Xrp1 is a transcription factor required for cell competition-driven elimination of loser cells

The elimination of unfit cells from a tissue is a process known in Drosophila and mammals as cell competition. In a well-studied paradigm “loser” cells that are heterozygous mutant for a haploinsufficient ribosomal protein gene are eliminated from developing tissues via apoptosis when surrounded by fitter wild-type cells, referred to as “winner” cells. However, the mechanisms underlying the induction of this phenomenon are not fully understood. Here we report that a CCAAT-Enhancer-Binding Protein (C/EBP), Xrp1, which is known to help maintaining genomic stability after genotoxic stress, is necessary for the elimination of loser clones in cell competition. In loser cells, Xrp1 is transcriptionally upregulated by an autoregulatory loop and is able to trigger apoptosis - driving cell elimination. We further show that Xrp1 acts in the nucleus to regulate the transcription of several genes that have been previously involved in cell competition. We therefore speculate that Xrp1 might play a fundamental role as a molecular caretaker of the genomic integrity of tissues.

to the removal of one copy of a haploinsufficient ribosomal protein gene, where, similarly to C/EBP homologs, it regulates its own expression via a positive autoregulatory loop, the expression of pro-apoptotic genes and that of other genes that were previously implicated in cell competition. In order to identify genes whose function is necessary for the elimination of RPG heterozygous mutant loser cells, we performed a forward genetic screen using ethyl methanesulfonate (EMS) in Drosophila melanogaster. We designed a mosaic system that allows direct screening through the larval cuticle for the persistence of otherwise eliminated RpL19 +/− loser clones (Fig. 1A). This enabled us to screen a high number of animals for mutations that either dominantly (anywhere in the genome) or recessively (on the right arm of the third chromosome) suppress cell competition. The induction of a single somatic recombination event between two FLP recognition targets (FRTs) generates a RPG heterozygous mutant cell that becomes homozygous for the mutagenized right arm of the third chromosome. Loser clones are induced at the beginning of larval development (L1). If no suppressive mutation is present, clones are efficiently eliminated over time and thus undetectable by the end of the third instar larval stage (L3) when the screening is performed (Fig. 1A). We screened 20,000 mutagenized genomes for the presence of mutations that prevent the elimination of loser clones. We retrieved 11 heritable suppressors (Fig. 1C) and focused our attention on three of the strongest suppressors that did not display any obvious growth-related phenotype. Figure 1B shows representative living larvae that were analyzed for the presence of RpL19 +/− GFP clones in the wing discs. RpL19 +/− clones are eliminated and little or no signal is observed. Their elimination, however, is prevented when cells are not heterozygous mutant for RpL19 or when different Xrp1 mutations (Xrp1 08 in the example) are additionally present. In the latter cases GFP signal is observed in wing discs.
Xrp1 suppressors did not belong to a lethal complementation group and the causative mutations were identified using a combination of positional mapping and whole-genome re-sequencing. In particular, three independent mutations in the introns of CG17836/Xrp1 were identified, all caused by substitutions of single nucleotides (Fig. 1C,D). These nucleotides are conserved within the Drosophila genus and inspection of the alignment revealed an embedment of these nucleotides in conserved sequence motifs (Fig. S1). Of particular interest are the polypyrimidine motifs containing the nucleotide mutations in Xrp1 20 and Xrp1 08 . These motifs flank the alternative first exon and are potential splice regulators. The CTCTCT motif in proximity of the 5′ splice site of Xrp1 has been identified as a putative intronic splicing enhancer (ISE) predicted to serve as binding site for the polypyrimidine-tract binding protein (PTB) splicing regulator 27 . The presence of these motifs prompted us to investigate the consequences of the Xrp1 08 allele on exonic