Regenerative neurogenesis is induced from glia by Ia-2 driven neuron-glia communication

A key goal to promote central nervous system regeneration is to discover mechanisms of injury-induced de novo neurogenesis. Glial cells might induce neurogenesis upon injury, but this is debated, and underlying mechanisms are unknown. A critical missing link is the identification of neuronal factors that could interact with glial NG2 to facilitate regeneration. Here, we used Drosophila genetics to search for neuronal partners of the NG2 homologue Kon-tiki (Kon), and identified Ia-2, involved in insulin secretion. Ia-2 is exclusively neuronal, and alterations in Ia-2 function destabilized cell fate. Injury increased ia-2 expression and induced neural stem cells. Using glial markers, genetic epistasis analysis and lineage tracing, we demonstrate that Ia-2 functions together with Kon and Dilp6 to induce de novo neural stem cells from glia. Altogether, Ia-2 and Kon initiate a neuron-glia interaction loop that coordinates de novo production of both neurons and glia for central nervous system regeneration.


INTRODUCTION
Humans cannot regenerate the central nervous system (CNS) after injury, but some animals do and regeneration most often involves de novo neurogenesis (Tanaka and Ferretti, 2009). Thereafter, newly formed neurons integrate into functional neural circuits. This enables the recovery of function and behavior, which is how CNS regeneration is measured (Tanaka and Ferretti, 2009). The brains of humans and most vertebrates continue to produce new neurons in response to the environment throughout life, they also integrate into functional circuits, and this constitutes one of the key manifestations of structural brain plasticity (Gage, 2019;Tanaka and Ferretti, 2009). Thus, even the adult human brain has cells that can respond to environmental challenge. If we could understand the molecular mechanisms underlying natural regenerative neurogenesis, we would be able to further enhance de novo neurogenesis to promote regeneration in the human CNS, after damage or disease.
The fact that regenerative neurogenesis is found in many diverse animals may reflect an ancestral, evolutionarily conserved genetic mechanism, which manifests itself to various degrees in fully regenerating and non-regenerating animals (Tanaka and Ferretti, 2009). On this basis, it may be possible to discover molecular mechanisms of injury-induced neurogenesis in the fruit-fly Drosophila, which is the most powerful genetic model organism.
Regenerative neurogenesis could occur through either activation of quiescent neural stem cells, de-differentiation of neurons or glia, or direct conversion of glia to neurons. Across many regenerating animals, new neurons originate mostly from glial cells (Falk and Gotz, 2017;Tanaka and Ferretti, 2009). Thus, unravelling the molecular mechanisms that switch glial cells into becoming neural stem cells or neurons is of paramount importance. In the mammalian CNS, glial cells can often behave like neural stem cells, even in normal development (Falk and Gotz, 2017). For instance, radial glia normally produce neurons during development, and in the adult brain, new neurons are produced daily in the hippocampus from these glial cells (Falk and Gotz, 2017). There is evidence that in the mammalian CNS, astrocytes and NG2-glia (also known as oligodendrocyte progenitor cells, OPCs), can produce neurons, most particularly upon injury Falk and Gotz, 2017;Valny, et al., 2017;Vigano and Dimou, 2016). NG2-glia are most relevant, as they are the only progenitor cell type in the adult brain, constitute 5-10% of cells in the total human CNS and remain proliferative throughout life . In the normal intact CNS, NG2-glia are progenitors of astrocytes, OPCs and oligodendrocytes, and upon injury they can also produce neurons Torper, et al., 2015;Valny, et al., 2017). They can also be directly reprogrammed into neurons that integrate into functional circuits (Pereira, et al., 2017;Torper, et al., 2015). The functions and diversity of NG2-glia are not yet fully understood, but they are particularly close to neurons: they receive and respond to action potentials integrating them into calcium signaling, they survey and modulate the state of neural circuits by regulating channels, secreting chondroitin sulfate proteoglycan perineural nets, and inducing their own proliferation to generate more NG2 glia, astrocytes that sustain neuronal physiology, and oligodendrocytes that enwrap axons Sakry and Trotter, 2016;Sun, et al., 2016). To what extent these functions depend on the NG2 gene and protein, is not known.
In the CNS, NG2 is expressed by NG2-glia and pericytes, but not by oligodendrocytes, neurons, or astrocytes. NG2 is a large transmembrane protein that can be cleaved upon neuronal stimulation by α− and γ−secretases, to release secreted and intra-cellular forms (Sakry, et al., 2014;Sakry and Trotter, 2016). The intracellular domain (ICD) -NG2 ICD -is mostly cytoplasmic, and it activates protein translation and induces cell cycle progression (Nayak, et al., 2018). NG2 ICD lacks a DNA binding domain and therefore does not directly function as a transcription factor, but it has a nuclear WW4 domain, nuclear localization signals, it was found in the nucleus and can regulate gene expression (Nayak, et al., 2018;Sakry and Trotter, 2016;Sakry, et al., 2015). NG2 is abundant in proliferating NG2-glia and glioma (Nayak, et al., 2018;Sakry and Trotter, 2016;Sakry, et al., 2015).
