The Transcription Factor Xrp1 is Required for PERK- Mediated Antioxidant Gene Induction in Drosophila

PERK is an endoplasmic reticulum (ER) transmembrane sensor that phosphorylates eIF2α to initiate the Unfolded Protein Response (UPR). eIF2α phosphorylation promotes stress-responsive gene expression most notably through the transcription factor ATF4 that contains a regulatory 5’ leader. Possible PERK effectors other than ATF4 remain poorly understood. Here, we report that the bZIP transcription factor Xrp1 is required for ATF4-independent PERK signaling. Cell type-specific gene expression profiling in Drosophila indicated that delta-family glutathione-S-transferases (gstD) are prominently induced by the UPR-activating transgene Rh1G69D. Perk was necessary and sufficient for such gstD induction, but ATF4 was not required. Instead, Perk and other regulators of eIF2α phosphorylation regulated Xrp1 protein levels to induce gstDs. The Xrp1 5’ leader has a conserved upstream Open Reading Frame (uORF) analogous to those that regulate ATF4 translation. The gstD-GFP reporter induction required putative Xrp1 binding sites. These results indicate that antioxidant genes are highly induced by a previously unrecognized UPR signaling axis consisting of PERK and Xrp1.


SUMMARY 1
PERK is an endoplasmic reticulum (ER) transmembrane sensor that phosphorylates 2 eIF2a to initiate the Unfolded Protein Response (UPR). eIF2a phosphorylation promotes 3 stress-responsive gene expression most notably through the transcription factor ATF4 4 that contains a regulatory 5' leader. Possible PERK effectors other than ATF4 remain 5 poorly understood. Here, we report that the bZIP transcription factor Xrp1 is required for 6 ATF4-independent PERK signaling. Cell type-specific gene expression profiling in 7 Drosophila indicated that delta-family glutathione-S-transferases (gstD) are prominently 8 induced by the UPR-activating transgene Rh1 G69D . Perk was necessary and sufficient for 9 such gstD induction, but ATF4 was not required. Instead, Perk and other regulators of 10 eIF2a phosphorylation regulated Xrp1 protein levels to induce gstDs. The Xrp1 5' leader 11 has a conserved upstream Open Reading Frame (uORF) analogous to those that 12 regulate ATF4 translation. The gstD-GFP reporter induction required putative Xrp1 13 binding sites. These results indicate that antioxidant genes are highly induced by a 14 previously unrecognized UPR signaling axis consisting of PERK and Xrp1.

INTRODUCTION 1
control, antioxidant response, and amino acid transport (Han et al., 2013;Harding et al., 1 2003;Walter and Ron, 2011). The Drosophila genome encodes mediators of all three 2 branches of the UPR, and the roles of the IRE1-XBP1 and PERK-ATF4 branches in 3 Drosophila development and tissue homeostasis have been established (Mitra and Ryoo, 4 2019;Ryoo, 2015). 5 The PERK branch of UPR draws considerable interest in part because its 6 abnormal regulation underlies many metabolic and neurodegenerative diseases 7 (Delepine et al., 2000;Ma et al., 2013;Pennuto et al., 2008). Stress-activated PERK is 8 best known to initiate downstream signaling by phospho-inhibiting the translation initiation 9 factor eIF2a (Harding et al., 1999;Shi et al., 1998). While most mRNA translation 10 becomes attenuated under these conditions, ATF4 protein synthesis increases to mediate 11 a signaling response. Such ATF4 induction requires ATF4's regulatory 5' leader 12 sequence that has an upstream Open Reading Frame (uORF) that overlaps with the main 13 ORF in a different reading frame. This overlapping uORF interferes with the main ORF 14 translation in unstressed cells. (Harding et al., 2000;Kang et al., 2015;Vattem and Wek, 15 GFP-positive signal was coming from non-photoreceptor cells ( Figure 2G). We therefore 1 conclude that gstD-GFP induction by ER stress occurs in a cell-type specific manner. 2 To further validate that gstD-GFP is induced by ER stress, we treated imaginal 3 discs with culture media containing dithiothreitol (DTT), which causes ER stress by 4 interfering with oxidative protein folding. Most cells of the larval eye and wing imaginal 5 discs did not express gstD-GFP under control conditions. However, 2 mM DTT treatment 6 for 4 hours induced gstD-GFP expression in a large population of cells in both the eye 7 and wing imaginal discs ( Figure S2). Similar to the outcome of Rh1 G69D overexpression, 8 DTT treatment induced gstD-GFP primarily in the non-neuronal cell layer ( Figure S2A-D). 