Macroarrays were probed separately with digoxigenin (DIG)-labeled mRNAs from worker or soldier R. flavipes (Figure 2a). Initially, 153 cDNAs with either clear or marginal differential expression between soldiers and workers were identified in the array screens. These clones were the basis of further analysis. Clones corresponding to each of the 153 positive array positions were sampled from frozen cultures in 384-well plates and grown as liquid cultures. Plasmid DNA was obtained from each culture and spotted on duplicate membranes in equal quantities. The resulting dot blots were probed with aliquots of the same DIG-labeled worker or soldier mRNA pools that were used to originally probe the array membranes. The secondary dot blots verified elevated transcript abundance in around 50% of instances (Figures 2b,c). In the dot blots shown in Figure 2, examples of worker- and soldier-specific cDNAs are in positions noted by squares and diamonds, respectively.

Dot blots were also used to determine the expression of two potential genes with similar expression levels between castes: these encoded NADH dehydrogenase and actin-related protein 66B (ARP 66B). These two transcripts were identified previously from expressed sequencing tag (EST)-sequencing studies 12 and were used as candidate control genes for expression. Similar expression levels of the two control genes can be seen in the upper left corners of the dot blots, with ARP 66B being circled (Figure 2b,c). In subsequent experiments it was determined that expression levels of these two genes were not robust enough to serve as controls in quantitative PCR. As a result, β-actin was used as the control in all subsequent experiments (see below).

Following verification on dot blots, 68 clones were confirmed with clear differences in expression between workers and soldiers. High-quality ESTs were obtained for each of these clones (Additional data file 1). Because the cDNA library was created with 5'-3'-oriented cDNAs, EST sequencing was only performed in one pass from the 5' ends of clones. EST sequencing results are summarized in Table 1 in order of BLASTx E-value significance scores. Lengths of high-quality, vector-trimmed sequence ranged from 66 to 632 nucleotides. On the basis of BLASTx database similarity searches, eight cDNAs occurred more than once in the array screen (Additional data file 1). These multiply-occurring cDNAs, shown in order of their identity-significance scores, correspond to genes encoding: termite salivary cellulases (3 total), termite endoglucanase (10), endosymbiont endoglucanase (3), endosymbiont cellulase (2), mitochondrial cytochrome oxidase I (24), muscle-specific LIM protein (2), mitochondrial 18K protein (2), and unknown #24 (3). For simplicity, only single clones representing the longest sequence are reported in Table 1. Clones 1-16 possess significant E-values of less than 1 × 10^-8 and therefore their identities are considered valid (Tables 1, 2). Because clones 17-25 have larger E-values, the identities of these clones, when available, are only considered putative.

Because the six ESTs noted above corresponded to multiple positive array positions, the total number of distinct/unique sequences present in the array-positive EST pool was 25. These 25 cDNAs can be categorized into 10 gene ontogeny categories following the summary presented by Adams et al. 13. These are: enzyme, cytoskeletal/structural, electron transport, transcription factor, mitochondrial organization and biogenesis, signal transduction, storage, DNA replication, ion channel, and five unknown (Table 1).

Relative expression patterns for 24 discrete gene transcripts, determined by quantitative PCR, are shown in Table 2. Raw data from quantitative PCR analyses are provided in Additional data file 2. There was only one gene transcript that could not be amplified and evaluated by quantitative PCR. This sequence encodes a putative cyclic GMP-gated ion channel (clone 20) with highly repetitive nucleotide sequences and high CG content 14. Because of these properties, no PCR primer pairs could be designed that would provide the high-fidelity PCR necessary for quantitative PCR analysis.

Transcripts having highest worker-associated expression were those encoding cellulase and endoglucanase enzymes involved in cellulose digestion. Cellulases and endoglucanases can have origins in either termites or their microorganismal endosymbionts. Cellulases and endoglucanases of apparent endosymbiont origin had substantially higher expression relative to those of apparent termite origin. Other transcripts with elevated expression in workers were either related to mitochondrial organization and biogenesis (mRpS2 and 18K protein-like; also elevated in soldiers), or had unknown identity (clones 18 and 24).

More than half of the transcripts evaluated had exclusive soldier-biased expression (13 of 24). These included two transcription factors, two signal transduction factors, four cytoskeletal/musculature proteins, one transcript associated with mitochondrial electron transport, one putative DNA replication factor, and three transcripts of unknown identity. Three additional transcripts had similar expression levels between workers and soldiers, yet had higher expression than the presoldier and/or nymphal intermediary developmental stages (mRpS2,18 K protein-like, and reverse transcriptase-like).