NG2 is required for OPC proliferation and migration in development and in response to injury, and it participates in glial scar formation (Biname, et al., 2013;Kucharova, et al., 2011;Kucharova and Stallcup, 2010). Potentially, NG2 may endow OPCs with plastic, homeostatic and repair properties in interaction with neurons Sakry and Trotter, 2016). However, whether NG2 itself may be involved in de novo neurogenesis remains unresolved. A critical missing link is the identification of neuronal partners that might interact with NG2 to induce regenerative neurogenesis.
The fruit-fly Drosophila is particularly powerful for identifying novel molecular mechanisms.
The Drosophila NG2 homologue is called kon-tiki (kon) or perdido (Perez-Moreno, et al., 2017;Schnorrer, et al., 2007). Kon promotes glial proliferation and cell fate determination in development and upon injury (Losada-Perez, et al., 2016). Kon works in concert with Notch and between these genes is also conserved in the mouse, where the homologue of pros, prox1, is critical for oligodendrocyte differentiation (Kato, et al., 2015). Together, Notch, Kon and Pros form a homeostatic gene network that sustains neuropile glial integrity throughout the life-course and drives glial regeneration upon injury (Hidalgo and Logan, 2017;Kato, et al., 2018).
Remarkably, we noticed that injury to the Drosophila larval CNS resulted in spontaneous, yet incomplete, repair also of the axonal neuropile (Kato, et al., 2011). This strongly suggested that injury might also induce improvements in neurons. These could correspond to axonal regrowth, or generation of new neurons. Here, we asked whether Kon may interact with neuronal partners that could contribute to regenerative neurogenesis after injury.

Ia-2 is a functional partner of Kon expressed in neurons
Genetic manipulation of glia induced axonal neuropile repair, and up-regulation of kon in glia was sufficient to induce CNS repair (Kato, et al., 2011;Losada-Perez, et al., 2016), implying that Kon might interact with neuronal factors during regeneration. To search for neuronal partners of Kon, we carried out genetic screens that aimed to identify genes expressed in neurons with non-autonomous effects on glia. We tested whether RNAi knock-down of candidate genes in neurons or glia rescued the extended ventral nerve cord phenotype of over-expressed full-length kon (Supplementary Figures S1 and S2). We tested factors predicted or known to interact with Kon and/or NG2 (Perez-Moreno, et al., 2017;Schnorrer, et al., 2007), and factors involved in Notch signaling, to validate the approach; phosphatases, as a relationship of Kon/NG2 had previously been reported for Prl1 and Ptp99A (Song, et al., 2012); and other transmembrane proteins expressed in neurons. Rescue by knock-down of known interactors, such as integrins (Perez-Moreno, et al., 2017), factors involved in Notch signaling (e.g. Mtm, Akap200), secretases (i.e. kuz, kuz-l) that cleave both Notch and NG2/Kon Trotter, 2016) andprl-1 (Song, et al., 2012), validated the approach (Supplementary Figure 1A-F). Amongst the novel hits, most prominent were genes encoding transmembrane protein phosphatases and insulin-related factors, including phosphatase LAR, Akt and phosphatase-dead ia-2 (Supplementary Figure 2A-D). Ptp2A negatively regulates insulin receptor signaling, maintaining neural stem cell quiescence (Gil-Ranedo, et al., 2019). LAR is involved in neuronal axon guidance, and is responsible for de-phosphorylating, and thus inactivating, insulin receptor signaling (Mooney, et al., 1997;Wills, et al., 1999). Akt is a key effector of insulin receptor signalling downstream (Van Der Heide, et al., 2006). Ia-2 is a highly evolutionarily conserved phosphatase-dead transmembrane protein phosphatase required in dense core vesicles for the secretion of insulin, insulin-related factor-1 (IGF-1) and neurotransmitters. It also has synaptic functions and influences behaviour and learning (Cai, et al., 2009;Cai, et al., 2011;Cai, et al., 2001;Carmona, et al., 2014;Harashima, et al., 2005;Henquin, et al., 2008;Hu, et al., 2005;Nishimura, et al., 2010).
Kon is required in glia, it influences gene expression, and loss of kon function prevents expression of glial differentiation markers (Losada-Perez, et al., 2016). Thus, to further test the functional relationship to kon, we used quantitative real-time reverse transcription PCR (qRT-PCR) on dissected larval CNS, to ask whether kon loss or gain of function affected the expression of genes identified from the genetic screens. Consistenly, kon knock-down in neurons (with kon c452 , elavGAL4>UAS-konRNAi) had no effect, whereas in glia (with kon c452 , repoGAL4>UAS-konRNAi) it resulted in a 3-fold increase in ia-2 mRNA levels (Supplementary Figure 3A). Conversely, overexpression of full-length kon in either neurons or glia down-regulated ia-2 mRNA levels by 25% (Supplementary Figure 3B). We validated these results by increasing the repeats of the most promising subset of genes (Supplementary Figure 3C,D), and this confirmed the strongest effect of kon loss and gain of function on ia-2 mRNA levels ( Figure 1A). This data showed that Kon prevents ia-2 expression. Next, we asked whether knock-down or over-expression of ia-2 in neurons (with elavGAL4) had any effect on kon mRNA levels, but none did ( Figure 1B). However, over-expression of ia-2 in glia (with repoGAL4>ia-2[GS11438]) decreased kon mRNA levels ( Figure 1B). These data indicate that Kon and Ia-2 restrict each other's expression to glia or neurons, respectively. Since both Kon and Ia-2 are transmembrane proteins, this effect is presumably indirect. Together, these data identified Ia-2 as a factor that interacts genetically with Kon.
