The experimental colonies were derived from source colonies in TX apiaries that had been treated with Apiguard (thymol gel) to reduce Varroa population one year (2009) prior to this study. The experimental colonies were established in March 2010, using 10 three-pound packages of bees from source colonies along with a caged Kona Italian queens in each package. Packages were installed into new bee equipment (10 frame hive body, bottom board and migratory cover, 10 frames with the new wax coated plastic foundations). New colonies were placed in a new apiary and fed high-fructose corn syrup and MegaBee protein patties on ad lib basis. Kona queens were replaced by Koehnen Italian queens in April, 2010. Five experimental colonies were selected at random for Varroa infestation; a half frame of sealed brood comb infested with Varroa mites (14.5 mites per 100 bees) was introduced in May 2010.

Samples were collected from the colonies on 13 July, 17 August and 8 September 2010: pink/purple-eyed pupae (n = 60), newly emerged adults (n = 72), and foragers (n = 72) as they entered the colony. With the exception of foragers, each insect was scored for the presence of Varroa mites and then transferred individually to a 1.5 mL vial and the fresh weight recorded ±1 μg accuracy. Samples were frozen and stored at −80 °C. In parallel, the level of Varroa infestation was determined using the alcohol shake method in samples of 280 bees per colony [20,21] and estimated as a number of mites per 100 adult bees.

The indices of immunological nutritional status were conducted on bees collected on 17 August and 8 September. By this date in the season, three of the 10 colonies had died out, and sufficient bees for analysis were available from 6 of the remaining colonies for all assays except for the gene expression study, which was conducted on 5 colonies.

Total RNA was isolated from each bee sample using TRIzol® (Life Technologies, Carlsbad, CA, USA) following the manufacturer’s protocol. To test for residual DNA contamination, 1 μL RNA was used as template in PCR, using primers that amplify a defensin 1 gene fragment that includes a 286 bp intron. Samples that tested positive for the presence of the intron were treated with DNA-free™ Kit (Ambion-Life Technologies, Carlsbad, CA, USA) until DNA was not detectable. Poly-A RNA was purified from the total RNA samples using MicroPoly(A)Purist™ (Ambion-Life Technologies, Carlsbad, CA, USA) following manufacturer’s protocol, and its purity was checked on a 1.0% denaturing agarose/formaldehyde gel before quantification using an Eppendorf BioPhotometer. Double-stranded cDNA was synthesized with 500 ng poly-A RNA template, SuperScript™ Double-Stranded cDNA Synthesis Kit (Invitrogen-Life Technologies, Carlsbad, CA, USA), and oligo-dT primers according to the manufacturer’s protocol.

The expression levels of defensin1, abaecin, hymenoptaecin, vitellogenin and hexamerin 70b, and the abundance of DWV [13] were determined by qRT-PCR normalized to ß-actin using CFX96™ Real-Time PCR Detection System (BioRad, Inc.), following evidence that this gene is stably expressed and the reference gene of choice [22,23]. Amplification was performed in 10 μL reaction volumes, containing 0.5 U of GoTaqR Flexi DNA polymerase (Promega Co., Madison, WI, USA) with the colorless 5× GoTaqR Flexi buffer, 0.38 mM dNTP mix, 5.0 mM MgCl₂, 0.35 μM of each primer (Table 1), 0.33 μL SYBR-Green (Invitrogen., 1/1,000 stock dilution), and 1 μL cDNA. Reactions were run for 40 cycles at 95 °C (5 s) and 60 °C (30 s) after initial denaturing at 95 °C for 3 minutes, and with a final extension of 72 °C (2 min) followed by a melt curve analysis. All experiments included two technical replicates of three biological replicates (for each developmental stage, colony and Varroa-category) andtemplate-free controls.The qRT-PCR data was analyzed using the comparative delta delta C_t method, with C_t values normalized to ß-actin. Fold change in expression of the target genes was quantified as by Pfaffl [24], using the equations for relative standard curves and relative efficiency plots calculated by GraphPad Prism software (GraphPad Software, Inc., La Jolla, CA, USA).

