Lgl1 controls NG2 endocytic pathway to regulate oligodendrocyte differentiation and asymmetric cell division and gliomagenesis

Oligodendrocyte progenitor cells (OPC) undergo asymmetric cell division (ACD) to generate one OPC and one differentiating oligodendrocyte (OL) progeny. Loss of pro-mitotic proteoglycan and OPC marker NG2 in the OL progeny is the earliest immunophenotypic change of unknown mechanism that indicates differentiation commitment. Here, we report that expression of the mouse homolog of Drosophila tumor suppressor Lethal giant larvae 1 (Lgl1) is induced during OL differentiation. Lgl1 conditional knockout OPC progeny retain NG2 and show reduced OL differentiation, while undergoing more symmetric self-renewing divisions at the expense of asymmetric divisions. Moreover, Lgl1 and hemizygous Ink4a/Arf knockouts in OPC synergistically induce gliomagenesis. Time lapse and total internal reflection microscopy reveals a critical role for Lgl1 in NG2 endocytic routing and links aberrant NG2 recycling to failed differentiation. These data establish Lgl1 as a suppressor of gliomagenesis and positive regulator of asymmetric division and differentiation in the healthy and demyelinated murine brain.

A dult oligodendrocyte progenitor cells (OPC) expressing the proteoglycan NG2 (CSPG4) continuously divide and generate differentiating oligodendrocytes (OL) throughout adulthood 1 . OPC divisions are frequently asymmetric and generate progeny of distinct fate, where only one of the daughter cells preserves NG2 expression and the opposite daughter cell downregulates NG2 2,3 . Thus far, asymmetric distribution of NG2 protein is the earliest immunophenotypic change amongst daughter cell pairs generated by asymmetric cell division (ACD). NG2-positive daughter cells proliferate at higher rates than NG2negative cells, showing that the early phenotypic asymmetry correlates with distinct short-term fate 2 . NG2 contributes to establishing this cell fate bias within newly generated cell pairs after ACD. NG2 binds platelet-derived growth factor-AA (PDGF-AA) and PDGF receptor alpha (PDGFRα) and thereby enhances PDGFRα signaling and promotes timely OPC proliferation [4][5][6][7] . Moreover, long-term fate tracking of the ACD progeny showed that early NG2 asymmetry permanently affects cell fate. The NG2-positive progeny of ACD retain OPC characteristics while the NG2-negative progeny upregulate CC1, a marker for commitment to the OL fate, and eventually differentiate 2,8,9 . Importantly, chemical-induced demyelination increases the rates of NG2 asymmetry 9 . Collectively, the data underline that asymmetric distribution of NG2 marks and also actively generates different OPC progeny. Furthermore, ACD balances OPC proliferation with differentiation in the normal brain and generates OL in demyelinated lesions to contribute to remyelination 2,3,9,10 . It is unclear how the downregulation of NG2 protein is achieved in the differentiating oligodendrocyte. This lack of mechanistic insights into ACD and early differentiation limits our understanding of brain homeostasis.
OPC give rise to glioma in genetically engineered mouse models [11][12][13][14] . When undergoing neoplastic transformation, OPC show higher rates of symmetric self-renewing divisions at the expense of ACD 2 . These data suggest that downregulation of NG2 in OPC progeny is critical for ACD, differentiation, and attenuation of tumorigenesis 2 .
Surface levels of the membrane-spanning NG2 protein are regulated by clathrin-mediated endocytosis in mouse embryonic fibroblasts 15 . Clathrin-mediated or receptor-mediated endocytosis is a multi-step process, whereby membrane-localized proteins are engulfed as cargo in clathrin-coated pits, which then bud off the membrane to form the early endosome. Cargo proteins are then sorted into either the recycling endosome and re-integrated into the membrane or targeted to the late endosome and subsequently the lysosome for degradation 16 . It is not known whether NG2 trafficking by the endocytic pathway is important for NG2 downregulation and OL differentiation.
The WD40 repeat-containing protein lethal giant larvae (Lgl) was initially characterized as a tumor suppressor gene, in Drosophila melanogaster 17 , and it was subsequently shown to mediate ACD of Drosophila neuroblasts 18 . Lgl is an evolutionary conserved protein that initiates cell polarity by recruiting proteins to membrane subdomains (for review, see ref. 19 ). As one of two mammalian Lgl genes 20 , Lgl1 is highly expressed in the brain 21 . Lgl1 knockout studies during mouse embryogenesis have revealed a function for Lgl1 in polarity and adherens junction integrity in neuroepithelial cells 22 , and in suppressing proliferation of dorsal telencephalon radial glial progenitors at early postnatal stages 23 . A bona-fide tumor suppressor role for Lgl1 in gliomagenesis is supported by studies showing that loss of tumor suppressor PTEN, a frequent and early driver mutation in glioblastoma 24 , inhibits Lgl1 activity 25 . Constitutively active Lgl1 inhibits invasion and induces differentiation of primary human glioblastoma cells 26 . Nevertheless, a tumor suppressive role for Lgl1 in OPC has not been investigated yet.
