Comparisons of Varroa-infected and non-infected pupae from the 4 colonies studied pinpointed the expression of 32 genes whose expression varied significantly between the two types of pupae. Among them, 15 (47%) of these cDNAs were significantly up-regulated and 17 (53%) down-regulated in bees exposed to Varroa. The magnitude of the difference of expression is small in all cases. The single exception was EST (BB160020A20G03) which was over-expressed 20-fold in bees infected with Varroa. BLAST searches indicated that this EST matches the honey bee virus (AY292384), deformed wing virus 19.

There were 116 cDNAs whose expression varied with bee genotype, 47 (40.5%) of them up-regulated and 69 (59.5%) down-regulated. Two genes were regulated both as a function of Varroa infection and bee genotype. Dlic 2 is down-regulated in Varroa-parasitized bees and up-regulated in tolerant bees, whereas Strn-Mlck is down-regulated both in Varroa-parasitized and tolerant bees. The full list of genes with significant differences in expression is presented in Additional file 1.

To add biological meaning to the relatively large amount of microarray-derived data, we searched the GeneOntology database for the putative biological processes and molecular functions for the genes that showed significant differences in expression. Among these 148 genes, 19 associated with the presence of Varroa and 68 associated with bee genotype had information in Gene Ontology.

To unravel functional differences related to Varroa parasitism and bee genotype, we analyzed sets of genes for overrepresentation in functional GO categories according to biological process and molecular function (Additional file 2). Regarding biological process, genes involved in the regulation of protein metabolism, embryonic development and reproduction show a significant enrichment among the categories tested for the genes that varied in expression with Varroa infestation. A set of genes involved in cytokinesis, nervous system development, behavior, signal transduction and transcription-DNA dependent emerged from the analysis of genes that varied with bee genotype. The cluster analysis showing genes sharing the same GO term is shown in Table 1. Most of the genes affected by Varroa infection are involved in protein metabolism and their function is related to transferase and catalytic activity. The set of genes differentially expressed between tolerant and sensitive bees are mainly involved in transcription and neuron development. These genes have a nucleic acid binding function and are associated with pyrophosphatase GTPase, transferase and catalytic activity.

To further evaluate our ability to detect significant variation in gene expression at low fold-difference levels, a set of 4 functionally annotated bee transcripts was subjected to RT-qPCR analysis (Table 2). Differently expressed levels of cDNAs were quantified and are shown in Fig. 1. In accord with microarray analysis, cDNAs from Varroa-parasitized bees were up regulated for genes baz and CG9520 (unassigned) (Fig. 1a and 1b); or from tolerant bees for genes Alh and Hr78 (Fig. 1c and 1d). In all cases, expression levels quantified by qPCR are represented by the mean of 3 measurements (maximum and minimum values are indicated in Fig. 1).

The pathways suggested by this study to be associated with honey bee response to Varroa parasitism are schematized hypothetically in Fig. 2. Responses to Varroa can be grouped in two main categories, one related to deformed bee adults occasioned by the presence of the Deformed Wing Virus (DWV) often associated with Varroa parasitism, and a second category of genes that are related to cognitive impairment.

One of the consequences of Varroa parasitism is a decline in immune capacity which appears to induce the proliferation of viruses such as deformed wings virus in bees 6. The down-regulation of the autophagic-specific gene 18 (Atg18) detected here is noticeable. Autophagy is an important mechanism for innate immunity against bacteria and viruses 2122. The candidate innate immunity gene poly U binding factor 68 Kd (pUf68) 23 is also down-regulated in Varroa parasitized bees. By decreasing autophagy and immunity processes in bees, Varroa might favor the proliferation of DWV. Interestingly, the Varroa-parasitized bees displayed high levels of DWV viral RNA (about 20-fold; Additional file 1). The boost of DWV multiplication might cause cellular and molecular damage, and thus the observed production of genes for protein repair (Pcmt) 24, and the labeling of proteins for degradation (Nedd8) 25. In contrast to these findings, we did not see a decrease on transcript abundance of immune pathway members 20 (Evans et al., 2006) found on this array. In fact, transcripts for the gene Rab7, a plausible regulator of immunity, were up-regulated in Varroa-parasitized bees.

