Four DGE-tag libraries were generated from our experimental groups: control bees without a pollen diet (V-P-), control bees fed with pollen (V-P+), varroa-parasitized bees without a pollen diet (V+P-) and varroa-parasitized bees fed with pollen (V+P+). In total, around 14 × 10⁶ tags (excluding the adaptor sequences) were sequenced from each library (see Table 1 for details). The libraries from varroa-parasitized bees contained more aligned tags (~7.3 × 10⁶ tags) than non-parasitized bees (~4.5 × 10⁶ tags). This higher number of aligned tags is due to the increase in tags matching virus sequences, which reached 1, 417, 481 in the V-P- library, 824, 212 in V-P+, 4, 258, 724 in V+P- and 5, 061, 614 in V+P+. These virus tags represent 32% of the total number of aligned tags in the V-P- library, 18% in V-P+, 60% in V+P- and 67% in V+P+. This clearly shows that varroa triggers a boost of pathogen loads in bees. This was likely induced by repeated varroa feeding on bees hemolymph that not only affect bee physiology but also contribute to the transmission of different viruses 19 21 27 28 . In addition, we noticed that healthy (non-parasitized) bees, which were fed with pollen, had an overall lower load of pathogen tags than bees restricted to a sugar diet. However, few viruses were present at significant different levels between each treatment group (see the Effects of pollen feeding and varroa parasitism on virus prevalence).

These distinct tags and their genomic count as well as raw data are available from NCBI Gene Expression Omnibus (GEO) database with the accession number: GSE25161. The number of distinct aligned tags was similar between each library (Table 1) and their distribution shows a consistent pattern across the 4 DGE libraries, with most of the distinct tags (34 to 40%) appearing only once and around 40% appearing between two and ten times (Additional file 1). The percentage of tags with a high count number dropped dramatically and only 1% of distinct tags have a count above 1, 000. Such pattern of DGE-Tag profiling was previously observed in honey bee heads 29 and in others insects and vertebrates 30 31 . In total, around 8, 000 honey bee genes were maps from the four abdominal libraries (Table 1), which represents approximately 60% of the genes so far identified from the honey bee genome 8 . As a comparison, 55 to 60% of the genes were also found to be expressed in the bee brain using DGE and microarray analysis 29 32 .

The number of genes, whose expression level was significantly affected by pollen feeding or varroa-parasitism at an adjusted P-value for multiple comparison (Q-value) <0.01 is given in Table 2 and the lists are provided in Additional file 2. While a similar number of genes were up- and downregulated by pollen nutrition in healthy bees, varroa-parasitism had a major inhibitory effect on gene expression. This inhibition at the transcription level might explain the loss of weight and the decreased metabolism induced by varroa in emerging bees 20 .

To assess the reliability of our DGE-tag profiling, we checked whether the regulation of some candidates genes was consistent with the pattern of expression previously reported in earlier studies and performed reverse-transcriptase PCR (RT-PCR) on those genes. Vitellogenin (Vg) is a yolk protein taken up by developing oocytes and associated to egg production in queens 33 34 but also has antioxidant functions that protect bees from oxidative stress and enhance longevity 35 36 . Therefore, Vg level can be assimilated to an indicator of longevity. Since, pollen nutrition promotes the development of fat body 12 17 and Vg is synthesized in the fat body 35 , we expected an increase in Vg level in bees fed with pollen as previously found 37 . That was both confirmed by the DGE and RT-PCR analysis (Additional file 2 and Figure 1). This suggests that one way by which pollen feeding increase longevity 7 is via the increase of Vg level. Interestingly, varroa parasitism decreased the Vg level in bees fed with pollen (see also 38 ), but this level was still significantly higher than in parasitized bees restricted to a sugar diet.

