Accumulation of differentiating intestinal stem cell progenies drives tumorigenesis

Stem cell self-renewal and differentiation are coordinated to maintain tissue homeostasis and prevent cancer. Mutations causing stem cell proliferation are traditionally the focus of cancer studies. However, the contribution of the differentiating stem cell progenies in tumorigenesis is poorly characterized. Here we report that loss of the SOX transcription factor, Sox21a, blocks the differentiation programme of enteroblast (EB), the intestinal stem cell progeny in the adult Drosophila midgut. This results in EB accumulation and formation of tumours. Sox21a tumour initiation and growth involve stem cell proliferation induced by the unpaired 2 mitogen released from accumulating EBs generating a feed-forward loop. EBs found in the tumours are heterogeneous and grow towards the intestinal lumen. Sox21a tumours modulate their environment by secreting matrix metalloproteinase and reactive oxygen species. Enterocytes surrounding the tumours are eliminated through delamination allowing tumour progression, a process requiring JNK activation. Our data highlight the tumorigenic properties of transit differentiating cells.

M aintenance of tissue homeostasis in the adulthood requires precise coordination of stem cell renewal and differentiation. Deregulation of these processes can lead to cancer. Stem cells live in a microenvironment and continuously receive signals from neighbouring heterologous cells composing the niche 1 . Stem cell niches are complex, heterotypic and dynamic structures 2 . Over the past few years, considerable progress has been made in elucidating how different niche factor promotes stem cell maintenance during homeostasis and contributes to tissue regeneration upon damage 3,4 . Stem cells usually divide asymmetrically to generate a self-renewing stem cell and a differentiating progenitor (or transit amplifying cell), which will eventually generate differentiated cells 5 . Recent studies in flies and mammals have begun to establish that these differentiating progenitors are not simply a passive intermediate between stem cell and differentiated cells, but play active roles in regulating stem cell activity and regeneration [6][7][8][9][10] .
Loss of proper differentiation is an important feature and likely a driver of cancer development 11 . Historically, mechanistic studies of human cancers and regenerative medicine have focused almost exclusively on stem cells 2,12 . The roles of the differentiating stem cell progeny in tumorigenesis remain largely unexplored 3,10 . In this study, we analyse how a defect in the differentiating program of stem cell progenies leads to tumours in the adult Drosophila intestine.
The adult intestine is continuously replenished by multipotent intestinal stem cells (ISCs) both in flies and mammals 4,12,13 . In the Drosophila midgut, ISCs differentiate into either large absorptive enterocytes or secretory enteroendocrine cells. This process involves an intermediate differentiating cell called the enteroblast (EB; Fig. 1a) [14][15][16] , analogous to the transit amplifying cell in mammalian intestines 17 . In this study, we show that Sox21a, a gene encoding a transcription factor of the SOX family, is required for EBs to become fully differentiated cells. Flies lacking Sox21a are viable but progressively develop intestinal tumours composed mainly of EBs. Using Drosophila genetics, we have provided a comprehensive dissection of cell-cell interactions that underlie the EB tumour initiation and progression as a result of this differentiation defect. Our data highlight a driving role of differentiating stem cell progenies in tumorigenesis. While the implication of stem cells in cancer has been the focus of intensive research, our data pinpoint the tumorigenic properties of transit differentiating cells. We speculate that the plasticity of these differentiating progenitors underlies their cancerous properties.

Results
Sox21a is necessary for EB differentiation. In an RNA interference (RNAi) screen for factors regulating stem cell differentiation, we identified Sox21a, a gene encoding a transcription factor of the SOX family with homologues implicated in stem cell regulation in mammals 18 . Silencing Sox21a with two independent RNAi constructs specifically in EBs using the conditional, temperature-sensitive Su(H)GBE-Gal4 UAS-GFP tub-Gal80 TS system (hereafter referred to as GBE TS ) 19 led to the accumulation of EBs in the adult (Fig. 1b,c;  Supplementary Fig. 1a-c). Consistent with this, Sox21a is specifically enriched in the midgut of adult Drosophila (Supplementary Fig. 1d). Moreover, examination of the cis-regulatory sequences of the Sox21a gene also revealed an intronic enhancer that drives reporter expression in both ISCs and EBs (referred to as progenitors; Supplementary Fig. 1e-h). The expression pattern driven by this enhancer is homogenous from the anterior to the posterior midgut, and is identical to the expression of Escargot (Esg), a transcription factor with well-defined expression in progenitors 20 .