9 gstD-GFP expression was also induced by a different ER stress-imposing chemical, the 10 N-linked glycosylation inhibitor tunicamycin ( Figure S2G-H). These results further support 11 the notion that gstDs are significant UPR targets in Drosophila tissues, with some degree 12 of cell-type specificity. 13 marked with armadillo-lacZ in GMR>Rh1 G69D eye discs. We examined two different 1 mutant alleles of cnc; the vl110 allele which has a deletion that spans significant parts of 2 the coding sequence (Mohler et al., 1995), and the k6 allele which has a nonsense 3 mutation that specifically truncates longer ROS-responsive Cnc isoforms including CncC 4 (Veraksa et al., 2000). Mutant mosaic clones of these cncC alleles failed to block the 5 induction of gstD-GFP by GMR>Rh1 G69D . Contrary to our expectations, cncC mutant 6 clones showed enhanced gstD-GFP expression as compared to the neighboring wild type 7 cells ( Figure S3A-C). This enhanced gstD-GFP expression in the mutant clones could 8 reflect increased proteostatic stress associated with cncC loss. Based on these data, we 9 conclude that cncC is not required for gstD-GFP induction in response to ER stress. 10 Consistent with these genetic experiments, we also observed that transcriptional 11 induction of gstDs in cultured Drosophila S2 cells was cncC-independent. When ER 12 stress was pharmacologically induced in Drosophila S2 cells by tunicamycin, we 13 observed robust transcriptional induction of gstD2, which shares the same 2.7 kb 14 enhancer as gstD1 (Tang and Tu, 1994;Toung et al., 1993). Such induction of gstD2 was 15 unaffected when cncC was knocked down in S2 cells prior to tunicamycin treatment 16 ( Figure S3D, E). The knockdown efficiency of cnc as estimated through q-RT-PCR from 17 these cells was 92.46%. As a control, we validated the known roles of cncC in antioxidant 18 response by utilizing the oxidative stressor paraquat, which leads to transcriptional 19 induction of gstD2 in S2 cells as well ( Figure S3E). Consistent with previous reports, 20 induction of gstD2 in response to paraquat was blocked after cncC knockdown in S2 cells 21 PERK is best known to phosphorylate eIF2a in response to ER stress. To test if 1 such kinase activity is required for gstD-GFP induction, we expressed a Perk transgene 2 with a mutation that disrupts its kinase activity (Malzer et al., 2010). While wild type Perk 3 expression robustly induced gstD-GFP, the kinase dead Perk (Perk KD ) transgene failed 4 to induce the reporter under otherwise equivalent conditions ( Figure S5B, C). To further 5 test if eIF2a phosphorylation causes the induction of gstD-GFP, we employed an RNAi 6 line against gadd34 (Malzer et al., 2010). gadd34 encodes a phosphatase subunit that 7 helps to dephosphorylate eIF2a, and therefore, the loss of gadd34 increases phospho-8 eIF2a levels ( Figure S5A). We found that knockdown of gadd34 using the eye specific 9 GMR-Gal4 and ey-Gal4 drivers induces gstD-GFP in posterior eye discs ( Figure S5D, E). 10 A striking induction of gstD-GFP was also observed when gadd34 was knocked down in 11 the wing disc posterior compartment using the hh-Gal4 driver ( Figure S5F, G). These 12 results support the idea that PERK-mediated phosphorylation of eIF2a promotes gstD-13 clones ( Figure 3H, I). The small number of the residual GFP signals came from Perk+ 1 mosaic clones, indicating that Perk's role in gstD-GFP expression is cell-autonomous 2 ( Figure 3I). Together, these results independently support Perk's role in gstD induction in 3 ER-stressed cells. in GMR>Rh1 G69D discs. For this, we employed the crc 1 allele, a strong hypomorph bearing 10 a missense mutation in the crc coding sequence (Hewes, 2000). To our surprise, loss of 11 crc did not impair gstD-GFP induction in the eye discs ( Figure 4B',C'). Rh1 G69D expression 12 levels were similar in discs with or without crc ( Figure 4B'',C''). As a positive control to 13 validate crc activation, we used Thor-lacZ, a lacZ-based enhancer trap inserted upstream 14 expression in eye discs ( Figure 4H, I). These data strongly indicate that the induction of 1 gstDs in response to ER stress is dependent on Perk, but not crc. 2 3 The induction of gstD genes and other antioxidants in response to ER stress 4 require the bZIP transcription factor Xrp1 5 We next sought to determine how PERK activation induced the expression of gstD-GFP 6 independently of crc, and turned our focus to another stress response transcription factor, 7 Xrp1. Although Xrp1 has no known association with the UPR, Xrp1 drew our attention as 8 it reportedly mediates the induction of the gstD1-lacZ reporter in response to ribosome 9 mutations that cause cell competition (Ji et al., 2019). 