Materials and methods

Array development
To generate the genetic arrays, cDNA clones were isolated from four subtractive libraries derived from worker- and queen-destined larvae collected in the third and fourth instars. We chose 144 clones from worker-biased libraries and 144 clones from queen-biased libraries. These clones were amplified by the polymerase chain reaction (PCR) using endogenous adaptor primers (primers Nest1 and Nest2r, described in [3]), then separated and verified by agarose gel electrophoresis. After being denatured for 5 min at 95°C in a final molarity 0.2 M NaOH, approximately 5 μg DNA from each amplified clone was fixed onto nylon membranes (Hybond N+, Amersham) using a 5 μl slot-pin library replicator (V and P Scientific) followed by UV irradiation. Each of 96 quadrants in a 384-cell printing contained three samples and a negative control. An aliquot of each cDNA clone was purified then sequenced using fluorescent-dye labeling and an ABI Prism 373 DNA analysis machine (Applied Biosystems).

Sample collection
Embryos, along with worker- and queen-destined larvae, were harvested from several colonies of A. mellifera ligustica of the same genetic stock maintained in the Bee Research Lab apiary. Two groups of 50 late-stage (2-day-old) embryos were harvested from standard (worker) cells and immediately frozen at -80°C. Four groups of 20 bipotential larvae aged 24 h post-hatching also were pooled and frozen. Fourteen worker-destined larvae aged 48, 72, 96, and 120 h post-hatching were also collected, along with 13 queen-destined larvae from the same age groups, raised in natural queen-rearing (swarm) cells. The ages of these queen larvae were determined by wet weight and body size comparisons with larvae of known age raised in artificial queen cells. For both worker- and queen-destined larvae, ages were likely to be accurate within 8-12 h, on the basis of wet-weight comparisons with larvae of known age [3]. Total RNA was extracted from these samples using the RNAqueous protocol (Ambion).

Labeling and hybridization
DNA probes were generated from 5 μg total RNA by reverse transcriptase synthesis (Superscript II enzyme, Life Technologies), using oligo-dT primers and biotin-labeled dUTP and dATP (in a 1:10 molar ratio to unmodified dTTP and DATP, respectively). Probes were cleaned by spin filtration, denatured, and then incubated in hybridization buffer (50% formamide, 6 × SSPE, 0.5% SDS, 5 × Denhardt's solution) with 0.1 pmol poly-(A)25 for 2 h at 42°C. Hybridization was carried out at 42°C, followed by a series of stringency washes, as in [3]. Membranes were washed, bathed in the Avidx alkaline-phosphatase conjugate then exposed to the chemiluminescent substrate CDP-star, according to the manufacturer's instructions (Tropix, Applied Biosystems). Membranes were next exposed to autoradiographic film for between 10 min and 8 h. Each membrane was used only once, giving an independent hybridization for 2 embryo replicates, 4 bipotential larvae, 13 queen, and 13 worker larval samples.

Data analysis
Cloned sequences were aligned using the CLUSTALW algorithm (Omiga 2.0, Oxford Molecular) to assess sequence quality and identify multiple captures of the same sequence. Each unique clone (167 out of 288) was compared against the GenBank database using the BLAST-X and BLAST-N algorithms at the site maintained by the US National Center for Biotechnology Information [21]. Matches showing a probability score of < 1.0 × 10-3 were treated as significant matches. Of the 167 unique clones, nine were dropped from the analyses as a result of evidence of multiple bands or inconsistent PCR amplification. Of the remaining clones, 64 showed a match to one or more sequences in GenBank and, of these, 63 could be placed into one of the gene function groups defined in [10] and the Gene Ontology database [22]. All gene fragment sequences used are available, along with their BLAST search results, at the Beenome Project website [23], and in the NCBI dbEST database [24].
Scanned images of exposed films were scored using the software program Zero Dscan (Scanalytics), giving a densitometry score for each cell in the membrane's 384-cell matrix. Spreadsheet macros (Microsoft Excel) were then used to generate an average intensity value for each of 158 unique clones that consistently showed a single band in the PCR amplifications. These values were normalized as a proportion of the strongest signal on each membrane, then were further normalized and centered on unity across genes and samples [18]. Centroid clustering was then carried out and visualized using the programs Cluster 1.0 and TreeView 1.0, respectively [25].