Lunatic fringe-mediated Notch signaling regulates adult hippocampal neural stem cell maintenance

Hippocampal neural stem cells (NSCs) integrate inputs from multiple sources to balance quiescence and activation. Notch signaling plays a key role during this process. Here, we report that Lunatic fringe (Lfng), a key modifier of the Notch receptor, is selectively expressed in NSCs. Further, Lfng in NSCs and Notch ligands Delta1 and Jagged1, expressed by their progeny, together influence NSC recruitment, cell cycle duration, and terminal fate. We propose a new model in which Lfng-mediated Notch signaling enables direct communication between a NSC and its descendants, so that progeny can send feedback signals to the ‘mother’ cell to modify its cell cycle status. Lfng-mediated Notch signaling appears to be a key factor governing NSC quiescence, division, and fate. DOI: http://dx.doi.org/10.7554/eLife.24660.001


Introduction
The ability of the hippocampal neurogenic niche to respond to ever-changing stimuli throughout the lifespan requires that it carefully maintain its finite stock of neural stem cells (NSCs) (Kuhn et al., 2005(Kuhn et al., , 1996. NSCs are the primary stem cells of the niche and are quite plastic in their responses: for example, social isolation seems to increase NSC self-renewal (Dranovsky et al., 2011), whereas running stimulates neurogenesis (van Praag et al., 1999), and seizures prompt NSCs to transform directly into reactive astrocytes (Sierra et al., 2015). Taking into account that nearly 80% of the newborn progeny of NSCs die by apoptosis (Sierra et al., 2010), the neurogenic niche needs to ensure not only that it responds properly to stimuli but also that NSCs, once they are activated, are cycling enough to produce sufficient progeny.
We hypothesized that, if Notch does facilitate communication between the mother NSC and its daughter cells, it might do so through the fringe proteins (Lunatic, Manic, Radical), which are known regulators of Notch signaling. Glycosylation of Notch receptors by fringe proteins affects the intracellular cleavage of the heterodimeric receptor complex and generation of the Notch1 Intra Cellular Domain (NICD) following ligand binding. Typically, NICD production increases upon binding by Delta-like (Dll) and decreases following Jagged1 (Jag1) binding (LeBon et al., 2014;Stanley and Okajima, 2010;Taylor et al., 2014;Yang et al., 2005); differential Notch cleavage ensures varying expression of downstream cell cycle genes (Chapouton et al., 2010;Isomura and Kageyama, 2014;Nellemann et al., 2001;Ninov et al., 2012;Yoshiura et al., 2007). To examine whether fringe proteins are present in NSCs, we systematically queried existing expression databases, such as the Allen Brain Atlas (Lein et al., 2007) and GENSAT (Gong et al., 2003), and discovered that Lunatic fringe (Lfng) appears to be selectively expressed in adult hippocampal NSCs. We confirmed this observation using Lfng-eGFP mice and further demonstrated the potency of Lfng-eGFP-expressing cells using a newly-generated Lfng-CreER T2 line for lineage tracing. The selective expression of Lfng in NSCs has enabled us to explicitly examine the role of Notch signaling in NSC regulation. Here, using several new transgenic mouse models, we unveil a novel Notch-based mechanism that mediates direct communication between NSCs and their progeny to control NSC quiescence and activation.

Results
Lfng-eGFP reporter mice specifically mark NSCs of the dentate gyrus Our database search indicated that Lfng might selectively label hippocampal NSCs, prompting us to fully characterize the Lfng-eGFP transgenic mouse. Spatially, Lfng-eGFP expression in the hippocampus was restricted to the subgranular zone (SGZ) of the dentate gyrus, where neural stem and progenitor cells reside (Seri et al., 2001) ( Figure 1A, left panel). Lfng-eGFP + cells closely resembled NSCs: their triangular soma was located in the SGZ, from which a single radial process extended orthogonally and spanned the granule cell layer, ending in fine arborizations within the molecular layer ( Figure 1A, right panel). In situ hybridization for Lfng mRNA confirmed that eGFP expression recapitulated the endogenous Lfng expression pattern in the adult dentate gyrus ( Figure 1B). To confirm that eGFP expression accurately reflects Lfng expression, we crossed Lfng-eGFP with Lfng Tm1Grid mice that carry beta-galactosidase (b-Gal) insertion in the Lfng locus (Zhang and Gridley, 1998). In the resulting Lfng Tm1Grid ; Lfng-eGFP mice, b-Gal and eGFP co-localized ( Figure 1C), demonstrating that the regulatory elements driving eGFP expression in Lfng-eGFP mice are active in the same cells that expressed b-Gal in Lfng Tm1Grid mice. Finally, Nestin, Sox2, Vimentin, and GFAP,  known to label NSCs, were all expressed in Lfng-eGFP + cells ( Figure 1D). Together, these diverse lines of evidence indicate that Lfng-eGFP is a marker of NSCs.
A small proportion of the eGFP-expressing cells in the SGZ had a round amplifying neuroprogenitors (ANP) -like morphology with no associated processes. To determine whether ANPs express eGFP, we crossed Lfng-eGFP with the Nestin-CFP nuc mice, which express CFP in the nuclei of both NSCs and ANPs (Encinas et al., 2006). In the double transgenic mice, all Lfng-eGFP + cells coexpressed CFP (Figure 1-figure supplement 1A), but only 7.4 ± 1.9% (mean ± SEM, N = 3) of the CFP-labeled ANPs contained eGFP. The eGFP + ANPs were in close proximity to dividing BrdU + NSCs and in all cases were in cytoplasmic connection with the mother NSC ( Figure 1E, left panel), which suggests that the asymmetric division had not yet finished. ANPs not cytoplasmically connected with the NSCs had little if any eGFP ( Figure 1E, left panel).
We next examined how long it took newborn ANPs to clear eGFP, using BrdU pulse-and-chase labeling. eGFP was cleared from Lfng-eGFP + ANPs, but not from Nestin-GFP + (Mignone et al., 2004) ANPs, between days 1 and 3 ( Figure 1E, left graph). The clearance correlates with the halflife of GFP (Corish and Tyler-Smith, 1999), suggesting that the eGFP in ANPs is likely from the protein retained by the first progeny of Lfng-eGFP + NSCs. Indeed, only 10% of Type 2a cells (Sox2 + early ANPs) were eGFP + in Lfng-eGFP mice compared to almost 95% in Nestin-GFP mice ( Figure 1E, right graph). No Type 2b cells (late ANPs) expressed Lfng-eGFP ( Figure 1E, right panel). Together, these data strongly suggest that most of the Lfng-eGFP + ANPs are the immediate progeny of NSCs that retained some eGFP due to sequestration of the protein during cell division. ANPs quickly lose eGFP, implying that Lfng-eGFP expression is selective for NSCs and that Lfng is active in NSCs but not in ANPs.