The most notable symptoms of Varroa-parasitized bees are disfigured, small adults with deformed legs and wings 46. Our results show that at the transcriptome level the presence of Varroa down-regulates two genes involved in developmental processes, slg and dlg1. It has been shown that the Drosophila sugarless (sgl) gene regulates wingless signaling 26, which has a critical role in developmental processes. The gene dlg1 has been implicated in the control of proliferation of Drosophila imaginal discs programmed to produce adult structures at metamorphosis 27. Although a down-regulation of these genes could have been induced by the presence of Varroa (via the DWV virus), a specific link with the development of deformed adults of virus-infected bees would need to be further investigated.

Varroa infestation does not always cause wing deformity. Adults may sometimes appear to be normal morphologically, but there are mite effects on adult bee behavior. In particular, mite-parasitized foragers display a decrease in learning capability 28, prolonged absences from the nest and a lower rate of return to the colony 29. A decrease in neuronal capacities involved in learning and navigation is a possible cause. Although the physiological mechanisms underlying reduced performance by bees in the presence of Varroa remain unknown, our results show that, the gene pale encoding tyramine hydroxylase is down-regulated in pupae parasitized by Varroa. Interestingly, the gene pale is needed for dopamine synthesis, which stimulates the nervous system and has many functions in the brain, including important roles in neural development, behavior and cognition, motor activity, motivation and learning. It is also interesting to note that the Dlic2 and Atg18 genes, both down-regulated in Varroa-parasitized bees, are enhancers of the blue cheese gene (bchs)30, which is up-regulated in tolerant bees (see below). The bchs gene has been reported as preventing progressive neural degeneration in aged flies 31. It is plausible, if these changes are chronic (still expressed in adults), that infested bees have a higher rate of neuronal apoptosis when aging, which might explain why foragers (the oldest bees in the nest) have difficulties in learning and orientating in flight.

Grooming behavior seems to be one of the main characters involved in the tolerance of bees to Varroa 4. Similarly, selection for tolerance to Varroa seems to be possible by selecting bees with a high level of hygienic behavior 32. Hygienic behavior is understood as the ability of bees to uncap and remove infected brood. Although the molecular mechanisms underlying this trait remain unknown, some insights are provided here based on the large-scale comparative analysis of transcriptome differences between Varroa-tolerant and sensitive bees.

A disproportionately high fraction of the genes differentially expressed between tolerant and susceptible bees are involved in the development of the nervous system (Fig. 2B). A large part of these genes are down-regulated in tolerant bees compared to sensitive bees, including the genes futsch, scratch (scrt), otk, Myosin heavy-chain-like (Mhcl), groucho (gro), kekkon-1 (kek1) 26333435363738. Fringe (fng), also down regulated in Varroa-tolerant bees, plays an important role in the positive regulation of Notch signalling pathway involved in the regulation of genes that control multiple cell differentiation processes during embryonic and adult life, such as neuronal function and development 39. Finally, a gene involved in locomotory behavior, single-minded (sim), is down-regulated in tolerant bees. Drosophila mutant sim flies are only able to walk in circles and this phenotype is due to defects in the central brain complex 40.

Several genes involved in neuron excitability are up-regulated in Varroa-tolerant bees compared to sensitive bees. Purity of essence (poe) (also named pushover) is involved in neuronal excitability and may play a role in responsiveness to environmental stimuli and behavior. For example, fly mutants display behavioral defects such as sluggishness, uncoordination and defective flight 41. This gene has been shown to be down-regulated in honey bees by the queen mandibular pheromone, which slows the behavioral maturation of workers, e.g. transition from inside hive work (nurse) to foraging activity(forager) 42. The GluClα gene, a glutamate-gated Cl^- channel specific to arthropods, which is also up-expressed in tolerant bees, is known to modulate neuronal membrane excitability 43. The paralytic gene (para), known to be important in the conducting of nerve action potentials in flies 44, is also up-regulated. Seen together, these genes suggest a mechanism by which Varroa-tolerant bees are more sensitive to external stimuli than Varroa sensitive bees. Also, the Dynein heavy chain 64C gene (Dhc64c), up-regulated in tolerant bees, is required for proliferation of mushroom-body neuroblasts 45. Mushroom bodies have an important role in insect cognition and are known to be involved in learning and memory, particularly for smell.

