Four genes were selected from each of three fly tubule expression classes: very highly enriched; uniformly expressed; and very highly depleted. The expression of each gene was verified by quantitative reverse transcription PCR (RT-PCR) and the data are presented in Table 3. The agreement between Affymetrix microarray and quantitative PCR determination is good, further increasing our confidence in the robustness of the dataset, and in the approximate correspondence between signal and RNA abundance as a population average. It should be noted that the absolute sizes of the ratios are quite variable; this is a property of dividing a large number by a very small one. Nonetheless, genes scored as enriched or depleted on the arrays are invariably similarly scored by quantitative RT-PCR (QRT-PCR).

These data can also be used to validate the use of the normalized Affymetrix signal as a semi-quantitative measure of RNA abundance (Table 1). If the QRT-PCR dataset of Table 3 is normalized against corresponding signals for rp49 (generally taken to be a ubiquitous gene with invariant expression levels in Drosophila), and compared with the globally normalized Affymetrix signal, the agreement is seen to be excellent (Figure 2), with a Spearman's r of 0.83 (p < 0.0001). With appropriate caution, the normalized Affymetrix signal can thus be taken as a reasonable estimate of expression levels between genes.

Table 1 shows the top 20 genes listed by mean Affymetrix signal intensity. Although this is only a semi-quantitative measure of transcript abundance, the identities of the known genes in the lists are illuminating, and persuade us that the approach has some informal value. Specifically, mRNAs for ribosomal proteins dominate the list, and transporters are conspicuous in the balance. For example, the V-ATPase that energizes transport by tubules is represented by one gene (other subunits are also abundant, but just below the cutoff for Table 1). The α-subunit of the Na⁺, K⁺ ATPase is also highly abundant: this is more surprising, and is discussed below. Two organic cation transporters are also very abundant. Alcohol dehydrogenase, long known to be expressed in tubules 2324, is also a major transcript. There are also surprises: the most abundant signal is for metallothionein A. This is entirely consistent with our classical understanding of tubule function: it has long been known as a route for metal sequestration and excretion 252627282930. However, in the entire literature on Malpighian tubules, we are not aware of a physiological investigation of the role of metallothionein, other than documentation of expression 3132. The microarray results can thus potently direct and inform future research.

Table 2 lists the 53 tubule-enriched genes that are enriched at least 25-fold, in comparison with the whole fly (the full list is provided as an additional data file). The conspicuous feature of these data is the extent to which tubule transcripts differ from any previously published profile. When comparing fly with tubule, there is a large set of genes that are downregulated and another large set of genes that are upregulated in tubule. The extent of the upregulation is also remarkable: the top gene is 99-fold enriched; the top 10 at least 50-fold enriched; and the top 100 at least 16-fold enriched in tubule compared to fly. The standard errors are also extremely low, meaning that we can be very confident (by two separate statistical measures) of the genes called significantly enriched in tubule.

Another prominent feature of the signal data in Table 1 is the relatively large fraction of novel genes (those for which there is not even a computer prediction of function) at the top of the list. Indeed, five of the top 10 genes by signal intensity are completely novel - that is, there are no known orthologs - and should provide tantalizing insights into tubule function. The 'phenotype gap' 3334 is a key problem in functional genomics; that is, the genetic models preferred for genomics are historically not the organisms selected by physiologists. This can lead to a log-jam in reverse genetics, which depends critically on a wide range of phenotypes to identify effects of the mutation of target genes 12. It has recently become possible to quantify the phenotype gap 35. The present dataset elegantly exposes the phenotype gap in Drosophila, and shows that the tubule phenotype may go some way to closing it. Around 20% of Drosophila genes have been studied in sufficient detail to attract names (beyond the standard 'CG' notation for computer-annotated genes). Figure 3 shows that the fraction of anonymous genes in the tubule-enriched list is far higher than would be expected. That is, previous work has tended to overlook these genes. Conversely, because it is possible to perform detailed physiological analysis in tubules, it is possible to close the phenotype gap for these genes. There is a general implication from these data: that functional genomics, in Drosophila and other species, will rely increasingly on the study of specific tissues, as it is only in this context that expression of genes will be either measurable or explicable.

Many microarray experiments merely classify enriched genes to their Gene Ontology families. However, the uniquely detailed physiological data available on the Malpighian tubule allows a much more informative approach. The dataset can be validated by inspection, based on known molecular functions in the tissue and new functions can be inferred from abundant or enriched transcripts in the dataset. As the array is relatively comprehensive (corresponding to the 13,500 genes in release 1 of the Gadfly annotation), the results are also relatively authoritative.

