Atf3 links loss of epithelial polarity to defects in cell differentiation and cytoarchitecture

Interplay between apicobasal cell polarity modules and the cytoskeleton is critical for differentiation and integrity of epithelia. However, this coordination is poorly understood at the level of gene regulation by transcription factors. Here, we establish the Drosophila activating transcription factor 3 (atf3) as a cell polarity response gene acting downstream of the membrane-associated Scribble polarity complex. Loss of the tumor suppressors Scribble or Dlg1 induces atf3 expression via aPKC but independent of Jun-N-terminal kinase (JNK) signaling. Strikingly, removal of Atf3 from Dlg1 deficient cells restores polarized cytoarchitecture, levels and distribution of endosomal trafficking machinery, and differentiation. Conversely, excess Atf3 alters microtubule network, vesicular trafficking and the partition of polarity proteins along the apicobasal axis. Genomic and genetic approaches implicate Atf3 as a regulator of cytoskeleton organization and function, and identify Lamin C as one of its bona fide target genes. By affecting structural features and cell morphology, Atf3 functions in a manner distinct from other transcription factors operating downstream of disrupted cell polarity.


Abstract
Interplay between apicobasal cell polarity modules and the cytoskeleton is critical for differentiation and integrity of epithelia. However, this coordination is poorly understood at the level of gene regulation by transcription factors. Here, we establish the Drosophila activating transcription factor 3

Author summary
Epithelial cells form sheets and line both the outside and inside of our body. Their proper development and function require the asymmetric distribution of cellular components from the top to the bottom, known as apicobasal polarization. As loss of polarity hallmarks a majority of cancers in humans understanding how epithelia respond to a collapse of the apicobasal axis is of great interest.
Here, we show that in the fruit fly Drosophila melanogaster, the breakdown of epithelial polarity engages Activating transcription factor 3 (Atf3), a protein that directly binds the DNA and regulates gene expression. We demonstrate that many of the pathological consequences of disturbed polarity

Introduction
Epithelia are sheets of highly polarized cells that represent the defining tissue type of metazoans. With their capacity to achieve various shapes and serve as selective barriers, epithelial tissues play vital roles in morphogenesis, tissue differentiation and compartmentalization, and intercellular signaling.
The integrity and function of epithelia rely on the interplay between key polarity determinants and a highly ordered yet dynamic cytoskeleton, which ensures tissue plasticity and the asymmetric distribution of cellular components along the apicobasal polarity axis [1].
Genetic studies have established a network of evolutionarily conserved signaling pathways and effector molecules that govern the organization of epithelial cellular architecture. In particular, the basic leucine zipper (bZIP) transcription factors, including Jun, Fos, and Atf3, are important regulators of epithelial function from the fruit fly Drosophila to mammals [2][3][4][5][6]. During Drosophila development, expression of atf3 is dynamic and under tight temporal constraints. Ectopic Atf3 activity in larval epidermal cells (LECs) disturbs epithelial morphogenesis of the adult abdomen, stemming from perturbed cytoskeleton dynamics and increased cell adhesion which prevent normal LEC extrusion [6].
The importance of finely tuned Atf3 expression during Drosophila development corresponds with the role of mammalian ATF3 as a stress-response gene regulated at the level of mRNA expression by various stimuli, including genotoxic radiation, wounding, cytokines, nutrient deprivation, Toll signaling, oncogenes and inhibition of calcineurin-NFAT signaling [7][8][9]. Independent transcriptome analyses of Drosophila epithelia have shown deregulated atf3 expression in wing imaginal discs lacking the conserved neoplastic tumor suppressor genes scribble (scrib) or discs large 1 (dlg1) encoding components of the Scribble polarity module [10], and in ras V12 scrib − tumors in the eye/antennal imaginal disc (EAD) [11][12][13]. These results point to loss of epithelial integrity as a novel trigger of atf3 expression and are congruent with studies linking Atf3 to processes involving transient controlled epithelial depolarization during morphogenesis and wound healing [14][15][16] as well as to pathological disturbances in polarity that characterize chronic wounds and tumorigenesis [5,9,[17][18][19][20][21]. However, which polarity cues induce atf3 expression and how Atf3 activity contributes to phenotypes associated with loss of polarity have yet to be determined.
In this study, we establish that loss of the Scrib polarity module is sufficient to increase the levels and activity of Atf3 via aPKC signaling. Increased Atf3 activity drives major phenotypic attributes of the Dlg1 deficiency as abnormal distribution of polarity proteins and differentiation defects in dlg1 mutant epithelial clones can be alleviated by removal of Atf3. Chromatin immunoprecipitation followed by high-throughput sequencing further revealed that Atf3 target genes are enriched for roles in cytoskeletal organization and dynamics. Thus, Atf3 links defects in the Scrib polarity module with gene dysregulation and subsequent perturbations in cellular morphology and differentiation.