Lfng-eGFP was not detected in neuroblasts/immature neurons, granule cells, or astrocytes ( Figure 1F). In a minority of cases (0.8%) we observed the endothelial cell marker CD31 overlapping with Lfng-eGFP ( Figure 1-figure supplement 1B). This is not surprising, as endothelial tip cells express Lfng during tip-stalk cell selection after exposure to angiogenic factors (Benedito et al., 2009). Further, we observed a few Lfng-eGFP + cells, some co-labeled with GFAP + or S100b + , randomly distributed in the dentate gyrus (0.9%; Figure 1-figure supplement 1C). The overall quantification of the cell-type expression of eGFP indicated that the majority of eGFP + cells in Lfng-eGFP mice were NSCs (81.7 ± 2.6%), while only 16.4 ± 1.5% were ANPs and 1.9 ± 1.1% were other cell types ( Figure 1G). In contrast, in Nestin-GFP mice, the majority of GFP + cells were ANPs (36.1 ± 3.7%), followed by other cell types (33.7 ± 2.8%) and NSCs (30.2 ± 2.4%). These data emphasize the specificity of Lfng-eGFP expression for adult hippocampal NSCs, compared to the low selectivity of Nestin-GFP expression for NSCs. At any given time, the majority of NSCs is quiescent; only about 1-4% of them divide under physiological conditions (Encinas et al., 2011;Knobloch et al., 2014). This property enables them to Dotted lines indicate the borders of SGZ. (C) In a double transgenic Lfng Tm1Grid ;Lfng-eGFP mouse, b-galactosidase (b-Gal) and eGFP are co-expressed, confirming that Lfng promoter guiding the eGFP expression is active in the same cells that express b-Gal. Lfng Tm1Grid =Lfng b-Gal mouse that carries b-Gal insertion in the Lfng locus. (D) Lfng-eGFP colocalizes with other NSC markers, such as Nestin, Sox2, Vimentin, and GFAP. Confocal photomicrographs of the representative examples and relative quantitation is shown. (E) eGFP is briefly retained in the first progeny of NSC. Left panel: Lfng-eGFP expressing cell (thick arrow) and its immediate ANP progeny (arrowhead) are both BrdU + and in cytoplasmic contact. BrdU + ANP not in the cytoplasmic contact with the BrdU + NSC has very low eGFP expression (thin arrow). Left graph: eGFP is cleared from BrdU + ANPs between 1 to 3 days following a single dose of BrdU, corresponding to the half-life of GFP protein (N = 3-5 per group, p<0.001 or p<0.0001 for pairwise comparisons of all timepoints, ANOVA with Tukey post-hoc test). Right graph: Lfng-eGFP is expressed only in minority of Type2a ANPs (Sox2 + GFAPcells in the SGZ; 10.58 ± 2.38, N = 3), unlike Nestin-GFP (94.7 ± 0.34, N = 4). Right panel: Late, Tbr2 + ANPs (arrowheads) do not express Lfng-eGFP. Arrow points to Lfng-eGFP NSC. (F) Lfng-eGFP does not co-localize with the markers of neuronal lineage (Dcx + neuroblasts and immature neurons, and NeuN + granule cells) nor S100b + astrocytes. (G) While in the Nestin-GFP (N) mice GFP is expressed in NSCs, ANPs, and other cell types throughout the dentate gyrus in approximately equal proportions (30.16 ± 2.43 NSCs, 36.12 ± 3.73 ANPs, 33.71 ± 2.81 other cell types), in the Lfng-eGFP (L) mice it is expressed predominantly in NSCs (81.68 ± 2.62% NSCs, 16.37 ± 1.54% ANPs, 1.95 ± 1.14% other cell types; N = 4 per genotype). Bars represent mean±SEM. **p<0.001, ***p<0.0001. Scale bars = 100 mm (A left, (B), 20 mm (A right, (C-F). See escape treatment with cytostatic drugs, such as temozolomide (Knobloch et al., 2014). Indeed, only 15.0 ± 3.0% of the eGFP + cells were lost in Lfng-eGFP mice following temozolomide treatment (N = 4, p=0.27), compared to 35.0 ± 4.0% in the Nestin-GFP mice (N = 4, p=0.0035). This suggests that the Lfng-eGFP + population consists mostly of quiescent NSCs, which we confirmed by independent quantification of NSCs and ANPs ( Figure 2A). Further, if Lfng-eGFP + cells are true NSCs, then both physical exercise and electroconvulsive shock (ECS) should activate them (Lugert et al., 2010;Madsen et al., 2000;Segi-Nishida et al., 2008;Suh et al., 2007;van Praag et al., 1999van Praag et al., , 2005Warner-Schmidt and Duman, 2006). We therefore tested whether Lfng-eGFP + NSCs respond to these stimuli. Mice were either administered a single ECS for four consecutive days or subjected to voluntary access to a running wheel for a week, followed by a single injection of BrdU. Sham controls either did not receive ECS or had a locked running wheel in their cage. Mice were sacrificed 24 hr after the BrdU injection and the BrdU + NSCs were quantified. A robust increase in the percentage of BrdU + Lfng-eGFP + NSCs in both conditions (N = 4 per group in ECS, p=0.0006; and N = 6 per group in PE, p=0.0271) indicated that Lfng-eGFP + NSCs are plastic, as expected ( Figure 2B).
We next assessed the net outcome of neurogenesis using BrdU pulse-and-chase labeling and compared Lfng-eGFP and Nestin-GFP mice to determine whether there were any gross differences in these two different transgenic lines. Over the course of 30 days, the total number of BrdU + cells in the dentate gyrus followed the same pattern of decline (R = 0.9992; Figure 2-figure supplement 1A). The decay was rapid between 1 and 3 days post-BrdU injection (dpi; 100-105 BrdU + cells lost per hour), slowed to 31-27 BrdU + cells/hr between 3 and 7dpi, and reached a stable rate of 3-4 BrdU + cells/hr between 7-15dpi, and 2-3 cells/hr between 15 and 30dpi. Neurogenesis in Lfng-eGFP mice thus follows the same survival pattern observed in Nestin-GFP mice (Kempermann et al., 2004;Kronenberg et al., 2003;Sierra et al., 2010). Further, the Lfng-eGFP + NSCs produced the same neurogenic progeny as Nestin-GFP + NSCs over 30 days, providing further assurance that there are no gross differences in these two transgenic lines ( Figure 2C). This is important, because it establishes the benchmark validity of the Lfng-eGFP mice for studies of neurogenesis.
Finally, adult neurogenesis declines with age (Kuhn et al., 1996), at least in part because of a decline in the NSC population (Encinas et al., 2011). Thus, if Lfng-eGFP + cells are NSCs, then their number should decrease over time. This is, in fact, what we observed; over an 18 month period the number of NSCs ( Figure 2D . This is a crucial finding because it signifies the utility of the Lfng-eGFP mouse model for specific studies of NSC biology during aging.

