Xrp1 and Irbp18 trigger a feed-forward loop of proteotoxic stress to induce the loser status

Cell competition induces the elimination of less-fit “loser” cells by fitter “winner” cells. In Drosophila, cells heterozygous mutant in ribosome genes, Rp/+, known as Minutes, are outcompeted by wild-type cells. Rp/+ cells display proteotoxic stress and the oxidative stress response, which drive the loser status. Minute cell competition also requires the transcription factors Irbp18 and Xrp1, but how these contribute to the loser status is partially understood. Here we provide evidence that initial proteotoxic stress in RpS3/+ cells is Xrp1-independent. However, Xrp1 is sufficient to induce proteotoxic stress in otherwise wild-type cells and is necessary for the high levels of proteotoxic stress found in RpS3/+ cells. Surprisingly, Xrp1 is also induced downstream of proteotoxic stress, and is required for the competitive elimination of cells suffering from proteotoxic stress or overexpressing Nrf2. Our data suggests that a feed-forward loop between Xrp1, proteotoxic stress, and Nrf2 drives Minute cells to become losers.

Introduction Cells within a tissue may become damaged due to spontaneous or environmentally induced mutations, and it is beneficial to organismal health if these cells are removed and replaced by healthy cells. During cell competition, fitter cells, termed winners, recognise and eliminate less-fit cells, termed losers, resulting in restoration of tissue homoeostasis [1][2][3]. Cell competition therefore promotes tissue health and is thought to provide a level of protection against developmental aberrations [4][5][6] and against cancer by removing cells carrying oncoplastic mutations [1,7]. However, an increasing body of evidence indicates that cell competition can also promote growth of established tumours, enabling them to expand at the expense of surrounding healthy cells [7,8].
Minute cell competition was discovered through the study of a class of Drosophila ribosomal mutations called Minutes [9] and initial work suggests that it is conserved in mammals [10]. While homozygous Rp mutations are mostly cell lethal, heterozygosity for most Rp mutations gives rise to viable adult flies that exhibit a range of phenotypes including developmental delay and shortened macrochaete bristles [9,11]. Rp/+ tissues display a higher cell-autonomous death frequency than wild-type tissues [12][13][14][15], and competitive interactions further elevate cell death in Rp/+ cells bordering wild-type cells, contributing to progressive loss of Rp/+ cells over time [14,16,17].
It was suggested that Rp/+ cells are eliminated by cell competition due to their reduced translation rate [3,[18][19][20][21][22]. However, we and others have recently shown that Rp/+ cells experience significant proteotoxic stress and this is the main driver of their loser status [13,14]. Rp/+ cells have a stoichiometric imbalance of ribosome subunits, which may provide the source of proteotoxic stress. The autophagy and proteasomal machineries become overloaded and protein aggregates build up in Rp/+ cells, leading to activation of stress pathways. This includes activation of Nuclear factor erythroid 2-related factor 2 (Nrf2) and of the oxidative stress response [23], which we have shown to be sufficient to cause the loser status [24]. Restoring proteostasis in Rp/+ cells suppresses the activation of the oxidative stress response and inhibits both autonomous and competitive cell death [13,14].
Genetic screening for suppressors of cell competition led to the identification of Xrp1 [20,25,26], a basic leucine Zipper (bZip) transcription factor. Loss of Xrp1 rescues both the reduced growth and competitive cell death of Rp/+ cells in mosaic tissues [20,25]. Consistently, loss of Xrp1 restores translation rates and abolishes the increased JNK pathway activity characteristic of Rp/+ cells [20]. Xrp1 forms heterodimers with another bZip transcription factor called Inverted repeat binding protein 18kDa (Irbp18) [27,28], and removal of Irbp18 also strongly suppresses the competitive elimination of Rp/+ cells in mosaic tissues [29]. Irbp18 and Xrp1 are transcriptionally upregulated and mutually required for each other's expression in Rp/+ cells, suggesting they function together in Minute cell competition [29]. Irbp18 forms heterodimers with another bZip transcription factor, ATF4 [28]. However, knockdown of ATF4 in Rp/+ cells reduces their survival in mosaic tissues, which is the opposite effect to knockdown of Xrp1 or Irbp18. This has been interpreted to suggest that the ATF4-Irbp18 heterodimer acts independently to the Xrp1-Irbp18 heterodimer [29].
