Dissecting Hes-centred transcriptional networks in neural stem cell maintenance and tumorigenesis in Drosophila

Drosophila neural stem cells (NSCs), also known as neuroblasts, are a well-studied system for stem cell self-renewal, differentiation and carcinogenesis. NSCs are born early in embryogenesis and, over the course of embryonic and juvenile (larval, early pupal) life, give birth to all neurons and glia that make up the central nervous system (Lee, 2017). They do so by undergoing asymmetric cell divisions that regenerate an NSC and also give rise to a more differentiated cell, usually a ganglion mother cell (GMC, Type I mode of division) or an intermediate neural precursor (INP, Type II) (Baumgardt et al., 2014; Homem et al., 2015). The GMC subsequently divides only once to give rise to two postmitotic cells (neurons or glia), whereas the INP can also self-renew, albeit for fewer rounds than an NSC, and generate GMCs. Similar asymmetric cell divisions have also been observed in vertebrate NSCs and the conservation extends beyond cellular phenomenology to the molecular players involved: many proteins that orchestrate fly NSC divisions have mammalian orthologues with similar functions (Paridaen and Huttner, 2014). This justifies the use of Drosophila to elucidate conserved features of neurogenesis and to pinpoint defects that can lead to runaway proliferation and potential malignant transformation of NSCs, which is the cause of many types of childhood and adult brain tumours.

One of the several consequences of the elaborate asymmetric cell division of the NSC is the unidirectional Delta-Notch signalling from the GMC/INP progeny to the NSC, achieved by asymmetrically segregating Numb, a Notch inhibitor, to the GMC (Knoblich et al., 1995; Spana and Doe, 1996). However, Notch signalling is not indispensable for NSC maintenance, as only minor defects are observed upon loss of Notch function: most Type I lineages are unaffected, whereas Type IIs terminate prematurely (Almeida and Bray, 2005; Wang et al., 2006; Zacharioudaki et al., 2012). Transcription factors (TFs) of the Hes family are common Notch targets across tissues and species. During Drosophila neurogenesis, the NSCs express dpn and at least three E(spl) genes out of the 11 Hes genes in the fly genome (Almeida and Bray, 2005; Bier et al., 1992; Zacharioudaki et al., 2012; see also Fig. S1). Double knockout of dpn and the E(spl) complex results in acute NSC loss by premature termination of both Type I and Type II lineages (Zacharioudaki et al., 2012; Zhu et al., 2012). Even though E(spl) genes as well as dpn are induced by Notch (Almeida and Bray, 2005; San-Juán and Baonza, 2011), we showed that Dpn expression persists in Notch loss-of-function (Zacharioudaki et al., 2012), justifying the weaker Notch phenotype. Therefore, the Hes TFs and, to a lesser extent, Notch are central to the maintenance of stemness in fly NSCs. The dependence of NSC maintenance on Hes activity and Notch has also been documented in mammals (Kageyama et al., 2019).

Unlike the mild loss-of-function effects of Notch, its overactivation has dramatic consequences on neurogenesis: many immature differentiating cells revert to NSCs, causing extensive NSC hyperplasia at the expense of differentiated neuronal progeny in the larval CNS (Bowman et al., 2008; Wang et al., 2006; Weng et al., 2010; Zacharioudaki et al., 2012). This effect is complete in Type II lineages, but it is also evident in Type Is, albeit with reduced penetrance. Overexpression of dpn or E(spl)mγ (E(spl)mγ-HLH) did not disturb Type I lineages and gave only mild hyperplasias in Type II lineages (Zacharioudaki et al., 2016, 2012). Other genetic insults can give rise to NSC hyperplasia in larval CNSs, typified by brat or pros loss-of-function (Kang and Reichert, 2015). Both of these gene products are asymmetrically segregated in the GMC and/or INP (only Brat in the latter, as pros is not expressed in Type II NSCs) and function to regulate transcription (Pros) or translation (Brat) (Choksi et al., 2006; Loedige et al., 2015; Reichardt et al., 2018). Compromising the function of the asymmetric cell division machinery [e.g. mutations in insc, pins, l(2)gl, dlg1] also often leads to NSC hyperplasia (Caussinus and Gonzalez, 2005; Gateff, 1978). These hyperplasias are prone to malignant overgrowth and metastasis to distant sites upon allografting to an adult host. It was not known whether the Notch-dependent hyperplasias also have a malignant potential.

Here, we revisited the ability of Hes TFs to cause NSC hyperplasia and tested the ability of both Notch and Hes induced hyperplasias to progress to malignant tumours. We show that, upon ectopically expressing pairs of Hes transgenes, we can elicit severe NSC hyperplasias. These hyperplasias, as well as those caused by Notch overactivation, are malignant upon allografting. Comparison of hyperplastic CNS transcriptomes showed high similarity between two different combinations and moderate similarity between these and the high-Notch hyperplasias. Much less similarity was found with other known NSC tumour transcriptomes deposited in databases, pointing to a Hes/Notch-type of tumour. As Hes proteins are known to act as repressors, we focussed on the downregulated targets which are, at the same time, bound by Dpn, shown using chromatin immunoprecipitation (ChIP). These were enriched for TFs, among which were pros and nerfin-1, known to prevent GMCs and early neurons from reverting to an NSC fate, as well as erm which does the same in Type II INPs. We chose to study two other Hes-downregulated TFs, Zfh1 and Gcm, in more detail, and will present their expression patterns and possible roles in modulating tumorigenesis.

