Centrosome loss or amplification does not dramatically perturb global gene expression in Drosophila

Centrosomes are the major microtubule (MT) organising centres (MTOCs) in many cell types and they are widely believed to have an important role in organising many cell processes such as cell division, the establishment and maintenance of cell polarity and the positioning of organelles within the cell. Centrosomes are also thought to act as “coordination centres” for various cellular pathways, as many cell cycle regulators and checkpoint proteins are concentrated at centrosomes (Hsu and White, 1998; Löffler et al., 2007; Tritarelli et al., 2004). Not surprisingly, centrosome defects have been implicated in a wide range of human diseases, most notably cancer, but more recently also in brain development and in conditions associated with defects in DNA damage repair (Alkuraya et al., 2011; Bakircioglu et al., 2011; Bond and Woods, 2006; Doxsey et al., 2005; Hsu and White, 1998; Lingle et al., 2002; Löffler et al., 2007; Pihan et al., 2003; Takada et al., 2007; Tritarelli et al., 2004).

Given their general importance, the recent demonstration that flies can proceed through the majority of development without centrosomes or with amplified centrosomes in the majority of their cells was very surprising (Basto et al., 2008; Basto et al., 2006). Somatic cell divisions were slowed when centrosomes were missing or amplified (as cells took longer than normal to assemble a bipolar spindle) but almost all cells appeared to ultimately divide normally in a bipolar fashion, although there was a small, but significant, increase in chromosome segregation errors in these cells. In contrast, asymmetrically dividing larval neuroblasts (NBs) appeared to have considerable difficulty in dividing accurately when centrosomes were either absent or amplified. These large stem-cell-like progenitors normally divide asymmetrically to generate another self-renewing NB, and a smaller ganglion mother cell (GMC) that usually divides only a few more times before terminally differentiating into either a neuron or glial cell (Doe, 2008; Knoblich, 2008). In NBs that either lack centrosomes or have too many centrosomes this division is symmetric ∼10–15% of the time (Basto et al., 2008; Basto et al., 2006). These symmetric divisions appear to give rise to two NB-like daughters (Basto et al., 2008; Lee et al., 2006a; Wang et al., 2006).

Intriguingly, brain cells that either lack or have too many centrosomes are predisposed to form tumours in abdominal transplantation assays (Basto et al., 2008; Castellanos et al., 2008). It is widely believed that this tumourigenesis is driven by the failure in asymmetric NB divisions (Castellanos et al., 2008; Gonzalez, 2007), and, in support of this possibility, mutations in genes encoding the asymmetrically distributed neural cell fate determinants Numb, Prospero and Brat also lead to tumourigenesis (Bello et al., 2006; Betschinger et al., 2006; Caussinus and Gonzalez, 2005; Lee et al., 2006a; Lee et al., 2006b; Wang et al., 2006). While a failure in asymmetric division can clearly lead to an expansion of the NB pool and so to over-proliferation, it is unclear how centrosome malfunction promotes malignant transformation. One possibility is that the relatively mild chromosome segregation defects induced by centrosome loss or amplification could facilitate the accumulation of mutations (Basto et al., 2008; Basto et al., 2006); however, mutations causing large-scale genome instability do not lead to brain tumour induction (Castellanos et al., 2008). In addition, it has been suggested that transcriptional and/or epigenetic changes may be required to promote malignant transformation in these brains (Knoblich, 2010). Interestingly, centrosome defects are known to increase cellular stress in vitro (Mikule et al., 2007; Srsen et al., 2006; Uetake et al., 2007), and stress can promote tumourigenesis in at least some Drosophila tumour models (Rossi and Gonzalez, 2011; Wu et al., 2010).

Thus, although flies with centrosome defects appear capable of progressing through most of development relatively normally, it remains possible that cell physiology could be significantly perturbed by these defects (Gonzalez, 2008). Here, we have attempted to address how centrosome loss or amplification might perturb cell physiology in vivo. We reasoned that any stress induced by these centrosome defects would likely induce physiological changes, at least some of which should be detectable as transcriptional changes. An analysis of global gene expression would, therefore, be a reasonable first readout of any possible changes in cell physiology. Indeed, many other Drosophila tumour models such as l(3)mbt, brat, aurA, lgl and aPKC^CAAX lead to large-scale changes in gene expression in situ (Carney et al., 2012; Janic et al., 2010).