To further study the function of Sox21a, we have generated two Sox21a mutations using CRISPR/Cas9-mediated genome editing 21 . Both mutants carry a small deletion in the DNA-binding domain of Sox21a, the HMG domain, resulting in reading frameshift and premature stop (Fig. 1d). Thus, these alleles should be considered as null alleles. Strikingly, Sox21a mutant flies are viable and fertile with no apparent defects. To confirm the role of Sox21a in EB differentiation, we performed lineage tracing using mosaic analysis with a repressible cell marker technique (MARCM) 22 . While the wild-type clones (positively marked by green fluorescent protein (GFP)) contain both enterocytes and enteroendocrine cells, cells in Sox21a mutant clones along the whole midgut remained undifferentiated, as revealed by the absence of GFP-positive cells expressing the enterocyte marker Pdm1 or the enteroendocrine cell marker Prospero (Fig. 1e,f). This differentiation defect is rescued by overexpressing Sox21a in the mutant clones (Fig. 1g). Quantification of clone size indicated that the Sox21a mutation reduces ISC division with a stronger effect in the posterior compared with the anterior midgut (Fig. 1h). Sox21a mutant clones generated in the posterior midgut barely grew, indicating a mandatory function of Sox21a for ISC division in the posterior midgut. In contrast, the existence of large Sox21a mutant clones in the anterior midgut indicates that Sox21a promotes to some extent stem cell division in this region but is less essential. The presence of a wild-type copy of Sox21a in the mutant clones also restored normal ISC division (Fig. 1h). To further confirm the differential impact of Sox21a on ISC division in the anterior and posterior regions, we compared ISC proliferation rate in wild-type and Sox21a mutant flies overexpressing the JAK/STAT ligand unpaired 2 (Upd2) in the enterocytes with Myo1A-Gal4 UAS-GFP tub-Gal80 TS system (hereafter referred to as Myo1A TS ) 23 . Unpaired are secreted proteins that have been shown to be potent inducers of ISC proliferation by activating JAK/SAT signalling in ISCs 23,24 . Overexpressing Upd2 strongly increased the number of mitotic ISCs in both the anterior and the posterior midgut in wild-type flies as revealed by the phospho-Histone H3 (PH3, a mitotic marker) count. In contrast, overexpressing Upd2 only increased the mitotic index in the anterior midgut of Sox21a mutant (Supplementary Fig. 4o,p). Collectively, our data show that Sox21a is essential for the differentiation of EBs into mature intestinal cells along the entire midgut. Its effect on ISC proliferation is more pronounced in the posterior midgut compared with the anterior midgut.
Overexpression of Sox21a drives EB differentiation. Since Sox21a is required to generate differentiated cells, we hypothesized that overexpressing this factor might force the progenitor cells to differentiate into mature intestinal cells. To test this, we created transgenic lines that enable its overexpression via the GAL4/UAS system ( Supplementary Fig. 1g). Strikingly, overexpressing Sox21a in the progenitor cells with esg-Gal4 UAS-GFP tub-Gal80 TS system (hereafter referred to as esg TS ) 15 was sufficient to induce their differentiation into enterocytes and cause the loss of progenitors (Fig. 2a-c). Although Sox21a is required for the differentiation of both enterocytes and enteroendocrine cells, Sox21a overexpression induced progenitors to differentiate into enterocytes (Pdm1 positive and polyploid), but not enteroendocrine cells (Pros positive; Supplementary Fig. 2a,b). To further delineate the role of Sox21a in progenitor differentiation, we overexpressed Sox21a either specifically in EBs using GBE TS or in ISCs using Dl-Gal4 UAS-GFP tub-Gal80 TS (Dl TS ) 19 . EBs were normally detected as small-nucleated cells with a polarized cell shape (Fig. 2d).
