 6) compared to bees inoculated with virus alone (virus). Mock-infected (mock) bees injected with buffer had no detectable GFP. (E) Boxplot of GFP fluorescence (510 nm, relative arbitrary units) in individual honey bee lysates (10 bees per condition). A Welch two-sample t-test was performed to compare honey bee lysates from each treatment group to virus-infected bees and determined statistically significant differences between bees co-inoculated with virus-specific dsRNA SINV p=0.023, with non-specific dsRNA DCV p=0.020, and with poly(I:C) p=0.0038, whereas co-injection of NTPs p=0.41 or dsDNA p=0.36 were not significantly different from virus alone (virus). Bees injected with virus only (virus) exhibited greater fluorescence compared to mockinfected controls (m); p=0.00001. (F) Relative abundance of SINV-GFP in individual bees (5 per condition) from different treatment groups was assessed by RT-qPCR. The bees treated with virus only had the highest SINV copy number 3 days p.i. (1.65x10 6 copies + 4,100 copies per 500 ng RNA, ~1.65x10 8 per bee), as compared to those in which either virus-specific SINV dsRNA (1.66x10 5 +1,050) or non-specific DCV dsRNA (2.32x10 5 + 1,070) was co-injected with virus. these criteria, 134 genes were induced and 112 genes exhibited decreased expression levels in response to virusinfection; dsRNA-treatment resulted in the induction of 171 genes and reduced expression of 286 genes ( Figure 3A). Seventy-five DEGs were shared between virus-infected bees and dsRNA-treated bees ( Figure 3B, Tables S3-S5).
To globally evaluate the putative function of transcriptionally regulated genes in response to virus-infection and dsRNAtreatment, Gene Ontology (GO) annotations were assigned to all Apis mellifera microarray probes with Drosophila melanogaster orthologs in terms of their associated biological processes, molecular functions, and cellular components (Tables S3-S4) [27,32,88]. Biological process assignments were further categorized into seven functional groups: immune response, including honey bee AMPs; transcription; splicing; rRNA processing and RNAi; signaling; trafficking; metabolism; translation and protein folding; and chromatin regulation ( Figure 4 and Tables S3-S5). This analysis indicated reduced expression of several immune genes and perturbation of cell signaling, trafficking and metabolism in virus-infected or dsRNA-treated honey bees ( Figure 4). Over-representation analysis, for pathways with more DEGs that would be expected by chance, identified pathways in virus-infected bees (i.e., the eicosanoid signaling pathway, p-value=0.009) and dsRNAtreated bees (i.e., oxidative phosphorylation, p=0.00005) ( Table  S6).
The most significant differentially expressed genes were the antimicrobial peptide (AMP) encoding genes apidaecin and hymenoptaecin, which exhibited decreased expression in both virus-infected and dsRNA-treated bees ( Table 1). Reduction of Apidaecin 1 was verified by qPCR analysis; expression in virus-infected bees was 20% + 6% that of mock-infected   Tables S3-4). Biological process assignments were further categorized into seven functional groups: immune response (red); transcription, splicing, rRNA processing and RNAi (blue); signaling (purple); trafficking (green); metabolism (yellow); translation and protein folding (orange); and chromatin regulation (teal). A small fraction of the DEGs (10-15%) belonged to other biological process gene ontologies (black) and a large fraction was not assigned (gray). Each pie chart represents 100% of the DEGs and gene numbers are reported in each section. *Each oligonucleotide pair was analyzed as a single gene and since multiple microarray probes corresponded to apidaecin variants the majority of annotated immune effectors in this Figure (7 of 10 in virusinfected bees and 8 of 12 in dsRNA) correspond with apidaecin 1.
controls and was similarly reduced to 14.5% + 3% in dsRNAtreated bees ( Figure 5A). As expected, the microarray probes corresponding to GFP were identified as the most induced genes in the virus (SINV-GFP) infected group, accounting for 11 of the top 15 induced probes identified (Table 2). This result, which was confirmed by SINV-GFP qPCR analysis, also validated our microarray analysis methods. Additional genes that were identified by microarray analysis and verified by qPCR include increased expression of unc-80 in virus-infected bees and lethal(3) in dsRNA-treated bees ( Figure 5).
