Stem cells reside in specialized microenvironments, called niches, that regulate their development and the development of their progeny. However, the development and maintenance of niches are poorly understood. In the Drosophila brain, cortex glial cells provide a niche that promotes self-renewal and proliferation of neural stem cell-like cells (neuroblasts). In the central brain, neuroblasts and their progeny control post-embryonic morphogenesis of cortex glia through PDGF-like ligands, and this PDGFR receptor tyrosine kinase (RTK) signaling in cortex glia is required for expression of DE-cadherin, which sustains neuroblasts. Thus, through an RTK-dependent feed-forward loop, neuroblasts and their glial niche actively maintain each other. When the EGFR RTK is constitutively activated in cortex glia, they overexpress PDGF orthologs to stimulate autocrine PDGFR signaling, which uncouples their growth and survival from neuroblasts, and drives neoplastic glial transformation and elimination of neuroblasts. These results provide fundamental insights into glial development and niche regulation, and show that niche-neural stem cell feed-forward signaling becomes hijacked to drive neural tumorigenesis.
In the mammalian central nervous system (CNS), neuro-glial stem/progenitor cells give rise to neurons and glia, and are maintained in distinct microenvironments called niches (reviewed by Bjornsson et al., 2015). In the niche, neuro-glial stem/progenitor cells are regulated by local and systemic cues that promote their self-renewal and modulate their proliferation and differentiation. Niche microenvironments are found in the subventricular zone and other neurogenic regions, and they are composed of multiple cell types (Bjornsson et al., 2015). This complexity endows niches with an ability to respond to local and systemic cues, but makes the interactions between niche cells and neuro-glial stem cells complicated to study. A deeper understanding of the contribution of various cell types to niche function and development may bring novel insights into the cellular origins of neurological diseases.
The Drosophila CNS, which has a simpler cellular composition than the mammalian CNS, is a powerful model for investigating interactions between neural-glial stem/progenitor cells and the neurogenic niche. The Drosophila brain and nerve cord are analogous to the mammalian brain and spinal cord, and are derived from stem-like neuro-glial progenitor cells called neuroblasts, which arise from embryonic neuroepithelia and divide asymmetrically to generate neurons and glia (reviewed by Doe, 2008; Homem and Knoblich, 2012). The Drosophila central brain is generated by neuroblasts that enter a first period of neurogenesis during embryonic development and a second period of post-embryonic neurogenesis during larval development (Fig. 1A), separated by a period of cell-cycle arrest. Post-embryonic neurogenesis is regulated by two classes of glia that contribute to the niche (Chell and Brand, 2010; Dumstrei et al., 2003; Sousa-Nunes et al., 2011; Spéder and Brand, 2014). One class of niche glia, subperineurial glia, creates the blood-brain barrier and responds to systemic nutritional cues (Spéder and Brand, 2014). Another class of niche glia, cortex glia, is large and extends processes to contact overlying subperineurial glia and wrap the cell bodies of individual neuroblasts and their daughter neurons (Fig. 1B). Cortex glia thereby create local microenvironments that are required for neuroblast maintenance and neuronal maturation (Coutinho-Budd et al., 2017; Dumstrei et al., 2003; Pereanu et al., 2005; Spéder and Brand, 2018).
In Drosophila, post-embryonic neurogenesis begins when neuroblasts exit from quiescence during early larval development, beginning in late 1st and early 2nd instar stages, to re-enter the cell cycle (Fig. 1A). Neuroblast cell cycle re-entry is triggered by expression of secreted insulin-like peptides (Dilps), such as Dilp6, by subperineurial glia in response to nutrient-dependent systemic signals (Chell and Brand, 2010; Spéder and Brand, 2018). Dilps activate insulin-like receptor (InR) in neuroblasts, leading to phosphatidylinositol 3-kinase (PI3K) signaling and cell cycle re-entry (Chell and Brand, 2010; Sousa-Nunes et al., 2011), and activate InR in cortex glia to initiate neuroblast wrapping (Spéder and Brand, 2018). During later larval development, cortex glia continue to express Dilp6, which is thought to sustain InR-driven neuroblast and glial proliferation (Avet-Rochex et al., 2012; Sousa-Nunes et al., 2011). In later larval stages, continued PI3K-dependent neuroblast proliferation is stimulated by cortex glia through secretion of Jellybelly (Jeb), which is a ligand for the Alk receptor tyrosine kinase (RTK) expressed by neuroblasts (Cheng et al., 2011). Yet how interactions between neuroblasts and glia regulate formation and function of the glial niche remains poorly understood.
To discover new pathways required cell-autonomously in glia for post-embryonic neurogenesis, I performed a Drosophila glia-specific RNAi screen. This approach identified a mechanism whereby the PDGFR ortholog Pvr and its ligands mediate paracrine signaling between cortex glia and central brain neuroblasts and their progeny to promote and sustain post-embryonic neurogenesis. I also found that, in the context of gain-of-function RTKs, Pvr signaling becomes hijacked to drive tumorigenic transformation of cortex glia, in a process similar to tumorigenic transformation of human glia. These results establish that PDGFR signaling plays evolutionarily conserved roles in glial development and tumorigenesis.