10 To assess a possible role of Xrp1 in UPR signaling, we performed clonal analysis 11 with an Xrp1 mutant allele, Xrp1 M2-73 , which bears a nonsense mutation truncating all 12 Xrp1 isoforms prior to both the AT-hook motif and bZIP domain (Lee et al., 2018). Loss 13 of Xrp1 had no effect on basal gstD-GFP expression levels ( Figure 5A). GMR>Rh1 G69D 14 discs prominently expressed gstD-GFP, but otherwise equivalent eye discs with Xrp1 15 mutant clones had marked reduction of the gstD-GFP signal ( Figure 5B, C). The small 16 patches of GFP positive cells in these discs were all within the Xrp1+ mosaic cells ( Figure  17 5C' D). 18 To validate the findings with the gstD-GFP reporter, we assessed the mRNA levels 19 of gstD1 and other select antioxidant genes from dissected eye imaginal discs through 20 qRT-PCR. A subset of the analyzed transcripts showed induction in GMR>Rh1 G69D disc 21 samples dependent on Xrp1 ( Figure 5E). These included gstD1, jafrac1 that encodes a by qRT-PCR analysis ( Figure 6E). These results indicate that Perk mediates the induction 1 of Xrp1 protein through a post-transcriptional mechanism. 2 To test if Perk's kinase activity is involved in Xrp1 protein induction, we compared 3 the effects of overexpressing Perk WT versus Perk KD . Xrp1 protein levels increased in 4 GMR>Perk WT eye discs, but not in GMR>Perk KD discs ( Figure 6F, G, H). To further test if 5 PERK's phosphorylation target eIF2a controls gstD-GFP expression, we knocked down 6 gadd34 in wing discs using the posterior compartment-specific hh-Gal4 driver. While 7 control wing discs showed a low basal anti-Xrp1 signal throughout the tissue, depletion 8 of gadd34 by RNAi resulted in distinct nuclear anti-Xrp1 signals throughout the posterior 9 compartment ( Figure 6I, J). These results indicate that Xrp1 induction is regulated by 10 eIF2a phosphorylation. 11 PERK's well-established downstream effector, ATF4, has a 5' regulatory leader 12 sequence with multiple upstream Open Reading Frames (uORFs) that allows its 13 translational induction in response to stress. To examine if there is an analogous 5' leader 14 in Xrp1, we used a bioinformatic program that predicts initiation codons 15 (https://atgpr.dbcls.jp). This approach did not detect initiation codons in the 5' leader of 16 Xrp1's shorter splice isoforms (e.g. isoform D and E). By contrast, two putative uORFs in 17 the 5' leader of the long splice isoforms of Xrp1 (e.g., isoform F and G) were identified, 18 with uORF1 predicted to encode a 124 a.a. length peptide, and the uORF2 with a 288 19 a.a. peptide. The uORF2 overlapped with the main ORF, but was in a different reading 20 frame ( Figure 7A). Such an arrangement is similar to that of ATF4's last uORF. 21 To assess the likelihood that Xrp1 uORFs are peptide coding sequences, we 22 performed Protein BLAST searches with the encoded peptide sequences. Xrp1's uORF1 23 did not have any homologous hits in other species. However, sequences homologous to 1 D. melanogaster Xrp1 uORF2 were identified in D. kikkawai (Percent identity = 71.85%), 2 D. persimilis (Percent identity = 48.67%), D. navojoa (Percent identity = 50.44%), D. 3 grimshawi (Percent identity = 49.48%) and D. busckii (Percent identity = 38.69%). The 4 homologous sequences in these other species were part of their Xrp1 main ORF N-5 terminal regions ( Figure 7B). The C-terminal regions of these main ORFs all encoded the 6 AT-hook and bZIP DNA binding domains homologous to that in the D. melanogaster's 7 Xrp1 main ORF ( Figure 7C). The phylogenetic conservation of D. melanogaster uORF2 8 at the peptide level supports the idea that this is a functional uORF that has been under 9 selective pressure during evolution. The similarities in the arrangements of the D. 10 melanogaster Xrp1 and ATF4 5' leaders, together with the observation that Xrp1 is 11 induced in response to eIF2a phosphorylation, suggest that Xrp1 and ATF4 share similar 12 mechanisms for their translational induction in response to stress. 13 14 Xrp1 binding sites within the gstD enhancer are essential for gstD-GFP induction 15 To determine if Xrp1 regulates gstDs directly, we examined the gstD 2.7 kb enhancer for 16 putative Xrp1 and ATF4 binding sites ( Figure 8A). To do so we used the Xrp1 position 17 frequency matrix derived from a previous Xrp1 ChIP-seq analysis (Baillon et al., 2018) 18 and a publicly available ATF4 position frequency matrix (see Methods). Our binding score 19 analysis predicted three putative binding sites each for Xrp1 and ATF4 in the enhancer 20 ( Figure 8B, C). Two of the three highest scoring Xrp1 binding sites overlapped with two 21 of the ATF4 sites, while one was predicted to be a unique Xrp1 site.