Each pupa or bee was thawed on ice, and then homogenized in 1 mL ice-cold buffer comprising 10 mM Tris, 1 mM EDTA pH 8.0 and 0.1% (v/v) Triton-X-100 with hand-held homogenizer, and centrifuged at 7,000 g at 4 °C for 10 min. The supernatant was used for analysis of protein, triglyceride and carbohydrates using coupled colorimetric assays with an xMark^TM microplate spectrophotometer, following manufacturer’s instructions (5 replicates per assay). The assay kits were the triglyceride assay kit of Sigma (catalogue number TG-5-RB); the Coomassie Brilliant Blue microassay method of BioRad (catalogue number 500-0201), with bovine serum albumin as standard (1.25–25 mg protein mL⁻¹)for protein; and the glucose assay kit of Sigma (catalogue number GAGO20) for glucose and, following trehalase (3.7 U mL⁻¹) and amyloglucosidase (2 U mL⁻¹) treatment, for trehalose and glycogen, respectively. The sugar analyses (glucose and trehalose) were restricted to pupae because datasets for adult bees were confounded by endogenous trehalase activity and sugars in the gut contents of some individuals.

For the amino acid analysis, 1 μL supernatant was brought to 20 μL in PBS, incubated on ice for 30 min with 20 μL 40 mM HCl, and centrifuged at 18,000 g for 15 min at 4 °C. The supernatant was filtered through a 0.45 μm filter plate (Millipore) by centrifugation at 1,500 g for 10 min, and 5 μL filtrate was derivatized with AccQ Tag (Waters), following manufacturer’s protocol, and injected into Waters Acquity UPLC with PDA detector and AccQ-Tag Ultra 2.1 × 100 mm column. The gradient was: 0–10 min, 99% A; 10–35 min, linear gradient to 65% A; 35–40 min, 65% A; 40–42 min, linear gradient to 100% B, where A is 90% AccQ-Taq Ultra Eluent in water, and B is Accq-Taq Ultra Eluent B. Amino acids were determined by comparison to standards, 1–100 pmol protein-amino acids μL⁻¹ (Waters amino acid hydrolysate standard #088122, supplemented with asparagine, tryptophan and glutamine).

All data were checked for normality by the Anderson-Darling test and homogeneity of variance by Levene’s test, and data that did not meet these criteria were analyzed by non-parametric methods. The abundance of Varroa in uninfested colonies and colonies experimentally infested with Varroa was compared by the MannWhitney U test for each of the four test months (July-September). The fold-change of gene expression in Varroa-infested bees relative to uninfested bees was compared to the null hypothesis of no change by t-tests, with Bonferroni correction for multiple tests.Preliminary analysis revealed no significant correlations between nutritional indices obtained for either pupae or hive bees in September (Pearson’s correlation analysis). Therefore, the insect weight and nutritional density data were analyzed by individual ANOVA tests, with three main factors: Vbee (whether the insect bore Varroa); Vcolony (whether the Varroa abundance was above the recommended treatment threshold of 15 mites per 100 bees); and individual Colony nested within Vcolony. The principal components analysis of free amino acids in the pupae and hive bees used a correlation matrix to standardize variables, with axes selected to display separation of Varroa-infested and uninfested bees. Linear regression analysis was used to investigate the relationship between nutritional indices of uninfested bees and Varroa abundance in the same colony, and the relationship between immune-related indices and nutritional indices.

Varroa was not detected in the mitocide-treated colonies set up in May 2010, but was present in all colonies tested in July, August and September, 2010. The median abundance of Varroa was significantly higher in untreated colonies than experimentally-infested colonies in July (Mann Whitney U: W = 38, p < 0.05), but did not differ between the two colony types in August (W = 18, p = 0.59) or September (W = 14.5, p = 0.72) (Figure 1).

The titer of Deformed Wing Virus (DWV) in pupae and newly-emerged bees was quantified in August and September, when there were sufficient Varroa-infested insects for analysis. DWV was detected in all samples, and the DWV titer was significantly elevated in Varroa-infested insects, relative to uninfested pupae and newly-emerged bees (Figure 2a, Supplementary Table S1), with differences ranging from less than two-fold to >300,000-fold. Significant among-colony variation in the ratio of DWV titer in infested/uninfested insects was obtained for pupae in August (F_4,14 = 21.90, p < 0.001) but not in pupae in September or in newly-emerged bees. Foragers in general bore a very low titer of the virus (data not shown).