Lfng-expressing NSCs generate diverse progeny
The claim that Lfng-eGFP + cells are functional NSCs capable of producing diverse progeny requires lineage tracing. Thus, we generated a Lfng-CreER T2 transgenic line using the same bacterial artificial chromosome used to generate Lfng-eGFP mice, modified by inserting CreER T2 in front of the transcription start site of the Lfng (RP23-270N2; Figure 3A). To verify that CreER T2 is selectively expressed in NSCs, we bred Lfng-CreER T2 mice to the AI14 (RCL-tdT) reporter line (Jackson Lab Stock no: 007908) (Madisen et al., 2010) and then to Lfng-eGFP mice. One day following induction, 89.1 ± 4.0% of tdTomato + cells were eGFP + and 37.9 ± 1.5% of Lfng-eGFP + NSCs were tdTomato + ( Figure 3B To further verify that Lfng-expressing NSCs are dividing, we crossed the Lfng-CreER T2 mice with the iDTR mice, in which the diphtheria toxin receptor (DTR, a.k.a. Hbegf, simian Heparin-binding epidermal growth factor-like growth factor) is conditionally expressed under the control of Cre-activated Rosa26 locus . Activation of this receptor by diphtheria toxin selectively kills DTR-expressing cells . Fifteen days following induction of DTR in Lfng-CreER T2 ; iDTR mice and activation by diphtheria toxin, we observed a significant reduction in both NSCs (36.8 ± 1.5%; N = 3-4 per group; p=0.0244) and the Ki67 + cells (57.3 ± 2.4%; p<0.0001) (Figure 3-figure supplement 1B). As neither DTR expression nor the high dose of diphtheria toxin alone cause cell death (Arruda-Carvalho et al., 2011;Buch et al., 2005;Gropp et al., 2005), our data confirm the , GFP + NSCs (middle panel) and GFP + ANPs (right panel) in 3 month-old Nestin-GFP and Lfng-eGFP mice (N = 4 per genotype) treated with temozolomide (TMZ). The difference between the two mouse models is most notable with respect to ANPs: while Nestin-GFP labels a large number of ANPs, Lfng-eGFP does not -it labels primarily quiescent cells. (B) Electroconvulsive shock (ECS) and physical exercise (PE) both activate Lfng-eGFP + NSCs (N = 4 per group in ECS and N = 6 per group in PE). (C) Lfng-eGFP NSCs produce neuronal progeny. The relative number of newborn, BrdU + progeny was quantified over a 30 day period (NSCs: Nestin-GFP + or Lfng-eGFP + cells with GFAP + radial processes; ANPs: GFAP -Dcx -NeuN -; NBs: Dcx + neuroblasts and immature neurons; GCs: NeuN + ; Other: BrdU + Dcx -NeuNcells outside the SGZ; N = 4 per genotype per timepoint). Cumulative BrdU paradigm (four 150 mg/kg injections given 2 hr apart) was used to increase the yield of labeled newborn cells. N=Nestin-GFP mice, L=Lfng-eGFP mice. (D) The number of Lfng-eGFP + NSCs declines over an 18 month period comparably to the number of Nestin-GFP + NSCs (left panel; N = 4 per timepoint per genotype). However, the contribution of GFP + cell types in the Nestin-GFP and Lfng-eGFP mice differs at different age (right panel). While Lfng-eGFP remains selective for NSCs during aging (p>0.15 for all timepoints, Tukey post-hoc test), Nestin-GFP labels Figure 2 continued on next page proliferative properties of Lfng-expressing NSCs and indicate that even partial elimination of NSCs leads to a dramatic decrease of cycling cells in the SGZ neurogenic niche.
Next, we examined the lineage of tdTomato + NSCs in Lfng-CreER T2 ;RCL-tdT mice ( Figure 3C-E and Figure 3-figure supplement 1C,D). We first focused on NSCs and their immediate progeny. Within 1-2 days after a single dose of tamoxifen (200 mg/kg), induced NSCs with radial processes and fine arborizations in the molecular layer appeared as tdTomato + cells. They expressed Sox2 and GFAP, further supporting recombination in NSCs ( Figure 3C, left panel). They divided both asymmetrically, to give rise to ANPs ( Figure 3C, middle panel) and symmetrically, to give rise to two Sox2 + cells connected via cytoplasm and both having prominent GFAP + radial processes ( Figure 3C, right panel). We then performed fate-mapping for two months, using a lower dose (120 mg/kg) of tamoxifen to sparsely induce cells and allow visualization of progeny morphology. Within 3 days of induction we detected tdTomato + Tbr2 + late ANPs (Type 2b cells), followed by Dcx + immature neurons 7 days after ( Figure 3D). Thirty days following induction, tdTomato + cells expressed the granule cell marker NeuN ( Figure 3D) and had extended processes reaching the molecular layer, prominent dendritic spines, and an axon extending throughout the hilus to the CA3 region ( Figure 3-figure supplement 1C). These data demonstrate that Lfng-expressing cells generate neurons and enable detailed studies of the neuronal lineage linked directly to a single NSC.
In addition to neuronal progeny, we also detected tdTomato + S100b + astrocytes in the granule cell layer ( Figure 3E, left panel) and tdTomato + Sox2 + GFAP + astrocyte-like cells in the hilus ( Figure 3E, right panel). These may represent either terminally differentiated NSCs or the progeny of Lfng-expressing NSCs; characterization of these cells is beyond the scope of this paper and will await future studies. Finally, as in the Lfng-eGFP mice, we found tdTomato + tip cells in Lfng-CreER T2 ; RCL-tdT mice (Figure 3-figure supplement 1D), further supporting the finding that the same regulatory elements control the expression of both eGFP and CreER T2 in these two different mouse models.
Quantitative evaluation of the lineage of Lfng-expressing NSCs provided additional information on the progeny and differentiation timeline ( Figure 3F). Initially, most of the population consisted of NSCs with some ANPs and astrocyte-like cells (GFAP + , S100b -). Over the course of 60 days, the contribution of NSCs to the population of tdTomato + cells diminished because of the growing contribution of ANPs, immature neurons and granule cells. Interestingly, the population of ANPs and immature neurons remained stable from 15 days on, indicating a continuous replenishment of these cell types. The granule cell number steadily increased from 15 days on and represented about 30% of all tdTomato + cells at 60 days following induction. Surprisingly, the number of astrocyte-like cells did not change over the course of 60 days, although there was one change: these cells were mostly GFAP + , S100bbetween 1 and 15 days and mostly GFAP + , S100b + thereafter. It may be that there is a steady conversion of NSCs into astrocyte-like cells that is independent of neurogenesis. Finally, we occasionally detected other tdTomato + cell types (mostly tip cells and their descendant endothelial cells), which constituted 6.8 ± 1.8% of the population at 60 days. In summary, these data demonstrate that Lfng-expressing cells in the SGZ are NSCs, able to self-renew and produce both neuronal (majority) and astrocytic (minority) lineage in the adult dentate gyrus. Collectively, these data establish Lfng-eGFP and Lfng-CreER T2 as novel mouse models enhancing the existing repertoire of tools . With these new mouse models, the field will be able to study the intrinsic properties of NSCs distinct from their progeny and gain a deeper understanding of the mechanisms of quiescence and lineage potential in the healthy brain, during aging, and in disease. The selective expression of Lfng in adult SGZ NSCs raises the question about its functional role in these cells. Surprisingly, the biological role of Lfng in adult SGZ NSCs has never been examined, despite its known function as a direct transcriptional target of Notch (Morales et al., 2002) and the importance of Notch signaling for adult NSC maintenance (Ables et al., 2010;Breunig et al., 2007;Ehm et al., 2010;Giachino and Taylor, 2014). Lfng N-glycosylates Notch receptors (Moloney et al., 2000), affecting the intracellular cleavage of the heterodimeric receptor complex and generation of the NICD following ligand binding: typically, Lfng modification of Notch elevates NICD production upon Dll1 binding, but decreases it following engagement to Jag1 (Haines and Irvine, 2003;LeBon et al., 2014;Stanley and Okajima, 2010;Taylor et al., 2014;Yang et al., 2005). This might lead to differential activation of cell cycle genes downstream of Notch, depending on which type of ligand dominates (Chapouton et al., 2010;Isomura and Kageyama, 2014;Nellemann et al., 2001;Ninov et al., 2012;Shimojo et al., 2008;Yoshiura et al., 2007).
In agreement with published data (Breunig et al., 2007;Lavado et al., 2010), we found the key elements of the Notch pathway expressed in the NSCs and surrounding progeny: Notch1 on the surface of NSCs, Jag1 on presumed ANPs surrounding NSCs, and Dll1 on granule neurons neighboring the apical parts of the NSC soma ( Figure 4A). In addition, Hes5 (an indicator of canonical Notch pathway; Imayoshi et al., 2010;Lugert et al., 2010) was present in some but not all Lfng-eGFP + NSCs ( Figure 4A).
