In order to investigate the molecular signature of the trx mutants in Drosophila melanogaster, we have compared whole-genome expression profiles of trx mutant third instar larvae and wild-type larvae (see Materials and methods). We designed two-color cDNA microarrays containing 12,120 genes annotated in RefSeq from D. melanogaster 37. The analysis of the microarray experiments identified 535 genes showing a statistically significant change (at least 2-fold change, p-value <0.05) in expression between mutant and wild-type samples (see Materials and methods). Of these, 260 were over-expressed and 275 were under-expressed in mutant larvae (Additional data file 1).

We mapped these deregulated genes to the fly genome (assembly dm2, April 2004) using the RefSeq 37 track of the UCSC genome browser 38, and the chromosomal distribution is shown in Table 1. The number of RefSeq genes annotated on each chromosome is also displayed. We mapped more co-expressed genes on chromosome 3L than on any other chromosome (30% of 535 deregulated genes; Table 1): 69 up-regulated genes (p-value <10^-2) and 94 down-regulated genes (p-value <10^-8). Chromosomes 2R and 3R are, however, richer in number of annotated genes (3,993 and 4,843 genes respectively, compared to 3,775 genes in chromosome 3L in Table 1).

Since chromatin modifications are typically associated with the coordinated expression of groups of nearby genes 3 and the analysis of different transcriptome datasets has shown that genes with a similar expression pattern are frequently located next to one another in the linear genome 2139, our next step was to determine whether deregulated genes in our trx mutants are located in close proximity in the fly genome (chromosomal clusters). There are many possible definitions of what a cluster of genes is (see 8 for a review). Here, we define a cluster as a group of genes located close to each other on the same chromosome in the genome, but not necessarily adjacent, that showed the same expression pattern (up-regulation or down-regulation) in the microarray experiment (see Materials and methods).

Chromosomal clusters can be identified computationally 2040. We detected 97 genes, organized in 25 genomic clusters, that are deregulated in trx deficient larvae (10 clusters of up-regulated genes and 15 clusters of down-regulated genes; Table 1), using the program REEF 41 with the following parameters: window length, 25,000 bp; window step, 1,000 bp; minimal number of co-expressed genes, 3; q-value ≤0.05. The chromosomal distribution of clusters and genes along the genome of D. melanogaster is shown in Figure 1 (up-regulated genes are depicted in red, down-regulated genes in green; the genomic position of each cluster is represented with the corresponding red or green triangle and each cluster is labeled with the same identifier used in Table 2). Clusters of genes deregulated in trx mutant larvae are not uniformly distributed along the genome: 15 out of 25 clusters (60%) are located on chromosome 3L (Table 1). Remarkably, the proportion of genes in clusters increases dramatically in chromosome 3L: 62 genes out of 163 deregulated genes mapped to this chromosome are clustered (38%), as opposed to only 35 genes out of 372 deregulated genes mapped to the other chromosomes (9%) (Additional data file 2).

The clusters reported here contain a total of 162 genes (97 deregulated genes and 65 genes whose change in expression was not significant), comprising in total 372,967 nucleotides, with an average gene density of 4.3 genes per 10 Kb. In contrast, the average gene density in the fruit fly genome is 1.6 genes per 10 Kb. The average length of the genes in clusters is 946 bp, while the length of the deregulated genes that are not clustered is 3,416 bp (the overall average for D. melanogaster is 6,976 bp). Since the REEF program approach is based on genomic proximity measured in number of nucleotides, this could favor artifactual cluster definition in gene-rich regions of the genome. To rule out this possibility, we have designed an alternative clustering algorithm based on measuring the number of co-expressed genes within a window containing a fixed number of annotated genes, rather than a fixed number of nucleotides (see Materials and methods for further details). Results obtained with our clustering strategy are highly concordant with those produced by the REEF program (Additional data file 3): 27 clusters were detected (22 identical clusters, 2 clusters with additional genes, 3 new clusters and 1 missing cluster). Therefore, the high gene density observed in our clusters is not the consequence of any computational limitation in the clustering method. Given the high concordance of the two clustering approaches and since REEF is the more standard approach, we have based our subsequent analysis and experiments on the REEF results (the list of the clusters and the genes that constitute each cluster are shown in Table 2).

As a control test to assess the statistical significance of the clustering, we repeated the analysis on 100 sets of genes that were randomly selected from the fly genome, preserving the gene distribution in the chromosomes that we observed in the set of genes deregulated in trx mutant larvae (see Materials and methods). The number of clusters identified on the random sets was very small (on average 1.7 clusters compared with the 25 clusters observed from the experimental data) despite containing the same proportion of genes on every chromosome (Figure 2a). In addition, we computed the Z-score of the number of clusters observed in our microarray, using the distribution of number of clusters found in the random sets as background distribution (see Materials and methods). This score is highly significant for trx clusters: 17.25 (Additional data file 4). Because of the small size of clustered genes, one could argue that the clustering described here is due to specific properties of short and active genes, and not related to a trxG characteristic. Therefore, we retrieved all small genes of the fly genome (that is, genes with the same range of sizes as the ones found in this work) and repeated the previous test (see Materials and methods). The number of clusters observed in the whole collection of fly small genes was significant: 107 clusters (including 21 of the 25 trx clusters; Z-score 9.75; Additional data file 4). The existence of clusters of small sized and active genes has already been established for many genomes and it is thought that this organization could favor coordinated and efficient gene expression 4243. However, the clustering tendency of genes regulated by TRX is stronger as the Z-score for trx clusters (17.25) clearly contrasts with the one measured in the whole fly genome (9.75). As an additional control, we generated 100 random sets of genes preserving the same size distribution observed in up-regulated and down-regulated genes (see Materials and methods). The number of clusters detected in trx deregulated genes is highly significant (10 and 15 clusters, respectively) in comparison to the average number of clusters identified on these random gene sets (0.9 and 1.4 clusters). This is strongly indicative that the clustering tendency observed here is a specific characteristic of TRX regulated genes, and not a general feature of short genes (Additional data file 5).