Overexpressing Sox21a in EBs for 4 days transformed most of them into large polyploid and round-shaped cells, indicative of a transformation into enterocyte (Fig. 2e,f). In contrast, overexpressing Sox21a with Dl TS in ISCs for 6 days did not induce ISC differentiation ( Supplementary Fig. 2c,d). Using other insertions of the UAS-Sox21a transgene, we sometimes observed clusters of two to four ISCs (positive for Dl-lacZ) when overexpressing Sox21a using esg TS (Supplementary Fig. 2e,f). Thus, besides inducing progenitor cells to differentiate, Sox21a may also have a role in stem cell division. We conclude that Sox21a is a critical factor required for the transition from EBs to mature intestinal cells in the adult midgut.
Sox21a functions downstream of the JAK/STAT pathway. The JAK/STAT pathway plays a major role in ISC proliferation and differentiation in Drosophila 23,25 . We therefore explored the relationship between Sox21a and JAK/STAT in stem cell differentiation.
Quantitative reverse transcription-PCR (qRT-PCR) experiments showed that Sox21a expression is regulated by the JAK/STAT signalling pathway. Sox21a expression in the midgut was lower when Stat92E was silenced by RNAi using the esg TS driver and was higher when expressing a gain-of-function allele of JAK (hop tumL ; Supplementary Fig. 2g). Previous study has shown that MARCM clone cells mutant for Stat92E were also negative for enterocyte marker Pdm1 ( Supplementary Fig. 2h), consistent with a mandatory role of JAK/STAT in progenitor differentiation 23,25 . Moreover, overexpression of Sox21a restored the expression of the enterocyte marker Pdm1 in Stat92E null mutant clones, confirming the role of Sox21a as a major downstream effector of this pathway in mediating differentiation ( Supplementary Fig. 2i). The position of Sox21a downstream of the JAK/STAT pathway and its role in EB differentiation were reinforced by two other observations. First, the expression of esg and its regulator miR-8, two genes encoding factors regulating the progenitor identity 16,20 , was not affected in Sox21a flies (Fig. 3a,b; Supplementary Fig. 3a,b). Second, the expression of Pdm1, a transcription factor specifically expressed in enterocyte 20 , was downregulated in Sox21a mutant EBs (see RNA sequencing (RNA-seq) experiment, described below).
Accumulation of EBs and formation of tumour in Sox21a flies. A striking feature of Sox21a flies is the presence of large clusters of progenitors in the anterior but not the posterior midgut ( Supplementary Fig. 4a-n). This regional difference is not surprising, since the Sox21a mutation has a differential effect on the ISC division rate in the anterior and posterior midgut. We have focused our subsequent studies on the anterior midgut to analyse the formation of these clusters. They contain both ISCs and EBs, but progressively become dominated by EBs, consistent with the function of Sox21a in EB differentiation (Fig. 3a-  . This suggests that ISCs in Sox21a tumours are still functional and most likely divide asymmetrically to generate another self-renewing ISC and an EB blocked at this stage 26,27 . Sox21a tumour growth relies on ISC division. We next explored the mechanisms by which a simple defect in the differentiation program of EBs leads to tumour formation. ISCs are the only proliferating cells in the midgut in normal conditions 17 . Although a study has reported that a small portion of EBs (o5%) displays mitotic activity upon enteric Pseudomonas entomophila infection, EBs do not divide under basal conditions 17 . Using PH3 staining, we analysed the identity of mitotic cells in Sox21a mutant expressing a GBE4GFP to mark EBs. While we detected many ISCs undergoing mitosis, no mitotic EB was observed (n4100; Supplementary Fig. 6a). We next explored whether tumour growth in Sox21a flies is sustained by stem cell division. Drosophila ISC division relies on the epidermal growth factor receptor (EGFR) signalling 28 . Blocking this pathway by expressing a dominant-negative form of EGFR in progenitors of Sox21a flies suppressed the formation of tumour (Supplementary Fig. 6b-d). The ingestion of enteric bacteria was previously shown to stimulate ISC proliferation in Drosophila and promotes tumorigenesis in other models 29 . Similarly, stimulating ISC proliferation by infecting Sox21a flies with bacteria increased the size and the numbers of tumours (Supplementary Fig. 6e-g).