We further examined Notch signaling in NSCs using CBF:H2B-Venus (JAX 020942) (Duncan et al., 2005), a Notch reporter mouse in which YFP variant Venus is expressed under the control of CBF1 promoter. Approximately half of the NSCs had active Venus + Notch signaling at any given time, whereas Venus expression declined in early (Sox2 + ) and late (Tbr2 + ) ANPs as well as in neuroblasts and immature neurons (Dcx + ) ( Figure 4B). Notch signaling was turned on again in mature granule neurons as most of the NeuN + cells were positive for Venus signal with various degrees of intensity ( Figure 4B), in agreement with the previous reports (Brandt et al., 2010;Breunig et al., 2007).
Although CBF:H2B-Venus mice are a good tool to evaluate the presence or absence of Notch signaling, the accumulation of Venus fluorescence protein and high fluorescence intensity in the GZ ( Figure 4B) make it hard to perform reliable intensity quantifications in the SGZ where NSCs reside. We therefore turned to NICD1 as a more reliable measure of Notch signaling intensity. To verify the reliability of NICD1 staining, we reasoned that NICD1 should be present in cells expressing Venus. Indeed, the NICD1 expression overlapped with Venus expression in CBF:H2B-Venus mice: over 90% in SGZ and over 95% in GZ ( Figure 4C).
These expression data provide the foundation for our hypothesis that Notch signaling, mediated by Lfng, represents a mode of communication between the progeny and the NSC to preserve the ancestor cell. Namely, in the resting state, the Dll1-expressing granule cells surrounding the NSC would promote NSC quiescence, so only a few stimulated NSCs would undergo division rather than the whole population. Once the NSC starts to divide in response to a given stimulus, the local accumulation of the Jag1-expressing ANP progeny would gradually shift the balance of Notch signaling Figure 3 continued expression of tdTomato + and eGFP + in induced Lfng-CreER T2 ; RCL-tdT mice (N = 3). Bars represent mean±SEM. (C) TMX-induced cells have NSC morphology and divide both asymmetrically and symmetrically. Left panel: tdTomato + NSC co-expresses GFAP and Sox2. Middle panel: tdTomato + NSC (arrow) divides asymmetrically to produce ANP (arrowhead). Right panel: NSC divides symmetrically to produce two cells (arrows) with prominent GFAP + radial processes. Scale bars = 20 mm. (D) tdTomato + NSCs produce new neurons through established cascade of cell types, from Tbr2 + late ANPs (Type 2b cells), through Dcx + immature neurons, to NeuN + granule cells. Scale bars = 20 mm. (E) tdTomato + NSCs also produce S100b astrocytes (or astrocyte-like cells) within the granule cell layer (left panel), as well as stellar Sox2 + GFAP + cells in the hilus (right panel) Scale bars = 20 mm. (F) Fate mapping of tdTomato + cells following TMX induction in Lfng-CreER T2 ; RCL-tdT mice reveals that NSCs form the majority (85.7% ± 0.9) of the induced cells at 1dpi, but progressively decline in ratio as they give rise to different progeny over the course of 2 months (N = 3-5 per timepoint). See  ANPs (type 2b cells) and granule cells are located. Jag1 is expressed in adjacent ANP, and Dll1 on adjacent granule cell-NSC boundary. Hes5, downstream target of canonical Notch signaling pathway, is present in some (arrowhead) but not all (empty arrowhead) NSCs. Scale bar = 10 mm (Notch1, Jag1, Dll1); 20 mm (Hes5). (B) Venus is expressed in some (arrowhead) but not all (empty arrowhead) NSCs (Sox2+ cell body and GFAP+ Figure 4 continued on next page within the 'mother' NSC, eventually halting Notch signaling and leading the NSC to exit the cell cycle. This would be beneficial on two fronts: it would prevent the niche from becoming overwhelmed by the perpetual production of ANPs, and it would preserve the mother NSC from exhaustion as there is only a finite number of divisions it can endure.

Lfng preserves NSCs by controlling their cell cycle
We first postulated that Lfng functions to preserve NSCs in both resting and active states, saving them from excessive mitosis. If this is the case, then lack of Lfng would lead to increased NSC division and their premature depletion from the niche. To examine the effect of Lfng ablation on NSC maintenance, we obtained Lfng tm1Grid mice (Lfng À/+ ; JAX 010619), in which most of the exon1 region of the Lfng is replaced by a targeting vector containing b-gal (Zhang and Gridley, 1998). Mice heterozygous for Lfng mutation are grossly normal. Homozygote mice typically have shortened trunks and malformed rib cages that impair respiration and cause premature death; the phenotype is variable, however, and some less severely affected homozygous mice do survive to adulthood (Zhang and Gridley, 1998). Because we could not consistently mate Lfng À/+ with Lfng-eGFP mice, we first developed a reliable method to quantify NSCs in the absence of eGFP reporter. We used a combination of markers known to be expressed in NSCs: Sox2 for labeling the NSC cell body (Suh et al., 2007) and GFAP for labeling the NSC radial process (Encinas and Enikolopov, 2008). We compared the NSC number in wild-type mice (using GFAP/Sox2 immunostainings) and Lfng-eGFP mice (using eGFP expression, GFAP + radial process and cell morphology) and found no significant difference between the two methods (N = 3; p=0.99; Figure 5A). Thus, for all subsequent studies where NSCs could not be identified because of the lack of fluorescent reporter, we used the GFAP/Sox2-based identification.
Mice lacking Lfng had markedly fewer NSCs than wild-type mice ( Figure 5B). This reduction in NSC number could be due to changes in the NSC cell cycle (Ables et al., 2010;Breunig et al., 2007;Ehm et al., 2010;Giachino and Taylor, 2014), as it has been suggested that NSCs undergo a finite number of divisions followed by their transformation into astrocyte-like cells (Encinas et al., 2011). We examined the ratio of cycling Ki67 + NSCs and BrdU + NSCs in mutant and wild-type mice and found that NSCs lacking Lfng were both cycling ( Figure 5C, left panel) and in S-phase ( Figure 5C, middle panel) in significantly higher proportions than wild-type NSCs (N = 4; p=0.0221 for Ki67 + NSCs; p=0.0017 for BrdU + NSCs).
Because these data were obtained in constitutive knockout mice, however, it is possible that they do not reflect the direct effect of Lfng. Therefore, to specifically delete Lfng from Lfng-expressing NSCs, we crossed Lfng flox/flox mice (Xu et al., 2010) with the Lfng-CreER T2 ; RCL-tdT mice. In the resulting conditional knockout mice (iLfng fl/fl ), the induced NSCs express non-functional mutant Lfng protein because exon 2 of the Lfng has been deleted (Xu et al., 2010(Xu et al., , 2012. These mutant clones can be visualized as they express tdTomato reporter. We then compared the relative number of BrdU + NSCs between Lfng-CreER T2 ;RCL-tdT control mice (tdTomato + GFAP + Sox2 + ), iLfng fl/fl wildtype clones (tdTomato -GFAP + Sox2 + ), and iLfng fl/fl mutant clones (tdTomato + ). Considerably more mutant NSCs were in S-phase than were either wild-type NSC clones or NSCs in control mice ( Figure 5C, right panel). Interestingly, iLfng fl/fl wild-type clones had a similar ratio of BrdU + NSCs as control mice ( Figure 5C, right panel), suggesting that the effect of Lfng in NSCs is cell-autonomous.