In the analysis presented here, we have used no information about homology between genes within clusters to control for overrepresentation of gene families. Many genomic clusters corresponding to gene families have indeed been previously identified 4445. Such genomic clusters could cause spurious co-expression because of probe cross-hybridization between highly similar genes. In fact, some of the clusters that we have computationally identified do contain members of the same gene family (Table 2). We have searched for regions of similarity between the sequences of the genes within each cluster but no significant pairwise sequence alignments were found for any cluster (see Materials and methods). Furthermore, we confirmed the reported change in the expression of these genes by quantitative real-time RT-PCR in two clusters (Figure 2b).

Finally, we used the specific set of 445 genes (302 RefSeq genes) that are basally expressed in larvae described by Arbeitman et al. 28 to measure the specificity of our results (see Materials and methods). We were not able to reproduce in this data set the organization in clusters found in genes regulated by TRX (only one potential cluster was found), indicating that this is not a general feature of the larval stage in D. melanogaster development.

To determine whether the chromosomal organization in clusters is also common to genes regulated by other proteins involved in chromatin dynamics, we performed a second microarray expression experiment with mutant larvae for another such factor, ASH2, and compared the results obtained in this experiment, as well as previous published results on the transcriptomes of NURF 46, dMyc 47 and ASH1 34, with the results obtained in the microarray analysis of the trx mutant. In all experiments, deregulated genes have been clustered on the D. melanogaster genome using the REEF program (Additional data file 6).

The ash2 gene (absent, small, or homeotic discs 2) is another member of the trxG involved in chromatin-mediated maintenance of transcription 4849. The microarray analysis identified 244 genes showing a statistically significant change (at least 2-fold change, p-value <0.05) in their expression between mutant and wild-type samples (see Materials and methods). According to their pattern of regulation, we identified 123 over-expressed genes and 121 under-expressed genes in the mutant larvae (Additional data file 7). As in previous studies 3233, we found the same proportion of up-regulated and down-regulated genes in the ash2 mutants. We also mapped these genes to the genome of D. melanogaster according to the RefSeq annotations in the UCSC genome browser, and identified eight clusters of co-expressed genes (six clusters of up-regulated genes and two clusters of down-regulated genes) using the program REEF (Table 3).

NURF is an ISWI-containing ATP-dependent chromatin remodeling complex 50. Badenhorst et al. 46 performed a microarray analysis using larvae from D. melanogaster lacking the NURF specific subunit NURF301. We mapped the list of 274 genes (265 RefSeq genes) that require NURF301 according to this experiment (the list of up-regulated genes has not been published) to the genome. We then identified seven clusters of down-regulated genes using the program REEF (Table 3).

Goodliffe et al. 47 reported that the Polycomb protein (Pc), a member of PcG, mediates Myc autorepression and its transcriptional control at many loci. In this study the authors used the Gal4 UAS system to express ectopic dmyc in embryos and performed microarray analysis to examine the effect on gene expression. We mapped the list of 272 genes (203 RefSeq genes) up-regulated in this experiment (the list of down-regulated genes is unavailable) and then identified 6 clusters of co-expressed genes using the program REEF (Table 3).

More recently, Goodliffe et al. 34 extended the studies on Myc function and reported a coordinated regulation of Myc trans-activation targets by Pc and ASH1. The ash1 gene (absent, small, or homeotic discs 1) is also a member of the trxG 48. In this work, the authors used RNAi to reduce the levels of ash1 and conducted microarray experiments 34. The analysis of these microarrays identified 398 genes with a substantial change in their expression (239 over-expressed RefSeq genes and 159 under-expressed RefSeq genes). We mapped these genes to the fly genome and identified eight clusters of co-expressed genes (seven clusters of up-regulated genes and one cluster of down-regulated genes) using the program REEF (Table 3).

Together, these results suggest that chromosomal organization in clusters is a distinctive feature of some genes controlled by chromatin regulators. To elaborate more on this hypothesis, we compared the clusters identified in the microarray experiments of trx with those identified in the experiments of the other factors at three different levels: common clusters, common genes in clusters and common genes in the transcriptome maps (see Materials and methods for further details). We consider that two clusters from two different microarrays are matching if and only if they are overlapping in at least one commonly deregulated gene. The results of the comparison are shown in Table 4 and, as an example, the regulatory gene profiles of trx, ash2, Nurf, dmyc and ash1 along the chromosome 3L and the clusters containing these genes are shown in Figure 3 (the regions of the chromosome harboring the same cluster at the same time in both the trx experiment and another microarray are indicated with gray).

Overall, between 50% (ASH1) and 100% (dMyc) of the trx clusters are also detected in the other chromatin regulators (71% on average; Table 4). This strongly suggests that there is high concordance between the trx clusters and the clusters inferred for the other chromatin regulators. There is not, however, an exact equivalence: clusters from different regulators that overlap in genome space with trx clusters may contain different regulated genes. Thus, the intersection between the genes deregulated by TRX and the genes regulated by other factors in the common clusters ranges from 38% (ASH1) to 75% (ASH2) of the genes (50% on average; Table 4). Nevertheless, this value dramatically decreases when the whole transcriptomes of each experiment are taken into account. In this case, the intersection between the set of genes deregulated in trx mutant larvae and any other set of genes whose expression was significantly affected by other chromatin regulators is lower than 20% on average (Table 4). These results suggest that the clusters identified in common form a group of gene targets directly or indirectly regulated by these chromatin regulators. In addition, this clustering is a specific feature of short and active genes: the average length of deregulated genes in these clusters is 1,135 bp, while the size of deregulated genes in these microarrays that are not clustered is, on average, 4,204 bp (Additional data file 8). These clusters overlap with clusters of small genes identified along the fly genome in the previous section: 75% of them for ASH2, 57% for NURF, 83% for dMyc, 75% for ASH1 (see Figure 3 for a graphical comparison on chromosome 3L).