How the Xrp1/Irbp18 complex contributes to the loser status is not clear. Given the recently identified role of proteotoxic stress in cell competition we sought to establish whether Xrp1/ Irbp18 and proteotoxic stress act independently or in the same pathway to contribute to cell competition in Rp/+ cells. We identify a feed-forward loop between Xrp1/Irbp18 and proteotoxic stress, which is required for downstream activation of the oxidative stress response and the loser status. Our data suggests a model in which the initial insult in RpS3/+ cells is ribosomal imbalance-induced proteotoxic stress, which is Xrp1 independent. Xrp1 is then transcriptionally activated downstream of proteotoxic stress, by increased phosphorylatedeukaryotic Initiation Factor 2α (p-eIF2α), and possibly by Nrf2. The Xrp1-Irbp18 complex then induces further proteotoxic stress, completing the feed-forward loop. This work provides new insight into the interactions between the stress signalling pathways active in Rp/+ cells and provides a mechanism for how the Xrp1-Irbp18 heterodimer mediates the competitive elimination of Rp/+ cells by wild-type cells.
To investigate the role of Xrp1 and Irbp18 in proteotoxic stress and the oxidative stress response, which are primary drivers of the loser status in Rp/+ cells [13,14,24], we expressed xrp1-RNAi specifically in the posterior compartment of RpS3/+ wing discs with the hedgehog (hh)-gal4 driver. Xrp1 knockdown significantly rescued the accumulation of p-eIF2α ( Fig 1A  and 1B), a marker of the integrated stress response, which is induced in response to proteotoxic stress [30,31] and is upregulated in RpS3/+ cells [13,14]. Xrp1 knockdown also strongly inhibited the oxidative stress response in RpS3/+ cells, as it reduced the expression of Glutathione S transferase D1-GFP (GstD1-GFP) (Fig 1A and 1C), a reporter of Nrf2 [32]. Irbp18 knockdown also rescued both p-eIF2α upregulation and GstD1-GFP upregulation in RpS3/+ discs (Fig 1D-1F). Refractory to sigma P (Ref(2)p), also known as p62, is an autophagy adaptor and cargo [33] and a marker of cytosolic protein aggregates [34], which accumulates in RpS3/+ cells due to proteotoxic stress overload [14]. The accumulation of p62-labelled aggregates in RpS3/+ cells was rescued both by xrp1-RNAi (Fig 1G and 1H) and by irbp18-RNAi (Fig 1I and  1J), further indicating that proteotoxic stress in Rp/+ cells is mediated by the Xrp1/Irbp18 complex. Together, these data show that Xrp1 and Irbp18 are required for, and act upstream of, proteotoxic stress and the oxidative stress response in RpS3/+ cells.
