Localized JNK signaling regulates organ size during development

A fundamental question of biology is what determines organ size. Despite demonstrations that factors within organs determine their sizes, intrinsic size control mechanisms remain elusive. Here we show that Drosophila wing size is regulated by JNK signaling during development. JNK is active in a stripe along the center of developing wings, and modulating JNK signaling within this stripe changes organ size. This JNK stripe influences proliferation in a non-canonical, Jun-independent manner by inhibiting the Hippo pathway. Localized JNK activity is established by Hedgehog signaling, where Ci elevates dTRAF1 expression. As the dTRAF1 homolog, TRAF4, is amplified in numerous cancers, these findings provide a new mechanism for how the Hedgehog pathway could contribute to tumorigenesis, and, more importantly, provides a new strategy for cancer therapies. Finally, modulation of JNK signaling centers in developing antennae and legs changes their sizes, suggesting a more generalizable role for JNK signaling in developmental organ size control. DOI: http://dx.doi.org/10.7554/eLife.11491.001


Introduction
Within a species, organ size is remarkably reproducible. While extrinsic factors like hormones are required for growth, classic transplantation experiments indicate that intrinsic factors within organs determine size (Bryant and Simpson, 1984). For example, embryonic limb buds transplanted from a large species of salamander onto a small species grow to the size characteristic of the donor (Twitty and Schwind, 1931). Similar findings have been made in quail and chick limbs (Iten and Murphy, 1980;Wolpert, 1978), rat hearts and kidneys (Dittmer et al., 1974;Silber, 1976), and mouse thymuses (Metcalf, 1963). Consistently, developing Drosophila wings transplanted into adult abdomens grow to the proper size, indicating that the information determining size is located within the developing organ (García-Bellido, 1965). Indeed, the Drosophila wing is a classic model system for studying organ size, as its size is highly replicable (García-Bellido and Merriam, 1971;García-Bellido, 1965), and all adult precursor cells are located within the pouch region of the developing larval imaginal disc (García-Bellido et al., 1973) (Figure 1A, grey). Despite extensive work, the molecular mechanisms underlying intrinsic organ size control remain unclear (Vogel, 2013). While morphogens direct both patterning and growth of developing organs (Tabata and Takei, 2004), a link between patterning molecules and growth control pathways has not been established (Schwank et al., 2011).
Here we show that localized JNK activity in the developing wing is established by Hedgehog (Hh) signaling and controls wing size through a non-canonical, Jun-independent signaling mechanism that inhibits the Hippo pathway.

Results and discussion
JNK is active in the developing Drosophila wing pouch Two independently generated antibodies that recognize the phosphorylated, active form of JNK (pJNK) specifically label a stripe in the pouch of developing wildtype third instar wing discs ( Figure (Glise et al., 1995;MacDonald et al., 2013;Martín-Blanco et al., 1998;Pérez-Garijo et al., 2013;Weber et al., 2000;Xu and Rubin, 1993).
The stripe of localized pJNK staining appeared to be adjacent to the anterior-posterior (A/P) compartment boundary, a location known to play a key role in organizing wing growth, and a site of active Hedgehog (Hh) signaling (Basler and Struhl, 1994;Tabata and Kornberg, 1994;Zecca et al., 1995). Indeed, pJNK co-localizes with the Hh target gene patched (ptc) ( Figure 1G; eLife digest A key challenge in biology is to understand what determines size. As an animal grows, signals are produced that control the size of its organs. Many of the signaling pathways that regulate size during normal animal development also contribute to the formation of tumors. Therefore, it is important to find out exactly how the signaling molecules that regulate size are linked to those that regulate tumor growth.
A protein called JNK activates a signaling pathway that triggers tumor growth. JNK signaling also stimulates cells to multiply in tissues that need repair, but it is not known whether it also regulates the size of organs during animal development. Here, Willsey et al. investigate whether JNK is active in the developing wings of fruit flies, which are commonly used as models of animal development.
The experiments show that JNK is active in a stripe across the developing wing and is required for the wing to grow to its proper size. A master signal protein called Hedgehog is responsible for establishing this stripe of JNK activity. Unexpectedly, rather than acting through its usual signaling pathway, JNK activates the Hippo pathway in the wing to control organ size during development.
Willsey et al.'s findings highlight potential new targets for cancer therapies. A future challenge will be to find out whether small patches of JNK signaling are found in the developing organs of other animals, and whether they can help explain how size changes between species.
ptc-Gal4, UAS-RFP). Expression of the JNK phosphatase puc in these cells specifically abrogated pJNK staining ( Figure 1H; ptc-Gal4, UAS-puc), as did RNAi-mediated knockdown of bsk ( Figure 1I and Figure 1-figure supplement 1L; ptc-Gal4, UAS-bsk RNA#i1or2 ). Together, these data indicate that the detected pJNK signal reflects endogenous JNK signaling activity in the ptc domain, a region of great importance to growth control. Indeed, while at earlier developmental stages pJNK staining is detected in all wing pouch cells (Figure 1-figure supplement 1A), the presence of a localized stripe of pJNK correlates with the time when the majority of wing disc growth occurs (1000 cells/disc at mid-L3 stage to 50,000 cells/disc at 20 hr after pupation, (Garcia-Bellido, 2009), so we hypothesize that localized pJNK plays a role in regulating growth.