Two soldier-biased troponin I-encoding transcripts had both differential expression and sequence variation (Additional data file 3). The first of these (Rf1-troponin I) aligns with a portion of the 3' region of the Drosophila troponin I cDNA whereas the second (Rf2-troponin I) aligns with the 5' region, and includes the ATG start signal. The two troponin ESTs have a 60-nucleotide overlap, but share only 65% amino-acid identity in the overlap region. These termite troponin forms have significantly elevated expression levels in soldiers, which are 1.7- and 10.5-fold elevated for the Rf1 and Rf2 forms, respectively, relative to workers.

Transcripts were also observed with high expression levels in the intermediary presoldier and nymph developmental stages. In all cases, transcripts that were identified with elevated nymphal expression were also elevated in workers (as with cellulases/endoglucanases and Rf BTB/POZ-1) or both workers and soldiers (as with mitochondrial 18 K protein-like). Finally, a transcript encoding the storage/hormone-binding protein vitellogenin had highest expression in presoldiers, a stage incapable of feeding itself, and that is also hypothesized to serve as a sink for juvenile hormone 15.

Two putative regulatory genes with significant homology to Drosophila and honey-bee (Apis mellifera) developmental genes were identified. Both these genes are transcriptional/translational regulators. One is Rfβ-NAC-1, which is related to the bicaudal locus of Drosophila, and the second is Rf BTB/POZ-1, which is related to the bric-a-brac (BAB) locus of Drosophila. Alignments of the respective translated amino-acid sequences are shown in Additional data files 4 and 5. A third related gene, Rf-PDL, has putative identity as a nanos-homologous posterior-determinant gene on the basis of weak homology to database sequences; however, hydropathy profiles for the translated amino-acid sequence bear striking similarities to Drosophila nanos protein and its homologs from other species (Additional data file 6).

Rfβ-NAC-1 is a Reticulitermes homolog of the ribosomal translation factor β-NAC. A partial open reading frame (ORF) of 420 nucleotides (140 amino acids) was included in our cDNA clone. Database searches using the Rfβ-NAC-1 translated amino-acid sequence revealed numerous proteins with significant homology, all of which are associated with ribosomal translational processes. Alignments of Rfβ-NAC-1 revealed 73 invariant amino-acid residues (Additional data file 4). Also observed was significant pairwise homology to β-NAC homologs from Drosophila (61.5%), A. mellifera (71.5%), and Caenorhabditis elegans (60.9%). These findings clearly establish identity of the Reticulitermes sequence as a β-NAC homolog.

Complete sequencing of the Rf BTB/POZ-1 cDNA clone elucidated a complete ORF of 1,320 nucleotides (439 amino acids). Database searches with the translated amino-acid sequence of the entire ORF did not reveal homologs with high degrees of similarity across their entire lengths; however, significant similarity to BTB/POZ domains of many BTB/POZ proteins was identified (Additional data file 5). The presence of the highly conserved BTB/POZ domain indicates that the sequence encodes a zinc-finger protein involved in transcriptional regulation. Alignments of the amino-terminal regions of several BTB/POZ domain-containing proteins revealed an insertion of 60-65 amino acids in the amino-terminal region of Rf BTB/POZ-1 that is not present in any of the homologous sequences, and the presence of three invariant sequence motifs in all BTB/POZ proteins shown (single-letter amino-acid code: LRWN, DVTL, LSACS). Overall identity between the BTB/POZ domains of termite Rf BTB/POZ-1 and Drosophila BRcore, Drosophila fruitless, A. mellifera pipsqueak, and Drosophila pipsqueak are 44.5, 43.1, 41.8, and 35.8%, respectively.

A third EST termed Rf-PDL (posterior determinant-like) shares modest similarity with nanos-homologous posterior determinant genes of Musca domestica (AAA87461) and Xenopus laevis (X72340) (part a in Additional data file 6). Striking similarities in hydropathy profiles exist, however, for sequenced portions of the termite cDNA versus nanos homologs from Drosophila, Musca, and Xenopus (part b in Additional data file 6). The nanos gene in Drosophila is a transcription factor that controls anterior-posterior polarity and is under direct control of β-NAC. Quantitative PCR results for Rf-PDL follow a similar caste and development-associated expression pattern to Rf β-NAC-1, implying a link between these two genes.

Two gene transcripts encoding signal transduction factors with established links to well-characterized developmental processes were also identified. The first of these genes, SH3-domain kinase, has significant homology to mammalian kinases occurring at the Ruk locus of mouse and rat (Additional data file 7), but interestingly, not to any sequences from insects. The second signal transduction factor, GAP, also has the highest degree of similarity to vertebrate sequences, although modest similarity also exists to a Drosophila GTPase-coding sequence (Additional data file 8). Relative to presoldiers, both of these transcripts showed substantial and significantly elevated expression in soldiers (132- and 4.5-fold, respectively), although expression was also significantly elevated in workers (86- and 2.3-fold, respectively).