Here, we report that Lgl1 expression is upregulated during OL differentiation and that Lgl1 protein is detected in committed OL in the adult murine brain. Lgl1 conditional knockout (cKO) OPC show a differentiation defect characterized by an aberrant coexpression of NG2 with OL commitment markers. Moreover, in Lgl1 cKO OPC rates of ACD are reduced while rates of symmetric, self-renewing divisions and proliferation are increased, in both intact and chemically demyelinated corpus callosum (CC). Lgl1 knockout synergizes with hemizygous Ink4a/Arf knockout in OPC to induce gliomagenesis. Time lapse imaging of surfacelabeled, endocytosed NG2 shows decreased co-localization with the lysosome in Lgl1 cKO OPC. Additionally, total internal reflection microscopy reveals that in Lgl1 cKO OPC NG2 bypasses degradation and is rather recycled to the membrane. Aberrant NG2 recycling is linked to defective OPC differentiation. Thus, we establish Lgl1 as a positive regulator of differentiation and ACD and suppressor of gliomagenesis. We suggest that Lgl1 promotes differentiation by supporting NG2 routing to the lysosome.

Results
Lgl1 expression is upregulated in OL differentiation. To unravel the mechanism for NG2 down-regulation, we searched the RNA-Seq transcriptome database 27 for conserved ACD regulator genes with altered expression in OL versus OPC. We noticed a four-fold increase of Lgl1 (Llgl1) transcript in newly formed OL versus OPC, suggesting that Llgl1 expression is increased upon OL differentiation. To directly test this notion, OPC were isolated from the white matter area of the CC of postnatal days 30-60 (P30-P60) mice and cultured in parallel under proliferation and differentiation conditions (Fig. 1a). We performed immunofluorescence staining for OPC marker NG2 and mature OL marker O4 to show differentiation. As expected, we found O4-expressing cells displaying a branched OL morphology only under differentiation but not under proliferation conditions (Fig. 1b). We used quantitative real-time PCR (qrt-PCR) to assess changes in Llgl1 transcript levels with differentiation and aging. Analyses reveal a significant increase in Llgl1 transcript under differentiation conditions (2.5-fold, p < 0.01) (Fig. 1c), along with differentiation markers Mbp, Slc1a1, Ndrg1, GPR17, and Mog. We noted that OPC markers, such as Cspg4 (NG2), Olig2, and Sox10 are unchanged. Llgl1 transcript levels increase with age, being higher at P60 in OPC, and reaching a peak at P180 in OL (Supplementary Figs. 1d, e).
Hence, Lgl1 expression increases with differentiation and overlaps largely with differentiating OL, suggesting therefore its potential role in a differentiation mechanism.
Taken together, these data show that upon Lgl1 knockout, OL aberrantly retain NG2 protein expression and expand as OPC marker positive cells, while failing to further differentiate and mature.
In contrast to postnatal and adult OPC, embryonic OPC are not committed to the oligodendrocyte lineage and reportedly also generate astrocytes or neurons 9,30,31 . To test if Lgl1 deletion affects the lineage commitment in young adult OPC, we quantified expression of neuroblast and astrocyte markers CD24 and GLAST, respectively, by flow cytometry (Supplementary Figs. 2a, b). As expected, only a small fraction of recombined cells expressed neuroblast and astrocyte markers (<2%) and the frequency of neuroblasts and astrocytes did not change significantly following Lgl1 deletion ( Supplementary Fig. 2b). We concluded that Lgl1 loss does not alter the potential of young adult OPC to generate astrocytes and neuroblasts, suggesting a specific role of Lgl1 in OL differentiation.
Demyelinating injury increases the rate of OL proliferation and differentiation when compared with non-lesioned murine brain 32,33 . To test if Lgl1 is also required for OL differentiation in demyelinating lesions, we quantified the frequency of mature OL marker Plp in lysolecithin-induced lesions in NG2-Cre ERT2 Ai14-TdTom Lgl1 fl/fl and Lgl1 wt/wt mice injected with tamoxifen at P52 (Supplementary Figs. 2e, f) Lgl1 is required for asymmetric OPC divisions. Postnatal and adult OPC divide asymmetrically to generate at a one-to-one ratio an NG2+ OPC and a CC1+ OL 2,3,10 . Given that Lgl1 knockout reduces the number of CC1+ cells (Fig. 2c), we hypothesize that Lgl1 regulates ACD. To test this hypothesis, we visualized and quantified ACD in Lgl1 NG2-Cre ERT2 Ai14-TdTom Lgl1 fl/fl and Lgl1 wt/wt mice that were injected with tamoxifen once at postnatal day 30 (P30) for clonal recombination 35 . To monitor proliferation, mice were treated with EdU for 14 days (Fig. 3a), and cell pairs were identified by co-immunostaining for Tom, CC1, and EdU 2,3 (Fig. 3b). Cell pairs in the corpus callosum were scored as (i) symmetric, self-renewing (both cells = CC1 negative), (ii) asymmetric (one CC1-positive and one CC1-negative cell) and as (iii) symmetric differentiating (both cells = CC1 positive; Fig. 3b). The average cell cycle length of OPC increases from~3 days in P20 to 9 days in P60 animals 1 . Assuming a linear increase in cell cycle lengths between P30 and P45, recombined OPC should have divided two to three times during this period. Indeed, we were able to detect recombined cell pairs at P45. Amongst these, ACD rates were reduced in the Lgl1 cKO (WT = 23.2% vs. cKO = 12.5%; p = 0.023). Vice versa, symmetric, self-renewing divisions were increased in the Lgl1 cKO (WT = 28.4% vs. cKO = 51.2%; p = 0.028). Noteworthy, symmetric differentiated (CC1+) cell pair rates were not significantly changed upon Lgl1 cKO (WT = 48.4% vs. cKO = 36%; p = 0.2) ( Fig. 3c and Supplementary Table 1).  Table 1).