The clustering organization reported here might be general for transcription factor target genes, and not a feature of genes regulated by chromatin remodeling factors. To rule out this hypothesis, we have collected microarray data for six transcription factors to extend the clustering analysis: fkh (fork head) 51, ey (eyeless) 52, spdk (spotted-dick) 53, gcm (glial cells missing) 54, Otd (Orthodenticle) 55 and lab (labial) 56. We mapped each set of genes to the fly genome, using the program REEF to identify putative clusters. In most cases, however, no clusters were detected (Additional data file 9), indicating that clustering is not a general characteristic of transcription factor target genes. The lack of clustering in these microarrays does not merely reflect the larger gene size for the targets of these genes (Additional data file 10).

Finally, we used the expression data published by Riedl et al. 57 as a negative control to qualitatively assess the significance of our results. The information has been obtained from two microarray experiments involving rover and sitter larvae to study foraging locomotion in the fruit fly 57. The intersection between these transcriptomes and the trx transcriptome is only slightly lower than that observed between TRX and the other chromatin regulators (6% and 9% for rover and sitter, respectively). However, only five clusters in total were detected among the genes regulated in the rover and sitter microarrays (2 and 3 clusters, respectively). Of these, only one mapped to chromosome 3L and none overlapped the trx clusters (Table 3).

The genomic structure of the gene clusters governed by chromatin regulators does not appear to be homogeneous. The average size of clusters in the trx mutants is 3.5 genes, while the genomic region that harbors such genes contains, on average, 6.7 genes (Additional data file 2). For instance, although the cluster shown in Figure 4a contains four genes down-regulated by TRX (depicted in green), there are five additional genes annotated in this genomic region (depicted in blue) for which no change in expression was detected in the microarray. In addition, the comparison of the clusters identified in the different microarrays indicated that, as already outlined, only about 50% of the genes in a cluster regulated by either TRX or another chromatin regulator are actually deregulated in both experiments at the same time (Table 4). In many cases, therefore, either genes in the equivalent clusters from different experiments do not show the same regulation pattern or the boundaries of the clusters are not exactly the same. For example, the same cluster containing eight genes shown in Figure 4a, b was identified by the program REEF in both the trx and the ash1 microarrays. However, there are three interesting differences: the gene boundaries of the clusters when considering only the regulated genes are not the same; the expression of the genes changes in the opposite sense (down-regulation versus up-regulation); and some of the clustered genes are not regulated by any of the factors.

We used the whole-genome expression data generated by Hooper et al. 30 to investigate whether all genes within the genomic expanse of the trx clusters, and not only those defining the clusters themselves, are co-expressed (there are 162 genes within the region of the trx clusters, but only 97 in the clusters). For this dataset, Hooper et al. measured the expression of genes during the first 24 hours of embryonic development in D. melanogaster (1 hour time points). We used the data between 4 h and 24 h to minimize the possibility that the maternal effect could mask zygotic expression (see Materials and methods). Co-expression was evaluated both by using only those genes that define the trx clusters and using all genes located within the boundaries of each cluster. Based on the expression data provided in 30, we computed the Pearson's correlation coefficient between each pair of genes within the same chromosome across the 20 time points. For each cluster, the level of co-expression was then defined as the mean of Pearson's correlation coefficients between all pairs of genes in the cluster (see Materials and methods). As a reference set, we calculated the same values for each possible artificial cluster of N consecutive genes in the genome (2 ≤ N ≤ 15).

The distribution of values obtained for the clusters containing only the genes deregulated in trx mutants, the clusters containing all genes mapped within the boundaries of the clusters and the artificial clusters of several sizes using the 4 h-24 h expression data set are shown in Figure 4c. Interestingly, the distribution of co-expression levels in randomly generated clusters of different sizes appears to be slightly positive (means ranging from minimum to maximum), probably suggesting an overall induction of transcription during the first stages of larval development. The distribution of co-expression levels computed within the boundaries of clusters, and, in particular, computed only from the regulated genes defining the clusters, is, however, clearly skewed to the right, indicating much stronger coexpression than expected at random. The bimodal shape of the distribution, more accentuated when considering only the genes defining the clusters, suggests the existence of a class of clusters with tight regulation of expression. The deviation from randomness in the trx clusters is perhaps more appreciable in the cumulative plots (Additional data file 11).

Therefore, genes present within the genomic boundaries of the trx clusters, including those not in the defined clusters, are overall co-expressed. There are several causes that can explain the existence of additional genes within the boundaries of a trx cluster. These genes might not have been included in the clusters either because they were not in the array (4 cases out of 65 additional genes), because the gene showed a different pattern of regulation (up-regulated instead of down-regulated or vice versa, 1 case), or because the expression intensity from the microarray was below the selected thresholds (60 cases).

The trxG members are known to be positive regulators of transcription 24. However, in our study, we found a similar number of up-regulated compared to down-regulated genes in the trx mutants. Similar results have recently been reported for ash2, ash1 and Nurf301 333446, suggesting that trxG proteins might also act directly or indirectly as repressors of certain genes. We once more applied the REEF clustering strategy, but this time considering all trx deregulated genes together, irrespective of the direction of their regulation. In addition to the 25 clusters previously detected, this method allowed us to identify six additional 'hybrid' clusters (with both up- and down-regulated genes). Moreover, we also enriched previously detected clusters with genes regulated in the opposite direction (Figure 5). In total, we identified 129 deregulated genes that were organized in 31 clusters.