These findings from both constitutive and conditional knockout models suggest that in the absence of Lfng, many more NSCs divide. This effect persists not only in physiological but also in pathological conditions: when Lfng À/+ mice were subjected to ECS, they had a significantly higher Figure 4 continued process) in CBF:H2B-Venus mice. Type 2b cells (Tbr2) and most of the neuroblasts and immature neurons are Venus -, whereas almost all granule cells (NeuN + ) are Venus+ with various degrees of intensity. Some of the mature astrocytes (S100b + ) have active Notch signaling (arrowheads). Quantification of Venus signal among various cell types reveals that most of the differentiating cells are devoid of active Notch signaling. As soon as the neurogenic differentiation finishes, Notch signaling is turned on again in the granule cells. Scale bar = 10 mm upper panels and 20 mm middle and lower panels. (C) NICD1 and Venus colocalize in CBF:H2B-Venus mice. NICD1 almost completely overlaps with the Venus signal in the SGZ (left inset), while it is lacking in the molecular layer (right inset). Scale bar = 20 mm. Quantification of Venus + cells among NICD1 + cells in GZ and SGZ verifies high degree of colocalization (N = 3; 90.79 ± 0.77% and 96.95 ± 0.57%, respectively). DOI: 10.7554/eLife.24660.008 The total number of GFAP + Sox2 + NSCs in 4-month-old wild-type mice does not differ from the total number of eGFP + NSCs in 4 month-old Lfng-eGFP (N = 3 per group; Student's t-test, p=0.99). Scale bar = 10 mm. (B) The total number of NSCs in 2month-old wild-type, Lfng heterozygote (Lfng À/+ ), and homozygote (Lfng À/À ) knockout mice shows Lfng dose-dependent decrease in the NSC Figure 5 continued on next page ratio of BrdU + NSCs than observed in either sham Lfng À/+ control (N = 3-4, p=0.0178; Figure 5D) or ECS-treated Lfng-eGFP mice ( Figure 2B). Loss of Lfng therefore causes NSCs to be more susceptible to both stimulus-independent and stimulus-dependent division. Indeed, Notch signaling, as measured by NICD1 immunofluorescence intensity, was much lower in individual mutant NSC clones in iLfng fl/fl mice two weeks after induction compared to controls ( Figure 5E). This suggests that decreased Notch signaling might contribute to loss of NSC quiescence in Lfng mutant NSCs.
Our findings also revealed an interesting phenomenon: the difference in the fold change of Ki67 + / BrdU + NSCs in mutant mice (1.21 fold) was lower than the Ki67 + / BrdU + NSCs in wild-type mice (1.81 fold). This means that most cycling NSCs are in the S-phase in the mutant mice. To delve deeper into the biology of Lfng effect on NSC cell cycle, we used sequential labeling with BrdU analogs, 5-chloro-2-deoxy-uridine (CldU) and 5-iodo-2-deoxy-uridine (IdU) (Breunig et al., 2007;Encinas et al., 2011;Lugert et al., 2010). To define the initial proliferative NSCs cohort, 3-monthold wild-type, Lfng À/+ , iLfng fl/fl , and Lfng-CreER T2 ; RCL-tdT control mice were injected with CldU. To mark NSCs that pass through the subsequent S-phase(s), we injected IdU 3 or 7 days after CldU. By determining the fraction of CldU/IdU double-labeled NSCs, it is possible to quantify the number of NSCs that re-entered S-phase 3 or 7 days apart (Encinas et al., 2011). Many double-labeled NSCs were observed in the wild-type mice, only a few were detected in the Lfng À/+ mice at either timepoint, but none were found in iLfng fl/fl mice ( Figure 5F). Lfng thus not only affects the NSC cell cycle but is also critically important for NSC cell cycle re-entry. Lack of double labeling could be due to intermittent division or to a genuine exit from the cell cycle. As Lfng À/+ mice had a decreased ratio of CldU/Ki67 double labeled NSCs 7 days following CldU injection (N = 4; p=0.0105; Figure 5G), these data strongly indicate that Lfng mutant NSCs either dilute BrdU by multiple divisions, die, or directly differentiate/ transform into another cell type over the course of seven days.
Thus, we examined the number of NSCs that retained BrdU ( Figure 5H, left panel). In Lfng mutant mice, after a two-hour pulse, only 5.2% of the initial BrdU + NSC population retained BrdU after 7 days (N = 4 per timepoint, p=0.0003). In control mice, however, 92.4% of the initial BrdU + NSC population retained BrdU after 7 days (N = 4 per timepoint, p=0.38). Further, we observed One-way ANOVA p<0.00001, Tukey HSD post-hoc test: p<0.0001 for wild-type vs Lfng À/+ or Lfng À/À , p=0.0484 for Lfng À/+ vs Lfng À/À ). (C) Lack of Lfng promotes increased division of NSCs. Left panel: NSCs lacking Lfng have a higher ratio of cycling Ki67 + NSCs compared to wild-type (N = 4 per group; Student's t-test p=0.0221). Middle panel: NSCs lacking Lfng have a higher ratio of actively dividing, BrdU + NSCs compared to wild-type (N = 4 per group; Student's t-test p=0.0017). Right panel: Lfng acts cell-autonomously and in a dose-dependent manner to control the NSC division (N = 4 per group). The ratio of BrdU + NSCs was compared between iLfng fl/fl mutant clones (tdTomato + ), iLfng fl/fl wild type clones (tdTomato -GFAP + Sox2 + ), and Lfng-CreER T2 ; RCL-tdT control mice (tdTomato + GFAP + Sox2 + ). N = 4; p=0.9978 for control vs iLfng fl/fl tdTomclones; p=0.0049 for control vs iLfng fl/fl tdTom + clones; p=0.0045 for iLfng fl/fl tdTomclones vs iLfng fl/fl tdTom + clones. (D) NSCs lacking Lfng are hyperactivated in response to ECS treatment (N = 3-4 per group; Student's t-test, p=0.0178). (E) Lack of Lfng decreases Notch signal intensity in mutant NSCs. Relative intensities of NICD1 staining are significantly lower in iLfng fl/fl mutant NSCs compared to control (N = 4 for control; N = 3 for iLfng fl/fl ; Student's t-test, p=0.0004). (F) NSCs lacking Lfng spend less time in the active state than wild-type NSCs. Lfng absence is associated with decreased S-phase re-entry 3 (left panel) and 7 (middle panel) days following the initial division compared to the wild-type NSCs (N = 4 per group; Student's ttest, p=0.0024 for 3d, p=0.0003 for 7d). CldU + IdU + cells represent NSCs that underwent first division at the time of CldU injection (day 0) and were in S-phase at the time of IdU injection (day 3 or day 7). Right panel: In iLfng fl/fl mice, no NSCs were found that re-entered S-phase 7 days after the initial division. CldU + IdU + cells represent NSCs that were induced at day 0, underwent first division 1 day post-induction (CldU + ) and were in S-phase 7 days post-induction (IdU + ; N = 3-4 per group; Student's t-test, p=0.0014). (G) NSCs lacking Lfng mostly exit cell cycle within a week of first division. CldU + Ki67 + NSCs represent NSCs that are actively cycling 7 days following the CldU injection (N = 4; Student's t-test, p=0.0105). (H) NSCs lacking Lfng give rise to astrocyte-like cells. Left panel: The number of BrdU-retaining NSCs 7 days after the BrdU injection is significantly reduced in Lfng À/+ mice compared to wild-type (N = 4 per timepoint; Student's t-test, p=0.0003). Middle panel: In iLfng fl/fl mice, significantly more tdTomato + astrocyte-like cells accumulate 7 and 30 days following induction compared to controls, while the number of tdTomato + ANPs significantly decreases (N = 4 per group, p=0.0003 and p<0.0001 for astrocyte-like cells, p=0.0024 and p=0.0049 for ANPs). Right panel: The number of BrdU-retaining astrocyte-like cells 7 days after the BrdU injection is significantly higher in iLfng fl/fl mice compared to controls, but the difference is lost at 30 days (N = 4 per group; Student's t-test, p=0.0302 for 7 days, p=0.4412 for 30 days). Bars represent mean ± SEM. *p<0.05, **p<0.001, ***p<0.0001. See fewer NSCs with multiple S-phase re-entry in Lfng mutant mice at both 3 and 7 days ( Figure 5F), arguing against BrdU dilution. To then rule out the possibility that NSC depletion is caused by a rise in apoptosis, we turned to the conditional iLfng fl/fl mice and quantified the number of apoptotic cells. Both immunostaining for activated caspase-3 and the ApopTag assay, which detects DNA strand breaks, indicated that iLfng fl/fl actually had fewer apoptotic cells than control ( Figure 5-figure supplement 1A,B). Most apoptotic cells in SGZ are ANPs and differentiating neuroblasts, with very few astrocytes or NSCs (Breunig et al., 2007;Sierra et al., 2010). Notch signaling in iLfng fl/fl mice is downregulated ( Figure 5E) and thus more apoptotic cells would be expected (Nakamura et al., 2000). The surprising decrease in the number of apoptotic cells could thus be due to the production of fewer ANPs in iLfng fl/fl mice.