Here we have compared the global transcriptome of brains and imaginal discs from animals that either lack centrosomes (due to mutations in the core centriole duplication proteins DSas-4 or DSas-6) (Basto et al., 2006; Peel et al., 2007; Rodrigues-Martins et al., 2007) or that have amplified centrosomes in the majority of their cells (due to the overexpression of Sak/Plk4) (Basto et al., 2008). Surprisingly, we found that these centrosome defects lead to very few changes to the global transcriptome, strongly suggesting that centrosome loss or amplification, per se, does not dramatically perturb Drosophila cell physiology in vivo.

To compare the transcriptome of normal cells to cells that either lack centrosomes or have too many centrosomes we used DSas-4 or DSas-6 mutant lines that lack centrosomes in >95% of their cells (Fig. 1A,B) (Basto et al., 2006; Peel et al., 2007; Rodrigues-Martins et al., 2007), or a Sak/Plk4 overexpressing line (hereafter SakOE) that has extra centrosomes in ∼70% of its cells (Basto et al., 2008) (Fig. 1A,C). In addition, we also compared the size of the brain lobes in these lines. As expected, brain size was slightly, but significantly, enlarged when compared to wild type in all three lines (Fig. 1D – note that these measurements were performed blind and compared tissues from experimental and control larvae grown in the same vials – see Materials and Methods), confirming that the centrosome defects were likely promoting some level of over-proliferation, presumably due to the previously described defects in asymmetric NB divisions (Basto et al., 2008; Basto et al., 2006). In support of this possibility, we could detect no significant difference in the size of the wing discs, which grow through mostly symmetric cell divisions, in any of these lines compared to wild type (Fig. 1E).

We decided to analyse the global transcriptome from larval brain and imaginal disc tissues for three reasons: (1) It is easy to isolate these tissues in a highly reproducible manner from tightly staged wandering 3^rd instar larvae; (2) Brains and imaginal discs are morphologically complex tissues that are polarised and are mitotically active, so might be expected to be particularly sensitive to centrosome defects; (3) As described above, DSas-4 mutant and SakOE brains are capable of forming tumours in abdominal transplantation assays (Basto et al., 2008; Castellanos et al., 2008) demonstrating that centrosome defects can, at least eventually, lead to perturbations in the normal physiology of these cells.

In an attempt to minimise variations that might simply be due to inherent differences between different inbred Drosophila strains, we backcrossed the DSas-4 and DSas-6 mutant lines and the SakOE line to a w⁶⁷ line (a wild type line carrying a point mutation in the white gene) that we used here as a wild type (WT) control. Our w⁶⁷ control stock was isogenised as well, so that all flies in this stock contained nearly identical sets of chromosomes. As additional controls, we also backcrossed a different WT strain (Oregon-R – [Or-R]) and a WT strain containing a TM6 balancer chromosome to w⁶⁷. All backcrossing was performed for at least 5 generations, and in each generation we selected for the Pw+ marker associated with the mutation or the transgene (or the endogenous w+ allele in the case of the WT Or-R stock, or the Tubby marker in the case of the TM6 balancer) and then backcrossed these flies to the original w⁶⁷ line (supplementary material Fig. S1). This allowed us to compare tissues from different genetic backgrounds that would be largely isogenic with respect to w⁶⁷ except for the region of the genome containing the P-element insertion that causes centrosome loss (in the case of the DSas-4 or DSas-6 mutation) or centrosome amplification (in the case of SakOE) (supplementary material Fig. S2).

For each genotype to be tested, brains and imaginal discs were dissected from 3 independent batches of 10 wandering 3^rd instar larvae and total RNA was prepared from each batch. We performed genome-wide expression profiling using Drosophila Genome 2.0 GeneChips (Affymetrix, Santa Clara, California, USA) and obtained lists of differentially expressed genes between WT and test strains that displayed at least 1.5-fold differential expression and adjusted P-values corrected for multiple hypothesis testing of ≤ 0.05. Array data-sets are deposited at Gene Expression Omnibus (accession no. GSE35240). We first compared w⁶⁷ to our Or-R WT control. The w⁶⁷ mutation was originally isolated by Lefevre in a WT Canton-S background (Lefevre and Wilkins, 1966) which was collected in Canton, Ohio by Calvin Bridges in the 1920s. These two strains have therefore been kept apart in various laboratories around the world for many decades. Encouragingly, we found no significant differences in the expression levels of any genes between these strains, suggesting that our isogenisation protocol had been largely effective in eliminating any variation due to inherent strain differences.