A screen for genes required for glial-specific regulation of post-embryonic brain growth
To identify new genes required in glia for post-embryonic neurogenesis, I performed an RNAi screen. I screened 553 Gal4-UAS-dependent RNAi constructs targeting 223 of the 243 kinases in the fly genome in order to enrich for signaling pathways (Read et al., 2013). RNAi constructs were expressed specifically in glia during development using the pan-glial repo-Gal4 transcriptional driver and tested for phenotypic effects on larval brain growth. Brain growth phenotypes were confirmed by confocal microscopy (Fig. 1C,D). RNAi constructs that yielded small brains in late stage larvae were subsequently screened for cell-type specificity: these RNAi constructs were overexpressed in neuroblasts and other proliferative neural cells to distinguish RNAi constructs that caused glial-specific growth phenotypes (Table S1; Fig. S1). From this screen, I identified 14 candidate genes for which restricted RNAi in glia preferentially reduced larval brain growth (Fig. 1E; Table S1). Candidate genes were classified using FlyBase, Gene Ontology and KEGG pathways (Table S1), which showed that kinases with core functions in PI3K signaling and/or in nutritional metabolism were most highly represented, consistent with known requirements for PI3K signaling and nutrient availability in neuroblast re-activation and post-embryonic neurogenesis (Chell and Brand, 2010; Sousa-Nunes et al., 2011; Spéder and Brand, 2018). For select candidate genes, the phenotypic effects of glial-specific kinase loss of function was confirmed by testing multiple RNAi constructs, dominant-negative constructs and/or mutant alleles, if available (Table S1). I found that glial-specific loss of Pvr, which encodes the sole Drosophila ortholog of the vertebrate VEGFR and PDGFR RTKs (Cho et al., 2002; Duchek et al., 2001; Heino et al., 2001), strongly reduced brain growth.
Pvr activity is required in cortex glia for neuroblast proliferation and neuron survival
Glial-specific Pvr knockdown (repo>PvrRNAi) impaired larval brain growth. Pvr mRNA expression was not detectable by qPCR on repo>PvrRNAi 3rd instar brains (not shown), demonstrating that Pvr mRNA was knocked down. Brain growth defects in repo>PvrRNAi animals were enhanced by Dicer co-overexpression (improves RNAi) and by a Pvr heterozygous mutant background, showing that Pvr gene function can be further reduced to further hamper brain growth (Fig. 2A). I did not observe obvious molting delays in repo>PvrRNAi animals (Fig. S2), indicating that defects were not due to organism-wide developmental delays. Brain size phenotypes caused by Pvr loss were glial specific: overexpression of Pvr RNAi constructs in neuroblasts or neurons caused no discernible gross defects (Table S1; Fig. S1B-E). This is consistent with observations that neuroblast clones homozygous for loss-of-function Pvr alleles showed no gross defects (not shown). Notably, glial-specific overexpression of dominant negative Pvr (PvrΔC, Brückner et al., 2004) in a Pvr heterozygous null mutant background reduced brain growth and reduced neuroblasts, neurons and glial cells compared with wild type (Fig. 2A,H,J; Fig. S3A,B).
To determine how glial-specific Pvr loss-of-function causes brain defects, I examined development of repo>PvrRNAi brains. About 8-10 h after hatching, in late 1st to early 2nd instar stages, neuroblasts normally undergo reactivation and re-enter the cell cycle to begin post-embryonic neurogenesis (Britton and Edgar, 1998; Chell and Brand, 2010; Sousa-Nunes et al., 2011). Brains of 1st instar animals showed no differences in size or glial morphology or glial cell numbers compared with wild-type controls (Fig. S4A-D,K), suggesting that embryonic brain development proceeds normally in repo>PvrRNAi animals, perhaps because of a lack of RNAi efficacy at early stages. Early 2nd instar repo>PvrRNAi brains appeared grossly normal, but showed defects in cortex glia cell morphologies, reduced cortex glia cell numbers and a slight, but not statistically significant, decrease in overall proliferation (Fig. 2B,C; Fig. S4A,B,E,F,L). By the late 3rd instar, during normal development, the brain dramatically enlarges by accumulation of new neurons born from neuroblasts and ganglion mother cells (GMCs), which are intermediate neuronal progenitor cells derived from neuroblasts (Fig. 1A). By the 3rd instar, glial-specific Pvr-RNAi or dominant-negative Pvr significantly reduced numbers of several cell types due to defects in cell proliferation and survival (Fig. 2B-L). Glial-specific Pvr-RNAi or dominant-negative Pvr significantly reduced proliferation in the central brain compared with controls, as revealed by EdU incorporation and phospho-histone-H3 staining cells, with repo>PvrΔC brains showing reduced numbers of mitotic neuroblasts compared with wild type (Fig. 2C-D). Apoptotic neuroblasts were also present in the central brain (Fig. 2I-L; Fig. S5), and comprised 3% and 2% of the TUNEL-positive cells detected in repo>PvrRNAi and repo>PvrΔC animals, respectively. TUNEL-positive immature neurons were more common (Fig. 2I-L; Fig. S5) and comprised 65% and 54% of the TUNEL-positive cells detected in repo>PvrRNAi and repo>PvrΔC animals, respectively. Of note, differential loss of type I or type II neuroblasts was not observed, determined by staining for markers such as Ase (not shown). Thus, glial-specific Pvr loss of function depressed brain growth through reduced proliferation and apoptosis of neuroblasts and immature neurons.