Localized JNK activity regulates global wing size
Inhibition of JNK signaling in the posterior compartment previously led to the conclusion that JNK does not play a role in wing development (McEwen and Peifer, 2005). The discovery of an anterior stripe of JNK activity spurred us to re-examine the issue. Since bsk null mutant animals are Figure 1. Localized JNK activity exists in the developing wing. (A) Schematic of wing precursor cells (grey) in the developing disc (A, anterior; P, posterior). (B-F) Antibody staining against active, phosphorylated JNK (pJNK, green; DAPI, blue) labels a stripe in wildtype (B-C) but not JNKK mutant (D-E, hep r75/Y ) third instar discs. Boxed region in (B) and (D) is magnified in (C) and (E), respectively. Weak pJNK signal is also detected along the dorsal/ ventral boundary. pJNK stripe staining is lost in JNKK mutant clones (F, hep r75 , clone is negatively marked in F'). (G-I) pJNK localizes to the same cells in which ptc is expressed (G, ptc>RFP, red) along the A/P boundary, and is lost following JNK phosphatase expression (H, ptc>puc, RFP, red) or RNAi-mediated knockdown of bsk within the ptc domain (I, ptc>bsk RNAi , RFP, red). Bar: 50 um (B-F, H-I) and 25 um (G). See also embryonic lethal, we thus conditionally inhibited JNK signaling in three independent ways in the developing wing disc. JNK inhibition was achieved by RNAi-mediated knockdown of bsk (bsk RNAi#1or2 ), by expression of JNK phosphatase (puc), or by expression of a dominant negative bsk (bsk DN ). These lines have been independently validated as JNK inhibitors (MacDonald et al., 2013;Martín-Blanco et al., 1998;Perez-Garijo et al., 2013;Weber et al., 2000). Inhibition of JNK in all wing blade cells (rotund-Gal4, rn-Gal4) or specifically in ptc-expressing cells (ptc-Gal4) resulted in smaller adult wings in all cases, up to 40% reduced in the strongest cases (  To test whether elevation of this signal can increase organ size, we expressed eiger (egr), a potent JNK activator (Igaki et al., 2002), within the ptc domain (ptc-Gal4, UAS-egr). Despite induction of cell death as previously reported (Igaki et al., 2002) and late larval lethality, we observed a dramatic increase in wing disc size without apparent duplications or changes in the shape of the disc ( Figures 2H-I and 2L; ptc-Gal4, UAS-egr). While changes in organ size could be due to changing developmental time, wing discs with elevated JNK signaling were already larger than controls assayed at the same time point ( In stark contrast to known developmental morphogens, we did not observe any obvious defects in wing venation pattern following JNK inhibition ( Figure 2A-B), suggesting that localized pJNK may control growth in a pattern formation-independent manner. Indeed, even a slight reduction in Dpp signaling results in dramatic wing vein patterning defects ( . Together, the molecular data are consistent with the phenotypic results indicating that pJNK and Dpp are separate programs in regulating growth. Consistent with our findings, during the revision of this manuscript, it has been suggested that Dpp does not play a primary role in later larval wing growth control (Akiyama and Gibson, 2015). Finally, we found that inhibition of JNK does not affect EGFR signaling (pERK) and that inhibition of EGFR does not affect the establishment of pJNK (Figure 2-figure supplement 3A-C and 4A).
A difference in size could be due to changes in cell size and/or number. We examined wings with reduced size due to JNK inhibition and did not detect changes in cell size or apoptosis ( Figure  . Importantly, this increased proliferation is not restricted to the ptc domain, consistent with previous reports that JNK can promote proliferation non-autonomously (Enomoto and Igaki, 2012;Pastor-Pareja et al., 2008;Ryoo et al., 2004;Sun and Irvine, 2011;Wu et al., 2010).