We examined the expression of genes involved in insect immunity, as well as additional genes identified in transcriptional studies of virus-infected fruit flies and mosquitos, in virus-infected and dsRNA-treated honey bees (Table S7) [22]. Specifically, our candidate gene analysis included genes in the Toll, Imd, and JAK-STAT signaling cascades, as well as those involved in dsRNA uptake and RNAi-mediated antiviral immunity in fruit flies and mosquitos [15,39,[41][42][43][44][45][46][47][48]. The majority of these genes (95%), even those that are transcriptionally regulated in other insects, were not differentially expressed in honey bees three days post SINV-GFP infection (Table S7). This analysis suggests that at three days post-infection only clathrin heavy chain, which is involved in endocytosis, and eater-like (nimrod C1), which encodes an EGF-motif containing protein involved in phagocytosis, were significantly activated in virus-infected honey bees (Table S7). Together the transcriptional data presented herein suggest that antiviral defense in honey bees may involve unique genes, alter or suppress known immune pathways, or may be predominantly post-transcriptionally regulated.

Discussion
Honey bees are agriculturally important eusocial insects that are readily infected with ssRNA viruses. Virus infections may be rapidly cleared or result in death of individual bees and entire colonies [3,6,89,90]. Although the outcome of infection depends critically on the honey bee antiviral immune response, the mechanisms involved in mitigating honey bee viral infections are not well-characterized. In this study we established a model of honey bee virus infection and Quantitative PCR (qPCR) was used to validate honey bee gene expression microarray results. (A) apidaecin 1 expression in virus-infected bees was 20% + 6 of mockinfected controls (112% + 30%) and dsRNA-treatment also reduced expression to 15% + 3% of mock-infected controls; ttest p=0.01 and p=0.01 respectively. (B) unc-80 expression in virus-infected bees (126% + 9%) was higher than mockinfected controls (101% + 4%) (p=0.046). (C) lethal (3) expression increased in response to dsRNA-treatment to 156% + 14 relative to its expression in mock-infected bees 101% + 6% (p=0.03. Standard SYBR qPCR was performed for apidaecin 1 gene expression, whereas Taqman probe and primer sets were employed for other assayed genes. qPCR reactions were performed as technical triplicates of five individual bee cDNAs (biological replicates) per condition; the expression of each gene-of-interest (GOI) was compared to the housekeeping gene (Rpl8) and then the expression level of each GOI in five individual bees per experimental condition was calculated as a percentage relative to its expression in individual mock-infected bees using the ΔΔC(t) formula (yaxis), error bars represent +/-standard error of the mean (SEM), and statistical significance was assessed using Student's t-tests. determined that co-administration of dsRNA, regardless of sequence specificity, reduced virus infection. Our results suggest that that honey bees activate antiviral immune pathway(s) in response to non-specific dsRNA. Understanding these mechanisms is critical to assessing the role of viruses and/or immune deficiencies in colony losses and may lead to the development of strategies that enhance honey bee survival.
There are several challenges to investigating honey bee antiviral responses, including genetic heterogeneity within and between colonies, the prevalence of pre-existing infections, and the lack of infectious honey bee virus clones. To address these challenges we used age-matched half-sisters in experiments, screened for pathogens prior to gene expression analysis, and established SINV-GFP as a model for honey bee virus infection. Sindbis virus is a +ssRNA virus that has been used to investigate antiviral immunity in fruit flies and mosquitos. Importantly, SINV does not encode a suppressor of RNAi [45,91,92], so although it is in a different family of +ssRNA viruses (i.e. Togaviridae) than honey bee viruses, it is a useful model for investigation of dsRNA triggered immune mechanisms. Furthermore, use of this model virus facilitated monitoring infections using the GFP-tag, ensured no prior infection, and reduced the effects of virus-induced immune repression often observed in co-evolved virus-host pairs.