Kon functions in concert with Notch and Pros during glial regeneration (Kato, et al., 2011;Losada-Perez, et al., 2016). Thus, to ask how ia-2 might relate to this regenerative gene network, we tested whether loss or gain of function of pros or Notch might affect the expression levels of ia-2 in dissected larval CNSs. Notch ts mutants caused an almost two-fold increase in ia-2 expression, whereas Notch ICD over-expression in glia (repoGAL4>Notch ICD ) caused a mild downregulation of ia-2 ( Figure 1C). So, Notch prevents ia-2 expression in glia. This resembles the effect of kon on ia-2, consistently with the fact that kon depends on Notch (Losada-Perez, et al., 2016). Ia-2 mRNA levels also increased in pros mutant larvae, but mostly when pros was overexpressed in glia ( Figure 1D). The loss of function phenotype is most likely indirect, as in glial cells Pros and Notch depend on each other (Kato, et al., 2011), so loss of pros causes the down-regulation of Notch, which would increase ia-2 expression. Instead, the stronger effect of pros gain of function on ia-2, and the fact that Pros is a transcription factor, indicate that Pros may directly regulate ia-2 expression. Importantly, pros is expressed, as well as in glia, in all ganglion mother cells and some neurons, raising the possibility that Pros may activate ia-2 expression during a cell-fate transition.
Altogether, these data show that ia-2 expression is repressed by kon and Notch in glia, and activated by pros. These data mean that ia-2 is functionally related to the kon, Notch, pros gene network that drives the regenerative response to CNS injury.
The above data suggested that ia-2 expression is normally repressed in glia. To test what cells normally express ia-2, we knocked-down ia-2 with RNAi in either neurons or glia and measured ia-2 mRNA levels with qRT-PCR in dissected larval CNSs. ia-2-RNAi knock-down in glia (with repoGAL4) did not affect mRNA levels compared to wild-type, however knock-down in neurons (with elavGAL4) down-regulated ia-2 transcripts to about 20% of wild-type levels, showing that ia-2 is expressed in neurons ( Figure 1E). To visualize ia-2 expression in vivo, we used a transgenic protein fusion of Ia-2 to yellow fluorescent protein (YFP). Ia-2YFP+ cells did not have the glial marker anti-Repo, nor anti-Deadpan (Dpn), which is the general neuroblast marker and also labels transit amplifying ganglion mother cells in type II neuroblast lineages (Boone and Doe, 2008), but all Ia-2YFP+ cells were Elav+ ( Figure 1G,H,J). This demonstrates that ia-2 is expressed exclusively in neurons.
Altogether, these data show that Ia-2 and Kon are restricted to neurons and glia, respectively ( Figure 1F), and that Ia-2 is a functional neuronal partner of Kon.

Ia-2 can regulate neurogenesis
Next, we carried out a functional analysis of ia-2 in the CNS. As kon knock-down increased ia-2 mRNA levels, we sought to verify this using Ia-2-YFP. We found that kon loss of function in glia (kon c452 /+; repoGAL4>kon-RNAi) increased the number of Ia-2-YFP+ cells along the midline (Figure 2A,B). The ectopic cells did not have the glial marker Repo ( Figure 2C). Midline cells were unaffected by kon over-expression in either neurons or glia (Figure 2A,B, elavGAL4>kon and repoGAL4>kon). Thus, in the absence of kon, ectopic Ia-2-YFP+ neurons were found at the midline. These results could also partly explain the increased ia-2 mRNA levels seen with kon loss of function.
To ask what function ia-2 might have in neurons, we altered ia-2 expression and visualized the effect using standard neuronal markers. ia-2 knock-down in neurons (elavGAL4>ia-2RNAi) had no obvious detectable effect on FasII or BP102 (Supplementary Figure 4A,B), and it did not change Eve+ neuron number either ( Figure 2D,E). As Pros activates ia-2 ( Figure 1D), we asked whether ia-2 might in turn affect Pros. Over-expression of ia-2 in either neurons or glia had no effect on Pros+ cells ( Figure 2F,G). By contrast, ia-2 knock-down in neurons (elavGAL4>ia-2RNAi) increased Pros+ cell number, and these cells looked small ( Figure 2F,G). Pros is normally found in neuropile glia, some neurons and all ganglion mother cells, suggesting that ectopic Pros+ cells might be ganglion mother cells or neurons, which are generally smaller than glia.
To test whether ectopic Pros+ cells originated from neural stem cells, we asked whether altering ia-2 function might affect the expression of deadpan (dpn), a general neuroblast marker. These data showed that interference with normal neuronal Ia-2 levels destabilizes cell fate, and induces ganglion mother cell and neural stem cell markers. This effect appeared to be nonautonomous, as neurons themselves were unaffected. As Ia-2 and Kon are functionally related and confined to either neurons or glia, respectively, this suggested that communication between neurons and glia might modulate cell fate stability.