To test for a disruption of ACD ex vivo, we generated Lgl1 KO and Lgl1 WT control OPC and plated them for pair assays, as described ( Fig. 3e) 2 . We found that the frequency of ACD was reduced in Lgl1 KO OPC (WT = 72% vs. cKO = 49%) ( Fig. 3f; p = 0.05).
These data show that Lgl1 does not prevent CC1 expression, but rather prevents downregulation of OPC marker NG2, thereby lowering rates of asymmetric distribution of NG2. Thus, Lgl1 is required for ACD and restricts symmetric, selfrenewing divisions in the healthy and injured brain and in the injured spinal cord.
Lgl1 knockout increases OPC proliferation and promotes gliomagenesis. Given that Lgl1 knockout cells fail to restrict expression of pro-mitotic NG2, we hypothesize that their proliferation rates might be increased. We therefore assessed proliferation in NG2-Cre ERT2 Ai14-TdTom Lgl1 fl/fl and Lgl1 wt/wt mice injected with tamoxifen at P30 and exposed to EdU at P45 ( Figs. 4d-f). These data indicate that Lgl1 restricts proliferation of OPC in the healthy and demyelinated CC. Increased proliferation, defective differentiation, and decreased rates of ACD are associated with gliomagenesis 2 . We therefore tested whether Lgl1 knockout promotes tumor formation. Immunostainings of normal adult mouse brain revealed that OL express the tumor suppressor p19 Arf (Supplementary Fig. 5a) 36 . We therefore bred Lgl1 floxed mice 21 with a NG2-Cre driver line 37 , and mice heterozygous for the Ink4a/Arf deletion (Ink4a/Arf+/−) 38 . Mice were monitored for neurological signs of tumor formation and survival. We observed that 100% of Lgl1 WT Ink4a/Arf+/− mice were still alive at 60 weeks postnatal age, whereas Lgl1 cKO Ink4a/Arf+/− mice had a significantly shorter median survival of 45.79 weeks ( Fig. 4c; p = 0.0108). Histopathologic analyses of Lgl1 cKO Ink4a/Arf+/− brains showed infiltrative tumors consistent with astrocytoma morphology (Fig. 4d). Expression analysis by qrt-PCR in tumor cells isolates showed a strong reduction of p19 Arf mRNA versus Lgl1 cKO Ink4a/Arf WT and Ink4a/Arf+/− cells, suggesting that the WT allele of p19 Arf is not expressed in the tumor ( Supplementary Fig. 5b). Fluorescent immunostainings of brain sections from tumor-bearing mice showed that tumors express the oligodendrocyte lineage and glioma markers Olig2 and NG2. Results are indicative of a mixed astrocytic and oligodendroglial high-grade glioma (Figs. 4d-f) 39 .
To determine whether our animal model findings were clinically relevant, we mined expression data from 99 patients from The Cancer Genome Atlas (TCGA) for a potential association of Lgl1 mRNA expression levels and glioblastoma patient survival. We found that low levels of Lgl1 mRNA expression correlated with poor patient survival ( Supplementary  Fig. 5c) 40 . In further support of a tumor suppressive function, Lgl1 protein expression was absent in six out of seven human glioblastoma xenografts as shown by immunoblotting (Supplementary Fig. 5d). Collectively, these data showed that Lgl1 attenuates OPC proliferation and synergizes with tumor suppressor Ink4a/Arf to suppress gliomagenesis.
Lgl1 cKO OPC show a defect in NG2 routing to the lysosome. A recent study showed that levels of surface NG2 are regulated by endocytosis 15 . We hypothesize that Lgl1 cKO OPC fail to route NG2 to the lysosome and thereby sustain aberrantly high NG2 levels. In order to test this, we transfected Lgl1 cKO and WT cells with lysotracker GFP (Lys-GFP) and labeled cell surface NG2 with an antibody specific for its extracellular domain (NG2-EC) and a fluorescent secondary antibody. Cells were consecutively monitored by fluorescent time-lapse microscopy for up to three hours under differentiation conditions. In both Lgl1 cKO and WT OPC, NG2-EC started to reach the lysosome one hour post imaging (Fig. 5a). Co-localization of NG2-EC with Lys-GFP peaked in Lgl1 WT cells three hours after starting the imaging. Interestingly, we observed a significantly higher colocalization of NG2 with lysosomes in WT OPC compared to Lgl1 cKO OPC (WT = 79.6%; cKO = 54.7%, p = 0.0047) (Fig. 5a) 00:00 suggesting that in differentiating OPC, Lgl1 acts as a positive regulator of NG2 trafficking to the lysosome. These results prompted us to stain for the lysosome associated membrane protein 1 (Lamp1) to investigate whether the lysosome is formed properly in Lgl1 knockout cells. We find that lysosomes are present in Lgl1 cKO OPC, based on Lamp1 positivity and subcellular localization. Noteworthy, lysosome shape appears to be altered due to Lgl1 knockout (Fig. 5c). Collectively, these results show that Lgl1 positively regulates NG2 trafficking to the lysosome.
Non-degraded NG2 in Lgl1 cKO OPC is recycled to the membrane. Given that NG2 is a transmembrane proteoglycan 41 , we suggest that it might be recycled to the membrane by the recycling endosome, a tubular structure continuous with the early endosome. To test this, we incubated Lgl1 cKO cells in differentiating conditions and live labeled NG2-EC. Cells were fixed at different time points after surface NG2 labeling and stained for recycling protein markers. At 45 min, we observed a higher colocalization of NG2-EC with the recycling regulator Rab11 in Lgl1 cKO compared to WT cells (Fig. 6a). colocalizes with ARF6, a GTPase critical for endocytic recycling, supporting the notion that NG2 is recycled to the membrane in Lgl1 cKO OPC (Fig. 6b). Additionally, NG2-EC overlaps with Rab4, a GTPase marker for fast endocytic recycling, hinting that NG2 recycling may not be restricted to the slow recycling pathway (Fig. 6c). We also observed co-localization of NG2-EC with the early endosome marker EEA1 (Fig. 6d), suggesting that NG2 is constantly uptaken and shuffled back to the membrane. Noteworthy, we observed a coexpression of Lgl1 and the early endosome-specific protein Rab10 both at the membrane and around the nucleus, further associating Lgl1 role with the endocytic machinery 42 (Supplementary Fig. 6).