To then determine the fate of the NSCs, we quantified tdTomato + NSCs (Sox2 + , single radial GFAP + process), ANPs (Sox2 + , GFAP -), and astrocyte-like cells (Sox2 + , multiple GFAP + processes emerging from the cell body). We observed a declining trend in the number of mutant NSCs (p=0.8215; 0.11; 0.06 for 1, 7, and 30 days, respectively), and a significant decline in the number of mutant ANPs (p=0.2817; 0.0024; 0.0049 for 1, 7, and 30 days, respectively) ( Figure 5H, middle panel). This is not surprising, as Lfng removal caused fewer NSCs to undergo multiple rounds of cell cycle, which could impede ANP production. Interestingly, we observed a notable increase in tdTomato + astrocyte-like cells at 7 and 30 days post-induction (p=0.0003 and p<0.0001, respectively). This warranted further examination to ensure that these cells are the progeny of induced mutant NSCs: we quantified the BrdU-retaining astrocyte-like cells and found a greater number of these cells in Lfng mutant mice, but only at 7 days (N = 4 per group, p=0.0302; Figure 5H, right panel). At 30 days, there was no difference compared to control mice (p=0.4412), suggesting that NSCs lacking Lfng differentiate more rapidly into astrocyte-like cells, depleting the NSC population more quickly.
In sum, in the absence of Lfng, many more NSCs divide at the population level compared to controls. Lfng-deficient NSCs also spend less time in the active state, produce fewer ANPs over time, and tend to differentiate into astrocyte-like cells more rapidly than controls, which depletes the NSC population prematurely. These effects appear to depend on the level of Lfng, as homozygous conditional knockout mice had more pronounced alterations.

Notch ligands Jag1 and Dll1 preserve NSCs by opposing effects on their cell cycle
We then set out to examine the effects of Dll1 or Jag1 loss of function on NSC recruitment, proliferation (cycling and S-phase), cell cycle duration, and generation of their immediate progeny. Our prediction was that lack of Dll1 and Jag1 would have opposing effects on these measures. Mice lacking Dll1 (Dll1 tm1Gos (JAX 002957; Dll1 À/+ )) (Hrab e e de Angelis et al., 1997) had far fewer NSCs than wild-type mice or those lacking Jag1 (Jag1 tm1Grid (JAX 010616; Jag1 À/+ )) (Xue et al., 1999) or Lfng ( Figure 6A, Figure 5B, respectively). This was accompanied by a substantial increase in the ratio of cycling, Ki67 + cells and BrdU + NSCs (N = 4 per group; p=0.0007 for Ki67, p<0.0001 for BrdU experiment; Figure 6B), suggesting that the dramatic decrease in the NSC population in mice lacking Dll1 might be due to increased cycling and S-phase entry, as observed in Lfng À/+ mice. Interestingly, no difference in the amount of NSCs was observed between wild-type and Jag1 À/+ mice (N = 4 per group; p=0.1663; Figure 6A), but there was an upward trend in the proportion of BrdU + NSCs (N = 4 per group; p=0.8393; Figure 6B, middle panel). There was, however, a significant increase in Ki67 + NSCs lacking Jag1 (N = 4 per group, p<0.0001; Figure 6B, left panel), suggesting that these NSCs may have a prolonged cell cycle. Indeed, the proportion of Ki67 + /BrdU + NSCs in Jag1 À/+ mice was higher compared to Dll1 À/+ mice (3.97 vs. 1.40 fold).
To further examine this observation and delete Jag1 specifically from the Lfng-expressing NSCs and their progeny, we then crossed Jag1 tm2Grid (Jag1 fl/fl ; JAX 010618) mice (Kiernan et al., 2006), in which exon 4 of the Jag1 is deleted (Kiernan et al., 2006), with the Lfng-CreER T2 ;RCL-tdT line. The induced mutant clones lack Jag1 and can be visualized by tdTomato. In these conditional homozygous knockout mice (iJag1 fl/fl ), the ratio of BrdU + mutant NSCs was significantly higher than in controls (Lfng-CreER T2 ;RCL-tdT; N = 4 per group, p=0.0028; Figure 6B, right panel). We hypothesized that the difference between constitutive heterozygous knockout (upward trend in BrdU + ratio, Figure 6B, middle panel), and conditional homozygous knockout mice (significant increase in BrdU + ratio, Figure 6B, right panel) could be due to Jag1's ability to control cell cycle duration but not the Figure 6. Notch ligands Jag1 and Dll1 regulate NSC cell cycle. (A) The number of NSCs is diminished in mice lacking Dll1 compared to both wild-type and mice lacking Jag1 (N = 4 per group; One-way ANOVA p<0.00001, Tukey HSD post-hoc test p<0.0001 for Dll1 vs wild-type or Jag1 À/+ , p=0.1663 for wild-type vs Jag1 À/+ ). (B) Lack of either Dll1 or Jag1 promotes increased division of NSCs. Left panel: Absence of Dll1 or Jag1 is associated with significantly higher proportion of cycling, Ki67 + NSCs (N = 4 per group, One-way ANOVA p<0.00001, Tukey HSD post-hoc test p<0.0001 for wild-type Figure 6 continued on next page recruitment of NSCs from the quiescent population. We tested this possibility by examining the number of NSCs that were still cycling and/or in S-phase 7 days following their initial division in both Jag1 À/+ and iJag1 fl/fl mice. The number of CldU/IdU double-labeled NSCs in both Jag1 À/+ and iJag1 fl/fl mice was notably higher than controls at both 3 and 7 days (N = 4 per group, p=0.0164 at 3d, p=0.0016 at 7d; p=0.0003 for iJag1 fl/fl mice; Figure 6C) and accompanied by a greater proportion of CldU + Ki67 + NSCs (N = 4 per group; p=0.005 for wild-type vs Jag1 À/+ ; Figure 6D). Assuming a typical NSC cell cycle duration of 24-36 hr (Brandt et al., 2012(Brandt et al., , 2010Encinas et al., 2011), these data indicate that absence of Jag1 causes NSCs to re-enter cell cycle. The increase in NICD1 expression in individual mutant NSC clones in iJag1 fl/fl mice two weeks after induction ( Figure 6E) supports this notion, in agreement with the reported conditional overexpression of NICD1, which increases cell cycle reentry of NSCs (Breunig et al., 2007).
Conversely, there was no difference in double-labeled NSCs between Dll1 À/+ and wild-type mice at 3 days, but at 7 days no CldU/IdU double-positive cells could be detected, suggesting that NSCs did not re-enter cell cycle (N = 4 per group, p=0.9624 at 3d, p=0.0016 at 7d; Figure 6C). These observations were then verified by the CldU + Ki67 + NSC analysis in Dll1 À/+ mice (N = 4 per group, p=0.0049; Figure 6D). Thus, whereas the absence of Jag1 causes NSCs to re-enter the cell cycle while maintaining their overall number, absence of Dll1 reduces NSC re-entry and substantially decreases their population.