To test if Xrp1 binds to this locus, we used the CUT&RUN approach (Skene et al., 1 2018) to pull down HA-tagged Xrp1 proteins from Drosophila larval tissues to examine if 2 putative Xrp1 target gene DNAs co-purify ( Figure 8D, see Methods). We specifically 3 employed an Xrp1 HA transgenic fly line that has an HA-tagged transgene inserted in the 4 endogenous Xrp1 locus (Blanco et al., 2020). Larval tissues incubated with DTT gave 5 strong q-RT-PCR signals for gstD regulatory DNA, while control tissues without DTT 6 treatment had minimal DNA recovered. Upd3 was previously reported as an Xrp1 target 7 gene, and as expected, increased Upd3 DNA was recovered in response to DTT 8 treatment. Tubulin DNA was used as a negative control, which had minimal DNA recovery 9 even after DTT treatment ( Figure 8E). These results support the idea that Xrp1 binds to 10 the gstD locus in larval cells under ER stress. 11 We then reconstructed new gstD-GFP reporters with (gstD WT -GFP) or without 12 (gstD ∆ATF4 -GFP) the predicted ATF4 binding sites. To generate an Xrp1 binding site-13 deficient gstD-GFP reporter (gstD Xrp1m -GFP), we introduced mutations in the remaining 14 Xrp1 binding site within the gstD ∆ATF4 -GFP reporter. To control for genetic backgrounds, 15 we targeted these reporters to a specific attP landing site to generate transgenic flies. 16 While GMR>Rh1 G69D effectively induced both the wild type and gstD ∆ATF4 -GFP reporters, 17 gstD Xrp1m -GFP was not induced under otherwise identical conditions ( Figure 8F-J). 18 Together, these results indicate that gstD genes are direct targets of Xrp1 in the context 19 of ER stress. 20 1 Here, we report that ER stress activates a previously unrecognized UPR axis mediated 2 by PERK and Xrp1. Specifically, we showed that gstD family genes are among the most 3 highly induced UPR targets in Drosophila, and that such induction requires Perk, one of 4 the three established ER stress sensors in metazoans. Surprisingly, the induction of gstD 5 genes in this context did not require crc, the ATF4 ortholog. Instead, we found that a 6 poorly characterized transcription factor Xrp1 is induced downstream of Perk to promote 7 the expression of gstDs and other antioxidant genes. 8 Our findings are surprising given that ATF4 is considered a major effector of 9 PERK-mediated transcription response (Karagöz et al., 2019;Walter and Ron, 2011). 10 ATF4 was the first PERK downstream transcription factor to be identified in part based 11 on the similarity of its regulatory mechanisms with that of yeast GCN4 (Dever et al., 1992;12 Harding et al., 2000). But more recent studies have shown there could be parallel 13 effectors downstream of PERK activation (Andreev et al., 2015;Baird et al., 2014;Palam 14 et al., 2011;Zhou et al., 2008). The functional significance of these alternative factors had 15 remained poorly understood. Our study here has led us to conclude that an ATF4-16 independent branch of PERK signaling is required for the expression of the most highly 17 induced UPR target in Drosophila. 18 As a potential mediator of this ATF4-independent PERK signaling, we first 19 considered cncC as a prime candidate for a few reasons: cncC is an established regulator 20 of gstD-GFP induction (Sykiotis and Bohmann, 2008), and previous studies had reported 21 that Nrf2 is activated by PERK in cultured mammalian cells and in zebrafish (Cullinan et al., 2003;Cullinan and Diehl, 2004;Mukaigasa et al., 2018). However, our results 1 reported here do not support the simple idea that gstD-GFP is induced by CncC, which 2 in turn is activated by PERK. Specifically, we found that the loss of Perk blocked gstD-3 GFP induction in this experimental setup, but the loss of cncC did not. While Nrf2/CncC 4 clearly regulates antioxidant gene expression in response to paraquat, our results indicate 5 that PERK mediates an independent antioxidant response in Drosophila. 6 Our data indicates that this ATF4-independent PERK signaling response requires 7 the AT-hook bZIP transcription factor Xrp1. Several pieces of evidence support the idea 8 that Xrp1 is translationally induced, analogous to the mechanism reported for ATF4 9 induction. First, our RNA-seq and qRT-PCR results indicate that Xrp1 transcript levels do 10 not change significantly in Rh1 G69D expressing eye discs. These results argue against the 11 idea that Xrp1 is induced at the transcriptional level. Second, we find that PERK's kinase 12 domain is required for Xrp1 protein induction. Third, knockdown of gadd34, which 13 increases phospho-eIF2a levels downstream of Perk, is sufficient to induce Xrp1 protein 14 and gstD-GFP expression. Finally, we find that Xrp1's 5' leader has a uORF that overlaps 15 with the main ORF, similar to what is found in ATF4's regulatory 5' leader sequence. 