In order to better investigate a potential NG2 recycling in Lgl1 cKO cells, we used total internal reflection fluorescence (TIRF) to monitor live-labeled NG2-EC docking to the membrane. Cells were plated in differentiation medium for four hours and then monitored. By this time, the majority of NG2 is degraded in Lgl1 WT cells and no significant NG2 recycling was observed (Fig. 6e,). To the contrary, NG2-EC was more abundant and continuously re-introduced to the membrane in Lgl1 cKO OPC (Figs. 6f, g; Supplementary Movie 1 see Methods).
Collectively, these data proof that in Lgl1 cKO cells, NG2 is mostly recycled to the membrane rather than degraded.
In order to investigate if early endocytosis is affected in Lgl1 cKO cells, we labeled surface NG2 with the NG2-EC antibody, and marked the early endosome by expression of Rab5a-GFP. Time-lapse imaging of cells for 30 min in differentiation conditions showed a prominent colocalization of NG2 with Rab5a only in mutant cells, suggesting that Lgl1 negatively regulates NG2 uptake (Fig. 7d). We corroborated these results by quantifying the endocytosis of transferrin using FACS, and found a ten-fold increase in transferrin incorporation in Lgl1 cKO OPC as compared to Lgl1 WT OPC (Supplementary Fig. 7).
Next, we tested whether NG2 is only recycled after failing to be degraded or also when prematurely internalized in Lgl1 cKO OPC. We used TIRF to monitor labeled NG2 docking to the membrane over the first 5 min of differentiation conditions. We observed that a significantly higher ratio of labeled NG2 was endocytosed and recycled to the membrane in Lgl1 cKO cells compared to WT cells (WT = 15%; cKO = 58%, p < 0.0001) (Figs. 7e-g). Taken together, these data show that Lgl1 negatively regulates NG2 internalization and recycling and establish Lgl1 as a regulator of NG2 trafficking by the endocytic pathway.
Differentiation defect in Lgl1 cKO OPC is linked to aberrant NG2 recycling. Our studies above suggest that Lgl1 knockout enhances NG2 internalization and recycling to the membrane, thereby preventing NG2 degradation and blocking OL differentiation. In order to dissect whether abnormal NG2 recycling has an effect on OL differentiation, Lgl1 cKO cells were incubated for 4 hours with monensin (10 μM), a drug widely used to interrupt protein recycling 43 . We tested if the drug has a specific effect on NG2 recycling in Lgl1 cKO cells by monitoring Tom+ cells under TIRF for up to one hour after monensin treatment. We observed that monensin treatment blocked NG2 recycling, validating the reported drug effect on recycling (Fig. 8a) 43 . Lastly, we assessed differentiation in treated and non-treated Lgl1 cKO cells kept in differentiation conditions for 5 days by O4 and NG2 co-immunostaining. A quantification of marker-positive cells showed that monensin treatment significantly increased the ratio of O4+ cells at the expense of NG2+ cells (treated/O4+ = 75% and NG2+ = 24%; non-treated/O4+ = 9% and NG2+ = 90%; p < 0.0001) (Figs. 8b, c).
These results link abnormal recycling of surface NG2 and defective OL differentiation.

Discussion
Adult OPC differentiate in multiple steps, beginning by committing to OL fate and ending by maturing and producing myelin. There is a wealth of studies on the later steps of OL differentiation and maturation 34,44 . Little is known, however, about the earliest decisions between OPC and OL fate. Thus far, the downregulation of proteoglycan NG2 is the earliest immunophenotypic change that distinguishes the two daughter cells within a newly formed OPC pair upon ACD. NG2 marks the OPC daughter, while the NG2-negative daughter acquires CC1 expression, which marks OL differentiation 2,3,9 . NG2 knockout OPC show delayed proliferation, which indicates that NG2 controls the timing of OPC proliferation 7 (reviewed by Sakry et al. 45 ). The NG2 knockout does not fully block OPC proliferation and differentiation and its effect is only temporary 7 . Thus, downregulation of NG2 is an important but not the only step for OL commitment. Indeed, PDGFRα is also asymmetrically distributed amongst newly formed OPC pairs 3 , further suggesting that cell fate determinants in addition to NG2 link ACD to differentiation 46 .
Due to the reported functional, morphologic, and neuroanatomical heterogeneity of OPC 47 , and its potential influence on cell division mode, cell cycle lengths, and differentiation 1,9,48,49 , we focused our studies on OPC in the corpus callosum. We analyzed OPC in mice between the ages P30 and P60 with the exception of tumor-bearing mice, which were older. The rationale for analyzing OPC at these stages was that the analyses of mice older than P60 would be technically very difficult due to a long OPC cell cycle time 1 . We expect to detect the maximum number of ACD between P30 and P60, because the rates of asymmetric OPC divisions are already high at postnatal stages (P20; ACD =~35%) and peak at P60 (ACD =~45%) before decreasing again with age (P120; ACD =~35%) 3 . Moreover, the rate of differentiation remains stable at 40% between P30 and P60 9 , which makes is a suitable range to identify potential effects on differentiation of a single gene knockout.