These observations indicate that ISC proliferation is essential for Sox21a tumour formation. This dependence on stem cell division again explains why tumours are only found in the anterior midgut but not the posterior midgut where Sox21a is required for both EB differentiation and ISC division. Interestingly, ISC proliferation was markedly increased in the neighbourhood of Sox21a tumours ( Supplementary Fig. 7a). Consistent with this, use of 10xStat-GFP D reporter gene reveals higher JAK/STAT activity in the tumour (Fig. 4a). In addition, Sox21a mutant cells generated via MARCM triggered Ras/MAPK signalling in neighbouring wild-type cells, as revealed by a staining of phosphorylated ERK (dpERK) (Fig. 4b). Thus, Sox21a mutant cells induce a local hyperplasia by stimulating division in adjacent ISCs ( Supplementary Fig. 7b). This suggests that EBs composing Sox21a tumours send a signal to neighbouring ISCs to drive their proliferation.
EB-derived Upd2 is essential for Sox21a tumour growth. ISC proliferation can be induced upon expression of secreted ligands of the EGFR pathway (Spitz, Vein and Keren), the JAK-STAT pathway (unpaired 1, 2 and 3) and the Wingless pathway (Wg) 6,23,24,28,30,31 . To identify the factor stimulating stem cell division in Sox21a tumour, we applied a candidate gene approach by knocking down genes encoding these ligands in either EBs with GBE TS or in enterocytes with Myo1A TS . Depletion of the JAK/STAT ligand upd2 by RNAi in EBs but not in enterocytes strongly reduced tumour formation in Sox21a flies ( Fig. 4c-e). Similarly, upd2; Sox21a-double mutant flies displayed a reduction in tumour burden (Fig. 4c). In contrast, inhibiting the other factors in EBs or enterocytes (Upd1, Keren and Wg) did not impair Sox21a tumour formation ( Supplementary Fig. 8a,b). It should be noted that upd3 mutation and to a lesser extent depletion of Spitz in EBs had a modest effect on Sox21a tumours ( Fig. 4c; Supplementary Fig. 8a). This indicates that ISC proliferation is mostly induced by Upd2 released from Sox21a EBs composing the tumour. The stimulation of ISC proliferation by EB-derived Upd2 produces more differentiation-defective EBs, providing a feed-forward loop underlying the expansion of Sox21a tumours. We hypothesize that tumours are initiated in Sox21a flies upon stochastic clustering of EBs, leading to local increase of the mitogen Upd2. This mechanism would explain the random localization of tumours observed in the anterior midgut of Sox21a flies. In addition to Sox21a mutation, loss of Notch signalling in progenitors has been shown to induce tumours in the Drosophila midgut 14,15,32 . In contrast to the Sox21a EB tumours, Notch tumours are composed of ISCs that fail to differentiate into EBs. We have investigated the role of Upd2 in Notch tumour formation. Simultaneous depletion of upd2 and Notch by RNAi in the progenitors also suppressed Notch tumours (Fig. 4f,g), further emphasizing the role of Upd2 in tumour growth. Importantly, silencing upd2 in progenitor cells in an otherwise wild-type background led to a decrease in ISC numbers, pointing to a role of Upd2 in basal level stem cell maintenance (Fig. 4h-i).
Mmp2 is required for Sox21a tumour progression. We then investigated how a defect in the differentiation program transforms EBs into an aggressive tumour. For this, we compared gene expression of fluorescence-activated cell sorting (FACS)-sorted EBs of wild-type and Sox21a flies by RNA-seq 33 (Fig. 5a). Of 1,080 differentially expressed genes (Po0.05, Robinson and Smyth Exact Test), 668 genes were reproducibly upregulated and 412 genes downregulated in Sox21a EBs compared with control (Fig. 5b). Gene ontology analysis of the RNA-seq data set revealed enrichment in genes involved in epithelia tube morphogenesis and redox homeostasis in Sox21a EBs (Fig. 5c). Many genes that were previously shown to be associated with tumorigenesis in other models were also identified in Sox21a tumour. For instance, the genes ImpL2, an insulin/insulin-like growth factor antagonist recently reported to mediate tumour-induced organ wasting 34,35 , and p53, which reprograms tumour metabolism 36 , were both upregulated in Sox21a EBs (Fig. 5d). Increased expression of breathless/fibroblast growth factor receptor (FGFR; btl) in Sox21a EBs was confirmed using btl-Gal4 UAS-actGFP (referred to as btl4actGFP; Supplementary Fig. 9a-f). While btl4actGFP never labels intestinal cells in wild-type midgut ( Supplementary  Fig. 9a), expression of the btl4actGFP reporter was observed in the anterior but not posterior midgut of Sox21a flies, in regions where tumours form ( Supplementary Fig. 9b-f). The observation that some but not all the Sox21a EBs express btl4actGFP  highlights the cellular heterogeneity of Sox21a tumours ( Supplementary Fig. 9c).