The hormonal control of fluid secretion is well understood. The major urine-producinig region of the tubule is the main segment 70, and is composed of two major cell types, principal and stellate cells 91371. Active cation transport in the principal cell is stimulated by the hormones calcitonin-like peptide and corticotrophin releasing factor (CRF)-like peptide, both of which act through cyclic AMP (cAMP). Another peptide family, the CAPA peptides, act through intracellular calcium to stimulate nitric oxide synthase and thus raise cyclic GMP (cGMP), an unusual autocrine role for nitric oxide 2072. In the stellate cell, the chloride shunt conductance is activated by leucokinin 1773, and a role for tyramine as an extracellular signal has also been proposed 74. So far, the CAPA and leucokinin receptors have been identified 7576; both are prominent among the receptors enriched in tubule (Table 10). The CAPA receptor appears much more highly enriched in tubule than the leucokinin receptor, which is consistent with our understanding of each: the tubule is the only known target of CAPA, whereas leucokinin receptors are widely distributed in the adult gut, gonad and nervous system 75.

There are many other receptors that are reasonably abundant and enriched in tubule. As well as candidate receptors for calcitonin-like and other neuropeptides, there are two glycine/GABA-like receptors that might be expected to form ligand-gated chloride channels, together with good matches to vascular endothelial growth factor-like, insulin-like and bombesin-like receptors. The localization of, ligands for, and functional roles of these receptors will be of great interest. It should be noted in this context that all hormones characterized so far act on one of the two main cell types in the principal section of the tubule. There are, however, six genetically defined cell types and six regions in the adult tubule 9, and it is likely that there will at least be ligands acting on the initial segment to stimulate calcium excretion, and others acting to regulate reabsorption by the lower tubule. If any of these receptors maps to these regions, they would be prime candidates for such roles.

Overall, the main surprise from these data is the sheer range of candidate ligands that could be inferred; this more than doubles the size of the endocrine repertoire so far postulated for insect tubules.

On a more general level, it is possible to trace out the key genes in all three intracellular signaling pathways that have been studied in detail in Drosophila tubule (Table 11). The results for signaling genes tend not to be as clear-cut as for transporters, as many are rather widely distributed, and so do not show enrichment, and many do not require high standing levels of protein (and implicitly mRNA) to achieve their effects. Nonetheless, it is possible to identify genes that are at least present, and frequently enriched, in tubule. For the cAMP pathway, it is possible to identify adenylate cyclases, protein kinase A catalytic and regulatory subunits, and a phosphodiesterase (dunce). For cGMP, there are both soluble and membrane guanylate cyclases, implying that the tubules may produce cGMP directly in response to novel ligands, as has recently been suggested 77. Both Drosophila genes encoding protein kinase G are expressed in tubule, and one is highly enriched. This is consistent with the renal phenotype observed both in foraging mutants 78, and in tubules in which protein kinase G is overexpressed 79. There is also a PDE11-like phosphodiesterase. For calcium, two genes for phospholipase C, one for calmodulin, and one for protein kinase C and for calcium/calmodulin-dependent protein kinase are apparent. There are also a number of interesting modulatory or anchoring proteins, such as 14-3-3 zeta, A-kinase anchoring proteins, and receptors for activated C-kinase (Rack1).

The developmental origin of the tubule has been reviewed in detail 808182. Briefly, four unique 'tip cells', specified by a cascade of neurogenic genes, control cell division in four outpushings (anlagen) of the hindgut, to form the Malpighian tubules. Late in embryogenesis the tubule is invaded by mesodermal cells, which intercalate between the future principal cells, and which then differentiate to form stellate cells 8. In the adult, there are known to be at least six cell types and six tubule regions 9. These regions are specified to great precision, and it is clear that each cell in the tubule has a precise positional identity. How does this identity persist throughout the lifetime of the animal? Presumably, combinations of transcription factors interact to provide both regional and cell-type coordinates and, after early establishment, these combinations must persist into adulthood. The microarray data allow the identification of transcription factors that are either highly abundant or highly enriched in tubule. Although this is by no means a complete list of transcription factors that are of importance to tubules, it is a good starting point. Furthermore, there are enhancer trap or reporter gene constructs available for many transcription factors. Accordingly, the top transcription factors and DNA-binding proteins were identified from the array dataset (Table 12).