Non-canonical JNK signaling regulates size
To determine the mechanism by which pJNK controls organ size, we first considered canonical JNK signaling through its target Jun (Ip and Davis, 1998). Interestingly, RNAi-mediated knockdown of , which is consistent with previous analysis of jun mutant clones in the wing disc (Kockel et al., 1997). Furthermore, in agreement with this, a reporter of canonical JNK signaling downstream of jun (puc-lacZ [Ring and Martinez Arias, 1993]) is not expressed in the pJNK stripe . Non-canonical JNK signaling regulates wing size. RNAi-mediated knockdown of Jun within the ptc stripe does not change adult wing size (A-B, red, ptc>jun RNAi compared to blue, ptc>). RNAi-mediated knockdown of jub does change global wing size (C-D, red, ptc>jub RNAi compared to blue, ptc>). Expression of yki in all wing cells (E-F, red, rn>yki, bsk DN compared to blue, rn>) or within the ptc stripe (G-H, red, ptc>bsk DN , yki compared to blue, ptc>) rescues wing size following JNK inhibition. RNAi-mediated knockdown or overexpression of yki in ptc cells decreases or enlarges wing size, respectively (I-J, red, ptc>yki RNAi , blue, ptc>, and K-L, red, ptc>yki, blue, ptc>).  ( Figure 1-figure supplement 1F). Finally, knockdown of fos (kayak, kay) alone or with jun RNAi did not affect wing size (Figure 3-figure supplement 1G-H; rn-Gal4, UAS-kay RNiA#1or2 and rn-Gal4, UAS-jun RNAi#1 , UAS-kay RNiA#1or2 ). Together, these data indicate that canonical JNK signaling through jun does not function in the pJNK stripe to regulate wing size.
In search of such a non-canonical mechanism of JNK-mediated size control, we considered the Hippo pathway. JNK signaling regulates the Hippo pathway to induce autonomous and non-autonomous proliferation during tumorigenesis and regeneration via activation of the transcriptional regulator Yorkie (Yki) (Bakal et al., 2008;Enomoto and Igaki, 2012;Sun and Irvine, 2011). Recently it has been shown that JNK activates Yki via direct phosphorylation of Jub (Sun and Irvine, 2013). To test whether this link between JNK and Jub could account for the role of localized pJNK in organ size control during development, we performed RNAi-mediated knockdown of jub in the ptc stripe, and observed adults with smaller wings ( It is important to note that the yki expression line used is wild-type Yki, which is still affected by JNK signaling. For this reason, the epistasis experiment was also performed with activated Yki, which is independent of JNK signaling (UAS-yki S111A,S168A,S250A.V5 ; (Oh and Irvine, 2009). Expression of this activated Yki in the ptc stripe caused very large tumors and lethality (data not shown). Importantly, inhibiting JNK in this context did not affect the formation of these tumors or the lethality (data not shown; ptc-Gal4, UAS-yki S111A,S168A,S250A.V5 , UAS-bsk DN ). Furthermore, inhibiting both JNK and Yki together does not enhance the phenotype of Yki inhibition alone ( Figure 3M-N  Mutants of the Yki downstream target four-jointed (fj) have small wings with normal patterning, and fj is known to propagate Hippo signaling and affect proliferation non-autonomously (Ambegaonkar et al., 2012;Harvey and Tapon, 2007;Strutt et al., 2004;Villano and Katz, 1995;Willecke et al., 2008). Although RNAi-mediated knockdown of fj in ptc cells does not cause an obvious change in wing size, it is sufficient to block the Yki-induced effect on increasing wing size ( Figure 3O-P and Figure 3-figure supplement 2G-H; ptc-Gal4, UAS-yki, UAS-fj RNAi and ptc-Gal4, UAS-fj RNAi ). However, overexpression of fj also reduces wing size, which makes it not possible to test for a simple epistatic relationship (ptc-Gal4, UAS-fj; Figure 3-figure supplement 2I-J). Overall, these data are consistent with the notion that localized pJNK regulates wing size not by Jun-dependent canonical JNK signaling, but rather by Jun-independent non-canonical JNK signaling involving the Hippo pathway.