Interestingly, genes encoding two matrix metalloproteinases, matrix metalloproteinase 2 (Mmp2) and to a lesser extent matrix metalloproteinase 1 (Mmp1) were upregulated in Sox21a EBs (Fig. 5d). Use of an endogenous Mmp2-GFP fusion 37 confirmed an increased expression of Mmp2 specifically at the tumour site (Fig. 6a,b). Mmp1 and Mmp2 are downstream effectors of the JNK pathway that mediate tumour invasiveness in an imaginal disc tumour model [38][39][40] . Inactivating the JNK pathway by expressing a dominant-negative form of JNK (basket, bsk), depleting Mmp2 (but not Mmp1) or expressing the tissue inhibitor of metalloprotease (timp) in EBs of Sox21a flies reduced tumour burden and growth towards the lumen (Fig. 6c-f). Of note, Mmp2 but not Mmp1 was previously shown to be required for the invasive growth of larva air sac/trachea into tissues 41 .
Tumour progression requires JNK activation in enterocytes. Tumour progression in Sox21a flies was associated with the elimination of neighbouring enterocytes, as shown by the progressive disappearance of 4,6-diamidino-2-phenylindole (DAPI)-stained polyploid cells (Fig. 3g-j; Supplementary  Fig. 5a-d). Tumour-induced elimination of normal cells has  ARTICLE been observed in other cases and probably reflects a common feature of aberrantly proliferating cells 42,43 . It is reminiscent of cell competition in Drosophila, a type of short-range cell-cell interaction, where the fitter cells eliminate the unfit cells by activating JNK signalling 44 . Using a puckered enhancer trap (puc-lacZ and puc-Gal4) as readout for JNK activity, we found that JNK signalling was induced in enterocytes surrounding Sox21a tumours from Sox21a flies or flies with EB-specific depletion of Sox21a by RNAi (Fig. 7a,b). Interestingly, JNK activation in enterocytes was coupled with the loss of the cell-polarity marker Discs large (Dlg) (Fig. 7a). In Notch ISC tumour 43 , flanking enterocytes are eliminated by delamination into the lumen. Similarly, confocal microscopy revealed the presence of delaminating enterocytes in the lumen of Sox21a flies at the vicinity of tumour (Fig. 7c). However, in Sox21a tumour, EBs were also found intercalated with enterocytes (Fig. 7d). Several observations indicate that JNK activation in enterocytes flanking Sox21a mutant tumours is essential for tumour progression. First, Sox21a flies lacking one copy of hemipterous (hep), a gene encoding the JNK kinase, have decreased tumour burden. In contrast, puc E69 / þ heterozygote flies with enhanced JNK activity display an increase of tumour burden of Sox21a flies (Fig. 7e). Second, inactivation of JNK signalling specifically in enterocytes by expressing a dominant-negative form of JNK greatly suppressed tumour formation and the presence of delaminating enterocytes (Fig. 7e). Moreover, the elimination of enterocytes nearby the tumours does not involve caspasedependent apoptosis, as tumour progression was not affected by expressing the caspase inhibitor P35 (Fig. 7e). Collectively, our data show that Sox21a tumour progression involves the elimination of enterocytes by JNK activation independent of caspase activation.