Some of these transcription factors are already known to be present in tubule, and their presence is confirmed: cut, which is known to be required for development of, and expressed in adult Malpighian tubules 83; and forkhead and homothorax, both implicated by expression or mutational analysis to be involved in tubule development 8485. Teashirt, which has recently been shown to be stellate-cell specific in the late embryo 8, is also present in the adult, with fairly high enrichment (4.6 times).

The array results also implicate a further set of transcription factor genes (ETS21C, CG4548, bowl, sequoia, tap, CG1162, pnt, shaven, forkhead domain 59A, sloppy paired 2, lim3) as important in adult. Significantly, these mainly encode transcription factors implicated in development of the nervous system (another ectodermal tissue), so their reuse in the adult tubule is not too surprising. Once the binding sites for these factors are known, it will be interesting to model gene expression in different tubule regions.

As transcription factors have been studied experimentally in some detail, they are relatively well represented by enhancer trap and other in vivo construct lines. Although individual lines do not necessarily represent the complete expression pattern of their cognate genes, a collection of such lines can provide a rapid first validation of a gene list (Table 12). Accordingly, representative reporter gene lines were ordered from the Bloomington Stock Center 86, and their adult staining patterns in tubule and gut are shown in Figure 4. The results are exciting: most lines showed patterned staining in tubule that is consistent with our original genetically derived map of the tubule 9. For example, homothorax marks out the initial, main and transitional segments of the tubule, whereas CG7417 marks the complementary lower tubule domain. The latter line is widely used as a highly specific mushroom body GAL4 driver line in brain, and it is interesting that the two known lower tubule GAL4 driver lines (c507 and c232) are both insertions in alkaline phosphatase 4, a gene which is only expressed in lower tubule and the ellipsoid bodies of brain (next to the mushroom bodies) 87. There is also a cell-type-specific transcription factor: corto is found only in stellate cells. Several other transcription factors show ubiquitous, rather than patterned, expression in the tubule, but this is nonetheless consistent with their identification in the microarray dataset.

Another interesting aspect of the data in Table 12 is the number of anonymous CG genes implicated in tubule function. These genes have been annotated as transcription factors because of DNA-binding domains, for example, but have not been characterized functionally. The epithelial phenotype gap is thus evident even in this most intensely studied group of genes.

The whole premise of microarray work is that an abundant or enriched signal indicates the importance of a gene product in a particular context. This hypothesis is normally both untested and unchallenged. The unusual depth of functional understanding of the tubule allows a more rigorous appraisal. In fact, the majority of the genes implicated in tubule function are found well up the list. There are, however, several conspicuous exceptions (Table 13). The calcium channels trp and trpl are normally considered to be eye-specific, and have an essential role in phototransduction 888990. It is thus not surprising to find both genes almost at the bottom of the gene list. We have shown, however, that fluid secretion is severely compromised by mutations in either gene. Similarly, nitric oxide synthase (NOS) is a major signal transducer in tubule 2072. Nonetheless, all three genes are within the 'bottom' 20 of the whole array, with signals that are barely detectable and significant depletion compared with the whole fly. This is a cautionary example: while abundant or enriched signals can be taken as reliable indicators of functional significance, the converse is not necessarily true.

Consequent to the demonstration of the phenotype gap, there are some intriguing, abundant and enriched genes which by virtue of their non-uniform expression, are likely to be important in (and best studied in) tubule. A systematic approach was taken by combining the tubule-enriched gene list with the homophila database of Drosophila genes with known human disease homologs. The results (Table 14) show the 50 human diseases with Drosophila homologs that are upregulated at least threefold in tubules. Intriguingly, several of these genes have human kidney phenotypes. Some are extremely well studied: for example, rosy (one of the first Drosophila mutations recorded) encodes xanthine oxidase, and mutation in either human or fly produces severe nephrolithiasis with concomitant distortion of tubules (reviewed in 12). The distension of tubules is remarkable (Figure 5). In both species, lethal effects can be ameliorated by a high-water, low-purine diet. Other diseases, although less well documented, have plausible renal phenotypes: for example, antenatal Bartter syndrome, a severe salt-wasting renal disease, associated with mutations in the ROMK channel (homolog ir); Dent disease, caused by mutation in ClC5 (homolog CG5284); proximal renal tubular acidosis, caused by mutation in the NDAE co-transport (homolog ndae1); nephrophatic cystinosis, caused by mutation in a lysosomal cystine transporter (homolog CG17119); mucopolysaccharidosis type IV, caused by mutation in galactosamine-6-sulphatase, an enzyme enriched in both human and fly kidney (homolog CG7402). Overall, there is a clear message that human and fly renal function may be relatively similar over quite a wide range of properties.