Hh sets up pJNK by elevating dTRAF1 expression
While morphogens direct both patterning and growth of developing organs (Tabata and Takei, 2004), a link between patterning molecules and growth control pathways has not been established (Schwank et al., 2011). pJNK staining is coincident with ptc expression (Figure 1G), suggesting it could be established by Hh signaling. During development, posterior Hh protein travels across the A/P boundary, leading to activation of the transcription factor Cubitus interruptus (Ci) in the stripe of anterior cells (Domínguez et al., 1996;Schwartz et al., 1995). To test whether localized activation of JNK is a consequence of Hh signaling through Ci, we performed RNAi-mediated knockdown of ci and found that the pJNK stripe is eliminated ( Figure 4A-B; ptc-Gal4, UAS-ci RNAi#1or2 ). Consistently, adult wing size is globally reduced (Figures 4D and 4G). In contrast, we do not observe a change in pJNK stripe staining following RNAi-mediated knockdown of dpp or EGFR (Figure 2-  figure supplement 4A-B). Expression of non-processable Ci leads to increased Hh signaling (Price and Kalderon, 1999). Expression of this active Ci in ptc cells leads to an increase in pJNK signal and larger, well-patterned adult wings ( Figures 4C,E, and 4G; ptc-Gal4, UAS-Ci ACT ). The modest size increase shown for ptc>Ci ACT is likely due to the fact that higher expression of this transgene (at 25˚C) leads to such large wings that pupae cannot emerge from their cases. For measuring wing size, this experiment was performed at a lower temperature (20˚C, see Experimental Genotypes) so that the animals were still viable. Furthermore, inhibition of JNK in wings expressing active Ci blocks Ci's effects, and resulting wings are similar in size to JNK inhibition alone ( Figure 4F-G; ptc-Gal4, UAS-Ci ACT , UAS-bsk DN ). Together, these data indicate that Hh signaling through Ci is responsible for establishing the pJNK stripe.
To determine the mechanism by which Ci activates the JNK pathway, we compared transcriptional profiles of posterior (red, hh+) and ptc domain cells (green, ptc+) isolated by FACS from third instar wing discs ( Figure 5A; Materials and methods). Of the total 12,676 unique genes represented on the microarray, 50.4% (6,397) are expressed in ptc domain cells, posterior cells, or both (log 2 normalized expression 6.5; Figure 5-figure supplement 1A-D; Supplementary file 1; Materials and methods). We thresholded on a false discovery rate <0.01 and fold change 1.5 and found that 5.7% (363) of expressed genes were upregulated in ptc cells and 3.8% (242) were downregulated (Figure 5-figure supplement 1D; Supplementary file 2; Materials and methods). Hh pathway genes known to be differentially expressed are identified ( Figure 5B). We next asked whether any JNK pathway genes are differentially expressed and found that dTRAF1 expression is more than five-fold increased in ptc cells ( Figure 5C), while other JNK pathway members are not differentially expressed ( Figure 5C; Supplementary file 1; Supplementary file 2).
dTRAF1 is expressed along the A/P boundary (Preiss et al., 2001) and ectopic expression of dTRAF1 activates JNK signaling (Cha et al., 2003). Thus, positive regulation of dTRAF1 expression by Ci could establish a stripe of pJNK that regulates wing size. Indeed, we identified Ci binding motifs in the dTRAF1 gene ( Figure 5-figure supplement 1H), and a previous large-scale ChIP study confirms a Ci binding site within the dTRAF1 gene (Chr2L: 4367100-4371393; [Biehs et al., 2010]). Consistently, a reduction in Ci led to a 29% reduction in dTRAF1 expression in wing discs ( Figure 5J; ptc-Gal4, UAS-Ci RNAi ). Given that the reduction of dTRAF1 expression in the ptc stripe is buffered by Hh-independent dTRAF1 expression elsewhere in the disc (Preiss et al., 2001), this 29% reduction is significant. Furthermore, inhibition of dTRAF1 by RNAi knockdown abolished pJNK staining ( Figure  Conversely, overexpression of dTRAF1 causes embryonic lethality (ptc-Gal4, UAS-dTRAF1), making it not possible to attempt to rescue a dTRAF1 overexpression wing phenotype by knockdown of bsk. Nevertheless, it has been shown that dTRAF1 function in the eye is Bsk-dependent (Cha et al., 2003). Finally, inhibition of dTRAF1 modulates the phenotype of activated Ci signaling (ptc-Gal4, UAS-dTRAF1 RNAi , UAS-Ci ACT ; Figure 5-figure supplement 2). Together, these data reveal that the pJNK stripe in the developing wing is established by Hh signaling through Ci-mediated induction of dTRAF1 expression.