Overexpression of activated Notch has been known to cause NSC hyperplasia in larvae (Wang et al., 2006) and our earlier work (Zacharioudaki et al., 2016, 2012) has shown that this is accompanied by induction of Hes genes dpn and E(spl)mγ and is suppressed by deletion of the E(spl) locus. We have also shown that the mild Type II hyperplasias caused by E(spl)mγ overexpression bypass the need for Notch activation. To test whether stronger defects can be observed by Hes gene overexpression we increased the transgene dose by expressing two copies of Hes genes. We selected to co-express either E(spl)mγ+E(spl)m8 or E(spl)mγ+dpn, as these are normally expressed in larval NSCs (Fig. S1) and we have previously shown that these combinations give strong DNA binding as heterodimers in vitro (Koumbanakis, 2007; Zacharioudaki et al., 2012). Both global NSC-lineage overexpression by grhNB-Gal4 and clonal overexpression by act>STOP>Gal4 were performed. We imaged control and hyperplastic CNSs using NSC markers Dpn and Mira, NSC/GMC markers Ase and Wor, GMC/neuron marker Pros (nuclear), neuron marker Elav and glia marker Repo. Using the grhNB-Gal4 driver (Fig. S2), we observed a large number of ectopic NSCs, not only at their normal superficial location but also invading deeper cortical layers, normally occupied by neurons (Fig. 1A-F). At the same time there was a deficit in Elav-positive neurons throughout the central brain and thoracic ganglion, regions in which grhNB-Gal4 is expressed. Overexpression of Hes genes was less disruptive than overexpression of ΝΔecd, an activated form of Notch lacking its extracellular domain: E(spl)mγ+dpn-overexpressing animals survived to pupation, although they almost never made it to adulthood; in comparison, NΔecd animals died as early larvae unless grhNB-Gal4 was kept inactive using a Gal80ts inhibitor until the last day of larval life.

Using random clonal expression (hs-FLP; act>STOP>Gal4) we could focus on the response of individual lineages. We shall henceforth use a shorthand notation for the three effector genotypes: N for UAS-NΔecd, MM for UAS-E(spl)mγ+UAS-E(spl)m8 and DM for UAS-dpn+UAS-E(spl)mγ. In these FLP-out clones, Type II lineages were severely affected (Fig. S3), exhibiting a massive size increase with almost complete absence of differentiating cells (Pros-, Elav- or Repo-positive); only a small number of escapers turned on Pros or Elav, and these escapers were more frequent in DM and MM than in N clones. In N clones almost all cells expressed the NSC marker Dpn; in MM there were many Dpn-positive cells, but also many Dpn/Pros-double negative cells. In DM clones all cells were Dpn-positive by design (UAS-dpn overexpression), although its levels varied greatly from cell to cell, pointing to post-translational regulation of Dpn protein. Unlike Type II lineages, Type Is showed NSC hyperplasia with variable penetrance and expressivity (Fig. 1G-K). About 30% looked like the wild type (wt), whereas 70% contained three or more (up to 176) Dpn-positive NSC-like cells; all clones contained differentiating cells (Pros/Elav-positive). Phenotype expressivity (number of Dpn-positive cells per lineage) was higher in N than in MM clones. As in Type II, DM clones expressed Dpn broadly, albeit at variable levels from cell to cell within the same clone. The ectopic NSC-like cells were dispersed and were sometimes located in deep cortical layers, away from the parent NSC, where postmitotic neurons normally reside. Elav and Pros were partially repressed. Elav repression was more effective in N than in MM or DM. As a rule, cells that accumulated high levels of Dpn downregulated Elav, although in MM clones Dpn-negative/Elav-negative cells were often seen. Nuclear Pros downregulation was less common, in fact many cells were found to co-express Dpn and nuclear Pros in N clones (but not in MM), suggesting that N misexpression turns on Dpn before turning off Pros (Fig. S4). These phenotypes are consistent with a scenario in which ectopic Notch or Hes activity causes dedifferentiation of some GMCs (adjacent to the NSC) and early neurons (in deeper layers) back to an NSC-like fate (Fig. 1L). Using combinations of two UAS transgenes to equalize the dosage, we found the phenotypic severity in Type I lineages to be N>DM>MM=2xmγ>2xdpn (Fig. S5). Some of the dedifferentiated clones contain NSCs as late as in pupal or even adult stages (Fig. S6), a time when normal NSCs have ceased proliferating and are not detectable anymore (Homem et al., 2015; Lee, 2017; Maurange et al., 2008; Sousa-Nunes et al., 2010). Contrary to the behaviour of NSC lineages, misexpression of N, MM or DM in post-mitotic mature neurons (single-cell clones located in deep layers, close to the neuropils or in the abdominal ganglion) did not result in loss of Elav (Fig. 1H-J, blue arrows), suggesting that Notch/Hes activity is effective in causing dedifferentiation only within a certain competence window after neuron birth.

To determine whether the NSC hyperplasias produced by persistent Notch/Hes activity are malignant we used an allograft assay (Rossi and Gonzalez, 2015). In Drosophila, a tissue is characterized malignant if it can overgrow upon transplantation to an adult host to the point of causing its premature death. We transplanted brain lobes or ventral nerve cords (VNCs) containing act>STOP>Gal4 FLP-out clones driving N, MM or DM into young adult female abdomens. Out of the flies that survived the transplantation process, 33-98% developed tumours, seen as increased GFP fluorescence throughout the body, away from the transplantation site; these flies had a significantly lower survival rate than siblings allografted with wt tissue (Fig. 2A,B). Tumour tissue dissected out of the body cavity of host flies was retransplanted into new hosts, whereupon tumour growth was observed in ∼100% of injected hosts, and this could be repeated for many rounds (E.Z., C.V., V.T., I.K.P., C.E., S.S.M. and C.D., unpublished). Therefore, persistent Hes expression or excessive Notch signalling are sufficient to cause malignant transformation to larval NSC lineages. The fact that we readily recovered transplantable tumours even from VNC allografts suggests that Notch and Hes can drive malignancies not only from Type II, but also from Type I lineages (Fig. 2A). Histological examination of tumours 10-20 days after transplantation showed broad expression of nuclear Dpn. In addition, several cells were positive for nuclear Pros (GMC/early neuron-like; Fig. 2C,D), while Elav-positive neuron-like cells were also observed more sparsely (examples shown later in Fig. 10). These more differentiated cells had generally low or undetectable levels of Dpn, although double-positive cells were also occasionally seen (Fig. 2C). These results indicate an NSC-like tumour phenotype with sustained proliferation (Fig. 2E) and little differentiation.