Rp/+ cells have recently been shown to have a stoichiometric imbalance in their ribosome subunits, suggesting that this is the initial proteostatic perturbation leading to proteotoxic stress. Specifically, Rp/+ cells have an excess of large-subunit (LSU) proteins and a reduced complement of small-subunit (SSU) proteins, relative to wild-type cells [13,14]. The data in Fig 1B, 1C, 1E, 1F, 1H and 1J indicate that Xrp1 and Irbp18 induce proteotoxic stress in RpS3/ + cells, therefore we asked whether the ribosomal imbalance in RpS3/+ cells is also downstream of Xrp1. Proteomic analysis revealed that removal of one copy of xrp1, which is (A-C) An RpS3 +/wing disc harboring the GstD1-GFP reporter (green) and expressing xrp1-RNAi (xrp1i) in the posterior compartment, immuno-stained for p-eIF2α (magenta) with nuclei labelled in blue (A). Quantifications of p-eIF2α signal intensity (n = 11; two-sided Wilcoxon signed-rank test) and GstD1-GFP signal intensity (n = 11; two-sided Wilcoxon signed-rank test) are shown in (B) and (C) respectively. (D-F) An RpS3 +/wing disc harboring the GstD1-GFP reporter (green) and expressing irbp18-RNAi (irbp18i) in the posterior compartment, immuno-stained for p-eIF2α (magenta) with nuclei labelled in blue (D). Quantifications of p-eIF2α signal intensity (n = 9; two-sided Wilcoxon signed-rank test) and GstD1-GFP signal intensity (n = 9; two-sided Wilcoxon signed-rank test) are shown in (E) and (F) respectively. (G-H) A wing disc of the same genotype as shown in (A), immuno-stained for p62 (grey) (G), with quantification of p62 signal intensity (H) (n = 7; two-sided Wilcoxon signed-rank test). (I-J) A wing disc of the same genotype as shown in (D), immuno-stained for p62 (grey) (I), with quantification of p62 signal intensity (J) (n = 11; two-sided Wilcoxon signed-rank test). (K) A bar graph showing the mean log fold change in all Small-subunit (SSU) and Largesubunit (LSU) ribosomal proteins detected by mass spectrometry in RpS3 +/and RpS3 +/-, Xrp1 +/wing discs relative to wild-type discs, as indicated (n = 29; two-sided Wilcoxon signed-rank test for comparison of SSU, n = 49; two-sided Wilcoxon signed-rank test for comparison of LSU, n = 29 and 49, respectively; sufficient to rescue Rp/+ cells from competition [20], rescues the excess of LSU proteins but does not affect the reduction in SSU proteins (Fig 1K and 1L). Thus, SSU protein imbalance in RpS3/+ cells is independent of Xrp1. This suggests that the initial proteotoxic stress experienced by Rp/+ cells is an SSU/LSU stoichiometric imbalance. This may provide the signal for Xrp1 induction, which in turn exacerbates proteotoxic stress, resulting in accumulation of LSU proteins.
Mutations in the E3 ubiquitin ligase encoding gene mahjong (mahj) lead to the loser status, and mahj -/cells are out-competed by wild-type cells in mosaic tissues [35]. Although Mahj is functionally distinct to ribosomal proteins, the gene expression signatures of mahj and RpS3 mutants significantly overlap, indicating a common mechanism leading to the loser status [24]. Indeed, mahj cells also show upregulation of p62 labelled aggregates suggesting that they experience proteotoxic stress [14]. Interestingly, Xrp1 knockdown rescued mahj-RNAi expressing cells from elimination in mosaic wing discs (Fig 2A-2C). We then induced larger patches of mahj-RNAi cells and found that, like Rp/+ cells, mahj-RNAi expressing cells upregulate p-eIF2α ( Fig 2D). Simultaneous Xrp1 knockdown rescued the accumulation of p-eIF2α ( Fig 2E and 2F), suggesting that Xrp1 is also upstream of proteotoxic stress in mahj deficient cells. The rescue was not due to the presence of a second UAS construct (UAS-xrp1-RNAi), which could have weakened the expression of UAS-mahj-RNAi by titrating Gal4, as the cells expressing mahj-RNAi in fact also carried a second control UAS construct (40D UAS ), which does not affect readouts of the loser status (S2 Fig). All further experiments in this study that compare the phenotype of expression of a single UAS construct to that of two UAS constructs use this strategy. Thus, Xrp1 contributes to the competitive elimination of cells with distinct loser backgrounds, Rp/+ and mahj, which are both linked to proteotoxic stress.