Injury up-regulates ia-2 expression and induces regenerative neurogenesis
Data had shown that altering normal ia-2 levels induced expression of the neural stem cell marker Dpn. CNS injury induced the up-regulation of kon expression (Losada-Perez, et al., 2016). Thus, we asked whether injury might affect ia-2 expression and, consequently, induce neurogenesis. Crush injury was carried out at 74-76h after egg laying (AEL) in early third-instar larval VNCs labelled with the endoplasmic reticulum GFP marker G9 ( Figure 3A). qRT-PCR in injured VNCs revealed a virtually 2-fold increase in ia-2 mRNA levels at 5-7h post-injury, which recovered homeostatically by 24h postinjury ( Figure 3C). Thus, CNS injury caused an increase in ia-2 expression.
Since increased ia-2 levels induced ectopic Dpn+ cells ( Figure 2I,J), and ia-2 was up-regulated in injury, we asked whether injury induced neurogenesis. We focused in the abdominal VNC only, which has 3 neuroblasts per hemi-segment in the early third instar larva, that occupy ventro-lateral positions. Crush injury in the abdominal VNC at 74-76h AEL resulted in ectopic Dpn+ cells by 5-7h later ( Figure 3A,B,D, in n=6/17 VNCs). These were more numerous than the normal developmental abdominal larval neuroblasts, and included cells located in dorsal positions not normally taken by them ( Figure 3B,D; see (Froldi, et al., 2015;Sousa-Nunes, et al., 2011)). The numerous Dpn+ cells could correspond to injury-induced divisions of neuroblasts normally found during larval development. To test whether injury might induce ectopic neural stem cells distinct from developmental neuroblasts, we next carried out crush injury at three later time points ( Figure 3E Figure 3H,J). These data show that injury induces ectopic neural stem cells. Since ia-2 levels increased upon injury, and ia-2 gain of function induced neural stem cells, this suggested that ia-2 was responsible for the increase in Dpn+ cells caused by injury.
These data raised the question of how might ia-2 induce neurogenesis.
dilp-6 is expressed in cortex and blood brain barrier CNS glia, and activates neuroblast proliferation following a period of quiescence in normal larval development (Chell and Brand, 2010;Sousa-Nunes, et al., 2011). Thus, we asked whether the increase in Dpn+ cells in ia-2 loss and gain function observed above involved dilp-6. We visualized dilp-6 expressing cells in wandering larvae using dilp6-GAL4 (Chell and Brand, 2010;Sousa-Nunes, et al., 2011) to drive expression of the nuclear reporter Histone YFP. Most dilp-6>YFP+ cells were also Repo+, but they did not surround the neuropile and lacked the neuropile glial marker Pros ( Figure 4A,B). Thus, most dilp-6 expressing cells in the abdominal larval VNC were cortex and surface glia, as previously reported (Chell and Brand, 2010;Sousa-Nunes, et al., 2011). Some dilp6>his-YFP+ cells were Repo-negative and Elav+, and thus were neurons ( Figure 4A,B). Thus, dilp-6 is expressed in a few neurons per VNC segment, and mostly in non-neuropile glia.
To ask what cells might receive Dilp6, we visualized the expression of its receptor, the insulin receptor (InR), using an available GAL4 line, InR NP2552 , to drive histone-YFP, and tested co-localisation with glial and neuronal markers. InR NP2552 >his-YFP+ cells were mostly Repo+ neuropile glia and a few were Elav+ neurons ( Figure 4C,D). We cannot rule out that InR may also be expressed in other cell types, and this profile could also be dynamic.
Using qRT-PCR, we found that ia-2 RNAi knock-down in neurons did not significantly alter the levels of dpn or elav mRNA, but decreased the levels of dilp-6 expression ( Figure 4E). The ectopic abdominal Dpn+ cells observed with ia-2 knock-down ( Figure 2H-J) were smaller and had lower Dpn+ levels than normal neural stem cells which were still abundant in the thorax, so any effect in dpn mRNA levels in this experiment would have been masked by this background expression, becoming undetectable. Over-expression of ia-2 in neurons increased the levels of dpn, elav and dilp-6 mRNA ( Figure 4F). These data were consistent with the in vivo data showing an increase in Dpn+ cells.
Importantly, these data showed that ia-2 function in neurons positively regulates dilp-6 expression, presumably indirectly. kon knock-down in glia reduced dilp6 mRNA levels even more ( Figure 4G), meaning that kon is prominently required for dilp-6 expression in glia. However, over-expression of full-length kon alone was not sufficient to increase the expression of neither dilp-6 nor dpn ( Figure   4H), perhaps because the full-length form does not get activated. These data showed that dilp6 expression depends partly on ia-2 from neurons, and mostly on kon from glia.
Altogether, in the third instar larva Dilp6 is produced by some neurons, but mostly by nonneuropile glia, and it is received at least by InR in neuropile glia. This suggested that an initial 'Dilp6 signal' originating from neurons was then amplified by cortex glia and received by neuropile glia.
Since both Ia-2 and Kon are transmembrane proteins, this raised the question of how this amplification loop may work.