Since ACD generate differentiated cells from stem and progenitor cells across species, it was attractive to speculate that conserved regulators of ACD, such as Lgl1, regulate early steps in OL differentiation. Previous studies showed that Lgl1 localizes to the lateral membranes and cytoplasm of embryonic neural progenitor cells 21,22 . Expression of Lgl1 protein in rat neocortex homogenates persists into adulthood 42 . Functional analyses in the adult brain, however, were limited to its role in SVZ neural stem cells, neurons, and astrocytes [21][22][23]42 .
Here, we show that Lgl1 cKO cells fail to lose NG2 expression and exhibit differentiation defects. Interestingly, we find that Lgl1 knockout in OPC increases the rates of cells in a transitional state and expressing both OPC marker NG2 and OL marker CC1, at the expense of mature OL. These data indicated that Lgl1 loss does not prevent the expression of CC1 but rather prevents downregulation of NG2, and thereby permanently attenuates differentiation.
Whereas normal OPC populate the brain sparsely and show little overlap in the areas that two adjacent OPCs occupy 50 , they are denser in demyelinated lesions, such as those typically found in chronic inflammation models in mice and in multiple sclerosis patients 32,33 . We found that Lgl1 is a critical regulator of OL differentiation during remyelination following chemical-induced demyelination. OPC are considered a critical source for remyelination and failed OPC differentiation are thought to be one important underlying cause for failed remyelination in chronically demyelinated lesions 51 . It is interesting to speculate that Lgl1 functions at an early step in the complex regulatory process of remyelination and thus may be a viable target for approaches aimed at promoting proper differentiation of OPCs in demyelinated lesions.  Pictures are acquired each 30 seconds as soon as cells were exposed to differentiation conditions. Each white puncta in focus represents labeled NG2-EC docked at the membrane. Individual recycled puncta are highlighted with dashed circles of different colors. Note that in Lgl1 cKO cells, the majority of the labeled NG2 is internalized and recycled to the membrane whereas in WT cells, the majority of NG2 puncta retains a stable localization at the membrane. Scale bars: 10 μm. g Graph representing the quantification of NG2 recycling at the membrane throughout the first 5 min of differentiation conditions. Data represented as mean ± s.e.m. n = 3 cells per genotype, three independent experiments (***p < 0.005, Student's t-test) Our analyses of cell division modes support the hypothesis that Lgl1 restricts NG2 during ACD as part of the mechanism to turn on OL differentiation in the NG2-negative progeny. Moreover, Lgl1 does not disrupt symmetric, differentiating divisions, which suggests that it specifically regulates differentiation in the ACD progeny. To date, only a few conserved ACD regulators have a known function in OL differentiation in addition to Lgl1. These include the polarity regulator partitioning-defective 3 (Par3) protein, which promotes myelination in the peripheral nervous system 52 , and the transcriptional regulator Prox1, which is required for OL lineage in the mouse. Par3, Prox1, and Lgl1 are in the same protein network that controls ACD in Drosophila neuroblasts 46 . A role for Par3 and Prox1 in ACD of OPC has yet to be determined.
Multiple studies have shown that OPC 13,14 and OL 36 are a cellular origin of glioma in genetically engineered mouse models. The neoplastic transformation of OPC and OL and formation of gliomas are associated with increased proliferation, decreased differentiation, changes in the tumor microenvironment 53 , and a decrease in ACD 2 . Lgl1 knockout during embryonic neurogenesis disrupts the asymmetric division of neuroepithelial cells and generates rosette-like structures reminiscent of PNETs but perinatal death prevented the analyses of tumor formation 21 . Lgl1 knockout in cell types known to only undergo symmetric but not asymmetric divisions, such as astrocytes, did not generate premalignant lesions 23 . Taken together with our data, these findings suggest that defective ACD might be an underlying cause for proliferation and differentiation defects in glioma cells. Our data support the notion previously observed in epithelial cells 54 that loss of polarity-albeit by itself not sufficient to induce carcinogenesis-contributes to tumor formation. We show that Lgl1 loss synergizes with Ink4a/Arf loss, which is a frequent and early event in glioma 24 , to induce gliomagenesis (Fig. 4). There is strong precedence for Ink4a/Arf heterozygous mice to be as susceptible as Ink4a/Arf nullizygous mice to gliomagenesis 55 . A likely explanation for this is, that the remaining Ink4a/Arf allele in tumors arising from Ink4a/Arf+/− mice is inactivated at high frequency, which is supported by our findings that p19 Arf mRNA expression is greatly reduced in Lgl1 KO tumors. Although 19 Arf and p16 Ink transcripts are reportedly low in OPC and OL, we find overlapping expression of the p19 Arf protein and OL marker CC1 ( Supplementary Fig. 5). This is interesting, because p19 Arf but not p16 Ink is an important tumor suppressor in a glioma mouse model study 36 . Collectively, the data suggest that tumor formation in NG2-Cre Lgl1 KO Ink4a/Arf+/− mice required inactivation of p19 Arf expression, hence the long latency ( Supplementary  Fig. 5). It is also plausible that Lgl1 suppresses tumor formation by restricting NG2 function as a co-receptor for receptor tyrosine kinases, since deregulated PDGFRα and EGFR signaling activity contributes to gliomagenesis 40,56 (for review see ref. 57 ). Endocytosis has a range of functions including regulating the cell surface expression and subcellular localization of proteins, remodeling the membrane and controlling signaling pathways 16 . Studies in Drosophila epithelia cells support a link of polarized protein localization and endocytosis by showing that loss-offunction mutants of the critical endosome adaptor AP2 leads to defects in polarity and differentiation similar to those observed in Lgl loss-of-function experiments 58 . Our study proposes an unprecedented link between defects in NG2 endocytic trafficking, ACD, and differentiation. Indeed, in Lgl1 cKO mutant cells, we find that NG2 is not properly routed to the lysosome and instead is recycled to the membrane. These observations show that endocytic pathway defects in Lgl1 cKO OPC disturb the regular intracellular itinerary of NG2 during differentiation and lead to its recycling rather than degradation. NG2 levels are permanently suppressed in OL most likely as a result of downregulation of the NG2 promoter in the OL daughter after ACD 3 .