Previous research in fruit flies and mosquitos has shown that RNAi is the primary mechanism of antiviral defense, while the role of other immune pathways in mitigating virus infection varies with the specific virus-host pair. The honey bee genome encodes the RNAi machinery and orthologs of the Toll, Imd, JAK-STAT, and JNK immune pathways [22,25]. RNAimediated gene knock-down of endogenous genes has been utilized as a reverse genetics tool for examining gene function [18,[93][94][95][96][97][98], although off target effects have also been reported [99]. RNAi has also been implicated in antiviral defense, although the molecular details require further investigation [50][51][52]73]. We investigated the role of dsRNA in honey bee antiviral immunity and determined that co-injection of both virus-specific and nonspecific dsRNA reduced virus production. This result was surprising in context of studies demonstrating virus-specific dsRNA mediated reduction of virus copy number and reduced honey bee lethality (i.e., [50][51][52]73]). However, careful examination of the experimental details revealed several differences between the work presented herein, which examined the effect of co-injection of a single dsRNA dose (1 μg / bee) on the replication of a model virus, and other Apis mellifera studies, which examined treatment of larva and adults with dsRNA prior to infection by feeding [50][51][52], higher doses of dsRNA (6 μg / bee) [50], different doses of semi-purified viruses [50,51], various evaluation methods at different timepoints post virus-inoculation [73], and one study that examined Apis cerana (Eastern honey bee) larva that were fed dsRNA prior to and during virus inoculation (5 μg dsRNA / larva) [52],; the experimental details may account for the differences observed. Clearly, additional studies are needed to more fully understand dsRNA-mediated immune responses in honey bees, which likely involve both virus-specific RNAi-mediated and non-specific dsRNA mediated mechanisms. Interestingly, the dsRNA-mediated response we observed in honey bees also differed from that observed in fruit-flies infected with the same model virus, Sindbis-GFP [15]. In fruit flies, co-injection of virus-specific dsRNA triggers an RNAimediated antiviral response, whereas co-injection of nonspecific dsRNA did not reduce virus titers [15]. In contrast, the general non-sequence-specific immune response to dsRNA that we observed in honey bees may be analogous to mammalian innate immunity. In mammals, several PRRs recognize dsRNA as a viral PAMP and signal innate antiviral responses, but to our knowledge the work described herein is the first report of non-specific dsRNA-mediated reduction of virus titer in adult insects. Interestingly, non-specific dsRNAtriggered antiviral immunity has also been reported in sand-fly cells [100], Chinese oak silk moth pupae [101], and shrimp (reviewed in 102], [13,[103][104][105][106][107], and dsRNA-triggered transcriptional responses in Bombyx mori larva suggest that dsRNA is recognized as a PAMP [108]. However, the PRRs, genes, and pathways involved in these processes have not been characterized. Intriguingly, expression of antiviral immune response genes (i.e., vago and vir-1) in Drosophila [60,109] and mosquitos [110,111] and up-regulation of Dicer-2 in response to Dcr-dsRNA [112] and to GFP-dsRNA in Bombyx mori [108] indicates that Dicer-2 may also serve as dsRNA (PAMP) sensor, response element, and/or mediator of innate immune signaling. Cumulatively, these studies suggest that non-specific dsRNA mediated immune responses are important in arthropods; however future studies are needed to determine the molecular mechanisms and relative importance of this response in specific host -pathogen interactions.
Transcriptional profiling of virus and dsRNA-treated honey bees facilitated analysis of previously identified immune gene orthologs (Table S7; [22]) and putatively identified genes involved in virus and dsRNA triggered innate immunity (Tables 1-2). In our study, the majority of canonical insect immune pathway members did not appear to be transcriptionally regulated 3-days post injection with either virus or dsRNA (Table S7) [22]. Interestingly, antimicrobial peptide (AMP) encoding genes apidaecin, hymenoptaecin, and abaecin, exhibited significant decreased expression in virus-infected and dsRNA-injected bees relative to mock-infected bees ( Table 1). These effector molecules are typically transcriptionally induced after insect immune challenge via Toll, Imd, or JNK signaling pathways [22]. Reduced AMP expression suggests that honey bee antiviral responses involve different effector molecules than those involved in antibacterial and antifungal immune responses. Intriguingly, parasitic mite infestation coupled with wing deformity and higher amounts of deformed wing virus (DWV) also reduced AMP and immune related enzyme expression in honey bees [72]. Likewise, AMP production was not induced by infection with acute bee paralysis virus (ABPV) [75] or in CrPV-infected fruit flies [59], whereas a positive correlation between hymenoptaecin expression and DWV titer was demonstrated in naturally infected honey bees [4]. The majority of genes that exhibited increased transcription in response to virus-infection or dsRNA-treatment were not canonical insect immune genes (Table 2 and Tables S3, S4 and S7). For example, virus-induced A. mellifera unc-80, whose function is unknown in A. mellifera, functions in vesicular trafficking and proper localization of NCA ion channels in C. elegans. Thus, we predict this protein functions similarly in honey bees. Lethal (3) expression increased in response to dsRNA and its function in honey bees is also unknown, but it may play a role in rRNA processing or splicing similar to its Drosophila and mammalian (NOP14 nucleolar protein) orthologs. It is noteworthy that several of the genes we identified by microarray analysis have multiple splice variants (i.e., DSCAM and fruitless). Interestingly, the induction and involvement of genes without previously characterized roles in immunity was also observed in Drosophila C virus (DCV)infected fruit flies [60] and O'Nyong-Nyong virus (ONNV)infected A. gambiae mosquitos [56]. Overall the transcriptional data presented herein suggest that antiviral defense in honey bees may involve unique genes and signal transduction cascades, alter or suppress known immune pathways, or may be predominantly post-transcriptionally regulated. Future studies involving more samples, additional time-points, and tissue-specific analyses are required to confirm the involvement of the genes identified herein, as well as further examine the potential role of canonical immune genes, in honey bee antiviral immune responses.