A positive neuron-glia communication loop boosts Dilp-6 production from glia
Signal amplification could occur if Dilp-6 were first secreted from neurons by Ia-2, to then activate InR in glia, and InR signalling in turn drove the Kon-dependent up-regulation of dilp-6 expression in glia. This would require that Kon functions in nuclei to regulate gene expression, but this is not known. Both in mammals and Drosophila, NG2 and Kon are responsible for glial proliferation (Kucharova and Stallcup, 2010;Losada-Perez, et al., 2016). In mammals, an intracellular domain of NG2 (NG2 ICD ) is generated upon cleavage of the full-length form (Nayak, et al., 2018;Sakry, et al., 2014). Cytoplasmic NG2 ICD interacts with components of the PI3K-Akt-mTOR pathway, to activate protein translation and cell cycle progression (Nayak, et al., 2018;Sakry, et al., 2015). NG2 ICD also positively regulates the expression of multiple genes, including downstream targets of mTOR, thus exerting positive feedback (Nayak, et al., 2018;Sakry, et al., 2015), but whether this requires nuclear localization of NG2 ICD is unknown. In Drosophila, Kon had been reported to lack a nuclear localization signal (Schnorrer, et al., 2007). Altogether, whether NG2 or Kon regulate glial proliferation and gene expression through cytoplasmic or nuclear events remained unsolved. Thus, to ask whether the intracellular domain of Kon (Kon ICD ) might function in the nucleus, we generated an HA-tagged form of Kon ICD . Glial over-expression of Kon ICD-HA (repoGAL4>UAS-Kon ICD-HA ) revealed HA colocalisation with the glial nuclear transcription factor Repo in neuropile glia cells, showing that Kon ICD was in nuclei ( Figure 5A). Next, to test whether Kon ICD is sufficient to induce glial proliferation, we over-expressed it in glia and automatically counted glial cells labeled with the nuclear marker histone-YFP, using DeadEasy software. Over-expression of full-length kon induces proliferation and increases glial cell number (Losada-Perez, et al., 2016). Over-expression of kon ICD in glia increased glial cell number too (UAShisYFP; repoGAL4>UASkon ICD , Figure 5B,C). Thus, Kon ICD can induce glial proliferation. These data showed that Kon can function in the nucleus to regulate dilp-6 expression, and strongly suggested that full-length Kon is normally cleaved, releasing Kon ICD to promote glial proliferation and regulate gene expression.
Next, to test whether Dilp6 activates InR in glia, we asked: (1) whether over-expression of dilp-6 could mimic the increase in glial cell number caused by Kon ICD , and (2) whether this could be rescued by over-expression of a dominant negative form of the insulin receptor (InR DN ) in glia. We found that over-expression of dilp-6 in glial cells increased glial cell number comparably to Kon ICD ( Figure 5B,C), and this was rescued with concomitant over-expression of InR DN in glia ( Figure 5B,C).
These data meant that Dilp-6 activates InR signaling in glia, and induces glial proliferation. Dilp6 and InR signaling also reactivate quiescent developmental neural stem cells (Chell and Brand, 2010;Sousa-Nunes, et al., 2011), but Kon functions in glia (Losada-Perez, et al., 2016). To further verify whether Kon function is restricted to glia, we asked whether Kon might also be required in neural stem cells. RNAi kon knock-down in neural stem cells with inscutable-GAL4 (ins-GAL4>UASkon-RNAi) did not affect the number or distribution of larval Dpn+ cells ( Figure 5D,E), meaning that Kon is not required in normal neural stem cells. Since glial proliferation depends on Kon (Losada-Perez, et al., 2016), the fact that dilp-6 could reproduce the increase in cell number caused by kon ICD , and this depended on InR in glia, strongly suggested that InR signalling can activate Kon cleavage downstream, specifically in glia.
To conclude, altogether these data suggested that Ia-2 triggers the release of Dilp-6 from neurons, which then is received by glial cells, where InR signaling activates Kon, which in turn induces glial proliferation and expression of dilp-6. Dilp6 secreted from glia in turn positively feeds-back on glia, further amplifying Dilp-6 production. Thus, a non-autonomous positive feedback loop between neuronal Ia-2 and glial Kon promotes glial proliferation, triggers dilp-6 expression and sustains Dilp-6 production in glia ( Figure 5F). This raised the question of whether neuronal Ia-2 in concert with the Kon-Dilp-6 glial loop only produced more glia, or whether they could also induce neurogenesis.

Ia-2 and Dilp6 can induce neural stem cells from glia
To ask whether Kon, Ia-2 or Dilp6 were responsible for injury-induced non-developmental neurogenesis, we over-expressed them in glia (with repoGAL4). Over-expression of full-length kon did not induce ectopic Dpn+ cells ( Figure 6A-D). By contrast, over-expression of ia-2 could induce ectopic Dpn+ cells prominently along the midline and in lower levels also in lateral locations surrounding the neuropile, which could be glia ( Figure 6A-D). Over-expression of dilp-6 had an even stronger effect, and there were many Dpn+ cells surrounding the neuropile ( Figure 6A-D). Dpn levels in ectopic cells were generally lower than in normal neural stem cells. These data showed that both Ia-2 and Dilp-6 can induce dpn expression. However, Kon-full-length alone can't (although this could be due to partial cleavage), meaning that insulin signaling is required to induce neural stem cells.