When we compared the transcriptome of w⁶⁷ to that of the DSas-6 mutant strain only two genes had statistically significant differences in their expression levels: DSas-6, which was down-regulated by ∼16× (FDR-adjusted P-value 0.005) and CG8245, which was up-regulated by ∼10× (FDR-adjusted P-value  =  0.004). DSas-6 was also one of only two genes identified when the DSas-6 mutant strain was compared to the WT Or-R strain: it was down-regulated by ∼16× (FDR-adjusted P-value  =  0.001), while CG32795 was up-regulated by ∼2× (FDR-adjusted P-value  =  0.05) (Fig. 2A). We performed quantitative RT-PCR (qRT-PCR) to test the reliability of these results. This analysis confirmed that DSas-6 was down-regulated compared to both WT strains (∼6 fold) (Fig. 2B). Interestingly, and in agreement with the chip data, CG8245 was significantly up-regulated (∼40×) in DSas-6 mutants compared to w⁶⁷ (Fig. 2C), but not compared to Or-R, while CG32794 was slightly up-regulated (∼1.8×) in DSas-6 mutants compared to Or-R, but not compared to w⁶⁷ (Fig. 2D). These results demonstrate that CG8245 and CG32795 are expressed at different levels in these strains, but this difference is unlikely to be caused by the lack of centrosomes in the DSas-6 mutant tissue as a similar difference is seen when the two WT strains are compared to each other. The different expression levels of CG8245 and CG32795 were also detected in the microarray comparison between w67 and OrR, although they were not scored as significant (data not shown).

These qRT-PCR experiments indicate that our chip analysis likely provides a robust and reliable comparison of the transcriptomes from these different strains. Remarkably, therefore, we conclude that there are few, if any, significant changes to the transcriptome of these DSas-6 mutant tissues when compared to WT.

We next compared the transcriptome of the DSas-4 mutant strain to w⁶⁷. We identified 19 genes that were up-regulated and 8 genes that were down-regulated in the DSas-4 mutant tissue (supplementary material Table S1). Again, we also compared DSas-4 mutant tissue to the other WT control, Or-R, and identified 15 genes that were up-regulated and 7 that were down-regulated (supplementary material Table S2). We assessed the overlap of these two lists and identified 14 genes that were differentially expressed when DSas-4 was compared to both WT controls (Table 1; Fig. 3A). Although these 14 genes are strong candidates for genes that are differentially expressed in DSas-4 mutant cells, they did not obviously cluster into any large functionally related group (Table 1). Moreover, it seems unlikely that these differences are directly caused by the lack of centrosomes in the DSas-4 mutant cells, as none of these genes were identified as being differentially regulated in DSas-6 mutant cells that also lack centrosomes. In addition, we noticed that several of the genes that are most differentially regulated in DSas-4 mutant cells were located very close to one other in the genome. For example, the two most up-regulated genes, Hsc70-2 and CG31157, are located directly adjacent to each other, while two of the most statistically significantly down-regulated genes (after DSas-4 itself and wallenda), Transcription factor TFIIFbeta and Mediator complex subunit 7 (MED7), are separated by only one gene, perhaps indicating that chromatin architecture may be altered in these regions in the DSas-4 mutant cells.

To confirm whether these genes were differentially expressed in DSas-4 mutant tissues we performed qRT-PCR experiments with the top 4 up-regulated genes (Hsc70-2, CG31157, CG31287 and CG9279) and the two most down-regulated genes (DSas-4 itself and wallenda). We designed qRT-PCR primers that would amplify across the region of each gene that was probed in our microarray experiments, as well as primers that would detect expression from a different region of each gene. As noted above, the two most up-regulated genes lie right next to each other in the genome and are transcribed in the same direction. To our surprise, we found that the 5′ end of HSC70-2 was not overexpressed, but the 3′ end (the region probed in our chip analysis) was strongly overexpressed by ∼185×, while both the 5′ and 3′ ends of CG31157 were overexpressed by ∼1400× and ∼130×, respectively in DSas-4 mutant tissue (Fig. 3B). We also analysed the expression levels of the small intergenic region between these genes and found that this was also overexpressed by ∼260× in DSas-4 mutant tissue. We conclude that an unusual transcript that starts in the second half of Hsc70-2 and reads through the CG31157 gene is highly overexpressed in DSas-4 mutants (Fig. 3B). We observed a similar phenomenon for CG9279 (Fig. 3C) and wallenda (Fig. 3D), as our qRT-PCR analysis revealed that the expression of the regions of the genes probed in our chip analysis were indeed mis-regulated, but the expression levels of a different region of each transcription unit were not. In contrast, the transcript levels of CG31287 (Fig. 3E) were not detectably altered in qRT-PCR experiments, suggesting that this is a false positive in our chip analysis. Down-regulation of DSas-4 mRNA in the DSas4 mutant was also confirmed by qRT-PCR (Fig. 3F).