Pvr knockdown induced prominent cortex glia cell loss and morphogenesis defects by the 3rd instar, resulting in reduced cortex glia membrane processes wrapping neuroblasts and immature neurons in the central brain (Fig. 2E-I; Fig. S4B-M). Similar phenotypes were observed in repo>PvrΔC; Pvrc02195/+ mutants (Fig. S2A,B). Previous studies indicated that cortex glia in the larval central brain undergo limited proliferation (Avet-Rochex et al., 2012), which was only slightly reduced by Pvr RNAi (Fig. S6). However, blocking cell cycle in glia does not induce glial morphogenesis defects or reduce brain size comparable with the repo>PvrRNAi phenotype (Read et al., 2009; Spéder and Brand, 2018). Instead, cortex glia were reduced by apoptosis: pyknotic cortex glia and TUNEL-positive cortex glia were evident in the central brain of repo>PvrRNAi and repo>PvrΔC animals (31% and 45% of all TUNEL-positive cells detected, respectively) (Fig. 2J-L; Fig. S4I,J; Fig. S5). Pvr loss also caused reduced expression of cell-type specific transcription factors, such as SoxN and PntP2, which are expressed in wild-type cortex glia (Fig. S7A-F) (Avet-Rochex et al., 2012). SoxN is the closest Drosophila ortholog of the mammalian SOX2 transcription factor, a marker for neuro-glial stem/progenitor cells (McKimmie et al., 2005), which, based on RNAi phenotypes, is required for development of cortex glia (not shown). Thus, Pvr function is required for survival and morphogenesis of cortex glia in the central brain.
To independently determine whether Pvr signaling is required for brain development, I examined phenotypes of loss-of-function mutants of Pvf1, Pvf2 and Pvf3, which encode Pvr ligands. The Pvfs are semi-redundant, and Pvf single mutants are viable and show no obvious brain phenotypes (not shown). Double Pvf2−/−;Pvf3−/− (Pvf2-3−/− dual deletion) and Pvf1ex3;Pvf2-3−/− triple mutants are early lethal, but this lethality is rescued by expressing Pvf2 or constitutively active Pvr (Pvrλ, Pvr fused to a lambda dimerization domain) specifically in hemacytes using crq-Gal4 (Parsons and Foley, 2013). Pvf2-3−/−crq>Pvf2 double mutants survive to adulthood, and show slight brain morphogenesis defects (not shown). Pvf1ex3;Pvf2-3−/−crq>Pvrλ and Pvf1ex3;Pvf2-3−/−crq>Pvf2 triple mutants survive to pupal stages, and showed neuroblast loss and reduced neurogenesis in the central brain relative to wild type during larval development (Fig. 2A,C,H; Fig. S3C,D). UAS-GFP reporters were included in these genotypes to confirm that crq-Gal4 was expressed in hemacytes and not in glia or neurons; some larvae showed crq-Gal4-driven GFP expression in scattered ensheathing glia and hemacytes in the CNS, but showed brain phenotypes (Fig. 2A). To label cortex glia in triple Pvf mutants, I used a Pax6-EGFP reporter (Metaxakis et al., 2005), which is expressed in cortex glia (Fig. S3C,D; Fig. S8A). Both Pvf1ex3;Pvf2-3−/−crq>Pvrλ and Pvf1ex3;Pvf2-3−/−crq>Pvf2 triple mutant 3rd instar larvae showed reduced cortex glial cell numbers, reduced cortex glial processes in the central brain and defects in brain morphogenesis (Fig. 2A; Fig. S3C,D). In contrast, larval brain phenotypes were not examined in Pvr mutants because no viable larvae were obtained from the five loss-of-function alleles tested, even when Pvr was rescued by crq-Gal4 (Parsons and Foley, 2013), and no central brain-specific Pvr homozygous mutant glial clones or Pvr-RNAi glial clones were recovered (not shown).
Pvr protein is strongly expressed in cortex glia in the central brain in early and late stage larvae (Fig. 3A-D). Pvr gene traps, with GFP or Gal4-containing transposons inserted into the Pvr locus, are also expressed in cortex glia (Fig. S9). To test whether Pvr functions specifically in cortex glia, I used cortex glia-specific Gal4 drivers (R108-Gal4 or R122-Gal4, Fig. S8B,C) to overexpress PvrRNAi, which reduced neuroblast and cortex glial cell numbers, and decreased larval brain size, which is indicative of fewer neurons (Fig. 3E-H). In R122>PvrRNAi central larval brains, cortex glial processes were reduced (Fig. 3E-I). R122>PvrΔC animals also showed similar defects, including neuroblast loss and defective cortex glia morphogenesis (Fig. S10A-C). Overexpression of Pvr RNAi constructs in other glia subtypes, such as astrocyte-like glia (alarm-Gal4), perineurial glia (NP6293-Gal4) or subperineurial glia (moody-Gal4, NP2276-Gal4) (Awasaki et al., 2008; Bainton et al., 2005; Doherty et al., 2009), combined with Dcr, yielded viable animals that had grossly normal brain and glial phenotypes (Fig. S11), indicating that effects of Pvr loss were specific to cortex glia.