Another interesting gene is malvolio (mvl), which encodes a manganese transmembrane transporter involved in sucrose responsiveness 39 . In honey bees, foraging workers that are more responsive to sucrose have a higher level of mvl mRNA in the brain than young nurses taking care of the brood inside the hive 40 . Pollen feeding induced an expression pattern opposite to Vg (Figure 1). Bees fed only with sugar had a higher level of mvl in the abdomen. Since old foragers feed mainly on carbohydrates and young nurses feed on both carbohydrates and pollen 41 , emerging bees that are restricted to a sugar diet tend to develop a mvl expression profile similar to foragers as compared to bees fed with pollen, which are more similar to nurse bees. Indeed, as opposed to poor diet, a high protein diet might stimulate nurse tasks via the development of hypopharyngeal glands used for secreting larval food 7 . Similarly, varroa parasitism induced an increase in mvl level (although not significant in bees fed with pollen), suggesting also a forager profile.

The expression level of the gene prophenoloxidase (PPO), involved in insect immunity 42 , was also confirmed by RT-PCRs (Figure 1) and previous data. Indeed, varroa parasitism inhibited PPO expression, which was consistent with the immune suppression previously reported 21 , but pollen feeding did not affect PPO level (see also 17 at the enzyme activity level).

Finally, we checked the expression level of the Drosophila ortholog spaetzle, which is a member of the Toll immune signaling pathway against fungi and bacteria 42 . The DGE-tag profiling of this gene, confirmed by RT-PCR, showed that pollen feeding significantly increased the number of spaetzle transcripts in healthy bees (Figure 1), which suggests that pollen nutrients can enhance immune function in bees. The expression of this gene was reduced by varroa parasitism in bees fed with pollen, supporting again the immune suppression induced by varroa. Altogether, these results validated the DGE-tag profiling of the four libraries.

Gene Ontology analysis was performed to explore which functional components are regulated by pollen feeding and affected by varroa parasitism. Although the pollen wall, composed of sporopollenin and cellulose, is poorly digested by honey bee digestive enzymes, 50 to 98% of pollen contents can be extracted from ingested pollen, depending on pollen type and age of the bee 10 . Bees of 9 days-old, close to the age of bees that we analyzed, can digest as much as 95% of pollen contents 41 . The upregulation of many metabolic functions in bees fed with pollen highlights the high digestibility of pollen (Figure 2A). Indeed, in response to pollen nutrients, composed mostly of proteins and lipids, genes involved in proteolysis, lipid metabolism and coding for peptidases were upregulated. Carbohydrate metabolism was also upregulated by pollen feeding, suggesting a digestive response to pollen starches. Besides the metabolic categories, "oxidative phosphorylation" and "generation of precursor metabolites and energy" showed significant enrichment among the tested categories. This might not be surprising since nutrients are used to produce energy, like ATP via oxidative phosphorylation.

Groups of genes implicated in translational processes and ribosome biogenesis were downregulated by pollen feeding (Figure 2B). In all eukaryotes, ribosome biogenesis and translation regulation are essential for tissue growth and are regulated by food intake. Proper regulation of those processes are required to ensure cell and tissue homeostasis upon changes in nutrition and are inhibited during starvation for dietary nutrients 2 . Therefore, an upregulation of ribosome biogenesis and translation was expected upon pollen feeding. However, the majority of pollen consumption and tissue development (e.g. hypopharyngeal glands, ovaries and fat body) occur within the first week after adult emergence 43 . The prior completion of physiological development before our analysis could explain the downregulation of translational processes and ribosome biogenesis.

The effect of pollen nutrition on honey bee physiology was dramatically altered by varroa parasitism. Whereas groups of genes involved in ribosome biogenesis and translation were also downregulated, metabolic categories were no longer upregulated (Figure 2A). Only small sets of genes involved in proteolysis, peptidase activity and carbohydrates metabolism were still upregulated by pollen nutrition. In contrast most of the genes implicated in protein metabolism were downregulated in varroa-parasitized bees (Figure 2B), demonstrating that varroa mites inhibited the digestion and/or the use of proteins. In addition, the "lipid metabolic process", "oxidative phosphorylation" and "generation of precursor metabolites and energy" categories were no longer enriched. Metabolic processes were also found to be mostly downregulated by varroa parasitism in prepupae 22 . Altogether, these results indicate that parasitized bees could not correctly assimilate and use the pollen nutrients required for their physiological development.