atf3 is a cell-polarity response gene activated by aPKC but not JNK signaling
Previous transcriptome profiling by our group and others [11][12][13] has suggested that disturbed cell polarity leads to upregulation of atf3 expression. Consistently, qRT-PCR from scrib 1 homozygous mutant larvae and adult heads bearing dlg1 G0342 homozygous mutant clones showed increased levels of atf3 mRNA (Fig 1A), thus confirming induction of atf3 transcription upon depletion of the Scrib polarity module. To extend this evidence, we tested whether Scrib or Dlg1 deficiency impacts the levels of an Atf3::GFP fusion protein expressed from a recombineered BAC (atf3 gBAC ) that is sufficient to rescue the lethality of the atf3 76 null mutants [22]. In control EAD, Atf3 was enriched in differentiated photoreceptors of the eye primordium and a subset of peripodial cells of the antenna ( Fig 1B). In wing imaginal discs, Atf3::GFP protein labeled the squamous cells of the peripodial epithelium and columnar cells of the outer ring that encircles the wing pouch ( Fig 1C). Knockdown of dlg1 (en>dlg1 RNAi ) or scrib (en>scrib RNAi ) in the posterior compartment of the wing disc resulted in a marked increase of the Atf3::GFP signal in the columnar epithelium of the wing pouch (Fig 1D, 1E).
Importantly, Atf3 was also upregulated in dlg1 G0342 loss-of-function mutant clones induced in the larval EAD (Fig 1F, 1G and S1A- C Fig).
Previous studies have shown that disruption of the Scrib module leads to activation of Jun-N-terminal kinase (JNK) signaling and to deregulation of the atypical protein kinase (aPKC) [23][24][25]. To determine if either one of these two pathways is sufficient to induce Atf3 expression, we assessed both Atf3::GFP protein levels and the activity of a synthetic Atf3-responsive element (ATRE) reporter. The ATRE construct expresses GFP or RFP under the control of four concatenated genomic segments, each of which is 22 bp long and contains an Atf3 binding site identified through ChIP-seq in this study (S1  Table). While activation of JNK signaling by expressing a wild type form of Drosophila JNKK (Hemipterous; Hep) induced a JNKresponsive TRE-DsRed reporter [26] (Fig 1J and S2G Fig), it failed to upregulate Atf3 expression when targeted to the posterior compartment of the wing imaginal disc (en>hep wt ) (Fig 1J''). Similarly, the ATRE-GFP reporter was insensitive to or only minimally activated by JNK signaling in the wing imaginal disc (dpp>hep wt ) and S2 cells, respectively ( Fig 1N and S2E Fig). Moreover, JNK signaling appeared dispensable for Atf3 induction caused by loss of polarity, as elevated Atf3::GFP signal persisted in dlg1 deficient EAD clones expressing a dominant negative form of the Drosophila JNK Basket (dlg1 G0342 bsk DN ) (Fig 1H, 1I). In contrast, expression of a membrane tethered aPKC (aPKC CAAX ) that causes mild overgrowth was sufficient to induce both Atf3::GFP and the ATRE reporter in specified wing disc compartments (en>aPKC CAAX and dpp>aPKC CAAX ) (Fig 1K, 1O). Although inhibiting JNK signaling suppressed the aPKC CAAX -mediated overgrowth, it did not prevent Atf3 induction (en>aPKC CAAX bsk DN ) ( Fig 1L). Signaling via aPKC has been shown to engage Yorkie (Yki), a transcriptional co-activator in the Hippo pathway [10,[27][28][29]. However, expression of a constitutively active form of Yorkie (dpp>yki act ) did not induce the ATRE reporter ( Fig 1P). Together, these findings highlight a role for Atf3 as a polarity response gene that is activated by loss of key polarity determinants downstream of aPKC but not JNK or Yki.