Taken together, these data suggest that the unequal distribution of NG2 in the newly formed progeny of ACD is achieved by NG2 degradation in the lysosome and is maintained by silencing the NG2 promoter in the OL daughter.
We can only speculate as to whether Lgl1 regulates NG2 trafficking directly or, indirectly, by regulating the endocytic machinery. Although not previously implicated as a negative regulator of endocytosis in mammalian cells, Lgl1 activates endocytosis protein and GTPase Rab10 by releasing the GDP dissociation inhibitor (GDI) in rat hippocampus to induce polarized axon growth 42 . In cells undergoing ACD, such as Drosophila neuroblasts, Lgl1 establishes cell polarity by mechanisms that regulate myosin II activity, directed vesicle transport and the subcellular localization of proteins (for review see ref. 46 ). Nonmuscle myosin II (NMII) is a critical regulator of clathrin-mediated endocytosis 59 and is a negative regulator of oligodendrocyte morphological differentiation 60 . Lgl1 associates with NMII and inhibits filament formation 61 and reduces the association of NMII with actin. Whether Lgl1 regulates endocytosis by affecting motor protein dynamics or through other mechanisms, such as regulating the subcellular localization of the GTPases 62,63 , remains to be determined. In further support of an indirect effect of Lgl1 on endocytic trafficking, Drosophila Lgl1 regulates vesicle acidification in the lysosome 64,65 . Noteworthy, our data indicate that lysosomes in Lgl1 cKO OPC have a tubular shape rather than the vesicular shape seen in WT OPC, hinting that lysosome maturation might be affected by Lgl1 knockout (Fig. 5).
Future studies should investigate further the potential link of defects in endocytosis and gliomagenesis.

Methods
Mice. All mouse experiments were approved by and performed according to the guidelines of the Institutional Animal Care and Use Committee of the University of California, San Francisco. Lgl1 lox/lox mice were courtesy of Dr. Vasioukhin 21 and were bred with NG2-Cre (Jax strain 008533) 37 and GFAP-Cre (Jax strain 004600) 66 , NG2-CreER TM (Jax strain 008583) 9 and Ai14 Cre reporter mice (Jax strain 007908) 29 mice or Ink4a/Arf hemizygous knockout mice 38 . Ascl1GFP 67 and NG2DsRed 37 mice from Jackson Laboratories were bred together to obtain Ascl1GFP-NG2DsRed double transgenic mice. FVB/N mice were obtained from Simonsen Laboratories, Gilroy. For survival studies, Lgl1 cre-conditional deleted mice were kept under observation until reaching the major endpoint, which was animal neurological symptom-free survival.
Induction of Lgl1 knockout and demyelination. Tamoxifen (Sigma-Aldrich) was dissolved in corn oil (Sigma-Aldrich) at 20 mg/mL and injected intraperitoneally (i.p.) at 1 mg to induce Cre activity Lα-lysophosphatidylcholine (lysolecithin = LPC, Sigma-Aldrich) was dissolved at 1% in physiological serum and 0.5 μl were injected to induce demyelination. For clonal labeling of NG2 cells P30-P52 NG2-Cre ERT2 , TdTom, Lgl1 fl/fl and NG2-Cre ERT2 , TdTom, Lgl1 wt/wt mice were injected with 1 mg tamoxifen twice within 24 h 35 . To assess proliferation rates and cell division modes of OPC, mice were subsequently treated with 0.1 mg/mL EdU in drinking water for 14 days and resected brains were processed for immunofluorescence. For differentiation NG2-Cre ERT2 , TdTom, Lgl1 fl/fl and NG2-Cre ERT2 , TdTom, Lgl1 wt/wt mice were injected with tamoxifen either once or twice daily for 4 consecutive days. Mice were analyzed 30 days post injection.
For broader deletion of Lgl1, P30-P52 NG2-Cre ERT2 , TdTom, Lgl1 fl/fl and NG2-Cre ERT2 , TdTom, Lgl1 wt/wt mice were injected with tamoxifen twice daily for 4 consecutive days. FACS isolated OPC were subjected at P45 to microarray analyses. To assess proliferation and cell division modes in demyelinated lesions, demyelination was induced by a single LPC injection into the CC or the spinal cord of mice previously injected with tamoxifen. For differentiation, demyelination was induced by LPC injection after the last tamoxifen injection and mice were sacrificed 10 days after to evaluate the effect on OPC differentiation. For spinal cord white matter demyelination, mice were anesthetized with isofluorane and demyelination was induced by LPC into the ventral or dorsal white matter funiculus at the level of T12 68,69 . For CC demyelination, LPC was injected in the right or the left hemisphere (AP+1 mm, L ± 1 mm, DV-2.1 mm) 70 . For EdU labeling, mice were injected i.p. with 10 mg/mL EdU twice a day, or EdU was mixed with drinking water at a low concentration (0.1 mg/mL) for 5-14 days. For long-term tracking of Lgl1 deletion effects, tamoxifen was injected at P30 for 5 consecutive days, and tissue analyzed by FACS at P120.