Increase of ROS activity at the vicinity of Sox21a tumour. Reactive oxygen species (ROS) have been implicated in JNK activation and induction of cell death in Drosophila. For example, feeding flies with the ROS-generating compound Paraquat causes cellular damage and JNK activation in the midgut 45,46 . A recent study reported that ROS from apoptotic cells propagate to and activate JNK in the nearby cells 47 . This raises the possibility that tumour-derived ROS contribute to JNK activation in surrounding enterocytes, facilitating their elimination. We therefore investigated the role of ROS in tumorigenesis in Sox21a flies. Gene ontology analysis of the RNA-seq data set revealed enrichment in genes involved in redox homeostasis in Sox21a EBs (Fig. 5c). For instance, Sox21a EBs display increased expression of several Cytochrome P450 genes, the NADPH oxidase Dual oxidase (Duox) and its regulator, the MAPK p38c (refs 48,49), as well as several genes encoding mitochondria and peroxisome components (Fig. 5d). Mitochondria and peroxisomes are two main sources of intracellular ROS 50 . Using reporter constructs (mitoGFP and peroxisome-GFP), we observed an increase in mitochondrial and peroxisome signals at the tumour site, evocating a shift of metabolism (Fig. 8a-e;  Supplementary Fig. 10a,b). Increase of peroxisomes in Sox21a EBs is further supported by the expression of Catalase (Cat), a protein localized to peroxisome (Fig. 8f,h). Consistent with these observations, in vivo ROS detection using dihydroethidium (DHE) revealed a gradient of ROS peaking at the periphery of Sox21a tumours (Fig. 8i). Surprisingly, the level of ROS at the tumour site was lower, suggesting that EB tumour cells were less exposed to ROS compared with neighbouring enterocytes. Reporter genes and immunostaining analysis showed that many enzymes involved in ROS detoxification, including Catalase, Glutathione S-transferase D1 (GstD) and Superoxide dismutase 2 (SOD2) are enriched in progenitors of Sox21a flies (Fig. 8g,  Supplementary Fig. 10c-e). These observations suggest that these progenitor cells have an increased capacity to deal with ROS, and explain the lower level of ROS in Sox21a tumour.
We then investigated the role of ROS in tumour formation by feeding Sox21a flies with an antioxidant, N-acetylcysteine amide (AD4). We observed that N-acetylcysteine amide-fed Sox21a flies have reduced tumour burden, although the difference with untreated control did not reach statistic significance ( Supplementary Fig. 8c). Interestingly, overactivation of the ROS-producing enzyme Duox specifically in EBs of wild-type flies led to increased JNK activity in the flanking cells, and often resulted in local hyperplasia (Fig. 8j-l). In these experiments, foci of JNK activation were not observed around individual EBs but only around clustered EBs (Fig. 8k). Clustering of EBs expressing Duox might thus lead to a local increase of ROS above a threshold sufficient to trigger JNK activity. While the relevance of ROS in Sox21a tumour progression requires further investigation, our data raise the possibility that tumour-derived ROS non-cell autonomously contribute to JNK activation and elimination of flanking enterocytes.

Discussion
In this work, we report the functional characterization of a new regulator of ISC differentiation, introduce a novel tumour model and provide mechanistic insights on how tumour may arise from a simple defect in the differentiation program of stem cell progenies (Fig. 9a,b).
Our data show that Sox21a, a previously uncharacterized transcription factor of the SOX family, plays a major role in the terminal differentiation of ISC progenies. Although the Drosophila genome encodes eight Sox genes, Sox21a is the only one enriched in the midgut. Sox21a is specifically expressed in progenitor cells along the entire midgut, and acts downstream of the JAK/STAT signalling to permit the transformation of EBs into enterocytes and enteroendocrine cells. Although Sox21a is required for both differentiated cell types, overexpression of Sox21a drives differentiation into enterocytes. It cannot be excluded that high level of Sox21a due to overexpression approach favours enterocyte rather than the enteroendocrine cell fate. Overexpression of Sox21a rescues the differentiation marker Pdm1 that is lost in JAK/STAT-deficient clones, demonstrating that Sox21a contributes significantly to EB differentiation downstream of this pathway. Consistent with this notion, our RNA-seq analysis demonstrates that Sox21a regulates a large set of genes including Pdm1, which encodes a transcription factor involved in the terminal differentiation of enterocytes 20 . Our study also shows that Sox21a contributes to stem cell division notably in the posterior midgut. This is similar to the JAK/STAT pathway that impacts both stem cell division and differentiation [23][24][25] . The observation that Sox21a flies are viable indicates that the role of Sox21a is likely restricted to the adult intestinal homeostasis. Moving on, functional characterization of Sox21a target genes and identification of Sox21a-binding sites are now required to better understand the role of this transcription factor in ISC proliferation and progenitor differentiation.