As both Hes overexpression and Notch overactivation produce NSC tumours, we wanted to find out to what extent the CNS transcriptomes following these insults are similar or different. For this reason we performed microarray analysis using RNA from hand-dissected CNSs exposed for 24h (by heat-inactivating the αtub-Gal80ts inhibitor of Gal4) to DM, MM or dpn (D) overexpression via grhNB-Gal4. We compared these to Notch- and GFP- (control) overexpressing CNSs using the same overexpression regimen (Zacharioudaki et al., 2016). For this analysis we decided to focus on the N and 2xHes combinations (DM and MM), as these gave the strongest hyperplasias. Out of 8479 probesets with detectable expression in control larval CNS, we found 1410 that were differentially expressed by at least 1.8× at a FDR≤0.05 in at least one of these three conditions (Fig. 3A; Fig. S7). MM and DM showed very similar transcriptomes – of the 1410 differentially expressed genes (DEGs), none showed an opposite trend between these two conditions (up in one and down in the other). The N transcriptome was more distinct, but even so, only a very small number of probesets (eight) showed an opposite trend between N and MM/DM (union of DEGs in MM and DM). Correlation analysis showed that Hes tumours, including the weaker D (dpn alone), are more similar to each other than to the Notch tumour (Fig. 3D). In conclusion, Hes gene overexpression causes dramatic transcriptome changes in the CNS, which bears similarities and differences to that caused by NΔecd overexpression.

To address whether our transcriptome changes simply reflect the change in cellular composition of the CNSs used, we asked whether they correspond to upregulation of NSC-specific genes and/or downregulation of neuron-specific genes. When we queried a list of genes differentially expressed in wt NSCs versus neurons (Berger et al., 2012), no such trend was observed. Instead, most neuron- and NSC-enriched genes were unchanged in the N/Hes CNSs and conversely most N/Hes differentially regulated genes belonged to neither the neuron- nor the NSC-enriched genesets (Fig. S8). Similar results were obtained with a more recent categorization of genes expressed differentially between (wt) NSCs and GMCs (Wissel et al., 2018). We conclude that the transcriptomic changes that we documented are mostly due to the deregulation of the Notch/Hes axis and less because of the accompanying over-representation of NSCs versus GMCs and neurons.

Other genetic insults have been reported to generate brain tumours in Drosophila (reviewed by Kang and Reichert, 2015). We recovered transcriptome data for two tumours formed by brat RNAi in NSC lineages (using insc-Gal4) (Neumüller et al., 2011), two caused by the loss of cell polarity genes l(2)gl and scrib (GEO GSE48852) and one caused by the loss of the nuclear architecture gene l(3)mbt (Janic et al., 2010). When we compared these with our N/Hes-induced tumours (Fig. 3D; Fig. S9), we saw little or no similarity. Essentially no correlation was observed between the Hes transcriptomes and any of the other tumours. The N transcriptome showed a moderate correlation with the two brat RNAi conditions, but not the l(2)gl^−/−, scrib^−/− or l(3)mbt^−/−. This is consistent with the fact that the N/Hes tumours consist mostly of NSC-like cells, whereas the l(2)gl^−/−and scrib^−/− tumours strongly express differentiation markers (Beaucher et al., 2007). l(3)mbt tumours stood apart from all others, in agreement with a different cellular origin (optic lobe precursors; Richter et al., 2011). In short, Notch overactivation partially resembles Hes overactivation and partially resembles brat loss-of-function, which are distinct from each other.

Upon GO term analysis of the N/Hes transcriptomes, a few terms showed statistically significant enrichment (Fig. 3B,C), mostly associated with neurogenesis. Interestingly, ‘DNA binding transcription factor’, was enriched in both the upregulated and downregulated subset of DEGs, which prompted us to focus on TF genes for the follow up analysis.

Transcriptomic analysis of complex tissues is inherently noisy from the contribution of numerous cell types. In order to reveal the direct targets of Hes proteins in the NSCs we performed ChIP using an anti-Dpn antibody on hyperplastic larval CNSs overexpressing UAS-dpn+UAS-E(spl)mγ driven by grhNB-Gal4. This enabled us to focus on the expanded population of NSCs, as there is little Dpn expression (endogenous or transgenic) in other cell types. From two ChIP-seq biological replicates we obtained a high confidence set of 229 regions displaying Dpn occupancy (Fig. 4A). Mining these 229 ChIP peaks for over-represented DNA motifs, we obtained an E_B/E_C box sequence and variants as the top hits (Fig. 4B), confirming that in an in vivo context Dpn binds sequences that conform to the well described in vitro binding consensus for Hes proteins (Jennings et al., 1999; Winston et al., 1999).

A large percentage of the Dpn peaks fell within or near genes (Fig. 4C). We extracted a list of 358 genes that contain at least one Dpn binding event within ±5kb of the gene body. These genes are enriched for factors involved in neural precursor development, Notch signalling and transcriptional regulation (Fig. 4D). When this list of Dpn binding-associated genes was overlapped with the DEGs from the microarray experiment, we obtained 50 genes, the transcriptional response of which is shown in Fig. 4E. Although Hes proteins are thought to act as repressors, only half of these 50 Dpn-bound genes showed repression in the Hes-induced tumours. This is not surprising, because of the heterogeneity of the tissue analysed and the complexities of transcriptional regulation. Genes bound by Dpn in the NSCs may also be expressed in cell types other than NSCs/GMCs/early neurons, in which we drive DM overexpression. Furthermore, mRNA levels respond to multiple transcriptional/post-transcriptional regulatory inputs, so that the overall effect of Hes overexpression (integrating direct and indirect effects) may not be repression, despite Dpn binding near these genes. We do not exclude the possibility that Hes proteins might behave as activators in certain enhancer contexts, although we have no evidence in support of such a scenario (Fig. S10). We focussed our subsequent experimental validation on a set of TF genes which appear to be prominent Hes targets, as they combine high Dpn occupancy with transcriptional repression in the Hes-induced tumours.