Loss of atf3 suppresses the effects of dlg1 deficiency on eye development
Imaginal disc clones lacking either Scrib or Dlg1 surrounded by normal epithelium are severely disorganized, suffer from disturbed vesicular transport, lose the ability to terminally differentiate and are eliminated through cell competition involving JNK signaling [23,25,30]. While JNK and its downstream transcription factor Fos are required for apoptosis and suppression of Yki-mediated hyperproliferation of scrib and dlg1 mutant cells [24,25,[31][32][33][34], aPKC is responsible for the aberrant morphology and differentiation of the clonal cells [23].
Because Atf3 upregulation results from depletion of the Scrib complex components as well as from ectopic aPKC activation (Fig 1D, 1E, 1K), we investigated whether this excess Atf3 might contribute to the differentiation and morphological defects of EAD clones lacking Dlg1. While EADs bearing dlg1 deficient clones (S1B, S1C Fig (Fig 2D'). An equivalent genetic interaction was observed between atf3 and the dlg1 m52 allele [35,36] (S3D- G Fig).
Importantly, adding a single copy of the atf3 gBAC transgene to animals with the double mutant mosaic EADs (atf3 76 dlg1 G0342 ;; atf3 gBAC /+) was sufficient to reinstate the aberrations to the adult eye morphology (S3C Fig), thus clearly showing that Atf3 is required for this dlg1 deficiency phenotype to develop.
Abnormalities were further exacerbated in adult eyes bearing dlg1 deficient clones in which Atf3 was overexpressed (dlg1 G0342 atf3 wt ) ( Fig 2E). It is important to note that mosaic overexpression of Atf3 alone disturbed the normal ommatidial arrangement in the adult eye ( Fig 2F). However, immunostaining of third instar EADs against a pan-neuronal marker Elav showed that, unlike cells lacking Dlg1, Atf3expressing clones (atf3 wt ) differentiated (S2A- D Fig). In contrast, clonal loss of atf3 alone (atf3 76 ) did not markedly impact adult eye morphology ( Fig 2B). These data provide causal evidence for the role of Atf3 downstream of disturbed epithelial polarity. They also indicate that while Atf3 is required for phenotypes caused by loss of dlg1, the gain of Atf3 alone does not fully recapitulate these defects.

Loss of atf3 restores polarity and differentiation to dlg1 deficient cells independent of JNK activity
Closer examination of third instar larval EADs revealed the presence of fewer dlg1 G0342 clones relative to control and atf3 76 mosaic EADs (Fig 2G-I . To rule out that clone elimination contributed to the observed rescue phenotype in the atf3 76 dlg1 G0342 adult eyes, we took two alternative approaches. To inhibit death of clonal cells, we expressed the baculoviral caspase inhibitor p35 in dlg1 G0342 and atf3 76 dlg1 G0342 EAD clones. To reduce competition elicited by neighboring cells, we utilized the EGUF/hid technique [37], which facilitates expression of a pro-apoptotic protein Hid in the non-clonal cells within the Glass multiple reporter (GMR) domain causing their elimination during pupal stages [38]. As expected, blocking apoptosis by To better characterize how loss of atf3 impacts dlg1 mutant phenotypes, we stained third instar larval EADs for markers of differentiation and cell adhesion. Unlike control clones, dlg1 G0342 mutant cells located posterior to the morphogenetic furrow of the eye primordium frequently lacked the panneuronal marker Elav. Many of them flattened and delaminated, pushing some of the non-clonal Elavpositive cells to the basal side of the epithelium (Fig 3A, 3B). In contrast, atf3 76 dlg1 G0342 clones remain columnar, contributing to photoreceptor and interommatidial cell differentiation ( Fig 3C).
Importantly, immunostaining for the apical determinant Crumbs, the adherens junction protein DEcadherin (DE-cad) and the lateral membrane marker Fasciclin III (FasIII) revealed a clear rescue of the apicobasal organization of the ommatidia in atf3 76 dlg1 G0342 clones compared to the disturbed architecture in clones mutant for dlg1 G0342 alone (Fig 3A-F). These findings uncover novel genetic interaction between Atf3 and the central Scrib polarity module component Dlg1 and establish Atf3 as a driver of the structural and differentiation defects stemming from the loss of dlg1.