Cell culture. Cells were isolated from the CC, which includes the subcallosal zone of P30-P60 mice, and cultured as non-adherent spheres in proliferative medium with NT3, CNTF, PDGFA, and FGF2 (Peprotech) 34 . Differentiation was induced by plating cells on poly-L-Lysine coated glass-bottom 96-well plates (CORNING) at 5000 cells/mL in medium containing CNTF, FGF2, and T3 2 . Cells were incubated with inhibitor of endocytosis Pitstop1 at 25 μM. To induce Lgl1 deletion in vitro, cells were incubated with adenovirus-cre-GFP (Vector BioLabs) and adenovirus-GFP 2 . Viability assay was performed using the Alamar Blue reagents 71 .
Pair assay. Cells were subjected to a pair assay by plating dissociated single cells at clonal density for 12-36 h. Cell pairs formed were fixed and subjected to immunostaining with primary antibody and fluorescence-coupled secondary antibody 2 . Nuclei were visualized using DAPI (Sigma). Unless indicated otherwise, for pair assay a minimum of 50 and up to 200 cell pairs per cell type and condition, respectively, were analyzed, using a fluorescent microscope.
Immunofluorescence and in vivo cell pair assays. NG2-Cre ERT2 ,TdTom, Lgl1 fl/fl and NG2-Cre ERT2 ,TdTom, Lgl1 wt/wt mice injected with tamoxifen were PFA perfused, brains were resected and frozen, sectioned and post-fixed for 10 min with 4% PFA, blocked with PBS 5% normal goat serum (NGS) or normal donkey serum (NDS) and 0.1% Triton X-100 for 1 h. For cell pair assays, sections were stained for anti-red fluorescence antibody (to detect the NG2+ cells and progeny with Lgl1 deletion) and anti-CC1 antibody overnight, washed and stained with fluorescence-couple secondary antibodies. Cell pairs were identified as two red fluorescence-positive (NG2Tom) and EdU-positive cells spaced less than 20 µm 2,3 . The entire CC was counted as three or more distinct brain sections per animal in 4-5 individual animals per group. Cell pairs with one CC1-positive cell were scored as an ACD.
The following antibodies were used at the given concentration: For O4 staining, cells were stained live in vitro: O4 was incubated 1 h (1:400) at 37°C in the corresponding culture medium (OPC proliferation or OL differentiation) before PFA fixation and further antibody staining.
EdU was visualized using the Click-it reaction kit (Life Technologies Corporation) following the manufacturer's instructions 71 .
Early endocytosis compartment was visualized using the BacMam technology following manufacturer's instructions (Life Technologies Corporation). Immunofluorescent images were acquired on a Nikon Ti-E microscope with Nikon perfect focus system equipped with a Hamamatsu Flash4.0 camera equipped with a 40×/0.75 Plan Fluor objective and/or Nikon Ti Yokogawa CSU-22 spinning disk confocal equipped with a Plan Apo VC 60×/1.4 oil objective. Images were analyzed with NIS-Elements 4.2 (Build 982) and ImageJ.
To perform absolute cell counts, single cell suspensions were transferred to tubes containing a calibrated number of fluorescent beads (TruCount tubes, BD Biosciences) 73 . Prior to FACS sorting propidium iodide (PI) or Hoechst 33258 was added to a final concentration of 2 µg/mL to label the dead cells. Cells were analyzed and sorted on a FACS ARIA II (BD Biosciences). The data were analyzed with FlowJo data analysis software (Tree Star, Ashland, OR, USA).
Immunoblotting. Glioblastoma xenografts proteins were extracted using cell lysis buffer (Cell Signaling) supplemented with proteinase and phosSTOP phosphatase inhibitor cocktail (Roche). Proteins were separated by SDS-PAGE and transferred onto polyvinylidene difluoride membranes, which were then probed with primary antibodies, followed by horseradish peroxidase-conjugated secondary antibody, and visualized by ECL (GE Healthcare). Antibodies specific for Hugl1 and tubulin as loading control were from Sigma.

RNA extraction and reverse transcription.
For all experiments, RNAs were extracted with QIAGEN RNeasy microkit and reverse transcribed into cDNAs using high capacity RNA to cDNA kit (Applied Biosystems).
Taqman low-density arrays. For the Taqman low-density array experiment, 200-500 cells/population were sorted from the CC of young adult Ascl1-GFP and NG2-DsRed double-transgenic mice. TaqMan ® Array Micro Fluidic Card were designed for neural stem cell, oligodendrocyte progenitor cell and differentiated oligodendrocyte markers, as well as well selected ACD regulators. A PreAmp pool was designed for these specific primers and a preamplification reaction was set up to 14 cycles according to manufacturer's instructions (Custom TaqMan ® PreAmp Pools Protocol, Applied Biosystems).