The ectopic Dpn+ cells were distributed along the midline and surrounding the neuropile, in locations that seemed to correspond to glial cells. Thus, to verify this, we tested whether Dpn colocalised with the glial marker Repo. Many of the lateral, ectopic Dpn+ cells observed with dilp-6 over-expression were also Repo+ ( Figure 7A), meaning they originated from glial cells. Dpn levels were lower than in normal neural stem cells. By contrast, the ectopic midline Dpn+ cells were not Repo+ ( Figure 7A, magenta cells). This meant that there are two distinct populations of ectopic Dpn+ cells: lateral cells that are neuropile glia, and midline cells which could originate from Repo-negative midline glia.
To further test whether the ectopic neural stem cells originated from glia, we used the celllineage marker G-TRACE. This GAL4-dependent tool results in the permanent labeling of UASexpressing cells and their progeny cells. Thus, repoGAL4>G-TRACE marks cells that were once glia, even if the repo promoter were to be silenced due to a change to a neuroblast cell state. Cells that were originally glia but may no longer be so would be labelled in green (GFP+), and recently specified glial cells would be labelled in red (RFP+). G-TRACE expression in all neurons with elavGAL4 or in glia with repoGAL4 together with dilp-6 caused larval lethality and thus could not be analysed. By contrast, over-expression of both G-TRACE and ia-2 in glia (repoGAL4>G-TRACE, UAS-ia-2) revealed G-TRACE+ Dpn+ cells around the neuropile ( Figure 7C,D). Most of these cells had GFP, but also RFP at somewhat lower levels ( Figure 7C,D). These data demonstrate that ectopic Dpn+ were once glial cells. Since RFP was also present, the data also showed that glial cell fate had not been suppressed, and instead suggested that glial cells may be in the process of de-differentiation.
Altogether, these data showed that Ia-2, Dilp-6, InR signaling with Kon can induce de novo formation of neural stem cells in neuropile associated glial cells.

DISCUSSION
A critical missing link to understand how to induce CNS regeneration in non-regenerating animals such as humans, had been to identify factors that might interact with NG2 to induce regenerative neurogenesis. NG2 is key because NG2-glia are the only population of progenitor cells that are present throughout life in the adult human brain Torper, et al., 2015;Valny, et al., 2017). Thus, they are the ideal cells to respond to injury and be manipulated to promote regeneration. Still, whether NG2-glia can give rise to neurons is highly debated, and if so, the underlying mechanism was unknown Falk and Gotz, 2017;Valny, et al., 2017;Vigano and Dimou, 2016). Here, using Drosophila in vivo functional genetic analysis we have identified neuronal Ia-2 and insulin signaling as the key interactors of the NG2 homologue kon, that can induce neurogenesis from NG2-like glial cells.
We provide evidence that Ia-2 is a neuronal partner of Kon responsible for inducing neurogenesis from NG2-like glia after CNS injury ( Figure 8A). We show that in the un-injured CNS, Kon and Ia-2 mutually repress each other's expression, confining each other to glia and neurons, respectively. Normal levels of ia-2 are required to prevent the non-autonomous induction of neural stem cells from neighbouring glial cells. A reduction in ia-2 levels non-autonomously increases kon expression, which up-regulates dilp-6 from glia; an increase in ia-2 levels up-regulates Dilp-6 secretion and production from neurons. Either way, the consequence is raised Dilp-6, which switches on a positive amplification loop from glia that results in further Dilp-6 production. Thereafter, Dilp6 can induce dpn expression in glia. Injury causes an up-regulation in ia-2 expression, as well as kon   Sousa-Nunes, et al., 2010). Ectopic Dpn+ cells are found upon injury and genetic manipulation of Ia-2 and Dilp-6. We show that these ectopic Dpn+ cells originate from glia. Altogether, this relay from neurons to cortex and then to neuropile glia enables concerted glio-and neuro-genesis, matching interacting cell populations for regeneration ( Figure 8B).
In vertebrates and also in invertebrates, neural stem cells found after development in the adult CNS and upon injury, are generally distinct from developmental ones, and can originate from hemocytes, but most often, from glial cells (Falk and Gotz, 2017;Simoes and Rhiner, 2017;Tanaka and Ferretti, 2009). Our findings that Dilp6 and InR can induce dpn expression are reminiscent of their functions in the induction of neural stem cells from quiescent progenitors in development (Chell and Brand, 2010;Gil-Ranedo, et al., 2019;Sousa-Nunes, et al., 2011). However, the Dpn+ cells induced upon injury and after development, are distinct from the developmental neural stem cells normally induced by Dilp-6. Firstly, in injuries carried out in third instar larvae, the induced neural stem cells were more numerous than normal neural stem cells. Secondly, in injuries carried out late in wandering larvae, Dpn+ cells were found after normal developmental neural stem cells have been eliminated through apoptosis (Bello, et al., 2003). Thirdly, in both cases Dpn+ cells were found in ectopic locations not normally occupied by developmental neural stem cells (Bello, et al., 2003). For instance, ectopic Dpn+ cells often occupied dorsal positions over the neuropile. Ectopic Dpn+ cells were located along the midline and surrounding the neuropile, in positions that correspond to, possibly midline, and often neuropile glia. Indeed, our data demonstrated that at least the nonmidline ectopic Dpn+ originate from glial cells: 1, all ectopic Dpn+ cells from genetic manipulations did not have Ia-2-YFP, which is expressed in all neurons. The fact that upon injury, a minority of ectopic Dpn+ cells were also Ia-2YFP+ could mean that upon injury ia-2 is over-expressed also in glia, or that as neurons acquire Dpn they down-regulate ia-2. In principle, regenerative neurogenesis could occur via direct conversion of glia into neurons, glial de-differentiation, or neuronal de-differentiation. Neuronal de-differentiation occurs both in mammals and in Drosophila (Froldi, et al., 2015). However, in mammals, neurogenesis after development and in response to injury most often originates from glial cells Falk and Gotz, 2017;Tanaka and Ferretti, 2009). In the adult mammalian brain, radial glia in the hippocampus respond to environmental challenge by dividing asymmetrically to produce neural progenitors, that produce neurons (Shtaya, et al., 2018). Astrocytes and NG2 glia can generate neurons in response to stroke or excitoxic injury, and genetic manipulations (Dimou and Gallo, 2015;Heinrich, et al., 2014;Peron and Berninger, 2015). Genetic manipulation can lead to the direct conversion of NG2 glia into neurons that integrate into functional neural circuits (Pereira, et al., 2017;Torper, et al., 2015). Importantly, injury creates a distinct cellular environment that prompts glial cells to generate different cell types than in the un-injured CNS. For instance, elevated Sox-2 is sufficient to directly reprogramme NG2 glia into neurons, but only upon injury (Heinrich, et al., 2014). And whereas during normal development NG2 glial cells may only produce oligodendrocyte lineage cells, upon injury they can produce also astrocytes and neurons (Dimou and Gallo, 2015;Huang, et al., 2018). What the injury cues are, is unknown.