junctions. The most prominent effect of this allele is a strong and consistent reduction in the expression of two similar Xrp1 transcripts, RC and RE (Fig. S1), which only differ in the composition of their 5′ UTRs. They share the transcriptional start site and contain the same long open reading frame that codes for the short isoform of Xrp1 (Fig. S1). We then checked the behavior of RpL19 +/− clones in the presence and absence of Xrp1 function. To this end we used the twin spot MARCM system, which enables us to differently mark twin clones generated by the same recombination event. In our genetic set up, mCherry expression marks loser clones whereas two copies of GFP mark wild-type twin clones ( Fig. 2A). As expected, RpL19 +/− loser clones are eliminated from the tissue (Fig. 2B). Elimination is also observed when RpL19 +/− cells within these clones are additionally mutant for Xrp1 08 but contain a transgene comprising the genomic region of Xrp1 (Fig. 2B'). Importantly, when Xrp1 mutations are not rescued cell competition-driven elimination of RpL19 +/− losers no longer occurs. In particular, we show that the Xrp1 08 intronic mutation retrieved from the EMS screen is able to prevent loser cell elimination (Fig. 2B") and that a similar result is obtained with a newly generated complete loss-of-function allele, Xrp1 61 ( Fig. 2B"'), as well as with Xrp1 26 (Fig. S4). Xrp1 61 contains a frame shift mutation upstream of the Xrp1 basic region-leucine zipper domain (b-ZIP), and is considered a null allele. Like other Xrp1 alleles analyzed it is homozygous viable and does not impair the development of mutant animals. To confirm that Xrp1 function is of general importance for the elimination of hRPG +/− cells, and not limited to RpL19 +/− loser cells, we tested the effect of Xrp1 mutations on RpL14 +/− loser clones (Fig. S2). Similarly to RpL19 +/− cells, RpL14 +/− cells are normally eliminated from wing discs during larval development. No elimination occurs if these cells express RpL14 from a transgene, or when Xrp1 is mutated (Xrp1 61 ) (Fig. S2).
Since Xrp1 is transcriptionally induced in response to various forms of stress [19][20][21][22] and since Xrp1 has been found to be upregulated in RpS3 +/− wing discs when compared to WT discs 25,28 , we hypothesized that its expression is induced in loser clones as a result of the loss of a haploinsufficient ribosomal protein gene. We therefore used a transcriptional reporter for Xrp1 -Xrp1 02515 , containing a lacZ P-element 20 -and found that Xrp1 expression is indeed upregulated in RpL19 +/− cells, indicating that the upregulation of Xrp1 might play a crucial early role in the elimination of loser cells (Fig. 3A-A'). In line with the recent report by Lee et al. 25 we found that Xrp1 is upregulated in wing discs that are lacking one copy of a ribosomal protein gene, indicating that Xrp1's role in cell competition does not depend on clonality (Fig. S3). In order to gain insights into this function we conditionally forced the expression of Xrp1 in the posterior half of the wing discs and observed a massive induction of apoptosis, as revealed by anti-cleaved caspase 3 staining (Fig. 3B,C).
Interestingly, unlike loss of Xrp1, blocking apoptosis by means of overexpression of dIAP1 or p35, or by abrogating the function of Dronc or Hid, does not fully suppress RpL19 +/− cell elimination, suggesting that Xrp1 does more than merely induce apoptosis. Only the co-overexpression of CycE, which promotes cell cycle entry, together with dIAP1, which suppresses apoptosis, lead to a degree of suppression of RpL19 +/− cell elimination comparable to that obtained with Xrp1 loss-of-function mutations (Fig. S4). This indicates that the combined effects of blocking cell cycle progression and promoting apoptosis are critical for the elimination of RPG +/− cells. Given the strength of the effect of Xrp1 mutations, Xrp1 may therefore additionally hinder cells to progress through the cell cycle. This is in line with Akdemir et al. 20 who found that Xrp1 expression induces cell cycle arrest in cultured Drosophila cells. Since Xrp1 possesses a sequence-specific DNA binding domain (Fig. S5), either one or both of these cellular functions might be directly regulated at the transcriptional level.