The clusters of genes detected here might be acting as transcriptional units with coordinated transcriptional regulation. One would therefore expect some level of conservation of cluster organization across species. The genomes of multiple species of Drosophila have been recently made available through the UCSC genome browser 38, allowing investigation of the conservation of trx clusters in other Drosophila species. Only three of these genomes have been completely assembled: D. simulans, D. yakuba and D. pseudoobscura 58. We have mapped all D. melanogaster genes to the genomes of each of these species using the BLAT alignments provided by the UCSC genome browser 59 (see Materials and methods). The number of genes annotated on each species using this method is shown in Table 5.

After mapping the up-regulated and down-regulated genes of the trx mutant from D. melanogaster to the other Drosophila genomes, we used the program REEF with the same set of parameters to identify putative clustering of these genes. The number of clusters detected in these species is shown in Table 5: 20 clusters in D. simulans (corresponding to 7 up-regulated clusters and 13 down-regulated clusters in the trx microarrays), 25 clusters in D. yakuba (11 up-regulated clusters, 14 down-regulated clusters) and 14 clusters in D. pseudoobscura (1 up-regulated cluster, 13 down-regulated clusters). We have compared the clusters obtained in D. melanogaster with the clusters identified in these three species: 24 out of 25 clusters (96%) identified in D. melanogaster were conserved in at least one other species (80% of the clusters were conserved in D. melanogaster and two more species, 36% of the clusters were conserved in all species). In contrast, the percentage of clusters identified in these species that was not detected in D. melanogaster was very low (0% in D. simulans, 16% in D. yakuba, 14% in D. pseudoobscura; Table 6), indicating that this set of deregulated genes is similarly organized in the genome of these species. The distribution of clusters on each genome is shown in Figure 6 (the clusters of D. melanogaster that are conserved in other species have the same identifier as in Figure 1).

Another genome of interest for the identification of homologous clusters potentially regulated by the trx gene is that of Anopheles gambiae 60. We obtained the list of putative Anopheles orthologs to the D. melanogaster genes using the ENSEMBL annotations 61. Less than 50% of the fly genes could be mapped to the mosquito genome in this way (Table 5). Consequently, only 7 clusters were identified. Most of these clusters, however, were conserved in D. melanogaster (Figure 6 and Table 6).

In the work presented here, we identified a set of 25 gene clusters in D. melanogaster that are phylogenetically conserved in other flies. However, given the strong synteny between the Drosophila genomes (see divergence time estimates in Table 6), we can not claim that the conservation of clusters that we observed is not simply a consequence of such an overall synteny. To discard such a hypothesis, for each cluster identified in D. melanogaster we examined the number of genes in common found in the corresponding cluster in each of the other Drosophila species (allowing for gene rearrangements and chromosome inversions inside the region; see Materials and methods for further details). We also analyzed the number of genes in common between the corresponding flanking areas of these clusters in order to compare the number of genes that are conserved inside and outside them; the results are shown in Table 6. While the genes that constitute the clusters of trx in D. melanogaster are mostly the same in the clusters of the other species (96% in D. simulans, 88% in D. yakuba, 96% in D. pseudoobscura), the number of conserved genes in the vicinity of each cluster decreases in more distant species (86% in D. simulans, 64% in D. yakuba, 58% in D. pseudoobscura). Additional statistical tests confirmed these observations (see Materials and methods). According to these results, we conclude that the overall synteny between the Drosophila genomes is not enough to explain the high level of conservation observed in the clusters of genes deregulated by TRX in D. melanogaster.

In order to characterize the clusters previously identified in D. melanogaster, we functionally annotated their constituent genes (Additional data file 12) using Gene Ontology (GO) 62. GO is a hierarchical dictionary of biological terms structured into three main categories: molecular function, biological process and cellular component. We also annotated the function of the full set of genes in our microarray and of the genes that were reported to be up-regulated or down-regulated to estimate the statistical significance of our results.

We analyzed the information available for the genes of each respective set (12,120 genes in the microarray, 535 deregulated genes, 97 genes in clusters) at the third level of the molecular function ontology (see Materials and methods). A graphical representation of the more abundant categories for each of the three gene sets is shown in Figure 7a. The clusters of down-regulated genes are significantly enriched in structural proteins involved in cuticle formation (p-value <10^-37; see Materials and methods). The over-representation is less relevant in the set of down-regulated genes, while it is not observed in the full collection of genes in the microarray (Figure 7). The clusters of up-regulated genes are also enriched in proteins with carbohydrate and pattern binding functions, as well as structural components of the peritrophic membrane (p-value <0.005; see Materials and methods). The set of up-regulated genes is also, albeit to a lesser extent, enriched in these categories, while no over-representation is observed in the complete set of genes in the microarray (Figure 7). A more detailed inspection of the functional annotations of the clusters reveals that this over-representation is due to the abundance of genes involved in the subcategory 'chitin binding' 62.

According to this, the clusters of genes deregulated in the trx mutant contain a significant number of genes involved in cuticle structure and other related functions. To confirm these results, we performed a whole-genome clustering of the genes from D. melanogaster according to their functional annotation in GO, in which no expression data were used (see Materials and methods). We focused, in particular, on the two functional categories over-represented in the clusters of genes deregulated in trx mutants: structural constituents of cuticle (GO:0042302) and chitin binding (GO:0008061).

We found 98 genes annotated as components of the cuticle in the genome of D. melanogaster. Using the program REEF with the same set of parameters, we identified 12 clusters of genes involved in cuticle structure (shown as blue stars in Figure 7b), 8 of which are located on chromosome 3L. We found 67 genes annotated as chitin binding proteins and identified 6 clusters, 5 of which are also located on chromosome 3L (shown as purple circles in Figure 7b). Of the 18 functional clusters, 8 overlap the clusters of genes regulated by TRX (shown as red and green marks in Figure 7b). In addition, most genes in the functional clusters are annotated as being in the same functional categories (Additional data file 13).