Localized pJNK controls antenna and leg size
Finally, we detected localized centers of pJNK activity during the development of other imaginal discs including the eye/antenna and leg ( Figures 6A and 6G). Inhibition of localized JNK signaling during development caused a decrease in adult antenna size ( Figures 6B-C and 6F; dll-Gal4, UASbsk DN ) and leg size ( Figures 6H-I and 6L; dll-Gal4, UAS-bsk DN ). Conversely, increasing JNK signaling during development resulted in pupal lethality; nevertheless, overall sizes of antenna and leg discs were increased (Figures 6D-E and 6J-K; dll-Gal4, UAS-egr). Together, these data indicate that localized JNK signaling regulates size in other organs in addition to the wing, suggesting a more universal effect of JNK on size control.
Intrinsic mechanisms of organ size control have long been proposed and sought after (Bryant and Simpson, 1984;Vogel, 2013). Our study reveals that in developing Drosophila tissues, localized, organ-specific centers of JNK signaling contribute to organ size in an activity level-dependent manner. Such a size control mechanism is qualitatively distinct from developmental morphogen mechanisms, which affect both patterning and growth (Zecca et al., 1995). Aptly, this mechanism is still integrated in the overall framework of developmental regulation, as it is established in the wing by the Hh pathway ( Figure 6M). Our data indicate that localized JNK signaling is activated by Cimediated induction of dTRAF1 expression. Furthermore, we discovered that it is not canonical Jundependent JNK signaling, but rather non-canonical JNK signaling that regulates size, possibly through Jub-dependent regulation of Yki signaling, as described for regeneration (Sun & Irvine, 2013) ( Figure 6M). As the human dTRAF1 homolog, TRAF4, and Hippo components are amplified in numerous cancers (Camilleri-Broët et al., 2006;Harvey et al., 2013), these findings provide a new mechanism for how the Hh pathway could contribute to tumorigenesis. More importantly, these findings offer a new strategy for potential cancer therapies, as reactivating Jun in Hh-driven tumors could lead tumor cells towards an apoptotic fate. Figure 6. Modulation of localized JNK signaling within the developing antenna or leg changes organ size. pJNK (green) staining of wildtype antenna/ eye (A) and leg (G) third instar discs. Inhibition of JNK in the developing antenna (B-C, F, dll>bsk DN ) or leg (H-I, L, dll>bsk DN ) leads to a smaller adult organ. Increased JNK activation within the antenna (D-E, dll>egr, RFP, red) or leg disc (J-K, dll>egr, RFP, red) causes an increase in disc size. (M) Model of how localized JNK signaling regulates wing size during development. Engrailed (En) controls Hh signaling, leading to a stripe of active Ci along the A/P boundary. Ci increases transcription of dTRAF1, activating JNK (pJNK, green). JNK acts in a non-canonical, Jun-independent manner to regulate Yki or Yki-dependent signaling. As the human dTRAF1 homolog, TRAF4, and Hippo components are amplified in numerous cancers, these findings provide a new mechanism for how the Hh pathway could contribute to tumorigenesis (Camilleri-Broët et al., 2006;Harvey et al., 2013). For box plots, whiskers represent maximum and minimum values (F, L). ****=p<0.0001. Bar: 100 um DOI: 10.7554/eLife.11491.017