Amongst the top downregulated/Dpn-bound transcription factors we noted two zinc-finger genes erm and nerfin-1, known to promote neural differentiation (Fig. 4E,H,I). erm is turned on in the immature INP progeny of Type II NSCs and restricts their self-renewal. Its loss-of-function is sufficient to cause dedifferentiation of INPs to highly proliferative ectopic Type II NSCs (Janssens et al., 2014; Koe et al., 2014; Weng et al., 2010). An enhancer recapitulating its expression was shown to bind stem cell repressors, including Dpn (Janssens et al., 2017) and coincides with one of the two peaks we detected (Fig. 4H). nerfin-1 is expressed in newly born neurons in both Type I and II lineages as well as in terminating NSCs after pupariation. Its loss can cause dedifferentiation and overproliferation (Froldi et al., 2015).

Besides, the well-studied erm and nerfin-1, there were another 13 TFs among the 27 downregulated/Dpn-bound TF genes. We chose two of these, zfh1 and gcm (Fig. 4F,G), as they had been implicated in differentiation (see below) and we had access to reagents, to ask whether their repression by the N/Hes system might play a role in NSC tumorigenesis. zfh1 encodes a large protein with a homeodomain and nine zinc-fingers; it has a broad tissue distribution and pleiotropic functions, e.g. embryonic mesoderm, neuron and glia differentiation (Layden et al., 2006; Postigo et al., 1999; Sellin et al., 2009). In third instar larval and early pupal CNSs we detected moderate expression in newly-born neurons in most brain (Type I and II) and VNC lineages (Fig. 5A,B, yellow arrows). Stronger expression was detected in dispersed older neurons located in deeper cortical layers (Fig. S11). In addition, a large number of glia, perhaps all glia, exhibited detectable zfh1 expression. Hyperplastic brains generated by overexpression of NΔecd or 2xHes using grhNB-Gal4 showed reduced zfh1 expression, especially in the superficial layers (Fig. 5C-F). The disorganization of the brain and shrinking of the neuropils did not allow us to assess the effect on mature neurons and glial cells.

The gcm gene encodes a transcriptional activator and founder of the Gcm-family of proteins, which are known to promote gliogenesis (Ragone et al., 2003), but also differentiation of some neurons, tendon cells and haemocytes (Cattenoz et al., 2016). In the larval CNS we detected gcm expression using the gcm^rA87-lacZ enhancer trap (Jones et al., 1995) (Fig. 6). Besides widespread expression in the optic lobe, there was sparse expression in the larval CNS (Fig. 6A,B). In deep areas (near the neuropils) lacZ was detected in some Repo-positive glial cells as well as nearby cells that could be neuropil glia precursors (Fig. S12). At the superficial levels of the central brain and thoracic ganglion, we did not detect lacZ expression in NSCs, but we did detect it in some INPs in both lateral and medial Type II NSC lineages (Fig. S12). This is in agreement with the fact that some of the Type II lineages produce glia as well as neurons (Viktorin et al., 2011, 2013). gcm-lacZ was also prominently expressed in two neuronal clusters, one on the ventral and one on the dorsal side of each brain hemisphere – these are Type I-derived neurons, but nearby NSCs were gcm-lacZ-negative (Soustelle and Giangrande, 2007). The only NSCs expressing gcm-lacZ were found at the medial edge of the medulla and at the optic lobe inner proliferation centre (Fig. S12A,C), probably gliogenic NSCs (Li et al., 2013). Upon grhNB-Gal4 overexpression of 2xHes or NΔecd, central brain expression of gcm-lacZ was downregulated in both neuron and glial precursors (Fig. 6C-F), with the exception of some deep glial precursors that retained gcm-lacZ expression. Optic lobe expression was unaffected, as grhNB-Gal4 does not drive expression in the optic lobe.

Both embryonic and post-embryonic phenotypes of the loss of gcm and zfh1 have been described. gcm loss-of-function in the larva results in the absence of most glial cells born post-embryonically (Awasaki et al., 2008). In the postembryonic CNS, zfh1 loss-of-function has been studied in the context of specific leg motor neuron lineages and it results in defects in neuron morphology, but not defects in their birth or neurotransmitter phenotype (Enriquez et al., 2015). To address their need in NSCs and their immediate progeny, we generated MARCM clones for null alleles (zfh1^75.26 or gcm^N7-4) 24-48 h after larval hatching and studied their effects 3-5 days later, namely in late larval or early pupal (12-24 h after puparium formation) stages. Unlike erm and nerfin-1, the loss of which causes supernumerary Type II and Type I NSCs, respectively, knocking out either zfh1 or gcm had no detectable effect in the cellular composition of either Type I or Type II lineages based on NSC, GMC and neuron markers (Dpn, Pros, Elav; Fig. S13).

We then turned to a gain-of-function approach to ask what would happen if we were to forcibly express these Hes-repressed factors in NSCs. For this reason we used an act>STOP>Gal4 driver and induced expression by hs-FLP in mid-larval stages. Overexpression of zfh1 in type I lineages resulted in complete elimination of Dpn within 1 day of clone induction (Fig. 7A-D,G). This was accompanied by shrinkage of the NSC, making it in most cases (76% of the clones) indistinguishable from other cells in the clone. Ase, another NSC (and GMC) marker, was not similarly repressed within the first day, although it too was extinguished at later times after clone induction (ACI) (Fig. S14). Interestingly, Elav and Pros were also partially repressed in zfh1-overexpressing lineages (Fig. 7C-F; Fig. S14): several cells lacked both Dpn and these differentiation markers in a large proportion of clones (Fig. 7H). For comparison, wt clones never contained Dpn/Pros-double negative cells; nuclear Pros was detectable in all NSC progeny (GMCs and neurons) and Elav gradually came on in neurons, even those that were still close to the NSC at the CNS surface. Thus overexpression of zfh1 was able to antagonize both NSC and neuron cell fates. These cells were not converted to glia, as the glial marker Repo was not induced in act>zfh1 clones (Fig. S14).