Interestingly, several genes involved in olfaction (smell impaired 21F smi21F, suppressor of the white-apricot gene su(w^a), poe, para, Rogdi) are up-regulated in Varroa-tolerant bees compared to sensitive bees 4647484950. However, the Down syndrome cell adhesion molecule gene (Dscam) and four wheel drive (fwd) also have a role in olfaction, and they are down-regulated in Varroa-tolerant bees 34. The differential expression of this group of genes depending on bee genotype is of major importance considering that in the hive, olfaction and neuronal sensitivity together may play a major role in the detection of Varroa infested cells. Observations made on two earlier generations of the bee stocks here studied have been shown that Varroa-tolerant bees have better ability to detect the mite 51. Previous studies showed that hygienic bees have a higher olfactory sensitivity and responsiveness than non-hygienic bees 525354. They are notably able to discriminate between odors of healthy and diseased brood at a lower stimulus level, suggesting that olfaction and responsiveness play a key role in hygienic behavior. In addition, observations made on the behavior of the brood, have shown open cells containing destroyed parasitized bee pupae in tolerant experimental colonies, which is in agreement with a bee hygienic behavior against Varroa. If these gene candidate pathways for Varroa behavioral tolerance can be further confirmed, the results suggest that they already set up during the pupae stage.

There were differences in expression between tolerant and sensitive bees for genes involved in increased resistance to toxins like Ahcy13 (Adenosylhomocysteinase at 13) involved in detoxification 55, and para involved in resistance to insecticide pyrethroids and DDT (Dichloro-Diphenyl-Trichloroethane) 56. Two genes linked to immunity, Dsam and otk, members of the Immunoglobulin gene superfamily in Drosophila 57 were down-regulated in Varroa-tolerant bees, while one galectin-family gene (an apparent ortholog to Drosophila CG32226) appears to be upregulated.

For most of the genes showing significant differences in expression in the present study, the magnitude of these differences was small. Recent reports have shown that microarrays can significantly underestimate gene expression changes and therefore if a severe cut-off is applied, this approach might miss important changes in gene expression. Recent reports have validated the capability of a microarray approach to detect small gene expression differences 58; 87% of a set of parasite-specific genes displayed changes 2-fold or less in D. melanogaster challenged to a protozoan parasite 17. Similarly, about 60% of genes of the lepidopteran, Spodoptera frugiperda, revealed around 0.5-fold changes in the transcript levels associated with virus infection 14. As one possible explanation for low level gene regulation, it has been suggested in Drosophila that much of the response to parasite attack probably does not involve de novo gene expression but post-transcriptional events 18. Another possibility is that many small differences in gene expression reflect subtle modulation by a large number of factors acting in cascade. The small changes reported here might also represent an underestimation of tissue-specific effects obscured by whole-body analysis. Future studies of Varroa effects on honey bees should analyze specific tissues, and our results suggest a focus on the brain would be fruitful. Previous studies have demonstrated extensive regulation of brain gene expression in conjunction with honey bee behaviour 42596061.

Both, sensitive and tolerant honey bee colonies here studied, belong to non related local strains of the same A. mellifera population which is bred in the Laboratory of Bee Biology and Protection, Institut National de la Recherche Agronomique, Avignon, France. Four colonies were used for this study; two (T222 and T757) had survived for 11 years with no chemical treatment against Varroa despite the widespread presence of this parasite in the locale, and displayed a very low rate of parasitism 10. The other two colonies (S21 and S38) showed a very high level of parasitism. No acaricide treatment was applied to avoid pesticide bias. Parasitism intensity was determined by counting the number of mites that died naturally and accumulated on the floor of the hive in each colony during the year (method described in Ellis et.al. 62). Estimations based on data for 5 years (4 measurements/year) showed that the infestation rate in colonies S21 and S38 was on average 10 times higher than in colonies T222 and T757.