The results described above suggest that Xrp1 functions upstream of proteotoxic stress and oxidative stress in RpS3/+ and mahj deficient cells, so we asked whether Xrp1 is sufficient to induce proteotoxic stress. We over-expressed the xrp1 long isoform [36] in the posterior compartment of wing discs with the engrailed (en)-gal4 driver and found this condition to be larval lethal before the 3 rd instar, which is consistent with previous reports that xrp1 over-expression induces expression of proapoptotic genes [25] and high levels of cell death [25,29,36,37]. To circumvent this lethality, we used a temperature sensitive Gal4 inhibitor, Gal80 ts , to prevent xrp1 expression throughout most of larval development. Shifting the larvae to the Gal80 ts restrictive temperature 24 hours before dissection allowed for a relatively short burst of xrp1 expression. Under these conditions, xrp1 over-expressing compartments accumulated GstD1-GFP (Fig 3A and 3B), p62 (Fig 3C and 3D) and had higher levels of p-eIF2α (Fig 3E  and 3F) than the wild-type, control compartments. Therefore, Xrp1 is sufficient to induce proteotoxic stress. Conversely, Irbp18 overexpression did not increase p-eIF2α or p62 (S3A- S3D  Fig) suggesting that Irbp18 alone is not sufficient to induce proteotoxic stress, which is compatible with the observation that Xrp1-, but not Irbp18-overexpressing cells, are eliminated from mosaic tissues [29]. Overexpression of an inert protein, GFP, also did not upregulate markers of proteotoxic stress (S3E- S3H Fig), confirming that the effects of Xrp1 are not due to overexpression per se. We next asked whether Irbp18 is required for Xrp1 to generate proteotoxic stress. Again, using Gal80 ts to control transgene expression, we found that a 24h burst of irbp18-RNAi expression in wild-type discs was able to reduce xrp1 expression (S4A Fig), two-sided Mann-Whitney U-test for comparison of SSU and LSU in RpS3 +/-, Xrp1 +/wing discs), error bars represent 95% confidence interval. (L) Mean log fold change in SSU and LSU ribosomal proteins detected by mass spectrometry (n = 2) in RpS3 +/and RpS3 +/-, Xrp1 +/wing discs relative to wild-type discs, as indicated. In this figure and throughout: scale bars are 50μm; dashed white or yellow lines mark compartment boundaries; each data point on the scatter plots represents one wing disc or one wing disc compartment and the horizontal line represents the median; all n values refer to the number of individual wing discs except for Fig 1K and  suggesting that Irbp18 can be effectively knocked down over this time window. Under this regime, xrp1 and irbp18-RNAi expressing cells had lower levels of p62 labelled aggregates than xrp1 expressing cells (S4B-S4D Fig), suggesting that Irbp18 is, at least partially, required for Xrp1 to activate proteotoxic stress. This is consistent with the earlier observation that the elimination of Xrp1 overexpressing cells from mosaic tissues can be rescued by a mutation in Irbp18 [29]. Overall, this data indicates that Xrp1, along with Irbp18, is responsible for inducing proteotoxic stress and the oxidative stress response in Rp/+ cells, which may explain why removal of Xrp1 or Irbp18 so effectively rescues Minute competition.
If Xrp1 and Irbp18 are required in cell competition because they induce proteotoxic stress, then inducing proteotoxic stress by other means should lead to the loser status in an Xrp1-and Irbp18-independent manner. To test this hypothesis, we induced proteotoxic stress by wellestablished means. eIF2α is phosphorylated in response to proteotoxic stress, leading to global attenuation of translation [30,31]. However, sustained increase in p-eIF2α has also been  shown to induce proteotoxic stress, by causing accumulation of aggregogenic stress granules [38,39]. Therefore, we sought to induce high levels of p-eIF2α. Growth arrest and DNA-damage-inducible 34 (GADD34) is a Protein Phosphatase 1 (PP1) regulatory subunit, which causes p-eIF2α dephosphorylation by providing PP1 with target specificity for p-eIF2α [40]. As expected, GADD34-RNAi increased the levels of p-eIF2α (S5A and S5B Fig). GADD34-RNAi expression in the posterior compartment of wing discs also led to higher levels of p62 (Fig 4A  and 4B) and of mono-and poly-ubiquitinated proteins (detected by the FK2 antibody; Fig 4C  and 4D) than in the control anterior compartment. As these are both markers of protein aggregates [34,41], these data indicate that sustained eIF2α phosphorylation induces proteotoxic stress and protein aggregation. GADD34 knockdown also upregulated GstD1-GFP (Fig 4E  and 4F) and p-JNK (Figs 4G and S5C). Thus, increased levels of p-eIF2α are sufficient to induce proteotoxic stress, the oxidative stress response, and JNK pathway activity, all of which are observed in Rp/+ cells.