The tubule phenotype may also prove highly informative for other genes implicated in disease. Recently, a small 10 kDa protein, bc10, was shown to be downregulated in the transition from early-stage to invasive bladder carcinoma 91. The normal function of this protein is not yet established, but its homolog (bc10) is highly abundant (893 ± 50) and moderately enriched (1.9 ± 0.09) in tubule, and a P-element insertion within the gene P{GT1}BG02443, is available from stock centers.

This comparative approach can be extended to non-human species. For example, CG4928 represents an abundant and enriched transcript (3,778, 13 times enriched), that is highly similar (1.9 × 10^-75) to the C. elegans gene unc-93 92. This is associated with a 'rubber-band' phenotype, in which motor co-ordination is sluggish; it is thus taken to be a myogenic or neuromuscular gene. The discovery that a close homolog is highly enriched in renal tissue opens new lines of investigation for this gene.

Drosophila melanogaster were kept on standard diet at 25°C and 55% relative humidity on a 12:12 h photoperiod. Malpighian tubules were dissected from 7-day-old adults, for compatibility with the extensive physiological literature on the tubule 10111315171920397075949596. At this stage, the tubules are in a relatively stable state after adult emergence, and their secretion parameters do not change detectably between 3 and 14 days post-emergence.

Tubules were dissected in batches of 1,000 by a group of eight experimenters. Tubules were aggregated into Trizol every 15 min to minimize the distortion of the transcriptome by the trauma of dissection and in vitro incubation. Care was taken to sever the tubules from the gut at the lower ureter so that no other tissue was included in the sample. For each experimental point, whole flies from the same culture were homogenized in Trizol in batches of 100, to permit a matched pair comparison. Six repeats were performed. RNA was extracted according to standard protocols, and quality was assessed with an Agilent RNA Bioanalyzer. Samples of 20 μg total RNA were reverse-transcribed, then in vitro transcribed, according to Affymetrix standard protocols. The quality of the ccomplementary RNA (cRNA) was also checked on an Agilent RNA Bioanalyzer, with a sample in which the broad cRNA peak exceeded the height of the low molecular weight degradation peak taken to be satisfactory. Samples were then run on the Affymetrix Drosophila genome array under standard conditions. Quality control was at several levels: the Affymetrix MAS 5.0 software provided evidence of successful sample preparation, with test genes providing a 3':5' signal ratio of less than 3. dChip 97 provided an alternative view, with a direct oligo-by-oligo view on the success of hybridization across the array surface; slides with both single-probe and probe-set outlier rates of less than 5% were taken as satisfactory. Only arrays in which both results were in range were accepted. In this case, 11 of 12 arrays were satisfactory; the first tubule array failed both MAS and dChip criteria, and so the first experimental pair was discarded to leave a five-sample paired design. As will be seen from the results, this design was sufficient to identify tubule-enriched genes with a high level of confidence. As sample collection extended over the whole day, array results from morning versus afternoon samples were compared (data not shown), but no difference was found between the two groups at this very broad time resolution.

Microarray samples were analyzed by two independent routes. The first was low-level analysis with the Affymetrix MAS 5.0 suite and identification of differentially expressed genes using the Affymetrix Data Mining Tool. The second was low-level analysis using dChip software 97 followed by assessment of significance using SAM software 98 followed by post-analysis by dChip. The MAS5 low-level analysis consisted of background subtraction followed by robust conversion of probe-level perfect match-mismatch (PM-MM) expression values into probe-set-level signals followed by linear multi-chip normalization (scaling). Tubule enrichment was based on an Affymetrix 'up' call, and a critical level of p < 0.05. In this analysis method, tubule and fly samples were taken as matched pairs, reflecting their biological origin. The dChip-based low-level analysis consisted of background correction followed by the multi-chip, 'invariant-set' nonlinear normalization at probe level followed by the calculation of model-based expression indices using PM expression values only. Differentially expressed genes between two groups of five replicates were identified within dChip by filtering data using the following criteria: lower 90% confidence bound of fold-change 21 > 2; difference between group means on antilog scale > 100 and p-value for t-test of equal group means < 0.01. Alternatively, the differentially expressed genes were identified using SAM software with 1,000 sample permutations and false-discovery rate cutoff of 1%. These were then post-filtered using two first criteria from the dChip analysis mentioned above. Fold change was calculated as a ratio of group means. Outputs were saved as Excel files, and parsed by hand-coded Perl scripts.