Disturbances to the endosomal trafficking machinery in dlg1 mutant epithelia require Atf3
Besides altered differentiation and mislocalization of polarity proteins, disturbed cellular trafficking [30,[39][40][41] is a hallmark feature of epithelial cells lacking the Scrib module components. Therefore, we tested if the distress of the trafficking machinery upon dlg1 loss was Atf3-dependent. Interestingly, neither dlg1 G0342 nor atf3 76  In addition, immunostaining of wing discs bearing Atf3-expressing clones revealed changes in levels and localization of polarity proteins whose proper membrane placement requires a functional trafficking machinery [42,43].
While the apical levels of Crumbs were markedly lower, the integrin subunit Myospheroid (Mys), which normally is restricted to the basal cell surface, was detected along the entire lateral membrane in Atf3-expressing clones relative to surrounding tissue (S9C, S9D Fig). These data demonstrate that Atf3 contributes to the alterations of the endosomal machinery upon loss of polarity and its overexpression is sufficient to mimic some of the hallmark features of dlg1-deficient epithelial cells.
Interestingly, despite the shift in apicobasal polarity markers Atf3-expressing cells, unlike those lacking dlg1, maintained their columnar shape and were not extruded from the epithelium (Fig 2R and   S8 Fig).

Atf3 binds a 12-nucleotide motif in genes fundamental to cytoarchitecture
As a bZIP transcription factor, Atf3 is expected to regulate gene expression through binding to specific DNA sequences. To capture a snapshot of genomic regions bound by Atf3, we employed the Atf3::GFP fusion protein expressed from the recombineered atf3 gBAC to perform chromatin immunoprecipitation followed by high throughput sequencing (ChIP-seq). The ChIP-seq using an anti-GFP antibody identified 152 genomic locations significantly enriched in samples prepared from Submission of this PWM to Tool for Motif to Motif comparison (TOMTOM) [45] determined that Atf3 recognizes DNA that is most similar to the Atf2 binding motif in Drosophila and ATF3/JDP2 motif in humans ( Fig 5B). Closer examination of the Atf3-response elements revealed an 8-nucleotide core that closely resembles the cyclic AMP response element (CRE) (TGACGTCA) (Fig 5B). We have previously reported that the bZIP domain of the Drosophila Atf3 protein indeed directly binds this DNA element [6]. Our present experiments using the ATRE and mATRE reporters further confirm that Atf3 can control expression from this site in cultured cells and in vivo ( Fig 1M' and S2 Fig).
The gene ontology (GO) analysis with FlyMine [46] revealed that structural and regulatory components of the cytoskeleton were significantly enriched among Atf3-bound regions. In total, 40 of the identified 121 genes containing Atf3 motifs belong to at least one out of five GO terms corresponding to cytoskeleton regulation ( Fig 5A, 5C and S1 Dataset). Figure 5D

Atf3 overexpression has a broad effect on gene transcription in imaginal disc epithelia
To complement the ChIP-seq approach and characterize the transcriptional response to excess Atf3, To identify additional putative Atf3 targets among genes misregulated in the RNA-seq dataset, we scanned the Drosophila genome with our experimentally derived Atf3 PWM using the Find Individual Motif Occurrences (FIMO) [47] and Peak Annotation and Visualization (PAVIS) [48] utilities. The Atf3 motif ( Fig 5B) was found within 5 kb upstream and 1 kb downstream of 1252 genes differentially regulated in atf3 wt mosaic EADs (Fig 5-source data 3).
Importantly, a significant proportion of transcripts altered by Atf3 expression in the mosaic EADs overlapped with genes that had been found deregulated in either scrib or dlg1 mutant wing imaginal discs [10] (S10 Fig and S3 Dataset). This intersection shows that the gain of Atf3 and the loss of the Scrib polarity module elicited a partly overlapping genetic response.