Pvf expression is required in neuroblasts and their progeny
To understand Pvr function in post-embryonic brain development, I determined which cells express Pvfs. Staining using a published validated Pvf1 antibody showed Pvf1 predominantly present in central brain neuroblasts and cortex glia (Rosin et al., 2004) (Fig. 4A). Pvf1 protein observed in cortex glia is likely receptor bound, as Pvf1 staining on cortex glia was reduced in repo>PvrRNAi animals (Fig. S12). Because no Pvf2 antibody was available, a Pvf2-lacZ promoter fusion reporter was used to determine expression patterns: Pvf2-lacZ is expressed in central brain neuroblasts and their neighboring daughter cells (Choi et al., 2008), likely GMCs (Fig. 4B-F). Pvf2-LacZ expression begins in late 1st instar/early 2nd instar co-incident with neuroblast cell cycle re-entry, as evidenced by EdU incorporation, and continues into 2nd and 3rd instar stages (Fig. 4B-F). In contrast, low to no Pvf3 staining was detected using published Pvf3 antibodies (Rosin et al., 2004). In lieu of antibody staining, I examined expression patterns of several Gal4 reporters with Pvf2 and Pvf3 promoter fragments (R77E03, R77E04, R77E08), which showed strong expression in overlapping populations of central brain neuroblasts and weaker expression in daughter neurons (http://flweb.janelia.org/cgi-bin/flew.cgi) (Li et al., 2014). Data from these Gal4 reporters are consistent with presence of neuroblast-specific regulatory elements, although they do not contain full-length Pvf2 and Pvf3 promoters so we can only draw limited conclusions. These data indicate that Pvf1 and Pvf2 are expressed by neuroblasts and more weakly by neighboring GMCs.
These results suggest that Pvfs are required in neuroblasts and their progeny to promote development of cortex glia. Confirming this hypothesis, overexpressing Pvf2 with dpn-Gal4, which drives transgene expression in neuroblasts and GMCs, partially rescued central brain neuroblasts and gross brain growth and morphology in Pvf1ex3;Pvf2-3−/−crq>Pvf2 triple mutants (Fig. 4G-M), but was not able to rescue lethality. Pvf2 expression does not have to be restricted to neuroblasts: repo-Gal4 driven Pvf2 expression rescued brain growth and neuroblast loss in Pvf1ex3;Pvf2-3−/−crq>Pvf2 mutants (Fig. 4G-M). These data show that loss of neural Pvf expression largely accounts for post-embryonic brain development defects in Pvf triple mutants.
Loss of glial Pvr causes loss of PI3K and DE-Cadherin signaling
In Drosophila, Pvf-Pvr signaling activates PI3K, Ras-Map kinase, Rho-Rac and Cadherin pathways (Brückner et al., 2004; Cho et al., 2002; Duchek et al., 2001; Garlena et al., 2015; McDonald et al., 2003). Genetic and cell biological approaches were used to determine which of these pathways are required for cortex glia development downstream of Pvf-Pvr. Cortex glia-specific loss of PI3K signaling, by overexpressing dp110 or dAkt RNAi or overexpressing dominant-negative mTor using cortex glia drivers, reduced larval brain size, reduced cortex glia cell numbers and induced morphogenesis defects, similar to Pvr knockdown (Fig. S10A-F). These results are consistent with published data showing that PI3K-mTor activity is required to promote cortex glia morphogenesis and support of neuroblasts (Chell and Brand, 2010; Sousa-Nunes et al., 2011; Spéder and Brand, 2018). In contrast, glial-specific knockdown of Ras-Map kinase or Rho-Rac pathway components failed to yield glial defects comparable with Pvr knockdown or induced early lethality (Fig. 1C,D; Fig. S10A,G; not shown).
Immunohistochemistry showed that glial-specific Pvr loss reduced PI3K pathway activity. To assess PI3K, I used the tGPH reporter, which is composed of GFP fused to a plextrin-homology (PH) domain that localizes to cell membranes in response to high PIP3 levels (Britton et al., 2002). repo>PvrRNAi and Pvf1ex3;Pvf2-3−/−crq>Pvrλ brains showed decreased membrane-associated tGPH fluorescence in cortex glia, neuroblasts and in immature neurons compared with wild-type controls (Fig. 5A,B; Figs S13, S14A,B). To confirm this, I examined eIF4E levels, which have been used as an indicator of PI3K signaling in Drosophila (Hsieh and Ruggero, 2010; Song and Lu, 2011), and eIF4E was reduced in neuroblasts in repo>PvrRNAi and Pvf1ex3;Pvf2-3−/−crq>Pvf2 mutants relative to wild type (Fig. S14C-D; not shown). Thus, Pvr loss reduced PI3K signaling cell-autonomously in glia and non-autonomously in neuroblasts and their progeny. Consistent with a requirement for PI3K signaling downstream of Pvr in glia, overexpression of constitutively active dp110 (dp110CAAX) partially rescued effects of repo>PvrRNAi on brain growth and neuroblast number (Fig. 5C,D,G; Fig. S15A-C).
During larval development, glia produce ligands that stimulate PI3K signaling in neuroblasts. The insulin-like ligand Dilp6 is expressed by subperineurial glia and cortex glia, and drives InR-PI3K signaling in neuroblasts, and Dilp6 expression requires mTor signaling in glia (Avet-Rochex et al., 2012; Chell and Brand, 2010; Sousa-Nunes et al., 2011; Spéder and Brand, 2018). At later larval stages, cortex glia also express Jeb, a secreted ligand for the Alk RTK expressed in neuroblasts (Cheng et al., 2011). Thus, the non-autonomous effects of Pvr loss on neuroblasts may be caused by reduced Dilp6 and/or Jeb expression. Consistent with this, qPCR showed a significant reduction in Dilp6 and jeb expression in larval brains from repo>PvrRNAi, repo>PvrΔC;Pvrc02195/+ and Pvf1ex3; Pvf2-3−/−crq>Pvrλ mutants compared with wild type (Fig. S16A,B). Dilp6 overexpression partially rescued neuroblast numbers and gross brain growth, and restored neuroblast proliferation in PvrRNAi animals (Fig. 5D,G; Fig. S16C), suggesting that activation of InR signaling, whether directly or indirectly, restores cortex glia and/or neuroblast function. In contrast, Jeb overexpression did not rescue neuroblasts or gross brain size in repo>PvrRNAi animals (Fig. S16C), indicating that Alk activation is insufficient to counteract Pvr loss.