Imaginal disc imaging
Imaginal discs to be imaged by confocal microscopy were mounted in Vectashield mounting media with DAPI (Vector Labs). Confocal images were taken with a Zeiss LSM510 Meta confocal microscope or a Leica TCS SP8 STEAD 3X confocal microscope with 405nm, 488nm, 561nm, and 633nm lasers. Both microscopes gave similar results. Measurements of disc size were performed from images of at least fifteen discs using NIH Image-J software.

Western blot analysis
Whole Canton-S and hep r75/Y larvae were lysed in standard RIPA buffer with protease and phosphatase inhibitors. Proteins were separated by SDS-PAGE using a 4-15% acrylamide gel (BioRad), transferred for 1 hr at 4˚C, and probed with primary antibodies: rabbit anti-pJNK (Calbiochem, 1:1000) and mouse anti-alpha tubulin (Sigma, 1:4000). HRP-conjugated secondary antibodies (anti-rabbit and anti-mouse) were used at 1:5000. ECL (Pierce) was used for detection with film.

Adult organ imaging
Adult wings, legs, or antenna were dissected in 70% ethanol, mounted in Permount mounting media (Fisher Scientific), and imaged with a Leica DFC300FX camera on a Leica MZ FLIII stereomicroscope. Measurements of wing size were performed from images of twenty to sixty female flies using NIH Image-J software. Wing images were false-colored and overlayed to scale using Adobe Photoshop CS3 software. Cell size was measured by dividing the number of hairs (1 hair/cell) by a set area using Adobe Photoshop CS3 software. Mean EdU signal was measured in Adobe Photoshop CS3. Measurements of antenna or leg size were performed from images of at least twenty male flies for each genotype using NIH Image-J software.

Statistical analysis
To determine whether differences in area were statistically significant, two-sided student's t-tests were performed using raw data values, matched for temperature and sex. Box plots were generated where whiskers represent maximum and minimum, a plus sign indicates the mean, a horizontal line within the box indicates the median, and the box represents the 25-75% quartile range. Both parametric and non-parametric analyses were performed, and p-values less than 0.05 were considered significant. Data are presented as relative to the mean of the matched Gal4-alone control.

Gene expression profiling
For each of three biological replicates, 200 pairs of wing imaginal discs were dissected from third instar larvae of the genotypes hh-Gal4; UAS-mCD8GFP or ptc-Gal4; UAS-mCD8GFP. Discs were stored in Schneider's Drosophila Media (21720, Invitrogen) plus 10% FBS (10438, Invitrogen) on ice for less than two hours prior to cell dissociation. Discs were washed twice with 1 ml cell dissociation buffer (Sigma, C-1544). Elastase (Sigma, E-0258) was diluted to 0.4 mg/ml in fresh cell dissociation buffer once discs were ready. Discs were incubated for 20 min at room temperature in 0.4 mg/ml elastase with stirring by a magnetic micro stirring bar. Undissociated tissue was spun out, cell viability was measured using the Beckman Vi-CELL Cell Viability Analyzer (>80%), and cells were immediately isolated using the BD FACSAria II system within the Stanford FACS facility. Dead cells labeled with propidium iodide (P3566, Invitrogen) were excluded during FACS, and purity of sorted cells was greater than 99% by post-sorting FACS analysis. Total RNA was extracted from sorted cells (RNeasy, Qiagen), quality was assessed with the Agilent Bioanalyzer 2100 (RIN > 7.0), and microarray analysis was performed in the Stanford Protein and Nucleic Acid Facility (Affymetrix D. mel Gene-Chip Genome 2.0 microarrays).