This was further confirmed by the effect of varroa on the bee transcriptome. Functional analysis showed that all of the previous categories, such as ribosome biogenesis, translation, metabolic functions and energy production, were inhibited by the varroa mite (Figure 2D). The inhibition by varroa of ribosome biogenesis and translation, essential to cell growth, might lead to lower tissue development. The lower concentrations of proteins and carbohydrates induced by varroa 20 might be explained by our results on the inhibition of metabolic genes and genes coding for endopeptidase. This reduction of metabolic reserves induced by varroa was reported at the emergence 20 , therefore the downregulation of metabolic genes lasting 8 days after the emergence in bees fed with pollen, demonstrates that pollen nutrients are not able to compensate those losses.

A large set of genes were regulated by pollen feeding both in parasitized and non-parasitized bees and presented a similar expression pattern (Table 3 part 1). However, a subset of genes was upregulated by pollen feeding in healthy bees and downregulated in parasitized bees. A similar pattern was observed regarding the effect of varroa on bees fed with or without pollen (Table 3 part 2). In order to decipher the biological meaning of those opposite regulations, we performed a functional analysis of those subsets. Among the subset upregulated by pollen feeding in healthy bees but downregulated in parasitized bees, groups of genes are involved in ribosome biogenesis, translation, protein metabolism and energy production (Additional file 3 part A). The subset upregulated by varroa parasitism in bees fed without pollen but downregulated in pollen-fed bees, showed significant enrichment for genes coding for oxidoreductase and involved in amino acid metabolism (Additional file 3 part B). Those results confirmed the negative effect of varroa on those previously reported functions. Indeed, the energy metabolism and nutrient demand of the mite is high, utilizing up to 25% of the nutritional reserves of pupae 44 . The decrease in metabolic reserves lasting in adults might come from the development of viruses transmitted by mites at the larvae or pupae stage.

Similar results to the functional analysis were found with the pathways analysis (Additional file 4). Using the later analysis, we also studied the effect of pollen and varroa on amino acids. Indeed, honey bees require ten essential amino acids for their adult development: arginine, histidine, lysine, tryptophan, phenylalanine, methionine, threonine, leucine, isoleucine, and valine 45 . This means those amino acids cannot be synthesized de novo by the bees and therefore must be supplied directly in the diet either as free amino acids or digested from proteins. We therefore looked at the amino acid metabolism pathways that were significantly enriched after pollen feeding and found that four of them (arginine, leucine, isoleucine, and valine) were upregulated (Additional file 4). Interestingly, five of those amino acid pathways were downregulated by varroa (lysine, tryptophan, leucine, isoleucine, and valine), confirming that varroa significantly reduces the development of bees. However, none of these essential amino acid pathways were significantly affected in varroa-parasitized bees fed with pollen compared to parasitized bees restricted to a sugar diet, suggesting that pollen nutrients compensated for the negative impacts of varroa on the metabolism of those amino acids.

The insulin/TOR pathway is a conserved signaling cascade that functions as a nutrient sensing pathway by linking food-intake to animal growth and metabolism, including reproduction and lifespan 46 47 . In honey bees, this pathway plays a major role in the regulation of aging of individuals 35 48 but also at a social level by regulating the division of labor between worker bees 49 50 . Less is known about the nutrient regulation of this pathway in honey bees. However, a recent study showed that the inhibition of Insulin receptor substrate expression in peripheral tissues (as opposed to central brain tissues) made forager bees collect pollen rather than nectar 51 . By analyzing the DGE-tag profiling of peripheral tissues (abdomen) we found that most of the genes from the mTOR signaling pathways were upregulated by pollen nutrition in healthy bees (Figure 3). The insulin-like peptide 2 (ILP-2) expression notably increased in response to pollen nutrition, as it was found in response to amino acids 48 , which might promote nurse physiology that extends bee longevity 52 . Altogether these results show that the insulin/TOR pathway responds positively to pollen nutrition and might trigger the pollen-induced physiological development. The opposite effect was observed regarding ILP-1, with a higher expression under poor nutrition conditions 50 . However, those inverse patterns might actually be explained by the functional differences of ILP-1 and ILP-2 52 . In varroa-parasitized bees, pollen nutrition tended to inhibit this pathway. This could be caused by the higher prevalence of viruses in parasitized bees fed with pollen compared to parasitized bees fed only with sugar (see Effects of pollen feeding and varroa parasitism on virus prevalence). Indeed, varroa parasitism (coupled with a higher viral load) induced a similar pattern of the TOR pathway (Figure 3). The parasite inhibits protein metabolism and decreases the nutrient reserve of bees (see above), which might be explained by the downregulation of the TOR pathway mediating tissue growth in response to nutrient intake. This is further confirmed by the inhibition of slimfast by varroa parasitism in bees fed with pollen. The downregulation of this gene, which encodes an amino acid transporter and is a downstream target of TOR pathway, causes a deficient global growth in Drosophila similar to that seen in individuals raised under poor nutritional conditions 46 53 .