16 Moreover, Xrp1's uORF2 encodes a peptide sequence that is phylogenetically conserved 17 in other Drosophila species. High sequence conservation at the peptide level enhances 18 confidence that uORF2 is a peptide coding sequence. 19 Xrp1 is known to respond to ionizing radiation, motor neuron-degeneration in a 20 Drosophila model for amyotrophic lateral sclerosis (ALS), and in cell competition caused 21 by Minute mutations that cause haplo-insufficiency of ribosomal protein genes (Akdemir et al., 2007;Baillon et al., 2018;Lee et al., 2018;Mallik et al., 2018). Interestingly, two 1 recent studies reported that these Minute cells induce gstD-GFP, and also show signs of 2 proteotoxic stress as evidenced by enhanced eIF2a phosphorylation (Baumgartner et al., 3 2021;Recasens-Alvarez et al., 2021). Although these studies did not examine the 4 relationships between Xrp1, gstD-GFP and eIF2a kinases such as Perk, our findings 5 make it plausible that the PERK-Xrp1 signaling axis regulates cell competition caused by 6 Minute mutations. 7 Despite the rising levels of interest in Xrp1 as a stress response factor, the identity 8 of its mammalian equivalent remains unresolved. Xrp1 is well conserved in the Dipteran 9 insects, but neither NCBI Blast searches nor Hidden Markov Model-based analyses 10 identify clear orthologs in other orders (Akdemir et al., 2007, Blanco et al., 2020Mallik et 11 al., 2018;Baillon et al., 2018). Such evolutionary divergence is not unprecedented in UPR 12 signaling: GCN4 is considered a yeast equivalent of ATF4, but they are not the closest 13 homologs in terms of their peptide sequences (Harding et al., 2000). Likewise, the yeast 14 equivalent of XBP1 (IRE1 effector, not to be confused with Xrp1 in this study) is Hac1, 15 but there is little sequence conservation between the two genes (Yoshida et al., 2001;16 Shen et al., 2001;Calfon et al., 2002). Yet, the UPR signaling mechanisms are considered 17 to be conserved due to the shared regulatory mechanisms. Along these lines, mammalian 18 cells may have functional equivalents of Xrp1. We consider among the candidate 19 equivalent factors those with regulatory 5' leader sequences that respond to eIF2a 20 phosphorylation (Palam et al., 2011;Zhou et al., 2008;Andreev et al., 2015). Based on 21 the emerging roles of Xrp1 in Drosophila models of human diseases, we speculate that those ATF4-independent PERK signaling effectors may play more significant roles in 1 diseases associated with UPR than had been generally assumed. 2 We note in our study that genes encoding cytoplasmic glutathione S-transferases 3 (GSTs) such as gstD1 and gstD9 are among the most prominently induced UPR targets 4 in our eye imaginal disc-based gene expression profiling analysis. Previous studies also 5 reported these as ER stress-inducible genes in Drosophila S2 cells (Malzer et al., 2018). 6 GSTs are cytoplasmic proteins that participate in the detoxification of harmful, often 7 lipophilic intracellular compounds damaged by ROS. These enzymes catalyze the 8 formation of water-soluble glutathione conjugates that can be more easily eliminated from 9 the cell (Low et al., 2010;Low et al., 2007;Sharma et al., 2004;Wilce and Parker, 1994). 10 It is noteworthy that ROS is generated as a byproduct of Ero-1-mediated oxidative protein 11 folding, and such ROS generation increases when mutant proteins undergo repeated 12 futile cycles of protein oxidation (Gross et al., 2006;Tang and Tu 1994;Tu and Weissman, 13 2004). Therefore, we speculate that cytoplasmic GSTs evolved as UPR targets as they 14 have the ability to detoxify lipid peroxides or oxidized ER proteins that increase in 15 response to ER stress. 16 In conclusion, our findings support the idea that an ATF4-independent branch of 17 PERK signaling mediates the expression of the most highly induced UPR targets in eye 18 disc cells. This axis of the UPR requires Xrp1, a gene that had not previously been 19 associated with ER stress response. The identification of this new axis of UPR signaling 20 may pave the way for a better mechanistic understanding of various physiological and 21 pathological processes associated with abnormal UPR signaling in metazoans.