Unlike zfh1, ectopic expression of gcm for 2-3 days did not greatly upset the execution of NSC lineages, other than converting some of the differentiated progeny into glia (Fig. S15; see also Fig. 9E), consistent with its well-known glia-promoting role.

We next wanted to address whether overexpression of zfh1 or gcm would influence N/Hes-induced tumorigenesis. If the depletion of these Hes targets contributes towards tumorigenesis in Hes/Notch CNS tumours, perhaps expressing them transgenically would reverse this effect. We therefore co-expressed either gcm or zfh1 with tumour-promoting transgene combinations, namely N, MM or DM, in random act>STOP>Gal4 clones.

Overexpression of zfh1 in a tumour background produced round clones with a conspicuous absence of Dpn in most lineages (unless UAS-dpn was used); only a few scattered escaper Dpn-positive cells were seen (Fig. 8). Therefore, the ability of Zfh1 to repress dpn overrides the activation of the latter by N or MM. Another NSC marker, Ase, responded more variably: it was fully repressed in UAS-N+zfh1 clones, whereas it showed variable penetrance in UAS-DM (or MM)+zfh1 clones (Fig. S16): in some clones it was repressed, whereas in others it was activated in many cells. More remarkable was the strong repression of Elav (Fig. 8) and Pros (Fig. S16) in UAS-N (DM, MM)+zfh1 clones, which was seen with full penetrance (in all clones) in the majority of the marked cells. Whereas both UAS-N (or DM or MM; Fig. 1) and UAS-zfh1 (Fig. 7) produced sporadic loss of these differentiation markers, their combination was far more potent (compare Fig. 8A,C,E with Fig. 8B,D,F). We excluded the possibility that the N/Hes+zfh1 expressing cells were massively converted to glia using the pan-glial Repo marker (Fig. S16). In summary, zfh1 overexpression overrides Dpn induction by N/Hes (Fig. 8G), but synergizes with the latter to maintain CNS progenitors in an undifferentiated state (which sometimes retain the Ase marker), generating cell clumps that stay separate and do not integrate with the cortex of differentiated neurons/glia.

Overexpression of gcm had a markedly different effect in N versus DM tumour backgrounds (Fig. 9; Fig. S17). In the N background it enhanced neuronal de-differentiation, with fully penetrant repression of Elav, but little or no repression of Pros (Fig. S17A,C). Repo was dramatically induced and often co-expressed with Dpn (Fig. 9A,B). More specifically, N+gcm clones consisted mostly of two types of cells: NSC-like Dpn-positive/Repo-negative cells and Dpn/Repo-double positive cells with lower Dpn levels. In contrast, gcm+Hes co-expression did not induce Repo as strongly; only a few cells were Repo positive in 10-20% of the clones, which is less than the frequency of Repo induction by misexpression of gcm alone (Fig. 9C-E). DM/MM+gcm also did not enhance Elav repression beyond that caused by DM or MM alone (Fig. S17B). Instead, it reduced the NSC-like cells, as evidenced by a reduction in Ase- or Dpn-positive cells compared with DM/MM alone (Fig. S17D-H). In conclusion, whereas gcm synergizes with Notch to generate precursor-like cells, many of which express the glia marker Repo, this cannot be recapitulated by co-expressing gcm with Hes factors, suggesting a different mechanism of Notch-Gcm synergy that may not involve the usual Hes targets of Notch.

To determine the malignancy potential of NSC tumours after co-expression of zfh1 or gcm, we transplanted brain lobes bearing N or DM clones into abdomens of adult females. Both zfh1 and gcm significantly decreased the number of allografts generating malignancies from ∼100% to 77% or less (Fig. 10A). Many of these tumours grew more slowly and allowed their hosts to survive longer (Fig. 10A). Therefore, both zfh1 and gcm have an anti-tumorigenic effect when co-expressed with N or DM. Upon histological examination of N+zfh1 tumours, we observed a large number of Elav-positive (neuron-like) cells, which were only sporadically seen in N tumours. Intriguingly, when the malignant tumour from the original allograft (T0) was transplanted to a new host, the resulting tumour (T1) contained significantly fewer Elav-positive cells and this was also true for T2, the next passage (Fig. 10B-G). Therefore, the N+zfh1 CNS lineages have a greater tendency to differentiate after allografting (despite their initial repression of Elav), but they lose this ability upon serial passaging.

During neurogenesis, NSCs divide asymmetrically and their more differentiated progeny sends a Dl-Notch signal back to the NSC. We have studied the NSC hyperplasias caused by overactivation of Notch or overexpression of its Hes targets, dpn, E(spl)mγ and E(spl)m8, in the larval CNS of Drosophila. Here, we show that such ectopic NSCs are highly prone to malignancy, as evidenced by their ability to be serially allografted and propagated in adult hosts. The proliferating larval NSCs and their progeny were already known to be prone to tumorigenesis if their asymmetric cell divisions are perturbed (Caussinus and Gonzalez, 2005; Gateff, 1978). Tumorigenesis has been more recently documented for other insults, among others overexpression of the early temporal TF chinmo (Narbonne-Reveau et al., 2016) or loss of the differentiation TF nerfin-1 (Froldi et al., 2015). It was suspected that Notch overactivity would also be tumorigenic based on the malignancy of numb loss-of-function lineages (Caussinus and Gonzalez, 2005), and we now firmly show that indeed Notch/Hes axis overactivity can lead to CNS tumorigenesis.