In summary, we established a virus infection model to investigate antiviral immunity in honey bees and determined that co-injection of dsRNA, regardless of sequence-specificity, mitigated virus infection. Our results suggest the involvement of an RNAi-independent immune mechanism. Future studies will focus on determining the mechanism and relative contribution of non-specific dsRNA-mediated antiviral immunity in honey bees.

Honey bees
Honey bees are not well adapted to the laboratory setting, therefore for our experiments we used newly emerged honey bees (~24 hours old). Frames of emerging bees were obtained with permission from privately owned, managed colonies in San Francisco, CA. No additional permissions were required since the managed honey bee colonies were privately owned colonies and Apis mellifera is not an endangered or protected species. Young female worker bees (all half-sisters) that emerged in the laboratory were collected daily and housed in modified deli-containers (similar to those described by Evans et al. 2009 [113]) at room temperature for the duration of the experiment. Water and sugar were provided and replenished daily through the duration of the experiment.

Sindbis virus (SINV-GFP) production and injection into honey bees
Sindbis virus expressing enhanced GFP (SINV-GFP) was produced by transfecting in vitro transcribed RNA from the pTE3'2J with GFP inserted into the XbaI site into BHK cells [15,77]. SINV-GFP was titered on BHK cell monolayers via plaque assay. SINV-GFP (3,750 plaque forming units (PFUs)) was diluted in buffer (2 μl of 10 mM Tris, pH 7.5) and injected into the thorax using a Harbo syringe (from Honey Bee Insemination Service; www.honeybee.breeding.com/ HarboAssembly.html) equipped with disposable borosilicate needles made from capillary tubes (0.8-1.10 x 100 m) using a micropipette puller (Sutter Instruments Model P-87). Honey bees were immobilized via incubation in a cold room (4°C) for 20 minutes and with insect pins and forceps during injection; after injection bees recovered at room temperature within 5 minutes. Honey bees were infected with 3,750 PFUs of SINV per bee. This virus dose is modest compared to Drosophila studies which typically utilize 250-2500 PFUs per fly; a newly emerged female worker honey bee (~150 mg) weighs ~ 200x more than an adult female fruit fly (0.8 mg) [15]. This dose allowed for a natural progression of infection over the course of the experiment and at 3-days post-injection (p.i.) virus abundance could be easily assessed by microscopy and Western-blot analysis. Virus was either injected alone or in conjunction with dsRNA (1 μg).

dsRNA preparation
dsRNA was generated by in vitro transcription with T7 RNA polymerase [15,18]. In brief, T7 promoter containing dsDNA PCR-products (1-10 μg), amplified with the primers listed in Table S8, served as templates for T7 polymerase (NEB) transcription (100 μl reactions: rNTPs (each 7.5 mM final), RNase OUT (40 units) (Invitrogen), buffer (400 mM HEPES pH 7.5, 120 mM MgCl 2 , 10 mM Spermidine, 200 mM DTT); reactions were carried out at 37°C for 4-12 hours. DNA templates were removed by incubating with DNAse I (1 unit; Fermentas) for 15 minutes at 37°C. ssRNA products were ethanol precipitated, suspended in 200 μl DEPC-treated water, and annealed by heating the reaction to 100°C for 10 minutes and then slowly cooling to room temperature. dsRNA products were purified by phenol:chloroform extraction followed by ethanol precipitation; dsRNA for injection was suspended in 10 mM Tris pH 7.5. dsRNA quality and quantity were assessed by agarose gel electrophoresis and spectrophotometer (NanoDrop 2000c, ThermoScientific).