The reproduction of Varroa mites can be affected by traits in developing bees 6364. We therefore examined gene expression in pupae. Honey bee pupae were collected from the four bee colonies at the blue-eye stage at the start of cuticle pigmentation, making it possible to determine the bee developmental stage 65. Capped brood cells were opened and two samples of 50 parasitized and non-parasitized pupae were collected. A pupa was considered to be parasitized if a reproductive Varroa was found with it in the honeycomb cell. The pupae were collected and snap frozen using N2 and stored at -80°C until RNA extraction.

We compared parasitized and non-parasitized pupae from both susceptible and tolerant colonies. To minimize the effect of genetic variation, pools of full-sister pupae (related by 75% due to haplodiploidy 66) were collected in equal numbers from each of the study colonies.

To identify full sisters, one hundred bees per colony were genotyped (50 parasitized and 50 not parasitized) using Ap53 and A107 microsatellite loci; these loci have been shown to be effective for this purpose 67. DNA was extracted from 2 legs/pupa, and DNA extraction and PCR amplification were as described 67, except that the PCR products were detected on a MegaBACE DNA Analysis System 1000 (Molecular Dynamics Inc., USA). The genotype data were used to assemble sets of 8 pupae belonging to the same full sibling group. For each of these, subsets of pupae of Varroa-parasitized and non parasitized bees were analyzed in microarray experiments. This procedure minimizes the effects of intra-colonial genetic variation on gene expression.

We compared parasitized and non-parasitized pupae from both susceptible and tolerant colonies using a direct loop design. Direct comparisons were made between parasitized and unparasitized bees from sensitive colonies and from tolerant colonies (Fig. 3). All comparisons were performed in dye-swap on 2 biological samples. A total of 8 samples were thus contrasted using 16 arrays in all (2 biological replicates and 2 technical replicates). Each sample consisted of a fraction of the total RNA obtained from a pool of 8 bees. Full-sister bees were assembled as pools from specific patrilines detected in each colony.

The sets of bees were freeze-dried and ground in liquid nitrogen. 30 mg (about 3% of the powder obtained) was used for each RNA extraction using the RNeasy Plant Mini Kit (Qiagen, Courtaboeuf, France) following the manufacturer's protocol for animal tissues. The total RNA solution was treated in liquid by RNase-Free DNase I and purified on column RNA Cleanup (Qiagen, Courtaboeuf, France) according to the manufacturer's protocols. The purity and concentration of RNA were determined by OD measurements in a spectrophotometer. RNA extraction was validated by specific PCR amplifications following RT-PCR of both the glyceraldehyde 3 phosphate dehydrogenase 1 (XM_397363) and the epsilon-tubulin 1 (XM_394700) genes. The quality of the extracted RNA (integrity and size distribution of total RNA) was verified by 1.2% agarose gel electrophoresis in TAE buffer and ethidium bromide staining.

10 μg total RNA was used for cDNA synthesis using fluorescent Cy3-dCTP or Cy5-dCTP (Amersham Biosciences, Orsay, France) and the Pronto!TM plus System kit (Promega, Charbonnières, France) for labeling. Dye-labeled cDNAs were purified on cleanup columns (Promega, Charbonnières, France). 55 pmol of the labeled cDNA per dye was used per slide. Prehybridation, hybridization and washing steps were performed according to 61. The slides were hybridized at 42°C and air dried by centrifugation for 2 min at 800 rpm at room temperature.

The arrays were scanned on a GenePix 4000A scanner (Axon Instruments, Foster City, USA) and images were analyzed by GenePix Pro 3.0 (Axon Instruments, Foster City, USA). For each array, the raw data comprised the logarithm (base 2) of median feature pixel intensity at wavelength 635 nm (red) and 532 nm (green). No background was subtracted. We excluded control spots (as described in 61) and spots for which duplicates did not pass quality controls standards, cDNAs were not found by spot-finding software, or those determined to be irregular by visual inspection. Intensity signals for cDNAs passing these filters were normalized for intensity- and position-dependent bias. Array-by-array normalization was performed to remove systematic biases. Then, we replaced the value of the spots that were considered as badly formed features with the value of the duplicate. We averaged the two values from each duplicated feature to obtain one value of the gene per array in each condition.