We have also attempted to characterize the expression pattern of genes in the trx clusters. Using data from the Li et al. 63 study, in which the tissue distribution of genes expressed 18 hours before the larval to pupal metamorphosis in D. melanogaster was determined, we characterized the expression of the members of three different gene sets (genes in the microarray, deregulated genes, genes in clusters; see Materials and methods for further details); results are shown in Figure 8a. As expected, the proportion of genes in our microarray expressed in different tissues is very similar to that reported by Li et al. 63: 28% are expressed in the central nervous system, 24% in wing discs, 20% in the midgut, 16% in salivary glands and 12% in the epidermis. Most genes regulated by TRX, on the other hand, are expressed in the midgut (30%, p-value <10^-5) and in the epidermis (27%, p-value <10^-6), while the genes that constitute the clusters regulated by TRX are abundantly expressed in salivary glands (41% of genes, p-value <10^-5). We compared the tissue-specific expression pattern of the genes in the trx clusters with the functional annotation of these genes as inferred through the GO annotations (see previous section) and results are shown in Figure 8b: 16 of 25 clusters (64%) contain genes that are either expressed in salivary glands or code for structural cuticle proteins. In addition, the clusters appear to be divided into those containing genes expressed in salivary glands and those containing genes coding for structural proteins. Together, these results demonstrate that TRX plays a key role in the regulation of clusters containing genes involved in the structure of the cuticle during the larval stages of D. melanogaster development.

The PcG and trxG complexes bind to sequences called PRE/TREs. However, not only is the genomic distance between well-characterized PRE/TREs and their target genes highly variable, ranging from a few nucleotides to more than 60 Kb in many cases, but they can be found both upstream or downstream of the gene 646566. Several methods have been proposed to detect PRE/TRE elements in genomic sequences. For example, Ringrose et al. 67 developed a computational approach to detect potential PRE/TREs and identified 167 candidates in the fly genome, some of which were validated experimentally 67. More recently, several ChIP-on-chip experiments have been performed to search for PcG targets. Among others, Schwartz et al. 68 determined the distribution of the PcG proteins Pc, E(z) and Psc and of H3K27me3 in the whole genome, Tolhuis et al. 69 constructed a map of binding patterns of the PcG proteins Pc, esc and Sce in chromosomes 2L and 4, and part of chromosomes 2R and X, and Negre et al. 70 analyzed the binding profile of the PcG proteins Pc and ph, and the GAGA factor in certain regions of chromosomes 2L and X.

We mapped the PcG target sites identified in each of these experiments (251 sites in Schwartz et al. 68, 131 sites in Tolhuis et al. 69 and 36 sites in Negre et al. 70) and the 167 PRE/TREs predicted by Ringrose et al. 67 in the fly genome (assembly dm2, April 2004). We then compared the location of the 25 clusters of genes deregulated by TRX in the fruit fly genome with the location of the PcG target sites and the PRE/TRE predictions (Figure 9). Since the distance between the PRE/TRE and its target gene can be highly variable, we decided to confine the search to the PcG binding regions or PRE/TREs that were located at most 100 Kb distant from our clusters. According to this restriction, we found 14 out of 25 clusters (56%) near one experimental evidence or a PRE/TRE prediction, five of which were supported by both a PcG binding site and a PRE/TRE (Additional data file 14). We performed ChIP analysis of third instar larvae, using both anti-TRX and H3K4me3 specific antibodies, to test the possible binding of the TRX protein to some of the predicted PRE/TREs 67 that are in close proximity to the trx deregulated clusters. Preliminary results seem to indicate that TRX is capable of binding to at least two of the PRE/TREs tested (Additional data file 15). In keeping with its role as a histone methyltransferase, TRX binding correlates with the presence of H3K4me3.

All Drosophila strains and crosses were kept on standard media with 0.025% bromophenol blue. The reference line used was the w¹¹¹⁸_iso; 2_iso; 3_iso isogenic line from the DrosDel Collection 93. To reduce the differences in the biological background between the alleles under study and the reference strain, we transferred chromosomes X, Y and 2 from the isogenic line to the TM6c-balanced ash2^I1, and to trx^B11 and trx^E3 alleles, which were used in trans-heterozygosity. Their genotypes were, respectively, w¹¹¹⁸_iso; 2_iso; ash2^I1/TM6c, w¹¹¹⁸_iso; 2_iso; trx^B11/TM6c and w¹¹¹⁸_iso; 2_iso; trx^E3/TM6c.

Microarrays were printed on Corning UltraGAPS slides (Corning, Corning, NY, USA) at the Plataforma de Transcriptòmica (SCT - PCB, Universitat de Barcelona, Spain) using the Drosophila Genome Oligo Set version 1.1 (Operon Biotechnologies Inc., Huntsville, AL, USA), a collection of 14,593 probes representing 13,577 Drosophila genes with Flybase ID (12,120 genes in RefSeq). The 70 mer Arabidopsis sequences from TIGR 94 and spots with no material or with buffer were also printed to be used as spike-in and negative controls, respectively. The microarray annotation is deposited in the Gene Expression Omnibus database with accession number GPL3797. Wandering blue-gut staged Tb+ early third instar larvae were selected in all cases to extract total RNA using the RNeasy Protect Mini Kit (Qiagen Inc., Valencia, CA, USA). At least two independent total RNA extractions were carried out. Quality was assessed in all samples using the Eukaryote Total RNA Nano Assay on a 2100 Bioanalyzer (Agilent Technologies Inc., Santa Clara, CA, USA). Total RNA from w¹¹¹⁸;+;+ larvae was used as a common reference. Four microarrays were hybridized for each experiment in biological replicate pairs, such that the total RNA from the sample used as a starting material came from different extractions. Both arrays from each pair were hybridized with the same amplified RNA from sample and common reference (obtained using the Amino-Allyl Messageamp II aRNA Amplification Kit from Ambion Inc., Austin, TX, USA) using dyes Cy3 and Cy5 (GE Healthcare UK Ltd, Buckinghamshire, UK).