Honey bee protein lysate preparation and analysis
Individual honey bees were homogenized in 400 μL buffer (10 mM Tris pH 7.5, 1X Complete Protease Inhibitor cocktail (Roche) in a 2 mL micro-centrifuge tube containing one sterile zinc-coated steel ball bearing (5 mm) using a TissueLyzer II (Retsch), for 4 minutes at 30 Hz. Lysates were clarified by spinning for 12 minutes at 12,000 x g. Fluorescent analysis of lysates was performed using a microplate reader (Safire, TECAN), and relative GFP abundance was measured by SDS-PAGE coupled with Western blot. For Western blot analysis, lysates were combined with Laemmli buffer (95°C for 3 min.), run on 4-12% acrylamide gels (Mini-PROTEAN TGX, BioRad), transferred to PVDF membrane (Immobilon-P), blocked with 5% milk in TBS buffer containing 0.1% Tween-20, incubated overnight with primary antibodies (anti-GFP sc-8334 or antiactin sc-1616; Santa Cruz Biotech.), HRP-conjugated antirabbit secondary antibody (ECL, GE Healthcare), and developed with SuperSignal West Pico Chemiluminescent Substrate (Pierce). ImageJ was used to quantify the sum of the pixels within actin and GFP regions in order to calculate and compare SINV-GFP levels in samples [20,21].

Honey bee RNA isolation and purification
TRizol reagent (Invitrogen) was added to individually homogenized bees and RNA was isolated according to manufacturer's instructions and further purified using Qiagen RNeasy columns, including on-column DNase Treatment (Qiagen). RNA quality was assessed using an Agilent 2100 Bioanalyzer.

Quantitative PCR (qPCR)
cDNA synthesis reactions were performed with SuperScriptIII (Invitrogen) and random hexamer primers according to manufacturer's instructions. qPCR was performed in triplicate wells using 2 μL of cDNA as template in 20 μl reactions composed of SensiFAST SYBR Mastermix (Bioline), MgCl 2 (3 mM), and forward and reverse primers (600 nM each) on a CFX Connect Real Time instrument (BioRad). The qPCR thermo-profile consisted of a single pre-incubation 95°C (3 min), 40 cycles of 95°C (5 s), and 60°C (20 s) for SINV and Apis m. Rpl8 primer sets (Table S8). Target SINV qPCR amplicons were cloned into the pGEM-T (Promega) vector and sequence verified. Plasmid standards, containing from 10 9 to 10 2 copies per reaction, were used as qPCR templates to assess primer efficiency and generate the SINV-specific standard curve used to quantify the viral genome copy number.