Taken together, these data strongly suggest that a small number of genomic regions are genuinely mis-expressed in DSas-4 mutant cells, but at least some of this mis-expression is due to the production of aberrant transcripts that are not normally found in WT cells. The significance of this is unclear (see Discussion), but, as these regions are not mis-expressed in DSas-6 mutant cells, it seems unlikely that these changes are a direct result of the lack of centrosomes in DSas-4 mutants.

We reasoned that our failure to identify any genes that are consistently mis-expressed in tissues lacking centrosomes might be due to a high rate of false negatives, perhaps because we were setting our statistical cut-off for significance at too high a level. To test if this was the case, we compared the overlap between the 100 most significantly differentially regulated genes in DSas-4 and DSas-6 mutant tissue, even though most of these genes had an adjusted P-value of more than 0.05. Only 4 genes were differentially regulated in both mutants, strongly suggesting that most of these genes are unlikely to be consistently mis-expressed in cells that lack centrosomes (Table 2). Of the 4 genes that did overlap, two seemed poor candidates for genes that might be differentially regulated in cells lacking centrosomes: CG13822 is up-regulated in DSas-4 mutants but down-regulated in DSas-6 mutants, while CG32541 is only down-regulated by −1.22 and −1.4 fold in DSas-4 and DSas-6 mutants, respectively. CG11357 and pathetic were both up-regulated in DSas-6 and DSas-4 mutant tissues (Table 2), but a qRT-PCR analysis again suggested that only the expression of the 3′ region of each gene (the region probed in our chip analysis) was affected, while the 5′ region was not. This analysis strongly supports our conclusion that a lack of centrosomes in these cells leads to the significant mis-expression of very few, if any, genes.

To assess the effect of having too many centrosomes in cells we compared the transcriptome of SakOE cells to cells from the WT w⁶⁷ and Or-R strains. We found 55 and 57 genes, respectively, to be up- or down-regulated in the SakOE tissues (supplementary material Tables S3, S4), and 32 of these genes were present in both comparisons (Table 3; Fig. 4A). No large functionally related group of genes appeared to be particularly enriched in this geneset (supplementary material Tables S5, S6).

To test whether some of these genes were really differentially expressed in SakOE cells we again performed qRT-PCR experiments with several of the most up- (CG32055, B52, CG14687, CG9279 and Gram-negative bacteria binding protein 2 (GNBP2)) or down-regulated (Brf, CG11999 and CG7900) genes. Five of these eight genes were confirmed by the qRT-PCR analysis, which revealed that both their 5′ and 3′ regions were mis-expressed in the SakOE cells compared to the WT strains (Fig. 4B–F). The 3′-region of CG9279 was up-regulated (in agreement with the microarray data) but the 5′ region was not (Fig. 4G) (and if anything appeared to be down-regulated) suggesting that an aberrant transcript may be produced from this genomic region in the SakOE cells, as we observed in several genomic regions in the DSas-4 mutant cells. The genes CG14687 and B52 appeared to be false positives in our microarray analysis as either no, or only a very small, change in expression levels was detected by qRT-PCR (Fig. 4H,I). These results indicate that at least some of these 32 genes are genuinely mis-expressed in SakOE cells, although, as these genes do not cluster into any obvious functional pathway, it is unclear whether this mis-expression is a direct consequence of the centrosome amplification in these cells (see Discussion). Nevertheless, it is clear from this analysis that centrosome amplification does not lead to a large-scale perturbation of the transcriptome and that only a relatively small number of genes, if any, are consistently mis-regulated in cells as a result of centrosome amplification.