GenePix Results (GPR) data files were obtained for each microarray with an Axon 4000B scanner and GenePix Pro 6 (Molecular Devices Corp., Sunnyvale, CA, USA). All GPR files were analyzed with the Limma package from BioConductor 9596 using the same criteria. The spots not fulfilling the quality thresholds (based on spot size, foreground versus background signals, saturation, coincidence between differently calculated ratio measures and R² of regression ratio) were eliminated from the analysis, and the data were background corrected with the normexp method and normalized using OLIN 97. Between-array normalization was carried out independently for each set of four arrays using the mad method from OLIN, and a linear model was fitted and corrected with False Discovery Rate (FDR) 95. We obtained lists of genes that were differentially expressed two-fold in the mutants compared to the isogenic strain (FDR-corrected p-value lower than 0.05). Raw and normalized data are deposited in the Gene Expression Omnibus database (user name: sergiba_rev_1, password: 2139861083) with the accession number GSE8783, which includes the data for trx (GSE8748) and ash2 (GSE8750) mutants.

We obtained the set of genes that are differentially expressed in the larvae of D. melanogaster from Arbeitman et al. 28. We selected only the genes that were up-regulated at least two-fold in the larvae (302 RefSeq genes). We extracted the gene expression data for Nurf301-/- from Badenhorst et al. 46 (only the list of down-regulated genes was published). The set of up-regulated genes in response to dmyc over-expression was taken from Goodliffe et al. 47. We extracted from Goodliffe et al. 34 the set of up and down-regulated genes when the regulatory activity of ash1 was repressed using RNA interference to study the interaction between the dMyc complex and ASH1. We used the expression data from two microarray analyses (rovers and sitters) about foraging locomotion from Riedl et al. 57 as a negative control in the clustering process. The set of genes expressed in five different tissues (midgut, salivary glands, epidermis, central nervous system and wing disc) studied in the larval stage of D. melanogaster was obtained from Li et al. 63.

Reverse transcription reactions with RNA independently isolated in Trizol from all mutant alleles and the reference were used to synthesize cDNA with M-MLV RT (Invitrogen Corp., Carlsbad, CA, USA) according to the manufacturer instructions. We used 1 μl of a 1/10 dilution of cDNA on each quantitative real-time PCR (qRT-PCR). The qRT-PCR was performed on the ABI PRISM^® 7700 following the recommended protocol (Applied Biosystems, Foster City, CA, USA). Each sample was replicated three times and average values were used for further analysis. Data were analyzed by the ΔΔC_T method, being normalized by subtracting the value of the control gene dia. TaqMan primers and probes designed and synthesized by Applied Biosystems for this analysis were: Dm02371023_s1 (CG30458); Dm02366349_s1 (CG30457); Dm02366353_s1 (CG10953); Dm01792445_s1 (CG14567); Dm01792458_s1 (CG14572); Dm01792469_s1 (CG14565); Dm01792478_s1 (CG14564); and Dm01811206_g1 (dia).

ChIP assays were essentially performed as previously described by Papp and Muller 98 with the following changes: 40 wandering third instar larvae were fixed with 1.8% formaldehyde solution for 25 minutes at room temperature, collected in 700 μl of lysis buffer (1% SDS, 50 mM Tris HCl pH 8.0 and 10 mM EDTA) and disrupted 6 times for 20 seconds at 30% of a Branson sonifier. Samples were centrifuged at top speed at 4°C and aliquotes of 100 μL of extract were used per immunoprecipitation. Ten percent of the immunoprecipitated extract was used as input, and a sample immunoprecipitated without antibody as a precipitation control. A 300 bp fragment corresponding to the transcription start of the Ultrabithorax gene, which has been previously described to present a TRX binding site 9899, was used as a positive control. TRX antibody was a gift from A Mazo and H3K4me3 antibody was obtained from Abcam Inc. The primers used for the PCR amplifications were: Ubx forward, 5' CATGCCCAGCGAGAGAGG 3'; Ubx reverse, 5' AACAGCACAGAAAGCGAGG 3'; cluster9 forward, 5' ACCCACTTTTGCGCCATCG 3'; cluster9 reverse, 5' ACAAAGCGGTTCCGTGTCG 3'; cluster25 forward, 5' ACGTCTGGCTATGGATCTGG 3'; cluster25 reverse, 5' GGACACCGATGTGACCACC 3'.

All gene sets were mapped in the genome of D. melanogaster using the RefSeq track of the UCSC genome browser 38 annotations (genome assembly dm2, April 2004). We considered only one transcript per gene (the first one present in the annotations). The file refGene.txt was used to retrieve the coordinates of each gene in RefSeq. The official name and the description of each gene were retrieved from the file refLink.txt. The statistics of gene distribution on each chromosome were computed using the 19,670 unique RefSeq genes contained in the file refGene.txt as well. To measure the statistical significance of the gene distribution, we randomly generated 1,000 rounds of sets of genes with the same size to evaluate the p-value of each result using a hypergeometric distribution. We used the BLAT alignments between the RefSeq genes from D. melanogaster and the genomes of D. simulans, D. yakuba and D. pseudoobscura to produce a catalogue of genes for these species (Additional data file 18). On each genome distribution, the file xenoRefGene.txt contains the best BLAT hits of the genes of other species in that genome 38. We selected only the best hit for each RefSeq gene from D. melanogaster in the corresponding genome distributions of D. simulans, D. yakuba and D. pseudoobscura. The sets of up- and down-regulated genes in the trx microarray were then mapped in those genomes using the appropriate catalogue of genes. We used the list of orthologous genes between D. melanogaster and A. gambiae provided by the Biomart tool 100 from ENSEMBL. The association between the ENSEMBL gene identifier and the official gene name in the fruit fly genome was used to cross this information with the set of RefSeq genes of fly annotated in the UCSC.