PI3K signaling is required in neuroblasts to maintain growth and proliferation throughout larval development (Chell and Brand, 2010; Cheng et al., 2011; Sousa-Nunes et al., 2011), and PI3K signaling in neuroblasts is required to anchor neuroblasts in the cortex glia niche through DE-cadherin-dependent adhesion (Doyle et al., 2017). DE-Cadherin is expressed in larval central brain cortex glia and localizes at the interface of cortex glia and central brain neuroblasts, and glial DE-cadherin (DEcad) function is essential for maintenance and proliferation of neuroblasts and their progeny (Dumstrei et al., 2003; Holcroft et al., 2013). In the ovary, Pvf1-Pvr signaling regulates DE-Cadherin subcellular localization and function, perhaps directly at the membrane (McDonald et al., 2003). As Pvr is localized in cortex glial processes, perhaps Pvr-PI3K signaling regulates DE-cadherin function in cortex glia to promote anchoring, survival and/or proliferation of neuroblasts and neurons. Consistent with this, Pvr knockdown reduced DE-cadherin levels in cortex glia and non-autonomously in neuroblasts (Fig. 5C,E,F; Fig. S17), but failed to similarly reduce DE-cadherin levels in neighboring immature neurons (Fig. 5E,F). Moreover, glial-specific overexpression of DE-cadherin partially rescued cortex glia, neuroblasts and neurons in PvrRNAi animals (Fig. 5D,G-L; Fig. S15). These results indicate that cortex glia-specific Pvr loss induces an autonomous and non-autonomous reduction in DE-cadherin levels, which contributes to neuroblast loss.
Pvr signaling drives glial neoplasia
Examining the effects of Pvr gain of function, I found that Pvf2 or Pvr overexpression increased cortex glia proliferation (Fig. 6A-D; Fig. S18). Pvrλ or wild-type Pvr (PvrWT) overexpression produced abnormal compact proliferative cortex glia that accumulate to enlarge the brain, which is consistent with neoplastic transformation (Fig. 6A-D). Moreover, abnormal cortex glia outcompeted neuroblasts during the 2nd to 3rd instar larval stages, such that neuroblasts were lost (Fig. 6B,E,F). To determine how loss of Pvr signaling contributed to transformation, I examined DE-cadherin expression, which was reduced in Pvrλ mutant cortex glia and remaining neuroblasts in R122>Pvrλ relative to wild type (Fig. 6E,F). Thus, in this context, neuroblasts may be lost due to reduced glial DE-cadherin function.
In the rodent brain, PDGFR signaling regulates proliferation and differentiation of oligodendrocyte precursor cells (OPCs) (reviewed by Funa and Sasahara, 2014), which are vertebrate glial progenitor cells, and constitutive PDGFR signaling can drive malignant glial transformation in glioblastomas (GBMs) (Assanah et al., 2006; Hambardzumyan et al., 2009). GBMs, which are the most aggressive primary brain cancers in humans, are thought to arise from neuro-glial stem or progenitor cells, such as OPCs, in response to mutations that cause constitutive activation of RTKs, such as EGFR or PDGFRA (reviewed by Furnari et al., 2007). In GBMs, ectopic PDGFRA signaling cooperates with gain-of-function mutations in other RTKs, such as EGFR, to drive tumor progression (Furnari et al., 2007).
Similarly, in a Drosophila model of GBM in which glial progenitor cells are malignantly transformed by constitutively active Drosophila EGFR (dEGFRλ) and dp110 (dp110CAAX), Pvr knockdown suppressed glial neoplasia and eliminated SoxN-positive tumorous glia (Fig. 7A-C; Fig. S19A,B). Thus, as in human GBMs, Pvr cooperates with dEGFR. This suggests that a large proportion of tumor cells in the EGFR-PI3K Drosophila GBM model originate from Pvr-dependent cortex glia cells. Consistent with this interpretation, dEGFRλ and dp110CAAX also drives neoplastic glial overgrowth when targeted to cortex glia: R122>dEGFRλ;dp110CAAX animals develop tumorous glia with neoplastic features (Fig. 7D,E).
As with Pvr-dependent transformation, abnormal cortex glia transformed by dEGFRλ; dp110CAAX outcompeted neuroblasts during the 2nd to 3rd instar larval stages (Read et al., 2009), such that neuroblasts were lost (Fig. 7D,E), likely due to reduced glial DE-cadherin expression (Fig. 7F,G). Thus, Pvr-dependent neoplastic cortex glia likely no longer rely on Pvf produced by neuroblasts, which are lost. Instead, neoplastic dEGFRλ;dp110CAAX glia express and require Pvfs: qPCR and the Pvf2-lacZ reporter showed that Pvf1 and Pvf2 become overexpressed in dEGFRλ;dp110CAAX brains, and Pvf1 RNAi partially suppressed EGFR-PI3K glial neoplasia (Fig. 7H,I; Fig. S19C). Thus, Pvf-Pvr create an autocrine loop that stimulates neoplastic transformation of cortex glia (Fig. 7J).