We then expressed GADD34-RNAi in a mosaic fashion to test whether it induces the loser status. GADD34-RNAi expressing cells were efficiently removed from wing discs in mosaic experiments (S5D Fig). Only a few fragments of cells remained, and these had been basally extruded from the epithelium (S5E Fig), consistent with competitive elimination. However, it was also possible that this was due to cell-autonomous activation of apoptosis. Thus, we designed an experimental strategy to obtain large GADD34-RNAi expressing patches of cells (S5F Fig) and directly compare the rate of apoptosis at the borders and centers of these patches, as increased border death is a hallmark of Minute cell competition [16,17,42]. We made use of Gal80 ts for conditional expression and placed larvae at the Gal80 ts permissive temperature after clone induction, to allow cells to expand without induction of transgene expression. We then activated GADD34-RNAi (and GFP) expression by moving larvae to the Gal80 ts restrictive temperature 24 hours before dissection (S5F Fig). This short period of GADD34-RNAi expression was sufficient to increase p-eIF2α (S5G Fig). Unlike control wild-type patches of cells, GADD34-RNAi expressing patches of cells had significantly higher levels of cell death at their borders than in the center, showing that they are subject to competitive elimination by wildtype cells (Figs 4H-4J and S6).
We next asked whether GADD34-RNAi induced cell competition depends on Xrp1. As Xrp1 and Irbp18 function upstream of proteotoxic stress in RpS3/+ cells (Fig 1), we were surprised to find that co-expression of xrp1-RNAi with GADD34-RNAi resulted in a strong rescue of competitive elimination (Fig 5A-5C). Thus, elimination of GADD34-RNAi expressing cells is mediated by Xrp1, suggesting that Xrp1 can also function downstream of proteotoxic stress. Altogether, these data show that Xrp1 functions both upstream and downstream of proteotoxic stress, suggesting that a feed-forward loop between proteotoxic stress and Xrp1 exists in Rp/+ cells.
Xrp1 knockdown also rescued the increased GstD1-GFP observed in GADD34-RNAi expressing compartments, bringing levels down to, or even slightly lower than, wild-type levels (Fig 5D-5F). Remarkably, Xrp1 knockdown was also able to partially rescue the increased p-eIF2α in GADD34-RNAi expressing compartments (Fig 5D, 5F and 5G), suggesting that removing Xrp1 breaks the feed-forward loop to proteotoxic stress, and therefore partially rescues the increased p-eIF2α levels in GADD34-RNAi expressing cells. Altogether these data suggest that Xrp1 is activated by proteotoxic stress. Consistently, we found that GADD34-RNAi the same genotype as in (A) immuno-stained for p62 (grey) (C) with quantification of p62 signal intensity (D) (n = 10; two-sided Wilcoxon signed-rank test). (E-F) A wing disc of the same genotype as in (A) immuno-stained for p-eIF2α (magenta) (E) with quantification of p-eIF2α signal intensity (F) (n = 8; two-sided Wilcoxon signed-rank test).
How might Xrp1 be induced downstream of proteotoxic stress? During ER stress, the UPR induces eIF2α phosphorylation, which mediates global translation repression and selective translation of a subset of transcripts, including that of ATF4, which, in mammals, mediates expression of chaperones and proapoptotic genes, including CHOP [30,31]. Although no clear mammalian Xrp1 homolog exists, sequence homology and functional data suggest that Xrp1 may be functionally homologous to CHOP [25,29,43]. Indeed, overexpression of ATF4 (also known as cryptocephal (crc) in Drosophila) was sufficient to upregulate both xrp1 transcription and GstD1-GFP (S7A and S7B Fig). These data therefore suggest that Xrp1 can be transcriptionally activated by the UPR in Drosophila. We then tested whether ATF4 translation is increased in RpS3/+ cells using the crc-5'UTR-dsRed translation reporter, which comprises the dsRed coding sequence   [44]. Activation of this reporter was clearly detected in GADD34-RNAi expressing cells (S7C and S7D Fig) but not in RpS3/+ cells (S7E and S7F Fig), suggesting that ATF4 translation is not activated in RpS3/+ cells. Furthermore, expressing a previously validated ATF4-RNAi line [45] did not rescue xrp1 transcriptional upregulation in RpS3/+ cells (S7G and S7H Fig). Therefore, this data suggests that, although ATF4 overexpression can activate Xrp1 transcription, ATF4 is not responsible for increasing xrp1 expression in Rp/+ cells.