How does the Notch/Hes axis contribute to NSC maintenance? Earlier we showed that Su(H), the TF that tethers cleaved (activated) Notch to its target enhancers, binds near all genes specifically expressed in the larval NSC, indicatively klu, ase, wor, pnt, Myc and mira, besides the Hes genes in the dpn and E(spl) loci (Zacharioudaki et al., 2016). This is consistent with Notch signalling promoting NSC maintenance and with the fact that one of the steps taken by the differentiating NSC progeny is to downregulate the Notch signal by accumulating high levels of Numb, a Notch inhibitor. In fact the phenotype of our act≫Gal4 (FLP-out) clones overexpressing NΔecd or 2xHes (Fig. 1) is almost identical to the phenotype of numb^−/− clones (Lin et al., 2010).

From our earlier analysis, we knew that Hes genes are needed for NSC maintenance (Zacharioudaki et al., 2012). In the present work, we have gained insight on how the Hes factors contribute to NSC maintenance. We have shown that Dpn binds and represses a number of TF genes, among which are some that promote differentiation. A typical one is gcm, a well-known glia-promoting TF, which was able to antagonize Notch/Hes-induced tumorigenesis (Fig. 10). Another Hes-repressed direct target is zfh1, which appears to reciprocally repress dpn (Fig. 7), although we do not know whether this is a direct or indirect effect. Previously, genetic interactions implicated Zfh1 in the action of the NURD chromatin remodelling complex, which was shown to shut off E(spl)mγ transcription in NSC progeny (Zacharioudaki et al., 2019). A plausible scenario is that Zfh1 acts with NURD to actively decommission many Notch-responsive target enhancers in NSC lineages, including dpn and E(spl). Besides repressing Hes genes, Zfh1 must also have other NSC targets, as its forced expression ameliorates DM tumorigenesis, despite exogenous expression of dpn and E(spl)mγ (Fig. 8; Fig. S16). Likely candidates for other Zfh1 targets are the remaining NSC stemness TFs, such as ase, wor and klu, although direct repression of other stem-cell functions, such as the cell cycle, is also possible.

Other TFs reported to repress NSC genes are Pros (Bayraktar et al., 2010; Cabernard and Doe, 2009; Choksi et al., 2006; Colonques et al., 2011; Liu et al., 2020; Southall and Brand, 2009; Vaessin et al., 1991), LolaN (Flybase: Lola-PP) (Southall et al., 2014) and Nerfin-1 (Froldi et al., 2015; Vissers et al., 2018) in Type I lineages and Erm in Type II INPs (Janssens et al., 2014; Koe et al., 2014; Weng et al., 2010). All four, like Zfh1, show mutually repressive relationships with the Hes factors: they have strong Dpn binding in their vicinity (Fig. 4) and most are downregulated in N/Hes tumours (Fig. 3A, although pros downregulation was less than the 1.8× cut-off we imposed). Repression by Hes factors has been reported for erm (Janssens et al., 2017; Li et al., 2017) and pros (San-Juán and Baonza, 2011; Zhu et al., 2012), and we now show it for zfh1. Together Pros, Zfh1, LolaN and Nerfin-1 may therefore constitute a group of TFs that switch off the expression of stemness markers, ensuring their specific accumulation only in the NSCs.

Curiously, these anti-stemness TF genes (zfh1, erm, nerfin-1, lolaN and pros) are also in the vicinity of experimentally detected Su(H) binding sites, even though Notch activity represses them (Zacharioudaki et al., 2016; this study). Perhaps high Notch signalling in the NSC primes these chromatin regions for activity, but transcription is prevented by the binding of Hes repressors that accumulate to high levels in the NSC. As Notch signalling diminishes in the NSC progeny, Hes proteins disappear and these genes are derepressed. They in turn would serve to repress residual stemness gene transcription in the early-born neurons, in which lower levels of Notch activity are still present, as evidenced by the need for Notch in neuron fate decisions (Truman et al., 2010) and the neuronal expression of Hey, another Notch target (Monastirioti et al., 2010).

Do these anti-stemness factors directly promote differentiation, like Gcm? We have evidence that Zfh1 does the contrary, namely antagonizes differentiation. We see this by the loss of early neuronal markers such as Elav upon Zfh1 overexpression (Figs 7 and 8). A plausible scenario is that after GMC divisions, newly-born neurons remain in a developmentally plastic state by virtue of expressing zfh1 (and probably also pros, lolaN and nerfin-1) and they only differentiate when these factors are downregulated at a later time. When in this plastic state, cells can be reprogrammed by increased Notch/Hes activity leading to ectopic NSCs. In contrast, mature neurons and glia never de-differentiate to NSCs upon the same Notch/Hes stimulation. A similar role in delaying differentiation has been described for Zfh1 and its mammalian homologue ZEB1 during myogenesis (Boukhatmi and Bray, 2018; Siles et al., 2013). We also noted a difference between the activity of Zfh1 and mammalian ZEB1: Whereas ZEB1 is well-known for promoting epithelial-to-mesenchymal transition and enabling tumour progression and metastasis (Brabletz and Brabletz, 2010; Rosmaninho et al., 2018), in our system, Zfh1 suppressed NSC lineage tumorigenesis.

In summary, Notch activity favours NSC fate via setting up a network of TFs that mutually cross-regulate and allow a gradual transition to differentiation as Notch activity dwindles. Why does persistent Notch activity lead to malignant tumours? Our data show that tumorigenesis is not a simple consequence of an inability to differentiate. N/zfh1 and DM/zfh1 CNSs contain more undifferentiated cells than N or DM, but they are also less tumorigenic. Also, normal NSCs, despite being undifferentiated, do not produce tumours when FACS-sorted and allografted (Landskron et al., 2018). Further studies are needed to characterize the transition of NSC lineages to malignancy.