Whole-genome microarrays. For the microarray experiments 200-500 Lgl1 knockout and wildtype OPC cells (PDGFRα and red fluorescent protein positive O4 negative), were extracted with the RNA plus micro kit (QIAGEN) and reverse transcription into cDNAs was performed using the High Capacity Reverse Transcription Kit (Thermofisher scientific). Sample quality was controlled using the Agilent Bioanalyzer and samples were loaded on GeneChipR Mouse Gene 2.0 ST Arrays (Affymetrix). Microarray data were analyzed using Partek ® Genomic Suite 6.6 with RMA normalization and differential gene expression detected with ANOVA. Ingenuity pathway analyses (QIAGEN bioinformatics) were used to identify targets and pathway differentially expressed between the Lgl1 WT and Lgl1 cKO samples. Transferrin uptake. The CC of GFAP-Cre, dTom lox/lox , Lgl1 wt/wt (n = 4) vs. Lgl1 ox/lox (n = 4) was extracted from young adult mice (P60). After enzymatic digestion with papain (10 min, 15 U/mL) 72 , cells were washed with cold wash solution (PBS containing 20 mM glucose and 1% BSA). Then Transferrin Alexa Fluor 488 conjugate (Molecular probes, T13342) 74 was added at 25 µg/mL and incubated at 37°C for 15 min. Non-internalized ligand was removed from the cell surface using three 5 min wash with ice-cold freshly prepared stripping buffer (PBS/0.15% BSA, pH 3.5 adjusted with HCL) and washed three times with cold wash solution. Data analysis to quantify changes in Transferrin fluorescence in dTom+ cells was performed using a FACS Aria II (BD Biosciences) and analyzed with FlowJo data analysis software (Tree Star, Ashland, OR, USA).
Live staining of cells. Plated Lgl1 cKO and WT OPC were incubated on ice at 4°C for 2 h with α-NG2-EC (Rabbit, courtesy of Dr. Stallcup) in Neurobasal medium (3% BSA). After two washes with neurobasal medium, cells were incubated for 1 h with an Alexa488 goat α-Rabbit. For lysosome tracking, cultured OPC were incubated with CellLight Reagent BacMam 2.0 (Molecular probe) 24 h prior to live imaging targeting the early endosome Rab5a (C10586) and lysosome-specific protein Lamp1 (C10507) at a concentration of 50 PPC.
Fluorescence time lapse microscopy. Time lapse movies were taken under a conventional widefield epifluorescence system (Nikon) equipped with a 37°C incubation chamber and CO 2 control for live cell imaging. Live-labeled cells were incubated in differentiation conditions and the system programed to take fluorescent pictures at the following frequencies: 1 frame/30 s to follow early NG2 endocytosis and 1 frame/1 h to follow its trafficking towards lysosomes.
TIRF microscopy. Live labeled cells (NG2-EC) were plated in differentiation conditions and placed on a coupled incubator for live cell imaging (37°C, 5% CO 2 ) under the TIRF microscope. Fluorescent pictures were taken at different frequencies: 1 frame/30 min or 1 frame/30 s. TIRF was used to provide a mean to selectively excite fluorophore very near the solid surface (<100 nm) without exciting fluorescence from regions farther from the surface. Therefore, any Alexa488 positive dot in perfect focus was considered as a docking NG2 on the membrane region that is being imaged. Different quantities of NG2 recycling over time were represented as total corrected fluorescence measured using ImageJ.
Recycling inhibition. Lgl1 cKO OPC were treated with differentiation medium containing 10 μM Monensin (TOCRIS, 5223) for 4 h. Cells were monitored under TIRF for NG2 recycling monitoring during the first hour of treatment or were switched back to normal differentiation medium for 5 days before being fixed and immunostained for O4 and NG2. Endocytosis of the proteoglycan NG2 was followed using a mouse-anti-NG2.EC antibody against the extracellular domain of NG2 (courtesy of Dr. Stallcup). OPC cultures were incubated 24 h prior to live imaging with CellLight ® Reagents BacMam 2.0 (Molecular Probes) targeting the early endosome-specific protein Rab5a (ref C10586) of lysosome-specific protein Lamp1 (ref C10507) at a concentration of 50 PPC. Cells were starved 2 h prior to imaging in NBA medium, and 50 ng/mL of PDGF-AA was added right before imaging to induce endocytosis. Time-lapse movies were taken over time spans of 30-45 min with 30 s interval between the images to follow early endocytosis of NG2 antibody and 2-3 h with a 30 min interval to follow its internalization to the lysosome. After live imaging, cells were fixed 10 min in PFA4%, stained with the appropriate antibodies and imaged with a Nikon Ti Yokogawa CSU-22 spinning disk confocal equipped with a Plan Apo VC 60×/1.4 oil objective.
Endocytosis inhibitor. Monensin (TOCRIS, 5223), a selective recycling endocytosis 75 , was used and was not toxic at a working concentration of 10 µM.
Univariate Cox model. The prognostic significance of high Lgl1 expression (≥median) on overall survival (defined as the time between diagnosis and death due to any cause) was retrospectively tested. The follow-up period ranged from 0 to 8 years (median = 1 year). The number of events that occurred during this period and the number of censored individuals are: 86 events, with 13 censored.
Statistics and analysis. Statistical analyses were conducted as indicated in the figure legends using GraphPad Prism 5.0 and Microsoft Excel. A Mann and Whitney two-tailed paired t-test was conducted to analyze for significant differences between two treatment groups. For Kaplan-Meier survival rate analyses, the number of surviving mice were recorded daily. The data were subjected to Logrank test in order to determine if significant differences existed in survival between the Lgl1 knockout and control groups. For statistical tests *p < 0.05,**p < 0.01,***p < 0.001. A univariate Cox models was fitted on the survival data of the 99 glioma patients (92 GBM and 7 LGG) to obtain p-values. R-function 'survfit' (package 'survival') was used to generate survival curves from the previously fitted Cox model.
Data availability. The data that support the findings of this study are available from the corresponding author upon reasonable request.