The linear standard equation generated by plotting the crossing point (Cp) versus the log 10 of the initial plasmid copy number for the qSINV primer set was as follows: SINV Cp = -4.3 +40.3, R 2 =0.977. The detection limit of this qPCR primer set was 10 3 copies and specific qPCR amplicons had Cp values of < 26. Values obtained from the no RT control reactions were below the detection limit of the assay and subtracted from treatment group values before graphing. For DWV qPCR plasmid standards containing from 10 4 to 10 7 copies per reaction were used as qPCR templates to assess primer efficiency and generate the DWV-specific standard curve used to quantify the viral genome copy number. The linear standard equation was: DWV Cp = -3.3 +32.0, R 2 =0.996. The detection limit of this qPCR primer set was 10 4 copies and specific qPCR amplicons had Cp values of < 25. An estimate of the number of viral genomes per bee can be obtained by multiplying the reported qPCR copy number values by 100. This estimate is based on the following: typical RNA yield was approximately 50 μg per bee, and each qPCR reaction was performed on cDNA generated from 500 ng RNA, therefore each well represents 1/100 th of an individual bee. qPCR using a host primer set, Apis m. Rpl8, was also performed on pooled cDNA templates from each treatment group; the Rpl8 primer efficiency was evaluated using a cDNA dilution series (Cp = -3.5 + 17, R 2 =0.996). Lastly, no-RT control reactions using pooled RNA as the template, and negative control reactions using water in place of template were performed in triplicate on each plate. Melt point analysis and 2% agarose gel electrophoresis ensured qPCR specificity. TaqMan® Fast Advanced Master Mix and Taqman® primer and probe sets were used for following Apis mellifera genes (target region; qPCR efficiency test): unc-80 (100-250; Cp = -3.2 +25.5, R 2 =0.99), lethal(3) (600-800; Cp = -3.8 +25.5, R 2 =0.99) (Applied Biosystems). The ∆∆C(t) method was used to calculate the expression of the gene-of-interest (GOI) in individual bees. In brief, ∆C(t) for each sample was calculated by subtracting the Rpl8 C(t) from the GOI C(t). The ∆∆C(t) was calculated by subtracting mock-infected ∆C(t) values from the ∆C(t) values associated with each treatment group (5 individual bee samples per condition). Lastly, the percent gene expression for each GOI was calculated using the following formula: 2^-∆∆C(t) x100 = % Gene Expression. Since apidaecin 1 gene has many repetitive elements that complicate qPCR primer design, we utilized primers developed by Evans et al. [22] for SYBR-green qPCR and modified thermocyling parameters (40 cycles of 95°C (30 s), 58°C (20 s), 72°C (60 s)); Cp = -3.7 +15.3, R 2 =0.99. Melt curve analysis indicated a single predominant peak, but gel electrophoresis revealed faint smaller bands in addition to the expected 493 bp product; minor product priming likely contributed to increased standard error of the mean (SEM) values in this assay.

Honey bee gene expression microarray (sample preparation and hybridization)
To minimize variability between samples all arrayed bees were obtained from a single brood comb from a naturallymated queen, therefore all the bees were age-matched halfsisters. The five representative bees from each condition (virus-infected, 3,750 PFU/bee; dsRNA-treated, 1 μg/bee; and mock-infected) were selected for microarray analysis were relatively pathogen free (assessed by APM and qPCR analysis, Table S1) [3]. To facilitate gene expression comparisons between multiple treatment groups we utilized a referencedesign strategy in which each Cy5-labeled experimental sample was hybridized with a standardized Cy3-labeled reference sample. A complex RNA mixture representing hundreds of bees of various ages exposed to different treatment groups, served as the "reference RNA" sample.

Honey bee gene expression microarray analysis
Microarrays were scanned using an Axon 4B scanner and GenePix Pro software (Molecular Devices) for gridding and initial analysis. We performed global analysis of the differentially expressed genes (DEGs) using Bioconductor's Linear Models for Microarray Analysis (LIMMA) [35] implemented in the R/Bioconductor platform [87]. In brief, microarrays were scanned and the data was loaded into the LIMMA program implemented in the R/Bioconductor platform. Fluorescent intensities were log 2 transformed and corrected for the global median background signal using NORMEXP, followed by LOESS normalization within each array, and AQUATILE-SCALE normalization between arrays [35]. DUPCOR from LIMMA was used to correlate data from duplicate array features and the data was condensed into a single value for statistical analysis [16,19]. The data from the five individual arrays in each treatment group were analyzed in a pairwise comparison to five individual arrays from the mockinfected group (for details see R-code) using a contrast matrix. The data was fit to a linear model using EBAYES and falsediscovery rate FDR p-value <0.05; an adjusted p-value <0.05 was selected to define significant differences between each experimental condition as compared to mock-infection. Experiments were designed to meet Minimum Information About a Microarray Experiment standards. All the protocols