The findings were opposite in mice lacking Jag1: the number of BrdU retaining NSCs was much greater 7 days following the BrdU pulse (1.6 fold, p=0.0028) than two hours after the pulse ( Figure 6F). Thus, NSCs lacking Jag1 may have undergone self-renewal. We then quantified the total number of tdTomato + NSCs and ANPs in iJag1 fl/fl at 1, 7 and 30 days following induction. Initially, both control and mutant mice had a similar number of tdTomato + NSCs and ANPs, but this number markedly increased at both 7 and 30 days in mutant compared to control mice (N = 4 per group, p=0.0263 and p<0.0001 for NSC at 7d and 30d, p=0.0033 and p=0.069 for ANP at 7d and 30d; Figure 6I). Interestingly, there was no difference in the number of tdTomato + NSCs between 7 and 30 days (N = 3-4, p=0.3903), whereas the number of tdTomato + ANPs dropped significantly. Removal of Jag1 therefore might lead to an initial increase in the accumulation of NSCs, previously interpreted as an indicator of self-renewal (Dranovsky et al., 2011). In addition, increased S-phase re-entry, observed in NSCs from both Jag1 À/+ and iJag1 fl/fl ( Figure 6C), might also lead to increased production of ANPs. Indeed, both the number of Type 2a ANPs and the size of the ANP clusters were significantly greater in Jag1 À/+ mice than controls ( Figure 6G). To determine whether apoptosis had any role in the increased number of ANPs in Jag1 mutants, we quantified the number of apoptotic cells using activated caspase-3 staining and ApopTag. Both methods demonstrated a prominent decrease in the number of apoptotic cells in iJag1 fl/fl mice compared to wild-type (Figure 5-figure supplement 1).
Altogether, these data indicate that Notch ligands, Dll1 and Jag1, are important regulators of NSC recruitment, cell cycle duration and exit, and generation of NSC progeny. Lack of Dll1 increases NSC recruitment, leading to overall larger proportion of dividing NSCs compared to controls, and causes dividing NSCs to be active for a shorter period of time -there is a tendency to produce fewer ANPs and the total number of NSCs falls dramatically following increased terminal astrocytic differentiation. This phenotype thus resembles premature exhaustion of NSCs. On the other hand, lack of Jag1 does not affect NSC recruitment, as the proportion of dividing NSCs did not differ from controls. Those that were dividing were active for a longer period and had increased cell cycle reentry, ultimately leading to greater ANP production and increased NSC self-renewal.

Discussion
In this paper, we demonstrate that Lfng, a known regulator of Notch signaling, is selectively expressed in adult NSCs, where it participates in their preservation by regulating their cycling. In addition, we provide evidence that distinct modes of Notch signaling from NSC derivatives expressing Jag1 and Dll1 ligands, most likely mediated by Lfng, directly affect the NSC cell cycle. The differential effects of NSC progeny on the ancestor cell could be the key for preserving NSCs and thus neurogenesis throughout life.
These data lead us to propose a model in which Notch signaling provides direct communication between the NSC and its descendants, such that the progeny send feedback signals to the mother cell to modify its cell cycle status (Figure 7). Lfng is central to this model, as it enables NSCs to distinguish between Dll1-expressing granule cells and Jag1-expressing ANPs. In the Notch signalreceiving NSC, Lfng glycosylation of the Notch receptor potentiates Notch signaling from Dll1 and attenuates signaling from Jag1: even though Lfng glycosylation of Notch does not affect Jag1 binding, the NICD is not cleaved from the Jag1-Notch complex. Consequently, Notch signaling decreases. In our model, the resting NSC is surrounded by mostly Dll-expressing granule cells. Thus, most Notch receptors are saturated with Dll1, which induces proteolytic cleavage and boosts the release of NICD in the NSC. The NICD is transported to the nucleus, where it activates a set of genes that ultimately maintain the NSC in a quiescent state with the potential to enter cell cycle if needed. Lfng-mediated Dll1-Notch signaling thus keeps majority of the NSCs in the quiescent state, protecting them from random activation and division.
Once a NSC is stimulated to divide (active state), it starts to produce ANPs. The first progeny (Type 2a) do not express Notch ligands and the NSC continues to actively divide. As ANPs mature into late ANPs (Type 2b), they start to express Jag1. Jag1 binding to the Lfng-modified Notch on NSC does not generate NICD, so the local accumulation of Jag1-expressing ANP progeny around the mother NSC gradually lowers the levels of NICD within the NSC and decreases Notch signaling, eventually causing NSC to stop dividing because it is surrounded by progeny. This signaling thus preserves both the spatially-enclosed SGZ niche from an overwhelming number of newborn cells and the NSC itself from over-dividing. Note that our model describes preservation of the NSCs in two contexts: in the resting state, Lfng-boosted Dll1-mediated Notch signaling preserves the NSCs at the population level (i.e., prevents global NSC activation to undergo division), while in the context of active division, Jag1-mediated inhibition of Notch signaling preserves the NSC on the single-cell level (prevents its perpetual division). Therefore, Lfng-mediated Notch signaling could be the key for preserving NSCs in both resting and active states.
Our model is supported by our data on the outcomes of Lfng, Dll1 and Jag1 loss-of-function with respect to the NSC number, cell cycle properties, NICD1 intensity, and fate. Lack of Lfng reduced NICD1 expression in NSCs and induced many more NSCs to divide at the population level; they spend less time in the active state, produce fewer ANPs and tend to differentiate into astrocyte-like cells, eventually depleting the NSC population. Similar observations were made when Notch signaling in NSCs was inhibited by targeting different components of the pathway. Removal of either Notch1 (Ables et al., 2010) or Rbpj-k (Ehm et al., 2010) caused the depletion of NSCs, which was attributed either to a failure of NSC self-renewal (Ables et al., 2010) or to an initial burst in the number of proliferating NSCs followed by their depletion (Ehm et al., 2010). The deletion of Rbpj-k caused a more robust and faster decrease in the NSC population compared to the removal of Notch1, possibly due to different mechanisms of Rbpj-k and Notch1 signaling. Namely, Rbpj-k is required not only for Notch1-mediated signaling, but also Notch signaling through other receptors (Andersson et al., 2011). Thus, deletion of Rbpj-k might cause a more severe depletion of NSCs as they could not remain quiescent. On the other hand, at least in the subventricular zone Notch1 seems to be required for the actively dividing NSCs but not when they are in the quiescent phase (Basak et al., 2012); thus, deletion of Notch1 affects only a subpopulation of NSCs.
Our data in Lfng mutants resemble both models, suggesting that Lfng mediates signaling in both quiescent NSCs (similarly to the Rbpj-k) and active NSCs (similarly to Notch1). Both modes of Lfngmediated action led to multiple rounds of NSC division, which, based on the literature, could be followed by either a return to quiescence (Bonaguidi et al., 2011;Lugert et al., 2010) or a terminal differentiation into astrocyte-like cell (Encinas et al., 2011). Removal of Lfng both diminishes the intensity of Dll-Notch signaling and augments the production of NICD following Jag-Notch binding  (GC, grey). Dll1 binds to the Lfng-modified Notch receptor on the NSC, which boosts Dll1mediated Notch signaling by producing more NICDs and the NSC is kept quiescent but ready to undergo cell cycle if stimulated. Once activated, NSC (red) starts to produce ANPs. The first progeny (Type 2a, small dark yellow cells) do not express Notch ligands and the NSC continues to divide. As Figure 7 continued on next page (Benedito et al., 2009;LeBon et al., 2014;Stanley and Okajima, 2010;Taylor et al., 2014;Yang et al., 2005). The diminished Dll-Notch signaling causes more quiescent NSCs to be recruited; once they start to produce Jag1-expressing ANPs, NSCs lacking Lfng will continue to be exposed to Notch signaling, which might lead to gliogenesis (Breunig et al., 2007;Chambers et al., 2001;Imayoshi et al., 2013;Tanigaki et al., 2001). Notably, we observed a greater number of astrocytelike cells in Lfng mutants, not increased apoptosis. Indeed, the number of apoptotic cells in the Lfng mutant mice was diminished, most likely due to production of fewer ANPs.