Atf3 controls the microtubule network and Lamin C downstream of loss of polarity
The requirement of Atf3 for the manifestation of cytoarchitecture and polarity defects in dlg1 deficient cells prompted us to further focus on its regulation of microtubules and LamC, the primary building blocks of the cytoskeleton and nucleoskeleton, respectively. Microtubules are classically associated with controlling cell shape and division, but they are also central to polarity by serving as railroad tracks for directed vesicular trafficking [49]. Nuclear lamins on the other hand provide structure and stiffness to the nuclear envelope [50,51]. The coupling between the cytoskeleton and the nuclear lamina via the LINC-complex is essential for maintaining the mechanical properties of the cell and signal transduction [52]. Immunostaining for LamC showed that the increase in LamC transcription in response to excess Atf3 ( (Fig 6A, 6D). Importantly, LamC protein levels also increased in the wing disc epithelium deficient for Scrib (en>scrib RNAi ) and in dlg1 mutant clones of the EAD (Fig 6B, 6F), whereas they were markedly reduced in atf3 76 dlg1 G0342 double mutant EAD clones compared to the surrounding tissue and control clones (Fig 6C, 6E, 6G). In contrast to LamC enrichment, expression of multiple microtubule genes (βTub60D, βTub85D and γTub37C) was downregulated in atf3 wt mosaic EADs (Fig 5G, Fig 5-source data 3). Interestingly, immunostaining for α-and β-Tubulin revealed a reduced microtubule network and markedly lowered levels of non-centrosomal γ-Tubulin in Atf3-overexpressing clones of the wing imaginal disc relative to surrounding epithelial tissue (Fig 7A-C). In addition, eye primordia bearing dlg1 mutant clones showed abnormal microtubule organization that was partially normalized by simultaneous clonal loss of atf3 (Fig 7D-F). Taken together, our molecular and genetic approaches identified microtubule encoding genes and LamC as new targets of Atf3. We demonstrate that the upregulation of LamC and disturbances to microtubule network are the common hallmark of epithelial cells in which Atf3 activity is enhanced either by transgenic overexpression or as a result of disturbed polarity. While LamC is a bona fide target of Atf3, changes to microtubule network likely arise indirectly as a part of the secondary response to ectopic Atf3 activity.

Discussion
Classical insults to epithelial polarization include disruptions of the Crumbs, aPKC/Par, or Scrib protein modules, which all alter tissue organization, proliferation, differentiation, and cell viability [1].
The engagement of transcription factors presents a significant but poorly understood consequence to compromised polarity [10,23,27,28,53,54]. In this study, we identify Atf3 as a novel polarity response gene induced by aPKC signaling upon loss of the neoplastic tumor suppressors of the Scrib polarity module. We show that aPKC upregulates Atf3 independently of Yki, as increased Atf3 levels were detected in dlg1 EAD mutant clones, where aPKC is active but Yki is functionally repressed by JNK signaling [24]. Moreover, activated Yki was not sufficient to induce Atf3 activity. Our results also exclude JNK activation stemming from loss of polarity as a driver of atf3 expression. Blocking JNK in dlg1 mutant clones or tissues with active aPKC signaling did not abrogate Atf3 expression. This study positions Atf3 as a major regulator in the aPKC circuit, as removing Atf3 from dlg1 deficient cells recapitulated the phenotypes conferred by reduced aPKC activity in dlg1 tissue, including restoration of a polarized, columnar morphology and differentiation in the eye primordium [23]. Importantly, Atf3 acts in a manner distinct from that of the bZIP protein Fos and the TEAD/TEF transcription factor Scalloped (Sd) and its co-activator Yki, operating downstream of disrupted polarity [23,27,28,32,54]. In scrib and dlg1 mutant cells, Fos and Sd/Yki promote apoptosis and proliferation, respectively; however they are not linked to disturbed cytoarchitecture [24,27,33].
Conversely, lack of Atf3 does not block JNK activity or the elimination of scrib/dlg1 mutant clones, whereas overexpression of dominant-negative forms of JNK or removal of Fos do [23,25,32]. As loss of atf3 did not affect dlg1 clonal abundance, we conclude that atf3 deficiency neither reduces the proliferative potential of dlg1 tissue nor interferes with JNK-mediated repression of Yki. Taken together, we propose that Atf3, Fos, and Sd/Yki act downstream of cell polarity insults in parallel ( Fig   8).
Impairment of the vesicular trafficking machinery has been recognized as a key feature of epithelial cells lacking polarity determinants such as Crumbs or the components of the Scrib complex [30,39,41,55]. However, the deregulation of endocytosis has been also shown to drive polarity defects and to alter signal transduction resulting in overgrowth or apoptosis depending on the tissue and genetic context [30,39,41,56]. Here, we show that while bulk endocytosis is not affected in dlg1 mutant clonal cells of the EAD, levels and distribution of specific components of the vesicle trafficking machinery and their cargo are disturbed and these defects depend on Atf3. However, despite the clear rescue of the Rab5 and Rab11 trafficking apparatus in atf3 dlg1 double mutant clones JNK signaling remains upregulated and cells are eliminated, suggesting that multiple mechanisms are in play to promote JNK activity upon loss of dlg1. Importantly, the disruption of Rab5 and Rab11-mediated trafficking routes can be recapitulated by Atf3 overexpression. The reduced levels of Crumbs, whose membrane delivery requires vesicular transport from the trans Golgi-network [30], and the ectopic expansion of Mys (βPS-integrin) along the lateral membrane, whose turnover relies on Rab5 and Rab21 vesicles [57] further highlight the impact of enhanced Atf3 activity on the trafficking and recycling of cellular and membrane constituents. Whether this contribution is direct or indirect via regulating cellular architecture remains to be determined.
Using unbiased ChIP-seq and RNA profiling our study provides the first in vivo insights into the binding specificity and transcriptional targets of Drosophila Atf3. The widespread transcriptional deregulation following Atf3 overexpression in epithelia indicates multiple cellular processes being affected, thus underpinning the complex phenotypes that arise in these cells. While a quarter of Atf3 ChIP-seq targets were misexpressed in atf3 wt mosaic EADs, they represent only a fraction of all differentially regulated genes. This relatively small intersection between ChIP-seq targets identified in adult flies and the EAD transcriptome could reflect a tissue specific context for DNA-binding and transcriptional regulation by Atf3. Based on our in silico analysis, nearly three quarters of the genes misregulated in atf3 wt mosaic EADs did not contain a consensus Atf3 site. It is therefore likely that the broad shift in the transcriptome profile also comprises an adaptive response to excess Atf3 activity.
Changes to the expression of microtubule genes which translate into the alterations of the microtubule cytoskeleton represent one such example.
Importantly, the transcriptional signatures of imaginal cells overexpressing Atf3 (this study) and those lacking scrib/dlg1 [10] significantly overlap, which could indicate common molecular and cellular mechanisms. A novel and potentially relevant addition to this mutual genetic program is the Atf3dependent upregulation of LamC that was elucidated through our genomic and genetic analyses.
Nuclear lamins constitute the filamentous meshwork that underlies the inner nuclear membrane [50,51]. Aside from their structural role, experimental evidence links lamins to transcriptional regulation through the formation of transcriptionally repressive lamin associated domains [58].
Furthermore, the physical coupling between lamins and the cytoskeleton is critical for cytoarchitecture organization and cell polarization [59,60]. Thus, it is tempting to speculate that the direct and strong Finally, it is important to note that the extent of Atf3 activity and the signaling milieu differ between dlg1 mutant epithelia and cells overexpressing Atf3, and therefore the two conditions cannot be expected to phenocopy each other. The model presented in Figure 8 summarizes the shared and distinct roles of Atf3 in these contexts as established in this study.