During post-embryonic Drosophila brain development, cortex glia create a microenvironment that promotes neurogenesis. I show that neuroblasts and their progeny express Pvfs, which signal through Pvr to stimulate cortex glia survival and morphogenesis, which then promotes PI3K-DE-cadherin-dependent proliferation and survival of neuroblasts and their progeny (Fig. 7K). Thus, through an RTK-PI3K-DE-Cadherin-dependent feed-forward loop, neuronal stem/progenitor cells and immature neurons actively maintain their microenvironment.
Pvf-Pvr signaling and cortex glia development and function
Previous studies of Drosophila neuroblast-glia interactions have focused on growth factors secreted by glia to promote neuroblast re-activation rather than on factors neuroblasts and associated progenitor cells use to shape the glial microenvironment. These studies show that the fat body, acting as an endocrine organ, produces a systemic relay signal that stimulates glial Dilp expression to drive neurogenesis in early larval stages (Britton and Edgar, 1998; Chell and Brand, 2010; Sousa-Nunes et al., 2011). My results show that neuroblasts and their progeny express Pvfs at the onset of post-embryonic neurogenesis and into late larval stages to promote Pvr-dependent niche morphogenesis and maintenance. Neuroblast Pvf expression could be initiated by Dilps or other relay signals, and understanding control of Pvf expression may reveal new mechanisms that govern post-embryonic neurogenesis.
My data show that Pvr loss-of-function impairs cortex glia morphogenesis and survival. This is consistent with previous studies showing that Pvr signaling is required for trophic support and morphogenesis in other cell types (Brückner et al., 2004; Garlena et al., 2015; Parsons and Foley, 2013). I found that loss of Pvf-Pvr function decreased PI3K signaling in cortex glia, that loss of PI3K pathway components partially phenocopies Pvf-Pvr loss of function, and that restoration of PI3K signaling in glia partially rescued the effects of Pvr knockdown. However, dp110CAAX overexpression did not fully rescue neuroblast loss with Pvr knockdown; this may be because receptor scaffolding is needed to facilitate localized interactions between dp110 and other signaling effectors in cortex glia, such as DE-cadherin. Alternately, other pathways downstream of Pvr may mediate interactions with effectors such as DE-cadherin. Consistent with my results, Spéder and Brand (2018) showed that reduced PI3K activity impaired cortex glia niche morphogenesis and neuroblast wrapping in the nerve cord. My results and those of Spéder and Brand seemingly conflict with those of Avet-Roche et al., which show that dp110 is not essential for cortex glia cell proliferation and/or survival. However, Avet-Roche et al. did not specifically examine the effects of reduced PI3K activity on central brain cortex glia, and instead they examined cortex glia cells throughout the brain or in the optic lobe. Differential requirements for PI3K signaling in central brain cortex glia may underlie the discrepancies between our results.
I found that cortex glia defects cause neuroblasts and immature neurons to die and/or exit the cell cycle. At later larval stages, cortex glia express Dilp6 and Jeb (Avet-Rochex et al., 2012; Cheng et al., 2011), and both Dilp6 and Jeb expression were reduced by Pvf-Pvr loss. However, Dilp6 overexpression, but not Jeb overexpression, largely rescued the cell growth and survival defects caused by Pvr knockdown. Whereas Alk is required for neuroblast proliferation, InR is required for both neuroblast proliferation and cortex glia proliferation and morphogenesis (Avet-Rochex et al., 2012; Chell and Brand, 2010; Spéder and Brand, 2018). It is possible that Dilp6 rescues Pvr loss by stimulating InR-PI3K signaling in both neuroblasts and cortex glia.
Previous studies show that extrinsic contact-dependent cues are required to polarize embryonic neuroblasts, which is essential for proper asymmetric cell division (Siegrist and Doe, 2006). DE-cadherin, a contact-dependent cue that is essential for neuroblast proliferation (Doyle et al., 2017; Dumstrei et al., 2003), showed reduced expression in cortex glia upon Pvr loss, and DE-cadherin co-overexpression largely rescued the effects of Pvr knockdown. In the Drosophila germline, DE-cadherin functions in homophilic adhesion between germline stem (GSCs) cells and niche cells, and this anchors GSCs to the niche to promote self-renewal and orient asymmetric cell division (Inaba et al., 2010; Song et al., 2002). Recent studies indicate that, similarly, DE-cadherin is required to mediate anchoring between neuroblasts and cortex glia (Doyle et al., 2017). Further investigation will be required to determine how DE-cadherin regulates morphogenesis and survival of cortex glia, and promotes neuroblast self-renewal. Of note, DE-cadherin levels were not as affected in immature neurons upon glial-specific Pvr loss, indicating that cortex glia may regulate neuronal survival via different mechanisms.