Our approach to investigate further the impact of Varroa on the immune status of the honey bees was to quantify the expression level of three antimicrobial peptide (AMP) genes, defensin1, abaecin and hymenoptaecin in pupae and newly-emerged adults sampled in August and September. Individuals bearing Varroa had significantly higher transcript abundance of these immune-related genes for at least two of the four datasets, with the mean fold-difference varying between two-fold and >30-fold (Figure 2b–d, Supplementary Table S1). No instance of a significant reduction in gene expression was identified in Varroa-infested bees. Across the full dataset, the expression levels did not differ between colonies with Varroa infestation above or below the threshold (15 mites per 100 bees) for intervention (p > 0.05) and significant among-colony variation was restricted to two AMPs, abaecin and defensin1 in August pupae (F_4,14 = 10.06, p = 0.002; F_4,14 = 21.90, p < 0.001).

The relationship between AMP expression and DWV titer was investigated by correlation analysis. The log₂ fold difference in DWV titer between Varroa-infested and uninfested pupaewas significantly correlated with the log₂-fold-difference in transcript abundance of defensin1, but not abaecin or hymenoptaecin in August (r = 0.998, p < 0.001). No significant correlations were identified in newly-emerged bees in August or September, or for pupae in September.

The weight of pupae and hive bees bearing Varroa was significantly depressed relative to Varroa-free individuals (Figure 3a,e, Supplementary Table S2a). For pupae, but not newly-emerged adults, mean weight was also depressed in colonies with high (i.e., above treatment threshold) Varroa abundance, irrespective of whether the individual bore Varroa, and the magnitude of this effect varied with colony.

The nutritional condition of the insects was significantly affected by Varroa (Figure 3, Supplementary Table S2), although the effects differed between pupae and newly emerged bees. The protein density (i.e., protein content per unit weight) was significantly depressed in Varroa-infested pupae, but elevated in Varroa-infested hive bees (Figure 3b,f), although this was not generally reflected in the expression levels of two key storage protein genes, hexamerin70b and vitellogenin (Figure 2e,f). The effect of Varroa infestation on triglyceride density varied with the Varroa status of the colony, being elevated in Varroa-infestedpupae (but not newly emerged adult bees) from colonies with high Varroa infestation, and reduced in Varroa-infested adult bees (but not pupae) from colonies with low Varroa infestation (Figure 3c,g).

The single most consistent determinant of the nutritional indices was the colony from which the insects were sampled; every nutritional index scored varied significantly with colony or the interaction between colony and whether the insect bore Varroa (Supplementary Table S2). Colony effects were particularly marked for glycogen density, which was significantly elevated in pupae and newly-emerged bees bearing Varroa in two of the four colonies (colonies 1 and 4) with low Varroa abundance (Figure 3d,h).

The metabolite analysis focused on two key sugars, glucose and trehalose, and the 20 amino acids that contribute to protein (Figure 4a–c). Individual colony effects were important for the sugars. Varroa-infestedpupae had significantly reduced trehalose density (Figure 4b), especially in colonies with low Varroa abundance; and the glucose density (Figure 4a) was significantly elevated in pupae bearing Varroa in two of the four colonies with low Varroa abundance (colonies 1 and 4: the same colonies with elevated glycogen contents). The total amino acid density of pupae was significantly elevated by Varroa infestation, at the level of both the individual insect and colony (Figure 4c); but no significant effect was obtained for newly-emerged adults (Supplementary Table S2).

Principal components analysis of the composition of the free amino acid pools of the insects (Supplementary Table S3) yielded near-separation along two axes between Varroa-infested and Varroa-free insects, for both pupae (Figure 5a) and newly-emerged adults (Figure 5c), accounting for 58% and 72% of the variance, respectively. Individual ANOVAs (with Bonferroni correction for 20 tests) (Supplementary Table S3) confirmed the inspection of the PC coefficients (Figure 5b) that a set of 11 AAs contributed to the elevated FAA titers in Varroa-infested pupae. This set includes 9 of the 10 essential amino acids (all but tryptophan). Exceptionally, aspartate was significantly depressed in Varroa-infested pupae. The amino acid titers of newly-emerged adult bees displayed a similar trend, although with weaker separation of Varroa-infested and uninfested bees and significant elevated titers of just two amino acids, arginine and methionine (Figure 5d, Supplementary Table S3).

To investigate the effect of Varroa abundance on the nutritional condition of the bees further, the nutritional indices for uninfested pupae, newly-emerged bees and foragers was regressed on Varroa abundance quantified in the same month and, for August and September, in the previous month. None of the regressions were statistically significant, indicating that the level of Varroa infestation in the colony does not have a consistent effect on any tested nutritional index of the bees. The relationship between nutritional indices and immune-related indices (DWV titer and AMP transcript abundance) was also investigated by correlation analysis: no significant correlation between any nutritional and immune-related index was identified.