We show here that centrosome loss or amplification does not dramatically alter the global transcriptome of Drosophila brain or imaginal disc cells. While we have identified a small number of transcripts (∼10–15) whose expression appears to be genuinely mis-regulated in DSas-4 mutant cells, several of these appear to be aberrant transcripts that do not encode normal proteins, and few of these genes appear to be linked by any known common function. We currently do not understand why these transcripts are mis-expressed in DSas-4 cells, but it seems unlikely that these transcripts are components of specific pathways that are activated or inactivated in response to the lack of centrosomes, as none of these transcripts are mis-expressed in DSas-6 mutant cells, which also lack centrosomes. Similarly, we identified a slightly larger subset of genes (∼30) that appear to be mis-expressed in cells that overexpress Sak and so have too many centrosomes. Again, however, few of these genes appear to be linked by any known common function. Unfortunately, we have no independent way of efficiently driving centrosome amplification in brain and imaginal disc cells (independent of Sak overexpression), to test whether centrosome amplification is directly responsible for this transcriptional mis-regulation or whether it is due to some other change in the SakOE cells, as appears to be the case in DSas-4 cells. The human homologue of Sak, Plk4, has been reported to phosphorylate Hand1, a transcription factor that controls cell fate (Martindill et al., 2007), perhaps explaining why more genes are differently expressed in SakOE cells than in cells that lack centrosomes.

While we cannot rule out that centrosome defects lead to significant post-transcriptional changes without affecting the transcription of many genes, our findings strongly suggest that centrosome defects do not induce a major stress response in these cells. This finding is important, as centrosome defects appear to be a source of stress to cells in vitro (Mikule et al., 2007; Srsen et al., 2006; Uetake et al., 2007), where they have been shown to activate a stress response via the p38 and p53 pathways. It is not known what effect the activation of these pathways might have on gene expression in Drosophila, but genome wide expression studies in mammalian cells have shown that p38 stress-activated protein kinase activation leads to changes in expression of many genes, some of which are transcription factors (Ferreiro et al., 2010). Our GO enrichment analysis does not show enrichment of any known stress response genes in response to centrosome defects. Thus, although cellular stress has been linked to tumourigenesis in some Drosophila models (Rossi and Gonzalez, 2011; Wu et al., 2010), it appears unlikely to be an important driver of tumourigenesis in flies with centrosome defects.

If centrosome defects per se do not result in dramatic changes in cell physiology, then why are Drosophila brain cells without centrosomes or with amplified centrosomes predisposed to form tumours (Basto et al., 2008; Castellanos et al., 2008)? It is known that spindle assembly is slow in cells that have lost their centrosomes or have extra centrosomes, and this leads to a relatively modest increase in chromosome segregation errors (Basto et al., 2008; Basto et al., 2006; Ganem et al., 2009). Aneuploidy has long been thought to contribute to tumourigenesis (Boveri, 2008) and although mutations that lead to large-scale chromosomal instability do not seem to drive tumour formation in flies (Castellanos et al., 2008), it is possible that the low-level of aneuploidy induced by centrosome defects is actually a more effective driver of cancer (Weaver et al., 2007). Alternatively, although most fly cells with centrosome defects appear to divide relatively normally, it has previously been shown that such asymmetrically dividing brain NB divide symmetrically ∼10–15% of the time, as the astral MTs generated by centrosomes in these cells help the spindle to efficiently align with cortical cell fate determinants (Basto et al., 2008; Basto et al., 2006). This appears to lead to an expansion of the NB pool, consistent with the increase in brain size that we report here.

Our transcriptional profiling experiments, however, show that this brain overgrowth is likely to be benign as it leads to few, if any, changes in gene expression. The NBs that accumulate in these brains are therefore unlikely to be malignant – whereas the tumours formed from these brains in abdominal transplantations clearly are malignant, as they show immortality, aneuploidy and the ability to metastasise (Basto et al., 2008; Castellanos et al., 2008; Caussinus and Gonzalez, 2005), all of which almost certainly require large-scale changes in gene expression (Ramaswamy et al., 2003; Rhodes et al., 2004; Scott et al., 2011). We conclude that in brain tissues with centrosome defects, an additional step must lie between the generation of overgrowing, but physiologically normal brains, and the tumours that can eventually be formed from these brains in transplantation experiments. An attractive possibility is that the overproliferation of NBs leads to a bigger stem cell pool and larger brains, but that the low level of chromosomal missegregation facilitates the generation of mutations that occasionally lead to the generation of a malignant stem cell, which ultimately drives tumour formation.