TOR is known to respond to the presence of amino acids and induces an upregulation of ribosome biogenesis, translation 2 54 55 56 57 and energy metabolism 58 59 required for tissue growth. Interestingly, the downregulation of TOR pathway was followed by the downregulation of each biological process (Figure 3), indicating that the same mechanisms of the nutrient sensing pathway are found in honey bees. The upregulation of the pathway in pollen-fed bees was followed by an increase in energy metabolism but the ribosome biogenesis and translation were both inhibited. As explained above, it is possible that tissue development that requires the activation of translational processes was already completed.

The expression level of genes did not always match well to the activator/inhibitor functions established in the Drosophila TOR pathway. The expression patterns correspond to global DGE-tag profilings of whole bee abdomens but many genes possess tissue-specific functions. Therefore, targeting specific tissues like the fat body, which has important storage functions associated with nutrition, should provide more accurate expression data for pathway analysis.

Since pollen nutrition and varroa increases and reduces the longevity of bees 12 60 , respectively, we looked at the expression pattern of genes that have been found to affect longevity in Drosophila. Those genes were found by using the functional analysis from DAVID 6.7 bioinformatic resources 61 and GoToolbox 62 . We found 20 genes including Vg, known to affect longevity, to be differentially expressed between the different treatments. A mutation at the loci of each of those genes causes a decrease in longevity except for Tor, whose inhibition increase longevity 63 . Among those genes, we expected to find a larger set to be upregulated by pollen feeding than downregulated and inversely regarding varroa effect. According to our expectation, 7 out of 10 genes regulated by pollen intake were upregulated in non-parasitized bees (Figure 4A). For example, Sod and Trxr-1, which encodes a cytoplasmic Cu-Zn superoxide dismutase and a thioredoxin reductase, respectively 64 , are known to increase lifespan 65 66 , thanks to the degradation of superoxide radicals and H₂O₂ or the neutralization of reactive oxygen species. The positive effect of pollen feeding on the expression pattern of both genes was confirmed by reverse-transcriptase PCR (RT-PCR) analysis performed on a different set of experimental bees (n = 8 pools of bees per treatment, Additional file 5). Those genes might be involved in the pollen-induced molecular mechanisms that enhance life expectancy of honey bees. However, Sod 2 (mitochondrial Mn superoxide dismutase), which is also involved in degradation of superoxide radicals and H₂O₂ 64 , was not affected by pollen feeding (Additional file 5), contrary to the downregulation found with the DGE analysis. Due to the higher level of Vg (Figure 1), a longer lifespan is expected in parasitized bees supplied with pollen compared to parasitized bees fed only with sugar (Figure 4A). However, except Vg, genes that were regulated by pollen intake were all downregulated. Whether pollen nutrition can affect varroa resistance remains to be tested but it might be possible that the effect of pollen on the genes affecting longevity is compensated by the increase in Vg. The picture was clearer with regards to the effect of varroa on the 19 genes affecting longevity (Figure 4A and Additional file 5). All of them were downregulated (including Sod 2 in presence of pollen, Additional file 5) supporting a decrease of longevity induced by varroa.