A total of 4,795 cDNAs passed initial filters; 3,045 of these were collapsed to genes and 1,750 ESTs were unassigned. We refer to the combined set of genes and unassigned ESTs as genes, although some redundancy might exist. ANOVA analysis was used to classify the differentially expressed genes according with the different factors considered (P < 0.05 was used to denote statistical significance).

Statistical analyses were conducted using R version 2.2.0 61 and the R/maanova software package version 0.98.8 6168. The gene-by-gene mixed effect ANOVA model, Yijkvr = μ + Ai + Dj + (ST)k + Vv + Eijkvr, was applied. Observation Yijkvr is the expression value of the gene in the biological replicate r studied on array i when the condition is labelled with dye j. The condition is defined by a modality of the genotype, say k, and the modality of the Varroa, say v. The residual of the mode Eijkvr and the term Ai are treated as random effects and the others as fixed effects. Statistical tests were performed with R/Maanova using the hybrid variance model, Fs, based on a James-Stein estimator 6970. To control the number of false-positives, the significance (nominal P-value) of genes was computed using a permutation test (pvalperm with 1000 permutations) 70. Permutation tests have the advantage of not assuming a parametric underlying distribution of expression values. A gene was declared differentially expressed if its adjusted P-value is lower than 0.05. The adjusted P-value used here made it possible to control the Family Wise Error Rate (FWER)

Microarray ESTs were annotated as described in 6071. Briefly, ESTs corresponding to microarray cDNAs were tested for near-perfect matches (98% identity) to coding (protein) sequence or to genomic sequence within or immediately downstream (500 bp) of predicted genes (using release 1 of the honey bee Official Gene Set 72). Redundant cDNA values were averaged (by using untransformedvalues), and resulting values were assigned to official gene names (which are all prefixed "GB"). Remaining cDNAs not associated with predicted sequences retained their EST identifiers and are presented here by EST accession number (prefixed "BB"). Genes were tentatively assigned molecular function terms based on annotation of the single best BLASTX match in Drosophila melanogaster.

We used Gene Ontology (GO) analysis to identify the biological function of the differentially expressed genes, as described in 60. We sought statistically overrepresented terms among the set of genes significantly regulated. This Enrichment of Gene Ontology (GO) category was tested with a Hypergeometric test followed by the Benjamini Hochberg correction for multiple testing using the analysis tool GOToolBox 60. Briefly, the analysis calculates the frequency of terms of a gene list (the significantly regulated genes) and compares these results with total frequencies of genes analyzed. We also grouped functionally related genes on the basis of their GO terms. Distances based on GO terms are calculated for all possible pairs from the gene list which are then used for clustering (e.g. the probability that they are functionally related). The functional clustering of genes was carried out in GOToolBox using the WPGMA algorithm. A minimum cluster size of 3 genes was applied.

The results of the microarray study were confirmed by measuring the expression of 4 differentially regulated genes (Table 2) using Real-Time quantitative PCR (RT-qPCR). Using an aliquot of the RNA extractions (pools of 8 bees) used for the microarray work, cDNAs were created using the ImProm-II™ Reverse Transcription System kit (Promega). This cDNA (three replicates) was used as template for the Real-Time PCR performed in a Roche LightCycler 480 Instrument platform. The reaction mix consisted of 10 μl of LightCycler 480 SYBR Green Supermix (Roche Laboratories), 5 μl (either 600 nM or 300 nM) of forward and reverse primers, 3 μl dH₂O and 2 μl cDNA template. Three reactions were performed and means calculated for each locus and treatment. To standardize the results, a housekeeping gene (Rp49) that did not vary in expression was used as control. The forward/reverse primer sequences are indicated in Table 2. The efficiency for each locus was determined by running a dilution series (1000x, 100x, 10x, 1x) in triplicate. The results were standardized using the 60 method. Efficiency of the amplicons obtained for each locus was adequately high and at least 97% (Table 2).