This model has obvious similarities to the cancer stem cell (CSC) theory. A correlation between CSC and chromosomal instability has been suggested (Lagasse, 2008; Li et al., 2009) and recent studies have indicated that aneuploidy and CSCs models of cancer are not mutually exclusive (Conway et al., 2009; Liang et al., 2010). It will be interesting to test whether low level aneuploidy in cells without centrosomes or with too many centrosomes leads to the accumulation of mutations in NBs in vivo. Transcriptional profiling experiments on single cells could answer this question in the future.

The following mutant alleles were used in this study: DSas-4^s2214 (Basto et al., 2006) DSas-6^c02901 (Rodrigues-Martins et al., 2007). Overexpression of Sak in the SakOE strain is caused by the ubiquitin-promoter driven expression of GFP-Sak (Basto et al., 2008). Laboratory w⁶⁷ and Oregon-R strains were used as wild type controls.

For immunofluorescent staining, brains of third instar larvae were dissected in PBS and fixed in PBS + 4% Formaldehyde in PBS for 20 minutes. Brains were transferred to 45% acetic acid for 15 seconds and then to 60% acetic acid on a coverslip for 3 min. Then the brains were squashed between a slide and the coverslip and flash frozen in liquid nitrogen. Coverslips were removed, and the slides were placed in 100% methanol for 5 minutes. Samples were rehydrated in PBT for 1 h and then incubated with primary antibodies under a mounted coverslip in a moist chamber overnight at 4°C. On the next day the slides were washed in PBT and incubated with the secondary antibodies for three hours. After incubation the slides were washed in PBT, stained for 10 minutes in Hoechst33258 (Life Technologies, Carlsbad, California, USA) and then mounted in mounting medium (85% glycerol and 2.5% n-propylgallate). Slides were observed on a Zeiss Axioskop 2 microscope (Carl Zeiss, Ltd., Welwyn Garden City, Hertfordshire, UK) with a CoolSNAP HQ camera (Photometrics, Tucson, Arizona, USA), using a 63×/1.25 NA objective (Carl Zeiss, Ltd., Welwyn Garden City, Hertfordshire, UK) with Immersol oil (Carl Zeiss, Ltd., Welwyn Garden City, Hertfordshire, UK). Images were acquired using Metamorph software (Molecular Devices, Sunnyvale, California, USA), imported into Photoshop CS2 (Adobe, San Jose, California, USA), and adjusted to use the full range of pixel intensities. Neuroblasts and ganglion mother cells were scored in prophase in order to ensure that centrosomes were duplicated but extra centrosomes would not be clustered at the spindle poles. These cells were identified using DNA morphology and dots were scored as centrosomes only if they co-stained for Cnn and Asl.

For quantification of centrosomes in brain cells, the following primary antibodies were used at a 1:500 dilution: rabbit anti-Asl (Conduit et al., 2010) and guinea pig anti-Cnn (Lucas and Raff, 2007). Alexa488 anti-guinea pig and Alexa568 anti-rabbit secondary antibodies were used at a 1:1000 dilution (Molecular Probes, Life Technologies, Carlsbad, California, USA). For stainings of spindle and centrosomes, rabbit anti-Asl and mouse monoclonal anti-a-tubulin (DM1a, Sigma-Aldrich, St. Louis, Missouri, USA) were used as primary antibodies (1:500), and Alexa568 anti-rabbit and Alexa488 anti-mouse (1:1000) as secondary antibodies (Molecular Probes, Life Technologies, Carlsbad, California, USA).

DSas-4, DSas-6 and SakOE strains were recombined with a Hand-GFP transgene that is expressed in the larval cardiac tissue (Han et al., 2006). w⁶⁷ fertilised females were put together with an equal number of either sakOE, Hand>GFP, DSas-6,Hand>GFP/TM6C or DSas-4,Hand>GFP/TM6C fertilised females and these females were allowed to lay eggs for a maximum of 12 hours. A few days later, brains and wing discs of wandering third instar larvae were dissected and pictures were taken with a Nikon DS-Fi1 camera mounted on a Nikon SMZ800 dissecting microscope (Nikon, Kingston Upon Thames, Surrey, UK); the circumference of the brain lobes and the width and height of the wing discs were then measured in ImageJ (Abramoff et al., 2004). After making these measurements, the genotype of the tissue that had been measured was checked by examining the larval carcass for Hand-GFP expression. In this way we ensured that the analysis was performed blind, and that the mutant (or Sak-overexpressing) and WT larvae were grown under identical conditions in the same vial. Brain and wing disc sizes of at least 20 animals were measured per data-point and the average and standard error were calculated. A Student's t-test was performed to calculate P-values.