We define a cluster as a group of neighboring genes located in a limited region of the genome, not necessarily consecutive, that show the same expression pattern (up-regulation or down-regulation) in the microarray experiment. Clusters are determined by the physical position in the genome, the number of co-expressed genes inside and the total number of genes in such a genomic fragment. The length of the clusters can be then defined in terms of nucleotides or number of genes annotated on that region. We follow two similar approaches to computationally detect the clusters in the sets of genes in our work: the program REEF and our own implementation of a clustering program. No significant difference was observed between the results obtained with both computational approaches.

To evaluate the significance of the number of clusters identified in our microarray, we generated 100 randomly generated sets with the same size and chromosomal gene distribution of the real sets of genes regulated by TRX. We then performed the clustering analysis in such data sets to find out how many clusters could be obtained by chance (expected clusters). The Z-score of the results observed in our microarray was thus calculated using the distribution of clusters identified in the random sets:

z = x − μ σ MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8GiVeY=Pipec8Eeeu0xXdbba9frFj0xb9Lqpepeea0xd9q8qiYRWxGi6xij=hbbc9s8aq0=yqpe0xbbG8A8frFve9Fve9Fj0dmeaabaqaciGacaGaaeqabaqabeGadaaakeaacaWG6bGaeyypa0tcfa4aaSaaaeaacaWG4bGaeyOeI0IaeqiVd0gabaGaeq4Wdmhaaaaa@384B@

where x is the number of observed clusters in the trx microarray, and μ and σ are the mean and the standard deviation of the distribution of number of clusters found in the random sets, respectively.

To evaluate whether regulation by TRX imposes a stronger clustering tendency compared to gene size, we extracted all genes from the fruit fly genome whose size is within the range delimited by the size observed in genes from trx clusters (average size 946 bp, standard deviation 788 bp). We thus mapped the 6,626 genes that match this condition on the genome to get their coordinates and identified 107 clusters, using the REEF program. In order to assess the statistical significance of the clustering (and to measure which parameter, trx regulation or gene size, is more discriminant in the clustering), we repeated the analysis on 100 sets of genes that were randomly selected, preserving the gene distribution observed in the dataset of short genes. As an additional control, we also counted how many clusters can be identified in 100 different random sets of 260 genes (number of up-regulated genes) and 275 genes (number of down-regulated genes), which preserve the same gene size distribution observed in both lists of deregulated genes in our trx microarray (average and standard deviation).

The program REEF 41 identifies regions of a genome enriched in specific features, compared with a reference landscape of feature density. The two input files are: the list of reference features (genes annotated in the fly genome) mapped on a genome sequence; and the list of selected features among the reference set (genes regulated by TRX in our microarray) with their genomic positions and the number and the length of the chromosomes in the genome under consideration. REEF scans the reference genome using a sliding window approach, calculating the statistical significance of each window using the hypergeometric distribution and the FDR. The windows are defined as fragments of a fixed genomic length. Consecutive significant windows form a cluster of regional enriched features. Using an approach similar to that used by Spellman and Rubin 20, we defined the optimal set of parameters comparing the growth function in the number of clusters obtained in random sets and in the real reference set for different values (Additional data file 19). The optimal configuration of parameters to scan the genome of D. melanogaster was: window length, 25,000 bp; window step, 1,000 bp; minimal number of co-expressed genes, 3; q-value ≤0.05s.

We have also implemented our own approach in Perl to cluster the genes regulated by a certain gene. Basically, we scan the genome following a sliding window approach combined with the hypergeometric distribution to evaluate the statistical significance of the clusters. The length of the windows is determined, however, in terms of genes rather than nucleotides. Thus, the clustering process is not limited by a fixed window length. Some parameters in this algorithm must be adjusted in order to maximize the sensibility and specificity. Firstly, only clusters of at least three genes showing the same sense of deregulation were considered. Another important parameter is the size of the window to apply. Using the results in random sets, a window size of seven genes was selected as the optimal length (the number of clusters found in trx data seems to grow logarithmically with the size of the window used, while the number of clusters found in random sets grows exponentially). We used our algorithm to scan the chromosomes of D. melanogaster using window lengths of seven genes, advancing one gene between two instances of the window so that all possible seven-gene windows were tested. To avoid multiple testing problems, we took a conservative p-value threshold of 10^-3. Consecutive statistically significant windows were merged up in only one cluster. Results obtained with our clustering strategy are highly concordant to those produced by the REEF program (Additional data file 3): 27 clusters were detected (22 identical clusters, 2 clusters with additional genes, 3 new clusters and 1 missing cluster).

We computed different values on the genes that constitute the clusters (using the file refGene.txt downloaded from the UCSC genome browser): the average gene length, the gene length distribution, the number of bidirectional/opposed genes on each cluster according to their strand, and the average intergenic distance. Such values were compared to those observed in the whole set of genes annotated in the fruit fly genome, to evaluate their significance. The comparison between the clusters of genes regulated by TRX and the clusters regulated by other chromatin regulators was performed at three different levels: common genes in the transcriptomic maps, common genes in clusters and genomic position of the clusters. The intersection between two transcriptomes is defined as the quotient between the number of common genes (twice) and the total number of genes in both experiments. Two clusters from two different microarrays are matching if and only if they are overlapping in at least one commonly deregulated gene. Thus, we calculated the clusters identified in a second experiment (ash2, Nurf, dmyc, ash1) that were supported by another cluster in the trx microarray results. Once a set of common clusters was identified in two microarrays, we computed the percentage of co-expressed genes that was present in each pair of equivalent clusters (the ratio between the number of common genes and the total number of genes on each cluster).