An unexpected observation of our work was that Sox21a flies developed tumours that increase in size and grow towards the intestinal lumen over time. Sox21a tumours are mainly composed of post-mitotic progenitors, the EBs that are blocked in their differentiation. Our study shows that the growth of Sox21a tumours relies on ISC division. The requirement for ISC proliferation to drive Sox21a tumours explains why tumours are not observed in the posterior midgut and are more frequent when flies are infected with bacteria, a condition that stimulates stem cell proliferation. Our results indicate that the release of a mitogen, Upd2, by accumulating EBs provides a feed-forward loop stimulating ISCs to divide and differentiate, leading to a further increase in the number of EBs. It is likely that Sox21a  tumours are initiated at random sites where EB clustering leads to a local increase of Upd2. Like other cancer model, Sox21a tumours also express matrix metalloproteinase, which probably shapes the tumour local environment to promote tumour progression. Accumulating EBs display a shift in metabolism with an increased expression of ROS-producing factors, such as Duox and a higher number of mitochondria and peroxisomes. This metabolic shift is likely to underlie the increase of ROS at the vicinity of the tumour. We observed that the progenitors express at high-level ROS detoxification enzymes. Thus, the concomitant high-level synthesis of ROS and detoxifying enzymes by accumulating EBs restricts high ROS levels to the tumour borders. It is likely that ROS production promotes Sox21a tumour growth by eliminating flanking enterocytes in a JNK-   11 . Our study highlights the cancerous properties of the differentiating stem cell progenies, which can stimulate stem cells proliferation and display a high cellular plasticity. Promoting the terminal differentiation of cancer cells has long been proposed and studied as a promising therapeutic strategy 58 . With increasing knowledge of genetic control of stem cell differentiation, it would be interesting to explore whether modulating progenitor cell differentiation can combat the progression of cancers.

Methods
Tumour-burden quantification and statistical analysis. Tumour burden of flies with different genotypes was quantified as follows. Flies were grown either at 25°C (for mutant analysis) or at 18°C (for overexpression analysis) until adulthood. Sorted genotypes were kept at 25°C (for mutant analysis) or at 29°C (for overexpression analysis) for 20-22 days. The midguts of flies were dissected and scored for tumour burden based on the criteria described below. Grade 0 midguts carried no tumour (Grade 0) and had progenitor cells that were evenly interspersed by large-nucleated enterocytes (Fig. 3g,k). Midguts with cluster of progenitor cells spanning 3-4 enterocytes were designated grade 1 (Fig. 3h,l). Large tumour mass and multilayering cells growing towards the lumen were features of grade 2 and 3 tumours. While enterocytes were still present in grade 1 and 2 tumours, grade 3 tumours were only composed of progenitors. Tumour border was defined as the place where the density of small-nucleated cells reduces to normal. The enterocytes surrounded by the grade 1 and 2 tumour cells were in the process of being eliminated through JNK activation, and we did not count them as part of the tumour. When a fly gut harboured several tumours, we assigned the final grade to the tumour with the highest grade. Overall, 20-30 flies were scored for each experiment and each experiment was repeated for at least three times. The results were pooled to generate graphs of tumour grade distribution in Excel. P values were calculated using w 2 -test, and indicated with *Po0.05; **Po0.01; or ***Po0.001. To test the effect of bacterial infection on tumour burden, Gram-negative bacteria Erwinia carotovora 15 (Ecc15) was orally ingested by 7-day-old Sox21a flies at OD 600 ¼ 100. The flies were returned to normal food after 2 days on Ecc15-containing medium, and scored for tumour burden after recovery for another 7 days. Other significance tests in the paper were done with Student's t-test using the Prism 5 software.
In most of our experiments, we used Sox21a 6 / þ flies as control. We did not see any difference between Sox21a 6 / þ flies and wild-type flies, indicating that Sox21a 6 is a recessive mutation.
Conditional expression of UAS-linked transgenes. The TARGET system was used in combination with the indicated Gal4 drivers to conditionally express UAS-linked transgenes 64 . Flies were grown at 18°C to limit Gal4 activity. After 3-5  Generation of Sox21a mutant and transgenes. Sox21a mutant flies were generated with the method described before 21 with the guide RNA sequence: 5 0 -GCTTTCATGGTCTGGTCGCG-3 0 . The alleles originally named Sox21a SK6 and Sox21a SK8 were referred to as Sox21a 6 and Sox21a 8 in the paper.