Lastly, we investigated whether Xrp1 also plays a role in Nrf2-induced cell competition. We have previously shown that proteotoxic stress induces expression of the Nrf2 reporter GstD1-GFP [14] and that over-expression of nrf2 is sufficient to turn otherwise wild-type cells into losers [24]. Consistently, nrf2 expressing cells were readily eliminated from mosaic wing discs, with only a few tiny patches remaining at the time of dissection (Fig 6A and 6C). xrp1-RNAi significantly rescued the growth of nrf2 expressing patches of cells (Fig 6B and 6C), indicating that Xrp1 functions downstream of Nrf2. Irbp18 knockdown also rescued nrf2 expressing cells from elimination (Fig 6D-6F) confirming that Xrp1 functions, along with Irbp18, downstream of Nrf2. This suggests that in Rp/+ tissues, Xrp1 is activated both by increased p-eIF2α, and by Nrf2.

Discussion
We have provided evidence that a feed-forward loop between proteotoxic stress, Nrf2 and the Xrp1/Irbp18 complex is operational in RpS3/+ cells (including in the absence of cell immuno-stained with anti-β-galactosidase (magenta) (H), with quantification of xrp1-lacZ signal intensity (I) (n = 7; two-sided Wilcoxon signed-rank test).
https://doi.org/10.1371/journal.pgen.1009946.g005 competition) and contributes to reducing their fitness during cell competition (Fig 6G). Our data suggests that an imbalance between SSU and LSU Ribosomal proteins generates an initial source of proteotoxic stress, independently of Xrp1. This leads to xrp1 transcriptional upregulation, likely via p-eIF2α. As we have shown, Xrp1, together with Irbp18, generates further proteotoxic stress, in a feed-forward loop. This causes LSU ribosome proteins to accumulate, exacerbating the stoichiometric imbalance between LSU and SSU subunit components in RpS3/+ cells. Knockdown of Xrp1 or Irbp18 rescues proteotoxic stress in RpS3/+ cells, suggesting that this feed-forward loop is essential for build-up of proteotoxic stress and to reduce the competitiveness of Rp/+ cells. We note that during the revision of this manuscript two other independent studies have reported relevant and complementary findings [46,47]. Nrf2 is also activated by proteotoxic stress and contributes to this feedback loop, either independently of p-eIF2α (as illustrated in Fig 6G), or downstream of p-eIF2α. Our data cannot distinguish between these two possibilities.
Our data indicate that xrp1 upregulation is likely mediated by increased p-eIF2α levels. p-eIF2α accumulates in Rp/+ cells [13,14], and increasing p-eIF2α in wild-type cells (by knocking down GADD34) leads to increased xrp1 transcription (Fig 5), suggesting that p-eIF2α does, at least partially, contribute to xrp1 transcription in Rp/+ cells. p-eIF2α induces many transcriptional targets via stabilization of the transcription factor ATF4 [30,31]. This suggested that ATF4 may activate xrp1. Consistent with this, we found that ATF4 overexpression is sufficient to upregulate an xrp1 transcriptional reporter in wing disc cells. However, we were surprised to find that xrp1 upregulation does not seem to depend on ATF4 in RpS3/+ cells. Indeed, ATF4 knockdown did not reduce xrp1 transcription in RpS3/+ cells. Furthermore, we were unable to detect stabilization of ATF4 in RpS3/+ cells using a translational reporter (S7 Fig). These observations suggest that p-eIF2α upregulates xrp1 transcription in Rp/+ cells by an unknown, ATF4 independent, mechanism. Alternatively, the role of ATF4 may be masked by other inputs onto the xrp1 promoter. For example, ATF4 knockdown could increase proteotoxic stress in Rp/+ cells, by inhibiting the UPR, and this may upregulate other pathways that act on the xrp1 promoter, thus masking any effect of ATF4 knockdown. This mechanism could involve Nrf2, since Nrf2 is also induced by proteotoxic stress [14] and since we have shown that Nrf2 induces cellular toxicity via xrp1 (Fig 6). However, it is also possible that other factors activate Xrp1 in Rp/+ cells.