To further explore this notion we set out to identify direct genomic targets of Xrp1 by chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) on wing imaginal discs 29 . In order to do this, we induced Xrp1 expression in wing discs. The top targets revealed by ChIP-seq comprise a number of genes that are already implicated in cell competition, cell cycle regulation and apoptosis 7,30-32 . Figure 4A shows a list of the most interesting genes that are bound by Xrp1. Among these we identified Xrp1 itself, suggesting the existence of a potential autoregulatory loop. To test this notion we overexpressed Xrp1 in the posterior compartment of the wing disc and checked the transcriptional behavior of Xrp1 with the aforementioned Xrp1-lacZ reporter. We observed the upregulation of lacZ expression in response to Xrp1 overexpression, indicating that Xrp1 can boosts its own expression in a positive autoregulatory loop (Fig. 4B-B'). We confirmed these observations by measuring mRNA levels of Xrp1 upon forced Xrp1 expression (Fig. 4C). With a similar strategy we also checked the response of other putative transcriptional targets from our ChIP-seq experiment. We could show that Xrp1 promotes the transcription of Dif (Fig. 4D-D' ,F), a Drosophila NFkB homolog gene that has previously been implicated in the cell competition-dependent induction of apoptosis via the induction of rpr transcription 7 . We also tested puc, Upd3, Nedd4 and rad50: all of these genes were upregulated upon induction of Xrp1 expression (Fig. 4F). puc, Upd3 and Nedd4 are involved in the JAK/STAT and Hippo signaling pathways, both of which have previously been implicated in cell competition 10,13,14,28,31 . Rad50 is instead required for double strand break repair 33 .
The most prominent sequence motif of Xrp1 derived from ChIP-seq data shows a strong similarity with the b-ZIP binding motif of the human C/EBP protein family. We therefore checked whether Xrp1 shows homology to C/EBP transcription factors, being itself a bona fide transcription factor. We found that Xrp1 shares a 40% identity with the human C/EBPs (PSI-BLAST). Phylogenetic reconstruction allowed us to recognize three Drosophila C/ EBP homologs, one of which is Xrp1 (Fig. S5). Interestingly, human C/EBP-alpha is retained in the nucleolus and binds to ribosomal DNA 34 , a feature that may be evolutionarily conserved since Xrp1 binds rDNA loci with high affinity (Fig. 4A). The encoded rRNA is found in the nucleoli.
We therefore propose a working model in which Xrp1, under normal conditions, sits on rDNA in the nucleolus. In the presence of genotoxic stress or of a ribosomal imbalance, as in the context of Minute cell competition, Xrp1 acts nuclearly as a C/EBP transcription factor that stimulates its own transcription and the expression of pro-apoptotic target genes (Fig. 5). When intermingled with wild-type cells, cells with only one copy of an hRPG are eliminated in a Xrp1-dependent manner. In our experimental system the deletion of one copy of the RpL19 gene is catalyzed by the Flp/FRT recombination system, which leaves no apparent lesion in the DNA 35 . Therefore, the initial recruitment of Xrp1 into the nucleus may not depend on DNA damage per se, but rather on the unbalanced physiology of the cell resulting from the loss of one copy of the hRPG. The nucleolus is the site of ribosome biogenesis and a major stress sensor organelle 36 . RpL19 +/− cells experience a related nucleolar stress, since their nucleoli are enlarged as revealed by anti-fibrillarin staining (Fig. S6). The most likely explanation for this is partially stalled ribosome assembly 37 . Since genotoxic stress triggers Xrp1 expression (Fig. S6B,C), we speculate that Xrp1 acts as a caretaker of genomic integrity. In support of this hypothesis, the growth of salvador −/− mutant tumor clones is suppressed by the concurrent loss of one copy of the RpL19 gene. However, this suppression fails in the absence of Xrp1 function (Fig. S6), indicating that the presumptive protective function that RPGs haploinsufficiency provides can also operate within tumorous cells. In addition, according to our Monte-Carlo simulation, the likelihood that one hRPG locus becomes heterozygous mutant before any other gene gets mutated to homozygosity is very high (Fig. S6). Together with the observation that hRPGs are broadly distributed within the genome 38 (Fig. S6), this further supports the potential role of Xrp1 as a caretaker of genomic integrity. Although further research is required to better elucidate this phenomenon, we nevertheless propose that RPG haploinsufficiency provides a simple, yet effective, mechanism to protect the organism from the emergence of potentially deleterious cells.    49 and inserted into the attP landing site ZH-attP-68E. The Xrp1 mutated genomic rescue was generated by inserting 5 bp (C > GATCCC at 3R:18925226 Dmel_r6.08) at the beginning of the second coding exon in the wild-type genomic fragment, which shifts the frame of all Xrp1 isoforms. Transgenesis was performed according to standard germ-line transformation procedures.