Next, we examined the effect of pollen intake and varroa on the expression pattern of immune genes, that have been previously annotated 42 . Few immune genes were upregulated by pollen feeding in healthy bees (Figure 4B). Three of them encoded antimicrobial peptides (AMPs) (Lyzozyme-2 and -3, Defensin-1), which might be linked to a higher content of fat body, the major site of synthesis of AMPs. Interestingly, in the absence of immune challenge (healthy bees), Spaetzle and the gene coding for the peptidoglycan recognition protein (PGRP)-LC, which are activators of the Toll and Imd pathways respectively, were both upregulated by protein feeding. The upregulation of those key immune genes confirmed by RT-PCR analysis (Figure 1 for Spaetzle and Additional file 5 for Defensin-1 and PGRP-LC) suggests that bees might be able to mount an efficient immune response to pathogen infections. In parasitized bees, only Imd, a specific gene for antimicrobial defense, was upregulated by pollen feeding. A slight increase of Imd expression was also observed with RT-PCR analysis although not significant (p = 0.058, Additional file 5). In healthy bees, Imd expression did not change after pollen feeding (Additional file 5). Since, protein and amino acid nutrition can enhance immune functions in humans 4 67 and insects 68 69 , an upregulation of immune genes was expected. The functional analysis showed that in varroa-parasitized bees fed with pollen, a significant number of genes involved in protein metabolism were downregulated compared to parasitized bees fed only with sugar (Figure 2B). A deficiency in protein metabolism might have affected the immune system of those bees. This immunosupression could be explained by the higher virus load in parasitized bees fed with pollen compared to parasitized bees fed only with sugar (see Effects of pollen feeding and varroa parasitism on virus prevalence). However, when fed with pollen, it is possible that parasitized bees invest more energy in other defense mechanisms than immunity which is inhibited by varroa, e.g. the antioxidant system (higher Vg level in bees fed with pollen).

Varroa induced the inhibition of most of the immune genes in bees fed with or without pollen (including PGRP-LC, Additional file 5). However, a few immune genes were upregulated, like the AMPs (Defensin-1, Hymenoptaecin, Apidaecin-1 and Lyzozyme-3) and cact-1, which controls the production on AMPs 70 . The upregulation of Defensin-1 by varroa parasitism was confirmed by RT-PCR analysis (Additional file 5). Those genes, except Lyzozyme-3 have already been found to be upregulated by a immune challenge 42 . The mechanisms triggering the immunosupression are still misunderstood but one explanation might be that the deficient protein metabolism induced by varroa and associated viruses (see above) block the development of immune function.

Immune genes that were upregulated by pollen feeding and varroa (Figure 4B) belonged to the antibacterial peptide family or were key activators of immune pathways. Since the characterization of immune defense in honey bee is still at his infancy, it is possible that the others immune genes don't actually play a major role in immune response in bees. Indeed, genes from immune signaling pathway have usually different roles, like developmental functions.

Using the DGE-tag libraries from honey bee abdomen samples, we performed a metagenomic analysis to determine the effects of pollen feeding and varroa on the prevalence of 10 honey bee viruses: Chronic bee paralysis virus RNA 1, Chronic bee paralysis virus RNA 2, Sacbrood virus, Deformed wing virus, Black queen cell virus, Acute bee paralysis virus, Kashmir bee virus, Kakugo virus, Varroa destructor virus 1 and Israel acute paralysis virus. All of them were identified in our DGE-tag libraries but very low numbers of tag copies of the Chronic bee paralysis virus RNA 1, Chronic bee paralysis virus RNA 2, Sacbrood virus, Black queen cell virus, Acute bee paralysis virus, Kashmir bee virus and Israel acute paralysis virus were found (from 1 to 65, see data deposited in GEO database: GSE25161) and no difference in their abundance was detected between treatments. However, tags mapping the deformed wing virus (DWV), Varroa destructor virus (VDV) and Kakugo virus (KV) were highly prevalent and significantly affected by the different treatments (Table 4). Varroa destructor is an efficient vector of bee viruses by promoting their transmission and development (DWV 19 21 , VDV 27 , KV 28 ) but the presence of viruses in control bees can be explained by a vertical transmission from queens and drones to worker offspring through fertilized eggs 71 .