ACKNOWLEDGEMENTS 2
We thank Nicholas Baker, Claude Desplan, Michael Garabedian, Moses Chao, Erika 3 Bach and Jessica Treisman for helpful comments. We also thank Heinrich Jasper, 4 Nicholas Baker, Konrad Basler, Donald Rio, Thomas Hurd, Temesgen Fufa, Robert 5 Hufnagel and Vikki Weake for fly strains, plasmids and technical advice. This project was 6 supported by NIH R01 EY020866 and GM125954 to H.D.R., T32 HD007520 and T32 7 GM136573 training grants support for B.B., K99 EY029013 to D.V., and the Cancer 8 Center Support Grant P30 CA061087 at the Perlmutter Cancer Center to the Genome 9 Technology Center. 10 We reared flies at ambient temperature on a standard cornmeal-agar diet supplemented 1 with molasses, and carried out crosses at 25ºC. We performed all gene overexpression 2 experiments using the Gal4/UAS binary expression system (Brand and Perrimon, 3 1993). We used the following fly stocks: w 1118 , gstD-GFP (Sykiotis and Bohmann, 2008), binding scores. We deposited this modified Python code in Github 1 (https://github.com/finnroach/transcription-factor-binding) 2 To identify putative ATF4 binding sites in the gstD1 upstream enhancer, we 3 utilized the position frequency matrix (PFM) information of human ATF4 available on 4 JASPAR (jaspar.genereg.net, Matrix ID:MA0833.1). The code outputs only binding 5 sites with scores that are greater than 76% of the optimal ATF4 binding sequence. This 6 cutoff both picks up two known ATF4 binding sites within the Drosophila Thor intron 7 sequence (Kang et al., 2017) and allows leeway to find slightly lower affinity binding 8 sites. Inputting the gstD1 enhancer sequence into this code revealed three unique 9 putative ATF4 binding half-sites with the following sequences: cgttccctcatac (77 percent 10 of optimal binding score), aatttcatcattt (83 percent of optimal binding score), and 11 tatttcatcaccc (86 percent of optimal binding score). 12 To score putative Xrp1 binding sites, we used information available from a 13 previous Xrp1 ChIP-seq study. The sequence logo of Xrp1 position frequency matrix 14 was reported previously (Baillon et al., 2018). The Xrp1 position frequency matrix itself 15 is available from the link (https://www.biorxiv.org/content/10.1101/467894v1). We show 16 in Figure 8 the outputs of sites with scores greater than 95% of the optimal Xrp1 binding 17 sequence. 18 digested pattB using InFusion cloning (ClonTech). To make gstD ∆ATF4 -GFP, we used 1 commercial gene synthesis (General BioSystems, Inc.) to reconstruct the entire reporter 2 region previously amplified by PCR for the creation of gstD WT -GFP, but with the predicted 3 ATF4 binding sites deleted. The three deleted sequences are (in the order of from that 4 closest to gstD2 towards the gstD1 coding sequence): ggtgatgaaata, aatgatgaaatt, 5 atgagggaa. 6 To generate the gstD Xrp1m -GFP reporter, we performed directed mutagenesis on 7 the gstD ∆ATF4 -GFP to mutate the remaining Xrp1 binding site. Specifically, the sequence 8 ttgtgaaatc was mutated to ttcccgggtc. The wild type and mutant gstD-GFP DNA were then 9 subcloned into the pattB vector, and sent to BestGene.Inc (Chino Hills, CA) for embryo 10 injection. Standard phiC31 integrase-mediated germline recombination approach (Groth 11 et al., 2004) was used to target the plasmids to an attP landing site at cytological position 12 51C (Bloomington Stock Center, #24482). 