utilized for cDNA synthesis, labeling, hybridization, analysis (Rcode for LIMMA) and all the raw microarray data obtained in this study was deposited at ArrayExpress [www.ebi.ac.uk/ arrayexpress (accession no. E-MEXP-3608)]. In addition, the complete microarray expression data is provided in Excel format in Tables S4-S5. Gene Ontology annotations were based on Drosophila melanogaster orthologs identified for 8,902 of the 14,400 unique probes (62%) on the microarray (Tables S2-S4). Sincẽ 25% of the genes on the array are represented by multiple probes, we used unique Entrez Gene identifiers to identify GO biological process (GOBPID) categories that were overrepresented based on the number of regulated genes as compared to the number of genes on the honey bee genome microarray in each GO biological process category. Of the 8,902 microarray features with Drosophila orthologs, Ingenuity Pathway Analysis (IPA) (Ingenuity Systems, www.ingenuity.com) mapped 5,168 (~36% of the microarray features). Since some Entrez Gene IDs corresponded with multiple array features, there were actually 3,282 unique genes involved in characterized pathways. These 3,282 genes served as a background for over-representation analysis (ORA) using a false discovery rate (FDR) threshold of p-value < 0.05. Using these criteria, the number of DEGs analyzed for virus-infected bees was 62, whereas for dsRNA-treated bees it was 162.  6). ImageJ was used to quantify the pixels in equal sized GFP and actin regions of each lane in order to calculate the GFP to actin ratio in individual bee lysates. As expected, the mock-infected (mock, lane 1) bee had no detectable GFP, therefore the ratio the pixel counts within the GFP and actin regions of lane 1 was set to 1(*) to facilitate comparisons between lanes 2-31. (EPS) Table S1. Pathogen test results for honey bees utilized for transcriptional analysis. Arthropod Pathogen Microarray (APM) results determined that SINV-GFP was the predominant infection in virus-infected bees, whereas dsRNA-treated and mock-infected (buffer-injected) controls were not appreciably infected with common honey bee pathogens; APM oligonucleotide list (worksheet 1), APM results (worksheet 2). APM evaluation is performed using a computer algorithm called "Bee-Predict", described in Runckel* and Flenniken* et al. PLoS ONE 2011, but the raw APM data is provided in order to support our assessment that the predominant infection in virusinfected bees was SINV-GFP, and that all bees were not appreciably infected with other honey bee pathogens. Since there are multiple spots on the array for each pathogen (see oligo list) one positive spot is not sufficient for pathogen detection, often these single postive spots are located adjacent to a highly positive control spot and therefore are not valid detections.

Supporting Information
(XLSX) Complete Honey Bee Gene Expression Microarray Results. Complete microarray results for virusinfected bees compared to mock-infected bees. Differentially expressed genes with adjusted p-value < 0.05 (worksheet 1) and all data (worksheet 2). Gene Ontology terms for biological process, molecular function, and cellular compartment were assigned and biological process assignments were further categorized into seven functional groups: immune response (red); transcription, splicing, rRNA processing and RNAi (blue); signaling (purple); trafficking (green); metabolism (yellow); translation and protein folding (orange); and chromatin regulation (teal). A small fraction of the DEGs (10-15%) belonged to other biological process gene ontologies (black) and a large fraction was not assigned (gray). Each pie chart represents 100% of the DEGs and gene numbers are reported in each section. *Each oligonucleotide pair was analyzed as a single gene and since multiple microarray probes corresponded to apidaecin variants the majority of annotated immune effectors in this  Table S4.
Complete Honey Bee Gene Expression Microarray Results. Complete microarray results for dsRNAtreated bees compared to mock-infected bees. Differentially expressed genes with adjusted p-value < 0.05 (worksheet 1) and all data (worksheet 2). (XLSX )   Table S5. Comparison of DEGs (adjusted p-value < 0.05) in virus-infected and dsRNA-treated honey bees using Venn diagram analysis (worksheet 1); virus only DEGs, RNA Only DEGs, and overlap between those two DEG lists. We further examined overlapping DEG list, by sorting based on virus list fold-change (FC) (worksheet 2), dsRNA-treated list FC (worksheet3), virus list adjusted p-value (adj.p.val) (worksheet 4), and dsRNA-treated list adjusted p-value (worksheet 5). (XLSX) Table S6. Pathways Over-Representation Analysis. Overrepresentation analysis was performed on the DEGs involved in canonical pathways; the 3,282 unique genes on the array that are involved in characterized pathways served as a background and a false discovery rate (FDR) threshold of pvalue < 0.05 was set. Using these criteria the number of differentially expressed (including both induced and reduced) genes analyzed virus-infected or dsRNA-treated as compared to mock treated controls was 62 and 162, respectively. Overrepresented pathways for each experimental condition are listed as well as the p-value associated with each test. (PDF)