Transcriptional profiles of wild type flies, DSas-4 flies, DSas-6 flies and SakOE flies were generated. Total RNA of brains and imaginal discs from 10 third instar larvae per sample was isolated using TRIzol (Life Technologies, Carlsbad, California, USA) chloroform extraction and isopropanol precipitation. For each strain three independent samples were analysed. Purity and integrity of the purified RNA was assessed on the Agilent Bioanalyzer 2100 (Agilent, Santa Clara, California, USA). Concentration was determined with a Nanodrop ND-1000 Spectrophotometer (Nanodrop Technologies, Wilmington, Delaware, USA). 500 ng RNA per sample was processed to labelled cRNA using the Affymetrix 3′ IVT Express kit according to the manufacturer's instructions and hybridised to Affymetrix Drosophila 2.0 GeneChips for 16 hrs at 45°C. Gene chips were washed and stained with streptavidin-phycoerythrin using the Affymetrix Fluidics Station 450 and scanned on an Affymetrix GeneChip 3000 scanner (all Affymetrix, Santa Clara, California, USA). Quality control of microarray expression data was performed using the Bioconductor package AffyPLM (Gautier et al., 2004). Probe intensities from Affymetrix image files (“.CEL” files) were normalised using quantile normalisation (Bolstad et al., 2003), and expression signals of all genes (probesets) were calculated using GCRMA (Guanosine Cytidine robust multiarray analysis) (Wu and Irizarry, 2007). Differentially expressed genes between WT and Experimental samples were identified using the Bioconductor package limma (Smyth, 2005). We obtained lists of genes that displayed at least 1.5-fold change in expression and a Benjamini-Hochberg FDR-corrected P-value of ≤ 0.05 (Benjamini and Hochberg, 2005). We analysed the list of differentially expressed genes of SakOE compared to w⁶⁷ for enrichment of Gene Ontology (GO) terms using DAVID, a program that weighs the enrichment of a specific GO term in a given data-set relative to the frequency of that term on the Drosophila 2.0 chip (Dennis et al., 2003; Huang et al., 2008). We were looking for enrichment in the GO FAT set, which attempts to filter the broadest terms so that they do not overshadow the more specific terms. Raw data can be accessed in Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo; accession number GSE35240).

For qPCR analysis RNA was isolated as described for the microarray experiments. 1 µg RNA was treated with DNAseI (Invitrogen life technologies) and used for reverse transcription with the Transcriptor High Fidelity cDNA Synthesis Kit (Roche Diagnostics Ltd., Burgess Hill, West Sussex, UK) using oligo(dT) primer. cDNA was diluted 1:2 and 1 μl diluted cDNA was used for qPCR analysis using the SensiMix SYBR No-ROX Kit (Bioline Reagents Ltd, London, UK) in a MJ Research PTC-200 thermal cycler with a Chromo4 detector (both Biorad, Hercules, California, USA). Genespecific primers for qPCR analysis were designed using PerlPrimer (Marshall, 2004) (supplementary material Table S7). Expression of the tested genes was measured in triplicates and gene expression levels for each individual sample were normalised expression of EF1a. At least 2 independent samples were measured per gene. Mean relative gene expression was determined and expressed as 2^−ΔCT (ΔCT  =  (CT_gene − CT_EF1alpha) × 1000) (Livak and Schmittgen, 2001).

We thank members of the Raff Lab for critical reading of the manuscript and stimulating discussions. We are grateful to Christopher Dooley and Markus Grohme for comments on the manuscript and experimental advice. We would like to thank Andrew Renault for sharing the Hand-GFP fly stock. This work was supported by a EPA Trust studentship to J.B., a National Institutes of Health Fellowship [F32 GM 80924-02] to M.P.L. and Cancer Research UK Programme Grant [10530] to J.W.R.