Outlook
A growing body of evidence points to striking similarities in the cellular and molecular events underlying loss of polarity and wounding [10,61]. In this context, the induction of Atf3 upon loss of polarity is in line with the early discovery of ATF3 as a gene rapidly induced in the regenerating rat liver [62] and recent studies showing Atf3 induction during epithelial wounding [15,16]. Here we demonstrate that Atf3 levels remain low as long as epithelial polarity is intact, whereas loss of polarity due to deficiency in tumor suppressors of the Scrib complex triggers Atf3 expression. Future investigations into the link between Drosophila Atf3 and cell polarity are likely to unravel the impact of ATF3 expression on epithelial homeostasis and on human pathologies arising from polarity breakdown.

Flies
The following fly strains were used: (a) y w 1118 ,(b) w 1118 ,(c) y atf3 76 Table. All crosses were carried out at 25 °C unless stated otherwise.

Quantitative reverse transcription-PCR (qRT-PCR)
For each biological replicate, total RNA was isolated from 10 larvae or 15 adult heads with Isol-RNA Lysis Reagent (5 Prime). After DNase I treatment (Ambion, Foster City, CA), cDNA was synthesized from 1µg of RNA using oligo(dT) primers and Superscript III (Life Technologies). PCR was performed in triplicate with BioRad 2x SYBR Green mix in the CFX96 real-time PCR system (Bio-Rad, Hercules, CA). All primers were designed to anneal at 62 °C (see S1 Table for oligonucleotide sequences). All data were normalized to rp49 transcript levels, and fold changes in gene expression were calculated using the ∆∆C T method [67].

Genetic mosaic analysis
Clones expressing atf3 in the wing imaginal disc were generated using a hsFLPout method described in [6]. Crosses were kept at 22 °C. Four days after egg lay, progeny were heat shocked for 30 min in a 37 °C water bath. Imaginal discs were dissected from wandering third instar larvae. Generation of mosaics in EADs using Mosaic Analysis with a Repressible Cell Marker method (MARCM) [68] was carried out as described [32].