To generate the enh::Sox21a construct, the following primers (5 0 -caccATG ACGAGCATCTCGGCCCTG-3 0 and 5 0 -TCAAATGATGTTTGGCGGACT-3 0 ) were used to amplify the 2.8-kb Sox21a-RA-coding regions together with the intronic Sox21a enhancer from the genomic DNA of a wild-type fly. The PCR product was first cloned into pENTR-D-TOPO (Life Technologies) vector, and then swapped into either pTW destination vector to make UAS-Sox21aEnh::Sox21a or pTRW destination vector to make UAS-RFP-Sox21aEnh::Sox21a. Transgenic flies were established by standard P element-mediated germline transformation. At least three independent lines of each construct were tested for expression level. Note that without the presence of a Gal4 activator, the UAS-RFP-Sox21aEnh::Sox21a transgene drives the RFP reporter under the control of Sox21a cis-regulatory sequence. Despite the presence of RFP, this construct can rescue the Sox21a mutation (Fig. 4c).
Histology and immunostaining. Flies were transferred overnight into a classical fly food vial containing a filter paper soaked with a solution consisting of 5% sucrose to clear the digestive tract. Then, guts of adult females were dissected in PBS, and fixed for at least 1 h at room temperature in 4% paraformaldehyde in PBS. They were subsequently rinsed in PBS þ 0.1% Triton X-100 (PBT), permeabilized and blocked in 2% bovine serum albumin PBT for 1 h, and incubated with primary antibodies in 2% bovine serum albumin PBT for overnight at 4°C. After 1-h washing, secondary antibodies and DAPI were applied at room temperature for 2 h. Phalloidin was added to the secondary antibody incubation step in some experiments. Histology and staining on cross-sectioned guts were done as described previously 65 .
For ROS measurement with DHE, guts were dissected and directly incubated in 30 mM DHE (Life Technologies) and DRAQ5 (BioStatus, 1:200) for 10 min at room temperature, washed twice and mounted for confocal microscopy immediately. GFP expressed under the control of btl4actGFP was used to identify the location of tumour, and live-cell dye DRAQ5 was used to visualize all the cells.
FACS and RNA-seq. Crosses were set-up at 25°C to obtain Sox21a/ þ (used as control) or homozygous Sox21a flies carrying the GBE4CD8::GFP transgenes. Eclosed flies were maintained at 25°C for 16-18 days. Around 150 flies for each biological replicate were dissected in ice-cold 1 Â PBS made with diethylpyrocarbonate (DEPC)-treated water under a dry-ice chilled dissecting microscope, within a 1-h time frame. Proventriculus, hindgut and midgut/hindgut junction were removed to collect only midgut GBE4CD8::GFP-positive cells. Trimmed midguts were split in the copper cell region of the middle midgut to obtain both the anterior midgut and the posterior midgut. Three biological replicates were performed. Cell dissociation, FACS sorting, total RNA isolation and messenger RNA amplification were performed according to the method described before 33 . Elastase solution with a final concentration of 1 mg ml À 1 was used for cell dissociation. Cells positive for GFP and negative for propidium iodide were directly sorted into a tube containing RNA extraction buffer on a BD FACSAria II cell sorter. RNA was isolated using the Arcturus PicoPure RNA Isolation Kit (Life Technologies). Messenger RNA was amplified using a two-step linear amplification protocol with the Arcturus RiboAmp HS Plus RNA Amplification Kit (Life Technologies), with 2 ng total RNA as input. We usually obtained 4100-mg amplified antisense RNA. Amplified antisense RNA integrity was determined with an Agilent 2100 Bioanalyzer before complementary DNA library preparation with the TruSeq RNA Sample Prep Kit-v2 (Illumina). RNA-seq was performed on a Hi-Seq2000 (Illumina) with 100 nt paired-end sequencing. RNA-seq data have been deposited in GEO under the accession number GSE71093.
RNA-seq data analysis. Pair-end reads were mapped to the Drosophila melanogaster genome (release version 6.02) using Bowtie2 (ref. 66) with forward and reverse composition. Each sequencing experiment generated in total 440 million raw reads, and 480% was successfully mapped for each experiment. Gene expression was quantified by the number of paired reads that fall into the exons. Normalization was carried out using the size factor formula 67 . Differentially expressed genes were identified using the method as described 68 . Over-represented gene ontology (GO) terms were computed by PANTHER 69 . GO terms were further filtered if the observed number of genes was o5% of the total input number of genes, and 25% of GO terms were discarded according to the ranking of fold enrichment.