13 approximately 100 larvae per sample in phosphate buffered saline (PBS), pH 7.4 with 18 0.1% Tween-20 (Sigma-Aldrich, cat. #P7949). We then washed dissected discs in ice-19 cold nuclear isolation buffer (10 mM HEPES-KOH, pH 7.5; 2.5 mM MgCl2; 10 mM KCl), 20 then lysed cells in 1mL of ice-cold nuclear isolation buffer in a 2mL Dounce 21 homogenizer (VWR, cat. #62400-595). We next filtered the homogenate through a 40µm Flowmi cell strainer (WVR, cat. #BAH136800040), after which we incubated a 1 20µL pre-isolation aliquot of the filtrate with 0.1µg/mL 4′,6-diamidino-2-phenylindole 2 (DAPI) (Millipore-Sigma, cat. #D9542) and mounted on a slide for imaging (Fisher, cat. 3 #12-550-433). We incubated the remaining filtrate with anti-EGFP-coupled protein G 4 Dynabeads (Invitrogen, cat. #10003D) for 1 hour at 4ºC with gentle end-over-end 5 rotation. Next, we collected the beads using a magnetic microcentrifuge tube holder 6 (Sigma, cat. #Z740155) and washed the collected beads with wash buffer (PBS, pH 7.4; 7 2.5mM MgCl2), then resuspended the beads in a final volume of 150µL of wash buffer. 8 We incubated an aliquot of the post-isolation sample with 0.1µg/mL DAPI and mounted 9 the sample on a slide for imaging. Finally, we suspended the post-isolation nuclei in 10 1mL of Trizol reagent (Life Technologies, cat. #15596018) for RNA extraction following 11 standard procedures. Prior to RNA precipitation with isopropanol, we added 0.3M 12 sodium acetate and glycogen (Invitrogen, cat. #AM9510) to facilitate visualization of the 13 using approximately 1.5ng total RNA per sample. For sequencing, we performed paired-1 end 50bp sequencing of samples on an Illumina NovaSeq 6000 platform (Illumina, cat.

Antibodies, and immunofluorescence and confocal microscopy 21
To generate a guinea pig anti-Xrp1 antibody, we expressed the Xrp1 long isoform 22 Tris pH 8.0 and 100mM NaCl). We ran the solubilized fraction containing His-Xrp1 5 through a Ni 2+ -NTA column and washed the column with reducing concentrations of Urea 6 in the denaturing buffer for purification and re-folding. We eluted His-Xrp1 from the 7 column with Tris buffer containing 400 mM imidazole, and sent the purified protein for 8 custom polyclonal antibody production (Covance Inc.). The crude antiserum was affinity 9 purified and used at 1:10 dilution for immunolabeling. 10 We immunoprecipitated EGFP-labeled imaginal disc nuclei with mouse anti-EGFP 11 (Roche, cat. #11814460001). For immunofluorescence, we used the following primary 12 and secondary antibodies at the indicated dilutions: rabbit anti-β-Gal (1:500, MP 13 with gentle rocking. Next, following three short rinses with PBTx 0.2%, we incubated discs 1 with primary antibodies in PBTx 0.2% for 1h at ambient temperature with gentle rocking. 2 Following primary antibody incubation, we washed discs for 3*10 minutes with PBTx 0.2%, 3 then covered from light and incubated with secondary antibody in PBTx 0.2% for 1h at 4 ambient temperature with gentle rocking. Following secondary antibody incubation, we 5 washed discs 3*10 minutes with PBTx 0.2%, then mounted in 50% glycerol (Millipore-6 Sigma, cat. #G5516) with 0.1µg/mL DAPI. For all confocal micrographs, we captured 7 images on a Zeiss LSM 700 confocal microscope (Carl Zeiss). For image acquisition, we 8 scanned all imaginal discs under a 40X water objective, and all isolated nuclei under a 9 100X oil objective. 10 11

S2 cell culture, RNAi and drug treatments 12