Identification of differentially expressed genes
All analyses were conducted in R version 3.1.1 (2014-07-10). Expression values were determined using the affy package (Gautier et al., 2004), available from BioConductor (http://bioconductor. org). The automatically downloaded Drosophila 2.0 CDF environment was utilized. Probe level data from the CEL files were imported using the function ReadAffy and converted to expression values using the function rma with default settings. This method implements robust multi-array average (RMA) for background correction followed by quantile normalization. PM correction was not performed. Probe level expression values were combined into probe set expression measures using medianpolish, the standard summary method employed in RMA (Irizarry et al., 2003). Expression values are log 2 transformed.
Post-normalization microarray quality assessment was conducted using the arrayQualityMetrics package (Kauffmann et al., 2009), available from BioConductor. Default settings were used, with ptc domain (ptc+) versus posterior (hh+) as the covariate in intgroup. Biological replicates cluster together in a dendrogram of inter-array difference, estimated as the mean absolute difference between the data of the arrays ( Figure 5-figure supplement 1A), indicating that biological effects are stronger than any batch effects. Similarly, principle components analysis also separates biological replicates into two clusters ( Figure 5-figure supplement 1B). Outliers were not detected by either of these methods.
Probe sets were mapped to genes using the drosophila2.db annotation package (version 3.0.0), available from BioConductor. 14,481 of 18,952 (76.4%) probe sets map to gene isoforms-12,676 (87.5%) of which correspond to unique genes (some genes are mapped by 1 probe set). In order to minimize technical artifacts, probe sets mapping to the same gene were not combined.
Based on the distribution observed in the density plot of normalized probe set expression values, probe sets (genes) with median log 2 expression value 6.5 in at least one condition (ptc+ and/or hh +) were considered to be expressed ( Figure 5-figure supplement 1C). According to these criteria, 7,228 of 18,952 probe sets (38.1%) are expressed. This corresponds to 6,854 of 14,481 gene isoforms (47.3%), which corresponds to 6,397 of 12,676 unique genes (50.4%, Figure 5-figure supplement 1D, Supplementary file 1).
To identify probe sets (genes) differentially expressed between ptc+ and posterior (hh+) samples, we used the samr package, an R implementation of significance analysis of microarrays (Tusher et al., 2001). This package is available from CRAN (http://cran.r-project.org/). Only expressed probe sets mapping to genes (6,854) were considered in this analysis. Differentially expressed probe sets were identified with the function SAM, using a two class unpaired response type, the t-statistic as the test statistic, and a false discovery rate (FDR) threshold of 0.01. The maximum number of possible permutations (720) was used. To ensure these results are biologically meaningful, we further trimmed this list to probe sets with a minimum 1.5 fold change between ptc+ and hh+ cells. Based on these criteria, 624 of 6,854 probe sets (9.1%) are differentially expressed, with 376 (5.5%) upregulated in ptc+ samples and 248 (3.6%) downregulated in ptc+ samples (Figure 5-figure supplement 1D, Supplementary file 2). A gene was considered differentially expressed if any mapped probe set was differentially expressed. Therefore, of the 6,397 unique expressed genes, 604 (9.4%) are differentially expressed, 363 (5.7%) upregulated and 242 (3.8%) downregulated. One gene, Tie, was mapped by probe sets both up-and down-regulated. The quantile-quantile plot in Figure 5-figure supplement 1D was prepared using the samr.plot function.

Transcription factor binding site analysis
The Drosophila Ci positional weight matrix from the BioBase TRANSFAC database was queried against the Drosophila melanogaster genome with a p-value <0.0001 (chosen based on known Ci binding sites within ptc) using FIMO (MEME) and aligned back to the UCSC genome browser.

Experimental genotypes
Crosses were maintained at 25˚C unless otherwise indicated