In our context in the Drosophila larva, Dpn may not be sufficient to carry neurogenesis through. Firstly, loss of ia-2 function resulted in ectopic Dpn+ and Pros+ cells that could be ganglion mother cells as well as neural stem cells, but gain of ia-2 function resulted only in Dpn+ cells but not Pros+. This suggested that Ia-2 and Dpn are not sufficient for neuroblasts to progress to ganglion mother cells. Secondly, alterations in ia-2 levels induced ectopic Dpn+ cells, but not ectopic Eve+ neurons. And finally, ectopic Dpn+ cells still had also Repo. These data suggest that to generate neurons, glia not only require to express dpn, but also to receive other yet unknown signals for neuronal differentiation ( Figure 8B). A limitation of injury in the larval VNC is time. Injury is best carried out in the late larva to avoid interference with developmental neuroblasts. However, just a few hours after injury, larvae pupate. Cells may not have enough time to go through neuronal lineages from neural stem cell to differentiated neuron. Pupariation and metamorphosis bring in a different cellular context that could interfere with regeneration. With our data, we can conclude that a neuron-glia communication loop involving Ia-2, Dilp-6, Kon and InR is responsible for the induction of the neural stem cell marker Dpn in glia.
Our work has revealed a key link between Notch, Pros, Kon and insulin signaling to drive regenerative neurogenesis from glia. Ia-2 has universal functions in dense core vesicles to release insulin and neurotransmitters (Cai, et al., 2011;Cai, et al., 2001;Harashima, et al., 2005;Kim, et al., 2008;Nishimura, et al., 2010). Dilp-6 reactivates quiescent developmental neural stem cells in Drosophila (Chell and Brand, 2010;Sousa-Nunes, et al., 2011), and in mammals, insulin-like growth factor 1 (IGF-1) induces the production of astrocytes, ologendendrocytes and neurons from progenitor cells in the adult brain, also in response to exercise (Mir, et al., 2017;Nieto-Estevez, et al., 2016). The transcription factor Sox-2 that can switch astrocytes to becoming neural stem cells and produce neurons, is a downstream effector of InR/AKT signaling (Mir, et al., 2017). Together, these findings mean that insulin signaling is involved in switching glial cells into becoming neural stem cells.
Thus, some Repo+ glia cells may produce only glia after injury, and those that become Dpn+ cells could give rise to both neurons and glia. Neuropile glia, identified by the alrmGAL4 driver and known as Drosophila astrocytes, uniquely express Notch, Pros and Kon, as well as InR, and we showed that ia-2 is functionally linked to these genes. In mammals, the combination of Notch1, Prox1 and NG2 is unique to NG2-glia, also after development, and is absent from astrocytes (Cahoy, et al., 2008). This would suggest that Ia-2 and Dilp-6 can only induce dpn in NG2-like glia bearing this combination of factors. Infact, dpn is regulated by both Notch and Pros: Notch activates dpn expression promoting stemness, and Pros inhibits it, promoting transition to ganglion mother cell and neuron (Babaoglan, et al., 2013;Bi and Kuang, 2015;San-Juan and Baonza, 2011;Vaessin, et al., 1991). Thus, only glial cells with Notch and Pros may be poised to modulate stemness and neuronal differentiation.