UAS constructs of dpn, E(spl)mγ and E(spl)m8 were made by PCR amplifying the coding sequences of these genes (primer sequences are listed in Table S1) from a cDNA template. The cDNA was synthesized from RNA extracted from 30 adult yw flies. PCR fragments were subsequently cloned into the pENTR3C entry vector (http://www.ciwemb.edu/labs/murphy/Gateway%20vectors.html) and subsequently N-terminally tagged with 3×HA epitope using the UASt destination vector pTHW by Gateway cloning technology.

These plasmids were injected along with helper plasmid (Δ2–3–transposase) (Karess and Rubin, 1984) into embryos of yw to generate transgenic UAS-HA-dpn, UAS-HA-E(spl)mγ and UAS-HA-E(spl)m8 flies.

Drosophila stocks are described in Flybase and were obtained from the Bloomington Drosophila Stock Center (BDSC) unless otherwise indicated. zfh1-lacZ is BDSC stock 11515 that carries the P {PZ}zfh1⁰⁰⁸⁶⁵ enhancer trap (Justice et al., 1995; Lai et al., 1991). gcm-lacZ is BDSC stock 5445 that carries the P{PZ}gcm^rA87 enhancer trap.

Overproliferating third instar larval CNSs were obtained by crossing UAS-HA-mγ and/or UAS-HA-dpn flies with tubP-Gal80ts; UAS-CD8GFP; grhNB-Gal4 (Zacharioudaki et al., 2016). Crosses were maintained at 18°C for 8 days, then transferred to 30°C for 24 h before dissection.

Stocks for act flip-out clones were generated as follows: Initially the act-FRT>STOP>FRT-Gal4 (BDSC stock 4780) was recombined with UAS-nlsGFP and subsequently combined with hs-FLP. Flies from the hs-FLP; act-FRT>STOP>FRT-Gal4, UAS-GFP stock were crossed with the appropriate UAS combinations for generating clones. UAS-zfh1 is BDSC stock 6879 and expresses the long isoform of zfh1 (RB). UAS-gcm was obtained from Fly ORF (Bischof et al., 2013). Control clones were obtained by replacing UAS-lines with yw. Progeny underwent heat shock for 45 min-1 h at 37°C at 72 h after egg lay (AEL). Phenotypes were analysed 3 days later, unless otherwise indicated.

FRT82 zfh1^75.26 (Leatherman and Dinardo, 2008) and FRTG13 gcm^N7-4 (Bernardoni et al., 1997) were crossed to appropriate FRT tubP-Gal80 counter-chromosomes (Lee and Luo, 1999) combined with hs-FLP, tubP-Gal4, UAS-GFP for generating clones. Progeny underwent heat shock for 1 h at 37°C at 72 h AEL and CNSs were dissected out from wandering third instar larvae 3 days post clone induction.

For each expression array, RNA was isolated from 300 larval CNSs dissected in ice-cold PBS using the RNeasy Plus Mini Kit (Qiagen). Synthesis of double-stranded cDNA and biotin-labelled cRNA was performed according to the manufacturer's instructions (Affymetrix). Fragmented cRNA preparations were hybridized to Drosophila genome oligonucleotide arrays [GeneChip Drosophila Genome 2.0 Array (3′ IVT Expression Affymetrix)] using an Affymetrix hybridization Oven 640, washed, and then scanned on a GeneChip Scanner 3000.Three replicate arrays were analysed for each genotype. Data from expression arrays have been deposited in Gene Expression Omnibus under accession number GSE141794.

Gene expression values were obtained using Affymetrix Transcription Analysis Console from CEL intensity files using MAS5 without normalization and further quantile-normalized and then filtered for low expression. Differential gene expression analysis was performed on 8479 probesets with detectable expression. For each of the four mutant conditions, log₂-fold-change values were calculated against wt samples, coupled with a Benjamini–Hochberg adjusted P-value following an unpaired two-tailed Student's t-test.

Differentially expressed genes were defined in the reduced set, based on moderately conservative criteria of absolute log₂(fold-change)≥0.85 (corresponding to fold-change values of >1.8 or <0.55) and corrected P-values of ≤0.05 (FDR 5%). The union of all genes differentially expressed in at least one of the studied conditions comprised 1410 genes. These were further clustered in six groups depending on their differential expression across conditions: overexpressed in Notch (114 genes), Hes (417) or both Notch and Hes (42), underexpressed in Notch (242), Hes (484), or both Notch and Hes (96). Further comparison of transcriptomes was performed through Pearson correlation calculations of differential expression values for the set of the 1410 inclusive DEGs.

Comparison of differential expression with public datasets was performed through the application of simple thresholds corresponding to the top/bottom 10% of genes being differentially expressed in FACS-sorted neural stem cells versus differentiated neurons (Berger et al., 2012). Differential expression values in the four studied conditions were then analysed for sets of ∼1600 genes corresponding to neuron and neuroblast-specific expression, respectively (Fig. S8). To compare our CNS transcriptomes with those of other CNS tumours, we obtained data for l(3)mbt (Janic et al., 2010), l(2)gl, scrib (Richardson lab, GEO: GSE48852) and two brat RNAi genotypes (Neumüller et al., 2011), and cross-analysed gene expression levels for our restricted 1410 DEGs (Fig. S9), and against the extended set of 8479 measured genes (Fig. 3D). This comparison was performed at the level of differential expression, which largely removes biases owing to genotypic variation, experimental procedures and processing, as all values refer to wt controls. Log(fold-change) data, being relative values, represent the best way for cross comparison of gene expression profiles. Clustering of gene expression profiles was performed on the basis of Spearman Rank Correlation for all 8479 genes. Spearman rank was preferred to linear measures (such as Pearson correlation coefficient), as our focus was on monitoring monotonic relationships between tumours that were unlikely to be linear owing to the inherent variability of the samples (coming from different experiments).

We used PANTHER (http://www.pantherdb.org/) (Mi et al., 2019) to assess statistical overrepresentation. Genes were provided as lists of Flybase FBgn IDs (Thurmond et al., 2019) to PANTHER and analysed at the levels of GO (Biological Process and Molecular Function). An adjusted P-value of ≤0.05 was set as the threshold for significance.