As expected, varroa provoked a boost of virus development (Table 4). Since AMPs have a broad-spectrum antimicrobial activity against bacteria, yeast, fungi, parasites but also viruses 72 , the upregulation of AMPs in varroa-parasitized bees (Figure 4B) might represent the immune response to the high viral infection. Different studies showed that pollen nutrients can help the bees to fight against parasites and pathogens 14 15 . For example, a recent study found that pollen nutrition reduce the prevalence of DWV in normal adult bees 16 . We obtained similar results in healthy (non-parasitized) bees with pollen nutrition reducing the load of VDV. However, we observed the inverse pattern with DWV and KV. Viruses utilize cell machinery to multiply, so higher number of cells available could contribute also to a better development of viral population. For example, DWV has been found to multiply in the bee fat body 73 and this tissue is more developed in pollen-fed bees. We also found that pollen feeding increased viruses prevalence in varroa-parasitized bees, indicating pollen nutrition is not effective for limiting virus loads in heavily infected bees. Larger physiological machinery might, to the contrary, favor their multiplication, but it might also help bees to withstand viral infection.

The experiment was performed with local hybrid honey bees (A. m. ligustica/A. m. mellifera). In order to obtain a large number of varroa-infected bees, an infested colony that was not treated with miticide was selected. The multiple-mated queen was placed in a queen cage to stop egg-laying for 10 days. After 10 days, all the brood produced before the queen confinement was sealed and the mites had no brood cell to parasitize. Then, the queen was transferred to a queen-excluder for 3 days with an empty frame. The brood produced by the queen on this frame was then uniformly parasitized by varroa mites. Three weeks after the queen laid eggs in the queen-excluder, the frame was removed from the colony and the sealed cells were opened carefully to remove the adult workers just before emergence and to check for the presence of reproducing varroa mites. Varroa-parasitized bees with deformed wings were discarded due to their extremely short lifespan.

Varroa-parasitized and non-parasitized bees (control) were placed in distinct cages in the dark at 32°C and 70% RH and fed ad libitum with candy (30% honey, 70% powdered sugar) and water. To simulate as much as possible the colony rearing conditions, cages were exposed to a Beeboost^® (Pherotech), releasing one queen-equivalent of queen mandibular pheromone per day. Half of the groups were also fed with a polyfloral pollen diet, which was replaced every day and prepared with fresh pollen obtained from our colonies in Avignon (France) and 1/10 water. Thus we obtained 4 experimental groups: control bees without a pollen diet (V-P-), control bees fed with pollen (V-P+), varroa-parasitized bees without a pollen diet (V+P-) and varroa-parasitized bees fed with pollen (V+P+). Since their pollen consumption diminishes when bees are 8-10 days old 41 , bees from each group were flash frozen in liquid nitrogen at the age of 8 days. One group (pool of 12 bees) per treatment was used for comparing the 'average' responses to the different treatments with the digital transcriptome analysis. Single replicate analyses have generated consistent results in previous studies 30 31 74 , but confirmation of some of the results obtained from digital expression analysis was performed with RT-PCR on four new experimental replicates (see below). The quantities of pollen consumed in the V-P+ and V+P+ groups were similar (8.2 ± 2 and 7.7 ± 2.6 mg per bee and per day, respectively).

Twelve abdomens per group were homogenized in Trizol (Invitrogen cat. No. 15596-026) and RNA extraction was carried out as indicated in the Qiagen RNeasy kit (Qiagen cat. No. 74104) for total RNA with on-column DNase I treatment (Qiagen cat. No. 79254).

Sequencing libraries were constructed using Illumina's DGE Tag Profiling DpnII Sample Prep kit (Illumina cat. No. FC-102-1007) according to the manufacturer's instructions (cat. No. 1004241 rev.A). Briefly, polyA+ RNA were isolated from 2 μg of total RNA using Sera-magnetic oligo-dT beads. Reverse transcription was performed on immobilized polyA+ RNA using SuperScript II Reverse Transcriptase (Invitrogen cat. No. 18064-14). Second strand was synthesized using DNA Polymerase I. Immobilized cDNAs were digested with DpnII. Digested fragments were washed off, and the retained cDNA fragments were ligated to GEX DpnII Adapter 1. Tags were released by MmeI digestion generating 20 bp tags that are dephosphorylated using CIAP and ligated to GEX Adapter 2. The adapter-ligated cDNA tags were enriched using 15 cycles of PCR, and purified on a 6% Novex TBE PAGE gel (Invitrogen cat. No. EC6265Box). 85-bp fragments were excised from the gel, eluted and precipitated. Tags were quantified on an Agilent Bioanalyzer using DNA1000 chip (Agilent cat. No. 5067-1504). Each library was diluted to 10 nM, denatured, and diluted again to 8 pM. 100 μl of the diluted library were hybridized on a lane of an Illumina's Flow Cell. Clustering and 36 cycles sequencing were performed according to Illumina's instructions.