Two clusters of genes identified in two different genomes are considered to be equivalent when the percentage of genes that are present in both clusters is 50% or higher. To calculate the percentage of genes that are common in two clusters, the genes that are located in a different relative order within the clusters are considered to be conserved. Clusters of genes that are conserved in different strands of the chromosomes are considered to be equivalent as well (allowing for chromosomal inversions in the flanking regions). The length of the left and right flanking areas of each cluster is equal to the number of genes of the corresponding cluster. To measure the statistical significance of these results, we randomly sampled 10,000 artificial clusters of seven genes in D. melanogaster (the average size of our clusters is 6.7 genes) and found the location of the equivalent cluster on the other genomes. We examined the number of genes shared between each artificial cluster of genes in D. melanogaster and its equivalent cluster conserved on each of the other Drosophila species. Each artificial cluster on each genome was constituted of the corresponding gene and the three genes annotated before and after in the same location (the average length of our clusters is 6.7 genes). The results are shown in Table 6. The conservation of the artificial clusters in terms of common genes is more similar to the conservation observed in the flanking area of our clusters (90% in D. simulans, 79% in D. yakuba, 65% in D. pseudoobscura). In fact, the difference between the number of conserved genes inside our clusters and inside the artificial clusters is significant in all the species (D. simulans p-value <10^-3, D. yakuba p-value <10^-3, D. pseudoobscura p-value <10^-6; Wilcoxon test). Similar results were obtained in the set of clusters detected in A. gambiae (Table 6).

Hooper et al. 30 generated whole-genome expression data for the first 24 h of embryonic D. melanogaster development by extracting RNA from overlapping 1 h time points for the first 6.5 h of development and non-overlapping 1 h time points for the rest. We generated a table containing all pairwise Pearson's correlation coefficients between all genes expressed between 4 and 24 h in the study by Hooper et al. 30. For each cluster identified in the trx mutants, we computed the mean of all pairwise correlation coefficients between the genes constituting the cluster to assess if they are co-expressed along the embryonic development. We also calculated the mean, including all the genes included in the boundaries of each cluster. As a reference set, we computed the same value for each set of correlative N genes in the genome (2 ≤ N ≤ 15). We then evaluated if each particular cluster was located in any of these tails and the total percentage of clusters that were found in these tails.

We downloaded the GO terms and relationships (file gene_ontology_edit.obo, release 1.2) and the associations between gene products from D. melanogaster and GO terms (file gene_association.fb, release 1.94) from the GO web site 62. We annotated different gene sets (full set of genes in the trx microarray, the up-regulated and down-regulated genes in the same experiment, the clusters identified in such co-expression sets) using the annotation available at level 3 of the 'molecular function' ontology tree. We climbed up in the ontology to obtain the corresponding parent term at level 3 for those genes that were annotated at deeper levels in the hierarchical tree. The statistical analysis was performed using hypergeometric distribution to test the probability of observing a given GO term significantly enriched in genes belonging to such clusters:

p = ∑ i = n m i n [ k , a ] ( A i ) ( G − A k − i ) ( G k ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2Caerbhv2BYDwAHbqedmvETj2BSbqee0evGueE0jxyaibaiKI8=vI8GiVeY=Pipec8Eeeu0xXdbba9frFj0xb9Lqpepeea0xd9q8qiYRWxGi6xij=hbbc9s8aq0=yqpe0xbbG8A8frFve9Fve9Fj0dmeaabaqaciGacaGaaeqabaqabeGadaaakeaabaGaamiCaiaad2dadaaeWbqcfayaamaalaaabaWaaeWaaeaaeaGabeaacaWGbbaabaGaamyAaaaaaiaawIcacaGLPaaadaqadaqaaqaaceqaaiaadEeacqGHsislcaWGbbaabaGaam4AaiabgkHiTiaadMgaaaaacaGLOaGaayzkaaaabaWaaeWaaeaaeaGabeaacaWGhbaabaGaam4AaaaaaiaawIcacaGLPaaaaaaaleaacaWGPbGaamypaiaad6gaaeaacaWGTbGaamyAaiaad6gadaWadaqaaiaadUgacaWGSaGaamyyaaGaay5waiaaw2faaaqdcqGHris5aaaaaa@4C3C@

In this case, G is the number of up- or down-regulated genes, A is the subset of G that are annotated in that GO term, k is the number of genes up- or down-regulated in clusters, and n is the subset of k annotated in the GO term.

We performed a whole-genome clustering of the genes from D. melanogaster annotated with the hierarchies based on the term structural constituents of cuticle (GO:0042302). No data about gene expression in the microarrays were used here. We first extracted the fly genes annotated with the following GO terms: structural constituent of cuticle (GO:0042302); structural constituent of chitin-based cuticle (GO:0005214); structural constituent of chitin-based larval cuticle (GO:0008010); structural constituent of pupal chitin-based cuticle (GO:0008011); structural constituent of adult chitin-based cuticle (GO:0008012); and structural constituent of collagen and cuticulin-based cuticle (GO:0042329). Then, we used the program REEF to identify clustering organization events in this gene set (the same configuration of parameters set in the case of co-expressed genes). We also performed a whole-genome clustering of the genes from D. melanogaster annotated with the hierarchies based on the term chitin binding (GO:0008061). We first extracted the fly genes annotated with the following GO terms: chitin binding (GO:0008061); polysaccharide binding (GO:0030247); carbohydrate binding (GO:0030246); and pattern binding (GO:0001871). Then, we used the program REEF to identify clustering organization events in this gene set.