ChIP-seq
For each genotype (see S2 Table), three biological ChIP replicates and one input replicate were generated. For each ChIP replicate, 1.2 g of adult males were collected and processed using a modified version of the protocol described in [69]. Anesthetized males were split into two batches of 600 mg and flash frozen using liquid nitrogen, pulverized with a mortar and pestle, and disrupted further via 20 strokes of a loose fitting pestle in a Dounce homogenizer containing 10 ml of Crosslinking Solution  Peaks were assigned to genes when located within a gene or 1 kb upstream of the transcriptional start site. Multiple genes were manually associated with a peak in the case of nested genes or dense gene regions. Enriched regions mapping to positions within the CH321-51N24 BAC (source of the atf3 gBAC ) were discarded (n=59), except for two peaks corresponding to the genes atf3 and CG11403, which contained the Drosophila Atf3 motif.

ChIP from eye/antennal imaginal discs
A custom protocol combining the methods described in [69,71]  Chromatin was decrosslinked and precipitated as described in the ChIP-seq protocol. Purified DNA was raised in 500 µl water. qPCR was performed in triplicate with BioRad 2x SYBR Green mix in the CFX96 real-time PCR system (Bio-Rad, Hercules, CA). All primers were designed to anneal at 62 °C (see S1 Table 1 for oligonucleotide sequences). Data were normalized to amplification values for ecd. Fold enrichment of chromatin was calculated using the ∆∆C T method [67].

RNA-seq
RNA was isolated from FRT82B atf3 wt mosaic EADs of third instar larvae as described [72]. Total RNA libraries were generated according to the Illumina protocol and single-end sequenced on an Illumina NextSeq 500 instrument at 75 bp read length. Image analysis and base calling were done with the Illumina RTA software at run time. Published sequence data from FRT82B mosaic EAD samples [11] were used as control. Data were processed using a high-throughput Next-Generation Sequencing analysis pipeline [73]. Basic read quality check was performed with FastQC (v0.10.1) (RRID: SCR_014583) and read statistics were acquired with SAMtools v0.1.19 (RRID: SCR_002105) [74]. Reads were mapped to the Drosophila reference assembly (version BDGP R5/dm3, April 2006) using Tophat v2.0.10 (RRID: SCR_013035) [75], and gene quantification was carried out using a combination of Cufflinks v2.1.1 (RRID: SCR_014597) [76], and the DESeq2 package v1.10.1 (RRID: SCR_000154) [77], with genomic annotation from the Ensembl database (RRID: SCR_002344), version 84. In all samples, the number of total reads exceeded 50 million, from which an average 83.6 percent could be mapped, and on average 97.5% of these mapped reads fulfilled the MAPQ≥30 criterion. The results were uploaded into an in-house MySQL database and joined with BiomaRt (RRID: SCR_002987) v2.26.1 [78] annotations from Ensembl, version 84. Lists of differentially expressed genes were defined by a final database export using 5 and 0.01 as cutoffs for DESeq2-based FCs and p-values, respectively. To identify genes differentially expressed under the respective conditions, the average of at least three biological replicates was calculated. Figure 5-source data 1 shows all transcripts whose expression differed ≥ 1.5-fold in FRT82B atf3 wt compared to FRT82B control.

Atf3 motif search
The experimentally derived Drosophila Atf3 PWM was submitted to the online FIMO utility (http://meme-suite.org/tools/fimo) (RRID: SCR_001783) to identify Atf3 motifs in Drosophila melanogaster genome (UCSC, dm3), with a p-value threshold set at 0.0001. FIMO results, in the form of a BED file, were subsequently submitted to the online PAVIS tool (http://manticore.niehs.nih.gov/pavis3), using default settings.

Plasmid constructs
To create the Atf3 expression reporter, oligonucleotides containing four intact or mutated Atf3 sites (see S1 Table for sequences) were cloned via Mlu1 and Not1 sites into the pRedRabbit and pGreenRabbit vectors [79]. To express N-terminally tagged GFP-or FLAG-Atf3 proteins from the UAST promoter, atf3 cDNA (see S1 Table for oligonucleotide sequences) was cloned into pENTR4 and subsequently recombined using LR Clonase II (11791-020, Life Technologies) into pTGW and pTFW vectors, respectively (T. Murphy, Drosophila Genomic Resource Center).