Notch also activates kon expression (Losada-Perez, et al., 2016). NG2 interacts with AKT and other downstream components of the InR signalling pathway to promote protein synthesis and cell cycle progression, and to regulate the expression of its downstream effectors in a positive feedback loop (Nayak, et al., 2018). Insulin signaling also activates dpn expression, by repressing FoxO, which represses dpn (Siegrist, et al., 2010). Consistent with these data, we have shown that in glia, the ability of Dilp-6 and InR to induce glial proliferation depends on Kon, and that Kon drives positive feedback on insulin signaling by regulating dilp-6 expression. Importantly, Kon does not appear to function in developmental neural stem cells. As Notch, Pros and insulin signaling are known to positively regulate dpn expression (Siegrist, et al., 2010;Babaoglan, et al., 2013;Bi and Kuang, 2015;San-Juan and Baonza, 2011;Vaessin, et al., 1991), and injury induces a Notch-dependent upregulation of Kon (Losada-Perez, et al., 2016), our data suggest that the Notch-Kon-InR synergy can trigger the activation of dpn expression. Induced neural stem cells thereafter would have the potential to generate only glia from daughter cells that have Kon and Pros, on which Repo and glial cell fate depend, or neurons, from daughter cells that lack Kon, and express Pros on which Ia-2 depends ( Figure 8B). Thus, upon injury, Ia-2, insulin signalling, Notch, Pros and Kon functioning together enable the regenerative production of both more glial cells, and of neural stem cells from glia ( Figure 8B,C).
To conclude, we have shown that Ia-2 triggers two distinct responses in glia: in glial cells with Kon, Ia-2 and insulin signaling boost Kon-dependent glial proliferation and amplification of Dilp-6. In NG2-like glial cells that also express Notch and Pros, their combination with Kon and insulin signaling in response to Ia-2 also unlocks their neurogenic potential, inducing neural stem cell fate.
As a result, these genes could drive the production of both glial cells and neurons after injury, enabling the matching of interacting cell populations, which is essential for regeneration.

MATERIALS AND METHODS
Fly stocks and genetics. Fly stocks used are listed in Table 1 below. Stocks carrying combinations of over-expression and RNAi, or RNAi and mutants, etc., were generated by conventional genetics. N ts mutants were raised at 18°C to enable normal embryogenesis, and switched to 25°C from larval hatching to the third instar larval wandering stage to cause N loss of function. For all experiments, larvae bearing balancer chromosomes were identified by either using the fluorescent balancers CyO Dfd-YFP and TM6B Dfd-YFP or using the balancer SM6a-TM6B Tb -, which balances both the second and third chromosomes, and discarded. For the genetic screens, larvae with fluoresencent VNCs (i.e. repoGAL4>UAS-FlyBow or elavGAL4>UAS-FlyBow) were selected. Crush injury in the larval VNC. Crush injury in the larval CNS was carried out as previously reported (Losada-Perez, et al., 2016), and only lesions in the abdominal VNC were analysed. Larval collections were staged by putting the Go flies in an egg laying chamber for 2h, then collecting the F1 larve 74h later. Crush injury was carried out: (1) at 74-76h after egg laying (AEL); after injury, VNCs were left to carry on developing at 25°C, and were dissected 5-7h later; (2) at 96 h, and after injury they were incubated at 25°C and dissected and fixed 6h later; (3) at 117h AEL, and after injury, were incubated at 25°C, and dissected 12 hours later. Dissected and fixed VNC when then processed for antibody stainings following standard procedures.

Quantitative real time reverse transcription PCR (qRT-PCR).
qRT-PCR was preformed according to standard methods and as previously described (Losada-Perez, et al., 2016), with the following alteration. For each sample, 10 third instar larvae were used per genotype per replicate. At least three independent biological replicates were performed for all experiments other than in Supplementary Figure S3A and B where two replicates were carried out on all candidates and those of interest where taken forward to carry out two further replicates. For a list of the primers used in this study please see Table 2 below:  Microscopy and imaging. Image data were acquired using Zeiss LSM710 and Leica SP8 laser scanning confocal microscopes, using a 25x lens, 1.25 zoom, resolution 512x512 or 1024x1024, step 0.96µm and 3x averaging for all samples except for cell counting with DeadEasy that have no averaging.
Images were analysed using ImageJ. Images of horizontal sections are projections from the stacks of confocal images that span the thickness of the entire VNC, using ImageJ. Transverse views were generated using the Reslice option. Images were processed using Adobe Creative Suite 6 Photoshop and compiled with Adobe Illustrator.

Automatic cell counting
Glial cells labelled either with anti-Repo or with repoGAL4>UAShistone-YFP were counted automatically in 3D across the thickness of the VNC using DeadEasy Larval Glia software, as previously described. Prospero+ and Dpn+ cells were counted manually in 3D (i.e. not in projections), as the signal was noisy for DeadEasy.

Statistical analysis
Statistical analysis was carried out using Graphpad Prism.  mRNA levels to 20%, whereas in glia it has no effect, meaning that ia-2 is expressed in neurons. One Way ANOVA p=0.0004, post-hoc multiple comparisons to control Dunnett's test. N=3 replicates.      showing the effect of kon gain of function. kon over-expression in either neurons or glia decreased ia-2 mRNA levels. The first two columns have been left cut out as they are controls with the increase in kon mRNA with kon over-expression, which are very high compared to the rest. (C,D) Further replicates were carried out for a selected group of genes, and they validate that kon prominently regulates ia-2 expression. N=4 replicates each. The first columns represent the very high increase in kon mRNA with kon over-expression, and they have been cut as they go well beyond this scale compared to the rest.    Ia-2 and Dilp-6 induce ectopic neural stem cells that do not express ia-2 and result from InR signaling in glia. Ectopic neural stem cells originate from glia