Materials and Methods
Mutagenesis and screen. EMS screens were performed according to standard procedure 50 . y w hs-FLP; M{3xP3-RFP.attP}ZH-36B; FRT82B starter line was first isogenized for the 3R cell competition screen. Isogenized males were fed with a 25 mM, 1% sucrose solution and crossed to tester virgin females. RpL19 +/− clones were induced in the resulting progeny. A total of 20,000 F1 larvae were screened for the persistence of RpL19 +/− GFP positive clones at the end of the third instar larval stage. 182 larvae showed persistence of GFP clones clearly above background noise. 125 of them gave rise to fertile adults and were further rescreened. 12 heritable suppressors were doubly balanced. For the Xrp1 coding sequence directed mutagenesis y w;; Xrp1 GS18143 /TM3.Sb males were fed with a 50 mM, 1% sucrose solution and crossed to tester virgin females y w ey-FLP; Act > y + > GAL4-w; M{3xP3-RFP.attP}ZH-86Fb. 10,000 F1 genomes were screened and 8 heritable suppressors were retrieved and balanced. A mutation in the Xrp1 coding region was identified in 5 of them. After the causative mutation was identified the upstream P{GSV6}Xrp1 GS18143 was removed using P element transposase and precise excision events were selected (direct sequencing of PCR amplicons) and recombined onto a FRT82B chromosome for clonal analysis. RpL19 knock-out was generated by mobilizing the P element P{lacW}RpL19 k03704 , imprecise excisions were selected based on the presence of the characteristic Minute bristle phenotype and the absence of the white + marker. The RpL19 IE-C5 1.09 kbp deletion (2R:24968426..24969517 Dmel_r6.08) was selected and characterized using direct sequencing of PCR amplicons. This specific excision removes RpL19 coding sequence and leaves neighboring genes unaffected. RpL19 +/− loser clones for in vivo screen were generated as follows: y w hs-FLP; M{3xP3-RFP.attP}ZH-36B; FRT82B mutagenized males were crossed to y w UAS-mCD8::GFP, hs-FLP; salE-GAL4, Df(2R)M60E; FRT82B, tubP-GAL80, M{RpL19 genomic}ZH-86Fb/ SM5a-TM6B tester virgin females.
Mapping the mutations. We initially mapped cell competition suppressors through meiotic recombinations coupled with DHPLC (Denaturing High-Performance Liquid Chromatography) for PCR amplicon analysis. The interval containing the suppressors Xrp1 08 and Xrp1 29 was narrowed down to a 106.5 Kb interval (3R:18872668..18979166 Dmel_r6.08). Sanger sequencing of the coding regions in this interval did not reveal the presence of any mutation. We then performed whole-genome sequencing on Xrp1 08 , Xrp1 20  3R:18920194 Dmel_r6.08), Xrp1 29 (G > A 3R:18921450 Dmel_r6.08). Other suppressors were roughly mapped to the second chromosome or to one of the arms of the third chromosome as indicated in the test complementation table. Minute mutants were identified on the basis of their characteristic bristle phenotype and developmental delay. warts and salvador mutants were identified on the basis of their clonal overgrown phenotypes and failure to complement independent loss of function alleles (warts m72 and sav 4 ). Note that for sup 88 the suppressive mutation is the Minute on the second chromosome and not the mutation in the salvador gene. Xrp1 suppressors isolated from the coding sequence directed mutagenesis were identified by direct sequencing of PCR amplicons: qRT-PCR. qRT-PCR was performed according to standard protocol. RNA was extracted with TRIzol Reagent and genomic DNA was digested with the Ambion DNase kit. RNA was isolated from third instar wing imaginal discs with the exception of the reaction to evaluate the expression of the different splicing variants in WT and Xrp1 08 mutant conditions. In this experiment we used the following primers (primer sequences are oriented 5′ to 3′). ChIP-seq preparation and analysis. Wing imaginal discs expressing HA-tagged Xrp1 (FlyORF-F000655) 48 were mass isolated and sorted, chromatin was immunoprecipitated and DNA libraries were prepared according to standard protocol 29 . Rabbit anti-HA ChIP grade antibody (ab9110, Abcam) was used. Libraries were sequenced on the Illumina HiSeq. 