We first filtered the raw data via the default Illumina pipeline quality filter, which calculates a Chastity score as the ratio of the highest of the four (base type) intensities to the sum of highest two. This score is used to remove clusters with low signal to noise ratio (often caused by clusters being too close to each other so their signals bleed into one another). The sequences were passed through the chastity filter of value > = 0.6. Then, adapters were trimmed from tag sequences by an in-house script. Reads displaying sequence ambiguity were excluded from further analysis. For each library, read counts were recorded for every unique sequence. Tag sequences were mapped to the honey bee genome (BeeBase Genome - release 2 http://genomes.arc.georgetown.edu/drupal/beebase/?q=home) with Eland v1.4, and only perfect tag matches of coding sequences were kept for statistical analysis. Mapping was also performed on different sequences of honey bee virus genomes (Chronic bee paralysis virus RNA 1: GenBank EU122229, Chronic bee paralysis virus RNA 2: EU122230, Sacbrood virus: AF092924, Deformed wing virus: AJ489744, Black queen cell virus: AF183905, Acute bee paralysis virus: AF150629, Kashmir bee virus: AY275710, Kakugo virus: AB070959, Varroa destructor virus 1: AY251269 and Israel acute paralysis virus: EF219380).

In the case of tag census, the probability distribution of the measured parameter 'tag abundance' is known a priori and follows a hypergeometric distribution. Accordingly, the probability that 2 census of the same transcript in 2 different conditions are significantly different can be calculated without replicate. Differentially expressed tags were therefore identified by a Fisher's exact test with Benjamini-Hochberg method, which controls the expected proportion of incorrectly rejected null hypotheses (type I errors), to compute the False Discovery Rate and a cut-off Q-value of 0.01 as implemented in the SAGE Genie resource http://cgap.nci.nih.gov/SAGE. The expression pattern of the tags was based on their sequence odds ratio between two pools A and B: (sequences in A/total sequences in pool A)/(sequences in B/total sequences in pool B).

We explored whether any particular molecular functions or biological processes from the differentially expressed gene lists were represented by larger numbers of genes than expected on the basis of chance. Drosophila melanogaster orthologs were identified by reciprocal best BLASTX match to bee genes, and Gene Ontology (GO) terms were assigned based on annotation of Drosophila genes. GOToolBox 62 was used to identify overrepresented terms (hypergeometric tests followed by the Benjamini Hochberg correction for multiple testing). Enrichment analysis of bio-pathways were performed using DAVID 6.7 bioinformatic resources 61 .

RT-PCR was performed on selected genes known to be affected by nutrition and varroa parasitism (vitellogenin (Vg), malvolio (mvl), Prophenoloxidase (PPO) and Spaetzle (spz)), three genes involved in lifespan (Superoxide dismutase (Sod), Superoxide dismutase 2 (Sod2), Thioredoxin reductase 1(Trxr-1)) and three immune genes (Peptidoglycan recognition protein LC (PGRP-LC), defensin1, Immune deficiency (Imd)). The RT-PCR analysis was carried out on a different set of bees obtained by using the same experimental procedure (see above). Four experimental replicates per treatment were performed and 2 pools of 3 bees per replicate (cage) were analyzed (n = 8 pools/treatment). The transcript abundance was measured for with an M × 3000 P QPCR Systems (Agilent) and the SYBR green detection method (Agilent cat. No. 600828). RT-PCR values of the selected genes were normalized by using an exogenous human RNA control (β-actin) spiked during cDNA synthesis performed with the SuperScript III (Invitrogen, France). Relative expression was calculated by raising 2 to the power of the difference in Ct values. Primer sequences are listed in Additional file 6.