RTK signaling and glial tumorigenesis
My results support an evolutionarily conserved role for PDGFR signaling in Drosophila glial development and tumorigenesis. In the mammalian CNS, PDGFRs are essential for development of neuro-glial stem/progenitor cells (Funa and Sasahara, 2014). In human GBM, PDGFRs drive tumor initiation and migration (Fomchenko and Holland, 2007). Previous studies show that, in Drosophila, ectopic Pvr signaling can drive neoplastic transformation and ectopic migration of retinal glia, and these studies have led to the identification of new pathways involved in migration of human GBMs (Kim et al., 2014; Witte et al., 2009). My results show that cortex glia are also subject to neoplastic transformation, and produce tumors in response to ectopic Pvr or EGFR activity. Human GBMs, including those with EGFR mutations, co-overexpress PDGFRs and PDGFs, and these redundant RTKs, and their ligands drive GBM tumorigenesis (Fig. S20A-E, Furnari et al., 2007). My results show that RTK-driven Drosophila glial tumors rely on similar autocrine Pvr signaling (Fig. 7K). Thus, the glial niche can originate brain tumors, and this process involves misregulation of pathways that control niche-neural stem/progenitor cell interactions.
In the presence of neoplastic cortex glia in the central brain, neuroblasts are eliminated. Neuroblasts may be lost through cell competition, a process in which oncogene-overexpressing cells actively kill their normal neighbors in response to differences in growth factor signaling or adhesion (reviewed by Amoyel and Bach, 2014), which in this case may be mediated by reduced DE-cadherin. In mammals, loss of normal stem/progenitor cells contributes to early stage tumorigenesis: in a mouse model of GBM, in the absence of different rates of proliferation, nascent neoplastic OPCs outcompete and eliminate normal OPCs to overtake the neurogenic niche prior to tumor emergence (Galvao et al., 2014). Although the mechanisms whereby neoplastic cells overtake the niche are unknown, elimination of normal cells may prime the microenvironment to support tumor growth (Galvao et al., 2014). Further study of interactions between normal and neoplastic neuroblasts and glia in the Drosophila central brain may reveal conserved mechanisms of development and tumorigenesis.
MATERIALS AND METHODS
Fly stocks, genetics and culture conditions
Drosophila stocks were obtained from the Bloomington Drosophila Stock Center unless otherwise noted as follows: Pvf1ex3-null mutant (a gift from M. Krasnow, Stanford University, CA, USA; Cho et al., 2002); UAS-Pvf2, UAS-Pvrλ and Pvf2-3−/− double null mutant and recombinants (gifts from E. Foley, University of Alberta, Canada; Parsons and Foley, 2013); Pvrc02195 mutant (a gift from D. Montell, University of California - Santa Barbara, CA, USA); wor-Gal4 (a gift from C. Doe, University of Oregon, Eugene, OR, USA); moody-Gal4 (a gift from R. Bainton, University of California - San Francisco, CA, USA); and alarm-Gal4 (a gift from M. Freeman, Vollum Institute, Portland, OR, USA). NP drivers were purchased from Kyoto Stock Center. R108-Gal4 and R122-Gal4 cortex glial drivers were isolated in my lab in an enhancer trap screen for new glial-specific Gal4 drivers (R.D.R., unpublished). The Pax6-EGFP promoter fusion reporter present in MiET Minos constructs (Metaxakis et al., 2005) was isolated in my lab as a specific reporter for cortex glia in the larval brain, and, to date, all MiET insertions tested show EGFP expression in cortex glia, labeling their cell bodies and processes. Kinome-wide RNAi stocks were obtained from VDRC and all stocks tested have been previously described (Read et al., 2013).
Flies were cultured at 25°C unless otherwise noted. Complex genotypes were established by standard genetics. Larval brain phenotypes were assessed as previously described (Read et al., 2009, 2013). Briefly, for larval brain size comparisons and neurogenesis experiments, embryos were collected for all genotypes side-by-side for 12-24 h and age-matched larvae at the same stages (1st, 2nd or 3rd instars) were collected for comparison by dissection, staining and imaging with confocal fluorescence microscopy. For late 3rd instar larvae, brains were collected at 130 h (5.5 days) after egg deposition (AED).
The screen was based on crosses that generated progeny containing a single RNAi construct exclusively expressed in GFP-labeled glia. Transgenes were overexpressed using the glial-specific repo-Gal4 transcriptional driver. Screening was performed using a fluorescence stereomicroscope to visualize GFP-labeled glia in living 2nd and 3rd instar larvae, and to compare RNAi brain size phenotypes with control brain size phenotypes. Those RNAi constructs that yielded obvious brain size reductions were confirmed with confocal microscopy. However, I cannot rule out the possibility that other RNAi constructs tested have subtler glial phenotypes. In the primary screen, each positive-scoring RNAi construct was tested at least twice with UAS-dcr in the background (to enhance knockdown), and at least twice without UAS-dcr in the background to confirm reproducibility and assess penetrance. To eliminate constructs that non-specifically interfered with glial cell development, positive scoring RNAi constructs were also tested for their effects on over-all brain growth in when overexpressed in neuroblasts or neurons using UAS-dcr; insc1407-Gal4, UAS-dcr; insc1407-Gal4; wor-Gal4, elav-Gal4c155; UAS-dcr, and/or UAS-dcr; ey-Gal4, and brains from resulting flies were scored on a fluorescence stereomicroscope and confirmed by confocal microscopy. Constructs that caused early organismal lethality were excluded from analysis. When available, dominant-negative and additional RNAi constructs were tested for candidate genes, and loss-of-function mutations were examined for brain growth phenotypes.