Nrf2 plays a pro-survival role in many contexts, by activating a battery of genes that enable the metabolic adaptation to oxidative stress [23]. It is therefore counterintuitive that Nrf2 overexpression should induce the loser status and, at high expression levels, cell death [24]. Our work suggests that the toxicity of Nrf2 is at least in part due to Xrp1 function, as elimination of Nrf2 expressing cells is rescued by Xrp1 knockdown. Whether additional Nrf2 target genes contribute to the loser status remains to be established.
Besides Xrp1 or Irbp18 knockdown, the only other condition known thus far to rescue xrp1 transcriptional upregulation in Rp/+ cells is an RpS12 point mutation, RpS12 97D [20,48]. However, the mechanism by which RpS12 affects xrp1 transcription remains elusive. It will be important in future work to establish whether RpS12 mutations rescue xrp1 transcriptional activation upstream or downstream of proteotoxic stress.
Our results provide compelling evidence that Xrp1 and Irbp18 are responsible for inducing proteotoxic stress in RpS3/+ cells. Firstly, knockdown of Xrp1 or Irbp18 rescues the accumulation of p62 labelled aggregates and rescues the increased p-eIF2α in RpS3/+ cells (Fig 1). Secondly, overexpression of Xrp1 is sufficient to upregulate markers of proteotoxic stress in wildtype cells (Fig 3). Third, the presence of Xrp1 in RpS3/+ cells worsens the imbalance of Ribosomal proteins, causing LSUs to accumulate (Fig 1). It will be crucial in future work to identify the relevant targets of Xrp1 that cause proteotoxic stress in Rp/+ cells. Xrp1 may alter expression of a single target, for example a gene encoding a component or regulator of the autophagy or proteasomal systems, which deregulates cellular proteostasis. Alternatively, several target genes may contribute to enhancing proteotoxic stress: if several subunits of multiprotein complexes are deregulated by increased Xrp1, this could lead to unassembled complexes, increasing the burden on the cellular degradation machinery in already stressed Rp/+ cells. There may also be Xrp1 targets that contribute to the loser status without affecting proteotoxic stress. It is remarkable that, in addition to rescuing competitive elimination of Rp/+ cells, loss of Xrp1 can rescue elimination of mahj deficient cells (Fig 2) and Nrf2 overexpressing cells (Fig 6). In mahj deficient cells, loss of Xrp1 was able to rescue the upregulation of p-eIF2α, suggesting that Xrp1 also promotes proteotoxic stress in mahj cells. It will be interesting to establish whether this is the case for Nrf2 expressing cells.
Xrp1 has been shown to play a role in a Drosophila model of Amyotrophic lateral sclerosis (ALS), a debilitating and lethal neurodegenerative disorder that can be caused by aggregogenic mutations in genes encoding RNA binding proteins, including TDP-43 and FUS, a member of the FET family of proteins [49]. TDP-43 and FUS also form cytoplasmic, ubiquitinated aggregates, in several other neurodegenerative disorders [50]. Drosophila cabeza (caz) is the single ortholog of the human FET proteins. Xrp1 is upregulated in caz mutants, and the pupal lethality, motor defects and dysregulated gene expression of caz mutants is rescued by xrp1 heterozygosity [51]. Therefore, it is possible that the feed-forward loop we have uncovered is also active in this context: formation of cytoplasmic proteotoxic aggregates could stimulate xrp1 expression, which could then induce further proteotoxic stress in a feed forward loop, resulting in neuronal toxicity. Understanding the relationship between Xrp1, proteotoxic stress and oxidative stress may thus be beneficial for the study of human proteinopathies.