Termites of multiple castes and developmental stages were obtained, and a polyphenic cDNA library synthesized as described previously 12. Total RNA was isolated from groups of whole termites using the RNaqueous total RNA isolation kit (Ambion, Austin, TX), and its integrity and quantity verified by formaldehyde-agarose electrophoresis 38. Poly(A) RNA was purified from the total RNA (Ambion), cDNA was synthesized using an oligo-dT₁₂ adaptor primer, and the library constructed (see 12). From the cDNA library, the Purdue University Genomics Core Facility prepared multiple identical 'macro' arrays, which are gridded bacterial colony filters. Each 22 × 22 cm filter was printed from 18,432 glycerol stocks contained in forty-eight 384-well plates. Duplicate filters were printed by using a Biorobotics robot (Total Array System, Woburn, MA) to spot samples from plates onto charged nylon membranes overlaid with LB agar containing 100 μg/ml ampicillin. The filters were grown overnight at 37°C, peeled away from the agar, placed on chromatography paper saturated with 5% SDS and 2x SSC for 2 min, heated in a microwave for 2.5 min, then UV-cross linked in a Stratalinker (Stratagene, La Jolla, CA).

Primary gene-expression comparisons between the soldier and worker castes were made as follows, using soldiers and workers isolated from laboratory colonies. mRNA was obtained separately from whole, pooled worker or soldier termites (Figure 1) using a commercially available protocol (Ambion), then digoxigenin (DIG)-labeled using a separate commercially available protocol (DIG Chem-link Labeling and Detection; Roche, Indianapolis, IN). Individual array membranes were probed with DIG-labeled mRNAs from either workers or soldiers (10 ml of 100 ng/ml). Autoradiograms of the membranes, showing the positions of specific clones, were produced using a commercially available protocol (Roche). Expression levels were compared between soldier and worker array autoradiograms by manual alignment.

For secondary verification, dot blots were performed using plasmid DNA affixed to nitrocellulose membranes (Millipore, Marlboro, MA). Liquid cultures were produced from array-positive cell lines, from which plasmid DNA was obtained using a commercially available protocol (Qiagen, Valencia, CA). Pure plasmid DNA (100-200 ng) was manually spotted onto duplicate membranes 38. Membranes were probed with labeled aliquots of the same worker- and soldier-derived mRNAs noted above, using an identical procedure, then were manually superimposed to compare relative expression levels between castes.

Array and dot blot-positive clones were sequenced from their 5' ends (T3 promoter) using a high-throughput EST sequencing procedure. For the putative regulatory genes Rfβ-NAC-1 and Rf BTB/POZ-1, longer sequences shown in alignments (Additional data files 4 and 5) were obtained by sequencing from both the 5' and 3' ends, and by designing internal sequencing primers in the case of BTB/POZ (forward = GCAACGGATGACTACATGGA, reverse = TGGTTGAAGCCTGATTCACA). All sequencing-related procedures were performed by the Purdue University Genomics Core Facility. After contaminating vector sequences were trimmed, the ESTs were used as query sequences to search the GenBank nucleotide and translated databases by BLASTx 39. Database searches were performed under default settings, with E-values for significance of identity being 1 × 10^-10 and smaller 40. Sequence alignments were performed using clustal analysis in DNAstar-Megalign software (Madison, WI).

Quantitative PCR was performed to precisely quantify relative caste-associated gene expression. cDNA templates for quantitative PCR were synthesized separately from DNase-treated total RNA of whole worker, presoldier, soldier and nymph termites. Total RNA isolation and DNase treatment were performed using commercially available protocols (Qiagen and Ambion, respectively). The quantity and quality of RNA was assessed before and after DNase treatment by spectrophotometry and formaldehyde-agarose electrophoresis 38. cDNA was synthesized from 5 μg total RNA using an oligo-dT₁₅ primer and 50 U Superscript II reverse transcriptase (Invitrogen, Carlsbad, CA) at 42°C for 50 min.

Primers for use in quantitative PCR were designed from EST sequence data by the MIT Primer 3 program 41. Primer sequences are provided in Additional data file 9. Quantitative PCR primers were designed to provide products of 100-200 base-pairs (bp) within putative ORFs; and had 45-55% GC content, T_m = 57-62°C, non-complementarity between forward and reverse primers, and pyrimidine nucleotides on their 5' and 3' ends 42.

Quantitative PCR was performed using an iCycler iQ real-time PCR detection system (Bio-Rad, Hercules, CA) with SYBR Green I product tagging. To verify the amplification efficiency for both target and control genes, quantitative PCR conditions were optimized for cDNA template dilution (1:5, 1:10, 1:25, 1:50, and so on) and annealing temperature (Tm ± 5°C). Before experiments, the production of single product bands and absence of 'primer-dimers' was confirmed by viewing products following agarose electrophoresis and ethidium-bromide staining. Simultaneous determination of relative expression levels for specific genes in workers, soldiers, presoldiers and nymphs was determined from quantitative PCR results in relation to the standard gene β-actin 43, using REST software 44 (see Additional data file 2).