S2 cell culture
Schneider 2 (S2) cells were cultured at 25 °C in Shields and Sang M3 insect medium (S8398-1L, Sigma-Aldrich) containing 8% fetal bovine serum (Gibco, Life Technologies) without antibiotics. Cells were transfected using X-tremeGENE (Roche Applied Science). Expression of UAS-driven genes was induced by co-transfection with a pWA-GAL4 plasmid expressing GAL4 under an actin5C promoter (a gift from Y. Hiromi). Cells were fed 24 hours after transfection. Cells were imaged or lysed 72 hours after transfection.

Tissue staining
Tissues from third instar larvae were processed as described previously [6].

Flow cytometry of eye/antenna imaginal discs
Third instar larvae were collected and washed 2x with PBS. EADs were dissected in PBS (on average sixty EADs /replicate/genotype) and transferred to 1.5 ml low binding microcentrifuge tubes (no more than one hour before dissociation). PBS was removed from EADs and was replaced with 100 µl dissociation solution containing 1 mg/ml collagenase I (Sigma, C2674), and 1 mg/ml papain (Sigma P4762). Samples were incubated for 60 min at room temperature and gently swirled every 15 min.
After the dissociation solution was removed, discs were carefully rinsed with 500 µl PBS, which was then replaced with 100 µl of PBS. The final dissociation was performed by passing the discs through a 27G insulin syringe (Terumo) five times. Additional 200 µl of PBS was added for a final volume of 300 µl (5 µl PBS per disc). Following sample filtration through a Filcone filter, propidium iodide was added to measure cell viability and samples were stored on ice until sorting. For each sample (n≥3 per genotype), 30,000 events were counted on a BD LSRFortessa Cell Analyzer in combination with BD FACSDiva software v8.0 (both BD Bioscience) using gates set to distinguish GFP+, GFP-, and PI-cells.

Sample Size Criteria
For sample size criteria, post hoc analysis of results presented in Figs 1, S1, S3, S5 and S9 using G*Power 3.1 (RRID: SCR_013726) [80] determined that the statistical power of all statistically significant differences (1-β) exceeded 0.94, given sample size, mean, and standard deviation for each condition.      genes that were differentially regulated by a factor ≥1.5 in mosaic EADs expressing Atf3 (eyFLP MARCM atf3 wt ). Three times as many genes were transcriptionally repressed (n=3666) than activated (n=1291).     in the eye. Homozygous dlg1 G0342 or dlg1 m52 EAD clones lead to smaller adult eyes of irregular shape (F,G) containing patches of undifferentiated tissue (A,D) relative to control (Fig 2A). Adult eyes derived from atf3 76 dlg1 G0342 or atf3 76 dlg1 m52 mosaic EADs exhibit mild or no differentiation defects (B,E) as well as restored eye size and shape (F,G). A single copy of a genomic atf3 gBAC reinstates differentiation defects to atf3 76 dlg1 G0342 adult eyes (C). Outlines of adult eyes from the indicated genotypes are presented vertically aligned along their midline (G). Adult eye measurements were performed from at least 17 biological replicates. Error bars depict 95% confidence interval; Unpaired Student's t-tests assuming unequal variance were used to calculate p-values: ***= p<0.001 ****p<0.0001. and dlg1 G0342 atf3 wt (n=5) clones were less abundant. dlg1 G0342 atf3 wt cells were less frequent relative to dlg1 G0342 alone. Blocking apoptosis raised the relative abundance of dlg1 G0342 (n=7) and atf3 76 dlg1 G0342 (n=6) cells, but not to control (n=4) levels. Unpaired Student's t-tests assuming unequal variance were used to calculate p-values. Error bars reflect the 95% confidence interval. ***= p<0.001. (B-C) An AP-1 reporter (TRE-DsRed) serves as a readout of JNK pathway activity and is upregulated in dlg1 G0342 (B') and atf3 76 dlg1 G0342 (C') EAD clones. (D) The eclosion rate of animals bearing atf3 76 dlg1 G0342 p35 EAD was less than control but was four times higher than animals bearing dlg1 G0342 p35 EAD. Four biological replicates were used for each genotype. Unpaired Student's t-tests assuming unequal variance were Atf3 did not bind the ATRE site in coracle which was occupied in samples from adults. Error bars indicate 95% confidence interval; Unpaired Student's t-tests assuming unequal variance were used to calculate p-values: **p=0.007 (LamC), *p=0.018 (dynactin), **p=0.008 (ude), **p=0.024 (arp1).