2500 v4 (Functional Genomics Center of the University of Zurich). Bowtie 2 (version 2.0.0-beta6) 51 was used to align the sequencing reads using default parameters. The dm3 Drosophila genome annotation was used as reference. The program findPeaks.pl with default parameters was used to identify enriched regions compared to the untreated control sample. The program MotifsGenome.pl (with the option size = 75) was used to identify predominant motifs. The sequence logo was generated with the PWM-Tools web interface (http://ccg.vital-it.ch/pwmtools/) from the SIB using HOMER's position frequency matrix output file.
Wing discs were imaged using a Leica LSM710 confocal microscope. Quantification and statistics. Statistical analyses were performed in Graphpad Prism 7 or Microsoft Excel. Depending on the distribution of data, t-test or Mann-Whitney test were used, unless differently indicated. Regarding RpL19 +/− loser clones for dissections and clone size quantification, we undertook a stringent comparative analysis based on the ratios between the areas of loser (mCherry) and winner (GFP 2+ ) clones. Areas were quantified with FIJI. We applied standardized statistical tests (Mann-Whitney test). In addition, we reasoned that a genuine suppressor of RPG mutant cell elimination should not only increase the mean size of RPG mutant clones but also restore a normal distribution of RPG mutant clones (in this case statistical analysis was performed by using the D' Agostino & Pearson normality test). For RpL14 +/− loser clones, GFP area was measured with FIJI and Mann-Whitney test was applied. Signal intensity calculation for Xrp1 targets was performed in FIJI with the mean gray intensity measurement tool. Statistical significance was calculated with a paired-ratio t-test.
Identification of Xrp1 homologs. In a heuristic approach, two iterations of PSI-BLAST 52 were performed using the bZIP domain of Css as a query. The COBALT constraint-based multiple protein alignment tool provided on the BLAST interface 53 was used to align all Drosophila Xrp1 protein sequences with the human C/EBPs family members identified with the PSI-Blast search. In a non-heuristic approach, BZip containing proteins from human and D. melanogaster were searched, aligned and trimmed according to the bZIP_2 motif from Pfam (PF07716) using probabilistic hmmer profiles 54 (hmmer.org). The resulting alignment was visualized with CLC Main Workbench and then used for phylogenetic reconstruction using the PhyML algorithm 55 with LG substitution models 56 , SPR topological rearrangements 57 and 100 bootstrap replicates. Phylogenetic tree was then mid-point rooted and displayed with the iTOL online tool 58 .
Drosophila RPGs map/gene density. Gene coordinates for each chromosome arm were retrieved using the cytosearch tool of Flybase. Gene positions were considered as the middle point between the start and the end of each gene. Gene density was calculated for 40 kbp bins and the final map was visualized using the radar chart type of Excel. The percentage of intragenic sequences was calculated as the complement of the total size of the genome minus the sum of the intergenic sequences downloaded from Flybase (Genome, FTP, r6.1).

Monte-carlo simulation for RPGs as caretakers of genomic integrity. The computational model
was realized with Phyton (detailed code is provided with the supplementary material). The Monte-carlo simulation was designed to determine the probability that a certain number of different genes is disturbed (both alleles are mutated) when a certain number of random mutations occur. It is assumed that each allele has the same probability to be hit by a mutation and that each mutation hits an allele.