We cultured Drosophila S2 cells in ambient conditions in S2 cell medium (Fisher,13 cat. #21720024) supplemented with 1% penicillin/streptomycin (Life Technologies, cat. 14 #15-140-122) and 10% heat-inactivated FBS (Fisher, cat. #10082147). For maintenance 15 of cell lines, we passaged cells every 3-4 days and utilized them for experiments between 16 passages 6 and 15. 17 We performed RNAi using a modified dsRNA bathing protocol (Ryoo et al., 2007). 18 We generated dsRNA against cncC, Perk, Xrp1 and crc mRNAs following an established 19 T7 in vitro transcription protocol (ThermoFisher, cat. #AM1334) (Armknecht et al., 2005). 20 On day 0, we added approximately 20µg of the indicated dsRNA to 1mL containing 10 6 21 cells in serum-free S2 cell medium in a 6-well dish. After 30 minutes of incubation at ambient temperature, we added 3mL of S2 cell medium with serum and incubated the 1 cells at ambient temperature. On day 3, we harvested the cells, resuspended them at 2 1*10 6 cells/mL in serum-free S2 cell medium, and subjected them to another round of 3 dsRNA bathing as described above. On day 6, we re-plated cells at a concentration of 4 ~2*10 6 cells/mL for drug treatments. For drug treatments, we exposed cells to either 5 20mM paraquat (Sigma, cat. #856177) or 10µg/mL tunicamycin (Fisher, cat. #351610) 6 for the indicated times, then harvested and resuspended the treated cells in 500µL Trizol 7 reagent. Following RNA isolation, we then treated the samples with Turbo DNAse 8 (Invitrogen, cat. #AM1907) for 25 minutes at 37ºC to remove traces of contaminating 9 genomic DNA. We utilized the following oligos to generate T7 double-stranded RNA: 10 We amplified the gstD-GFP reporter from transgenic fly genomic DNA 1 for InFusion cloning into BglII/NotI-digested pattB using the following Gstd1-R: GTTGAGCAGCTTCTTGTTCAG 5 6 Gstd9-F: TTGCCGTTCCATCCTGATGAC 7 8 Gstd9-R: GCTTAAGATGCTCGCCGGCAT 9 10 Gstd10-F: TGCCGCTCTGTTCTGATG 11 12 Gstd10-R: Quantification of gstD-GFP pixel intensity changes 24 We used ImageJ (https://imagej.nih.gov/ij/index.html) to calculate average pixel 25 intensities of gstD-GFP signals from eye disc microscopic images. We specifically 26 measured pixel intensities of the GMR-Gal4 expressing region in posterior eye discs. To 27 calculate fold change of the gstD-GFP signal, we divided the average pixel intensities of 28 interest with that from control discs. Statistical significance was assessed through two 29 tailed t tests unless otherwise stated. We plotted graphs using the ggplot2 package in the 30 Approximately, 100 tissues including eye, wing, leg discs and brains for each 1 experimental condition were isolated from third instar wandering larvae (Genotype: hspflp; 2 FRT42/Cyo;Xrp1[HA-1]/TM6B). The isolated tissues were soaked in 2mM DTT for 6 hrs. 3 in S2 media at RT in shaking condition. After incubation, samples were centrifuged at 4 1200 rpm for 5 mins at 4 degree and 1X trypsin was added to the pellet to isolate single 5 cell suspension. Trypsin digestion was continued for 4 hrs. at RT in a nutator. After 6 digestion, the reaction mixture was centrifuged at 1200 rpm for 5 mins and the isolated 7 cells were re-suspended in cold wash buffer (2% FBS in 1X DPBS without Ca/Mg and 8 EDTA). The cells were kept on ice from this step. The cell suspension was filtered using 9 a 40µm cell strainer. Isolated cells were counted in haemocytometer and approximately 10 250,000 cells/sample were used for further processing using CUTANA ChIC/ CUT & RUN 11 kit (Epicypher, 14-1048) according to the manufacture's protocol. For IP reaction of HA-12 tagged Xrp1, ChIP grade anti-HA antibody (Abcam, ab9110) was used. As a control 13 isotype, ChIP grade control IgG was used (Abcam, ab171870). An equivalent region labeled with anti-Elav antibody that labels photoreceptors (magenta). 5 Scale bar = 10µm (G) A magnified view of GMR>Rh1 G69D eye discs in planar orientation 6 with gstD-GFP (green) and anti-Elav labeling (red). Individual channels are shown 7 separately in (G' -G'''). Scale bar = 2.5µm