Immunohistochemistry and confocal imaging
Larval brains were dissected, stained and imaged as previously described (Read et al., 2009). All cell nuclei were visualized with DRAQ7 DNA dye, and actin filaments were visualized with phalloidin-Alexa-555 for low-magnification whole-brain images. Primary antibodies: rabbit anti-Pvr [1:500, a gift from D. Montell (McDonald et al., 2003)], rabbit anti-Pvr [1:500, a gift from J. Brugarolas, University of Texas Southwestern Medical Center, Dallas, TX, USA (Tran et al., 2013)], guinea pig anti-Dpn (1:500-1:1000, a gift from J. Skeath, Washington University, St Loius, MO, USA), rat anti-Dpn (1:2, a gift from C.-Y. Lee, University of Michigan, Ann Arbor, MI, USA), rabbit anti-eIF4E (1:400, a gift from P. Lasko, McGill University, Montreal, Canada), rat anti-Pvf1 [1:50-1:100, a gift from B. Shilo, Weizmann Institute of Science, Rehovot, Israel (Rosin et al., 2004)], rabbit anti-Repo (1:500, a gift from H. Reichert, University of Basel, Switzerland), mouse anti-Repo (1:10, DSHB), chicken anti-GFP (1:500, Aves), and goat or donkey anti-HRP (1:100-1:200, Jackson Immunolabs). For EdU labeling experiments, dissected brain ventral nerve cord complexes were incubated for 1 h in 10 μM EdU in Schneider's medium, fixed for 30-45 min in 4% formaldehyde 1×PBS, permeabilized in 0.3% Triton 1× PBS, followed by detection with an Alexa-647-labeled probe as per manufacturer's instructions (Click-iT EdU Imaging Kit, Life Technologies). TUNEL-labeling experiments were performed according to manufacturer's instructions. Otherwise, antibody staining was performed according to published protocols (Read et al., 2009). Fixed tissues stained for fluorescence microscopy were mounted dorsal side upwards in Vectashield (Vector Laboratories), and whole brains were imaged using confocal microscopy.
Fluorescent images were acquired with a Zeiss LSM 510 or LSM 700. Confocal projections were prepared using Zen. Images were subject to standard processing using Zen or Photoshop (Read et al., 2009). Nuclear Dpn stains (guinea pig antibody) were corrected for particulate background fluorescence in Zen using the median filter. For experiments in which protein levels and/or localization were compared between genotypes (tGPH, eIF4E and DEcadGFP), brain samples were isolated from age-matched animals grown side by side, and brains were prepared and imaged in parallel using the same microscope and confocal imaging settings, and processed using Zen and Photoshop in the same manner. All experimental and control brains shown in each figure panel grouping are from age-matched representative animals grown alongside each other, unless noted otherwise. For figures with low-magnification whole larval brain images, brains were imaged separately and panels were assembled in Photoshop or PowerPoint. Image quantification procedures are described in the supplementary Materials and Methods.
RNA sequencing and human tumor genomics
Primary glioblastoma (GBM) neurosphere cultures were isolated from surgical specimens donated for research with informed consent from patients, and were collected and used according to recognized ethical guidelines in a protocol (IRB00045732) approved by the Institutional Review Board at Emory University. GBM cultures were maintained as per published protocols (Read et al., 2013). RNAseq was performed for each culture as per commercial procedures (Beckman Coulter), and primary sequencing data were subject to bioinformatic processing (Winship Bioinformatics Core), assembly, and gene and transcript expression analysis using TopHat and Cufflinks. mRNA expression data for primary glioblastoma tumor tissue specimens and for lower grade glioma tumor tissue specimens (grade II and III gliomas) were obtained from TCGA datasets through the cBio Portal (www.cbioportal.org, PanCancer Atlas dataset). Additional details are provided in the supplementary Materials and Methods.
Drosophila brains were dissected away from the body, ventral nerve cord and all other larval tissues. RNA was prepared from Trizol as per manufacturer's instructions (Life Technologies), cDNAs were prepared as per manufacturer's instructions (Bio-Rad), and qPCR was run on a BioRad CFX96 Touch as per manufacturer's instructions. Primer sequences for Drosophila loci were obtained from DSRC FlyPrimerBank (www.flyrnai.org/flyprimerbank). For each comparison, each experiment included technical replicates (in duplicate) and, when possible, two separate primers were used for each gene to confirm expression data (bar graphs show only results for one primer). Results from gapdh and rp49 were used as controls to normalize relative gene expression levels as per standard ΔΔCq methods as described by the manufacturer's instructions.
Experimental data were compared pairwise using Student's two-tailed t-test or, for the membrane intensity datasets, paired Student's t-test. Graphs show averages for each value presented. Error bars show s.e.m. unless otherwise noted.
I thank Coston Rowe for technical assistance, Jeffrey Olson for GBM tissues, and Brendan Parsons and Edan Foley for Pvr and Pvf mutants.
Conceptualization: R.D.R.; Methodology: R.D.R.; Validation: R.D.R.; Formal analysis: R.D.R.; Investigation: R.D.R.; Resources: R.D.R.; Data curation: R.D.R.; Writing - original draft: R.D.R.; Writing - review & editing: R.D.R.; Project administration: R.D.R.; Funding acquisition: R.D.R.
This research was funded by the National Institute of Neurological Disorders and Stroke (R00 NS065974). Deposited in PMC for release after 12 months.
The RNA-sequencing data have been deposited in Gene Expression Omnibus under Accession Number GSE122679.
The author declares no competing or financial interests.