Fly husbandry
Fly food composition is: 7.5g/L agar powder, 50g/L baker's yeast, 55g/L glucose, 35g/L wheat flour, 2.5% nipagin, 0.4% propionic acid and 1.0% penicillin/streptomycin. Eggs were collected for 24 hours in a 25˚C incubator and experimental crosses were then maintained in either an 18˚C incubator, a 25˚C incubator, or in a water bath set to a specific temperature. Mosaic wing discs were generated with the hs-FLP transgenic line by heat shocking crosses three days after egg laying in a 37˚C water bath. For experiments using temperature sensitive Gal80 (Gal80 ts ) to control the timing and level of transgene expression, conditions were optimized for each experiment. All experimental conditions are listed in the Genotypes Table (Table 1). All Drosophila strains used are listed in the Key Resources Table (Table 2). Wing discs were dissected from wandering third instar larvae. For all experiments, egg collections, heat shocks, temperature shifts, dissections, and imaging were done in parallel for control and experimental crosses. All experiments comparing the effect of one UAS driven transgene to that of two UAS driven transgenes included an additional blank UAS, 40D UAS , to control for possible Gal4 titration effects, except for the experiment in S4B-S4D Fig, where 40D UAS was not included due to the complexity of the genetic crosses. For mosaic competition experiments, all dissected larvae were of the same sex for both the control and experimental crosses. For half-half experiments, where the anterior compartment and posterior compartment were compared, sexes were not differentiated.

Immunostaining
Wandering third instar larvae were dissected in phosphate buffered saline (PBS) and hemi-larvae were fixed in 4% paraformaldehyde for 20 minutes at room temperature. Tissues were permeabilized with three 10-minute washes in PBST (0.25% triton in PBS) and blocked for 20 minutes in blocking buffer (4% fetal calf serum in PBST). Samples were incubated with primary antibodies diluted in blocking buffer at the concentration indicated in the Key Resources Table (Table 2) overnight at 4˚C. Samples were washed three times in PBST for 10 minutes and incubated with secondary antibodies and Hoechst diluted in blocking buffer at the concentration indicated in the Key Resources Table (Table 2) for 45-minutes at room temperature. After a further three 10-minute washes in PBST, wing discs were dissected from hemi-larvae and mounted in Vectashield (Vector laboratories) on borosilicate glass sides (no 1.5, VWR international).

Proteomics
Sample preparation and Tandem Mass Tag (TMT) mass spectrometry were performed as described in [14].

Image acquisition and processing
Images were acquired using a Leica SP8 confocal microscope with a 40x 1.3 NA P Apo Oil objective. Wing discs were imaged as z-stacks with each section corresponding to 1μm. Images were processed using Photoshop (Adobe Photoshop 2020) and Fiji (Version 2).

Quantifications
Pouch coverage, cell death quantifications and fluorescence intensity quantifications were carried out using custom built Fiji scripts. All analysis focused on the pouch region of the wing disc. For measurements of pouch coverage, the percentage of the volume of the pouch occupied by GFP-positive cells was determined. For cell death quantifications the border is defined as any cell within a 2 cell-range of the boundary of the GFP-positive patch. Cell death measurements were normalized to the respective volume of the GFP-positive patch border or center, as measured in Fiji. For all scatter plots the horizontal line represents the median.

Statistics and reproducibility
All data represented by the scatter plots including details of the specific statistical test used for each experiment are provided (S1 Data). Statistics were performed using GraphPad Prism (Prism 8). Univariate statistics were used to determine P-values. The statistical tests used were the Mann Whitney U-test for non-paired data, and the Wilcoxon matched-pairs signed rank test for paired data. P-value corrections for multiple comparisons were not considered due to the low number of comparisons. For experiments comparing across wing discs a minimum of three biological repeats were performed. For experiments with an internal control, a minimum of two biological repeats were performed. Experiments performed to validate reagents (e.g., testing efficacy of RNAi lines) were carried out at least once.   This source data file includes all raw data and analysis presented in Figs 1B, 1C,  1E, 1F, 1H, 1J, 1K, 1L, 2C, 2F, 3B, 3D, 3F, 4B, 4D, 4F, 4H, 5C, 5E, 5G, 5I, 6C, 6F, S1B, S1D,  S1G, S1J, S2B, S2C, S2E, S2F, S3B, S3D, S3F, S3H, S4D, S5B, S5C, S7B, S7D, S7F and S7H. (XLSX)