Larval CNSs of different genotypes were dissected and kept in Trizol, where they were homogenized with a plastic pestle and incubated for 10 min on ice. RNA was extracted using chloroform and precipitated in isopropanol overnight at −20°C. RNA was resuspended in RNase-free water and subsequently treated with DNA-free™ DNase (Invitrogen, now Thermo Fisher Scientific, AM1906) for removal of genomic DNA. Then 700 ng of RNA was reverse transcribed with oligo(dT)₁₅ primers (Promega C1101) using SuperScript III Reverse Transcriptase (Invitrogen, now Thermo Fisher Scientific, 18080044). The cDNA products were subsequently diluted 1:5 and 2 μl were used as a template in each qPCR reaction. qPCR was performed using KAPA SYBR Green fast Master PCR Kit (Roche, SFUKB-KK, 04707516001). Generation of specific PCR products was confirmed by melting-curve analysis. Ct values for all genes were normalized to the levels of GAPDH. For data analysis, the delta-delta Ct values was applied. Primer sequences are listed in Table S2.

For each ChIP analysis, chromatin was prepared from 500 third instar larval CNSs dissected from the cross grhNB-Gal4×UAS-mγ; UAS-dpn. Tissues were cross-linked with 1% formaldehyde, quenched with 200 mM glycine. After homogenization in Nuclear Lysis Buffer [50 mM Tris-HCl (pH 8), 10 mM EDTA, 1% SDS and protease inhibitor], chromatin was fragmented using Diagenode Bioruptor for 7 cycles with 30 s OFF/30 s ON to obtain 200-500 bp length. Fragmented chromatin was immunoprecipitated with anti-Dpn antibody (see Supplementary Materials and Methods for production method) overnight at 4°C. Complexes were incubated with Protein A/G agarose beads (SC-2003) and precipitated chromatin was eluted using IP Elution Buffer (100 mM NaHC0₃, 1% SDS) and purified using Qiagen PCR purification kit. Validation of enrichment was performed over specific chromatin regions (Primer sequences in Table S3).

Two biological replicates of anti-Dpn ChIP and one input were sequenced on an Ion Torrent Proton platform with Ion 540 ChIP kits (Thermo Fisher Scientific, #A27765). Libraries were prepared according to manufacturer's protocol using the Ion Plus Library Kit (Thermo Fisher Scientific, #4471252).

Fastq reads were mapped to University of California, Santa Cruz (UCSC)/dm3 genome (archive-2015-07-17-14-30-40) downloaded from iGenomes using bowtie2 (version 2.2.8, --very-sensitive) (Langmead and Salzberg, 2012). Bedgraphs of mapped reads were generated by bedtools genomecov (v2.25.0) and uploaded on UCSC/Berkeley Drosophila Genome Project R5/dm3 genome browser for visualization. Reads that overlapped with dm3 repeat elements (downloaded from UCSC) were removed from bam files before peak calling using bamtools intersect (default). Peak calling was performed with macs2 over input (-p 0.05, version 2.1.0.20140616) (Zhang et al., 2008). A Dpn binding consensus list of 229 genomic regions was generated using peaks with fold-change>2 over input in both biological replicates. Motif enrichment analysis was performed in cistrome (Liu et al., 2011) using the SeqPos motif tool, searching against known Drosophila melanogaster transcription factor binding motifs from Jaspar. Genome wide peak distribution was performed with the pavis tool (Huang et al., 2013) using the Flybase R5.57/dm3 genome assembly and a 5kb upstream and 5kb downstream length window. Peak to gene annotation was performed by bedtools intersect using the Dpn binding consensus and the RefseqFlat.txt with 5kb upstream and 5kb downstream window extensions for each Refseq transcript. Gene ontology analysis of the genes with peaks in their proximity was performed using flymine v47.1 (Lyne et al., 2007). ChIP data have been deposited in Gene Expression Omnibus (GSE141794).

Fixation and immunohistochemistry of larval tissues were performed according to standard protocols (Daskalaki et al., 2011). Details of primary and secondary antibodies are provided in Tables S4 and S5. Samples were imaged on a Leica SP8 confocal microscope at the Confocal Facility, Institute of Molecular Biology and Biotechnology (IMBB) of the Foundation for Research and Technology Hellas (FORTH).

Transplantation assays were performed as previously described by Rossi and Gonzalez (2015). Donor larval brains were made using FLP-out GFP clones or a tubGal80ts grhNB-Gal4 UAS-stinger-RFP combination to overexpress different transgene combinations. CNSs were dissected, sliced into single brain lobes or single VNCs, loaded into a fine glass needle and implanted into the abdomen of female host w1118 flies using a nanoinjector (Nanoject II Auto-Nanoliter Injector, Drummond Scientific Company, 3-000-205A). Host flies carrying allografts were kept at 25°C (GFP allografts) or 30°C (RFP allografts) and examined daily for viability and the presence of GFP/RFP in their abdomen and other tissues. Malignant GFP/RFP-positive tumour pieces (T0) were dissected out of the abdomen of host flies and either re-transplanted into new host flies (T1) or fixed with 4% formaldehyde (for 25 min at room temperature) and used for immunohistochemistry experiments according to standard protocols (Daskalaki et al., 2011). Alternatively they were stored in Trizol for RNA extraction and qRT-PCR.

We are grateful to Sarah J. Bray and Cayetano Gonzalez for hosting S.S.M. and training him in chromatin immunoprecipitation and tumour allografting. We thank Ioannis Livadaras and Alexandros Babaratsas for fly injections and maintenance. Microarray hybridizations and high-throughput sequencing were performed at the IMBB genomics facility. Many thanks to Erika Bach, Stephen DiNardo, Angela Giangrande, Yuh Nung Jan, Ruth Lehmann, James Skeath and Isabel Guerrero for providing fly stocks and antibodies.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.191544.reviewer-comments.pdf