As the Lfng-modified Notch receptor enables the cell to distinguish between Dll1 and Jag1 ligands to activate differential downstream targets (Chapouton et al., 2010;Isomura and Kageyama, 2014;Nellemann et al., 2001;Ninov et al., 2012;Yoshiura et al., 2007), it is not surprising that Dll1 and Jag1 have opposing effects on NSC biology. Absence of Dll1 caused a greater proportion of NSCs to undergo division and accelerated their exit from the active state, leading to differentiation into astrocytes and premature exhaustion of the NSC population. On the other hand, absence of Jag1 exerted its effect on the active NSC population rather than the quiescent one. We base this claim on the following: (i) the total NSC populations in Jag1 À/+ and wild-type mice did not differ, so the quiescent NSCs were not exhausted because of increased recruitment; (ii) the two-hour BrdU incorporation rate was slightly greater in NSCs lacking Jag1 compared to controls. On the other hand, there was a significant increase in the proportion of BrdU + NSCs in iJag1 fl/fl mice compared to controls. This may be due to the dosage difference between the homozygous conditional and heterozygous constitutive Jag1 knockout mice, which might result in an increase in the actively dividing NSC population rather than increased recruitment of NSCs from the quiescent pool. The extended cycling of NSCs agrees with previously reported extended cycling of NSCs that overexpress NICD (Breunig et al., 2007); indeed, NICD1 expression is higher in iJag1 fl/fl NSCs. Thus, removal of Jag1 lowers the chance of NSC to stop dividing via sustaining the Notch signaling intensity, which suggests that Jag1 expressed on the ANP progeny is able to send feedback signals to the ancestor NSC and downregulate its Notch signaling. In addition, our data suggest that in the absence of Jag1, NSCs undergo self-renewal in addition to producing large amounts of ANPs. Intuitively, more ANPs should give rise to more new neurons, but conditional ablation of Jag1 has been shown to cause aberrant neuronal lineage formation (Lavado and Oliver, 2014); an increase in the number of ANPs should therefore be interpreted cautiously with respect to neurogenesis. Nevertheless, increased self-renewal with no significant rise in neuronal population has been observed in social isolation (Dranovsky et al., 2011). Thus, it will be interesting to further investigate the mechanisms that mediate self-renewal in the future, as these processes might be exploited for targeted manipulation of NSCs.
Altogether, our data represent a substantial advance in our understanding of adult hippocampal neurogenesis, with critical implications for the preservation of adult NSCs. Namely, Lfng in NSCs along with Dll1 and Jag1 ligands in the progeny are an important part of the regulatory machinery that governs NSC maintenance, controlling their recruitment (preventing global activation), division (both the number and the termination of active state) and terminal fate (preventing over-transformation into astrocyte-like cells), both in physiological and pathological states associated with the NSC impairment.

Generation of the Lfng-CreER T2 mouse
The RP23-270N2 BAC parent plasmid containing the Lfng locus was modified by recombinering to insert a CreERT2 sequence in frame at the transcriptional start site. Lfng-CreER T2 mice were generated by pronuclear injection of the modified BAC construct into fertilized FVB/n embryos. To trace progeny of Lfng-expressing cells, the mice were crossed with Ai14 (RCL-tdT) reporter mice (Madisen et al., 2010).

Immunohistochemistry
Animals were deeply anesthetized by injection of 4% Avertin per body weight and perfused transcardially with 30 ml of phosphate-buffered saline (PBS) followed by 30 ml of 4% (w/v) ice cold paraformaldehyde (PFA) in PBS. Brains were removed longitudinally; post fixed with 4% PFA for 4 hr at room temperature, and the PFA was then replaced with PBS and the tissue kept at 4˚C. Free floating serial sagittal sections of 50 mm thickness were cut using a vibratome. Sections were incubated with blocking and permeabilization solution (0.3% Triton-100X and 3% BSA in PBS) for 2 hr at room temperature, followed by overnight incubation at 4˚C with the primary antibody diluted in the same permeabilization-blocking solution. Sections were then washed three times with PBS and incubated with fluorochrome-conjugated secondary antibodies diluted in permeabilization-blocking solution for 2 hr at room temperature. Sections were then washed three times with PBS and mounted on coated slides with DakoCytomation Fluorescent Mounting Medium (DakoCyomation, Carpinteria, CA) as anti-fading agent.
In the experiments where sections were treated with HCl for BrdU, CldU, IdU, and Ki67 detection or with citric acid-based antigen retrieval solution for Dll1, Jag1, Hes5, Notch1, and NICD1 detection, we had to use antibodies against GFP or RFP to detect GFP or tdTomato, as they rapidly fade due to the acidic treatment of the tissue. The eGFP signal from Lfng-eGFP and Nestin-GFP mice, or tdTomato signal in Ai14 crossed control and conditional knockout mice was amplified with antibodies against GFP (chicken anti-GFP (Aves Labs Cat# GFP-1020 RRID:AB_10000240) at 1:1000), or with antibodies against RFP rabbit anti-RFP (Rockland Cat# 600-401-379 RRID:AB_2209751) at 1:500); goat anti-RFP (SICGEN, Cantanhede, PORTUGAL) at 1:200). Anti-GFP or anti-RFP antibodies were not used for other analyses.

Confocal microscopy and stereology
To estimate the total number of different cell types or cells that are positive for different antigens without falling into a spatial bias, 50 mm free-floating sagittal sections spanning through the whole dentate gyrus were collected in five parallel sets using vibratome. Each set contained 13-14 sections 250 mm apart from each other. Previously described modified optical dissector method (Encinas and Enikolopov, 2008) was used for unbiased quantification of absolute number of different cell types in the whole brain. Briefly, an observer blind to experimental groups counted immunoreactive cells that had the aforementioned markers and morphological properties in every fifth sagittal section throughout the dentate gyrus. NSCs were identified as cells with triangular cell body in the SGZ, a GFAP + radial process originating from a Sox2 + nuclei in SGZ and spanning through granule cell layer in mice without the fluorescence reporter (C57BL/6, Lfng À/+ , Jag1 À/+ , Dll1 À/+ , tdTomatoclones in iLfng fl/fl , iJag1 fl/fl , and Lfng-CreER T2 ; RCL-tdT control mice). In mice with a fluorescence reporter (Lfng-eGFP, Nestin-GFP, and tdTomato + clones in iLfng fl/fl , iJag1 fl/fl , and Lfng-CreER T2 ; RCL-tdT control mice), a triangular cell body located in the SGZ and a terminal ending with fine arborizations in granule cell layer-molecular layer boundary was also included as the criteria for NSC identification. Type2a and Type2b cells were identified as GFAP -Dcx -Sox2 + or Tbr2 + , respectively, round cells in the SGZ without a process cells. Neuroblasts and immature neurons were identified as Dcx + cells with single or multiple processes. Granule cells were identified as NeuN + cells with prominent dendritic arborizations. Astrocytes were identified as S100b + cells with stellar morphology. Astrocytelike cells in iLfng fl/fl mice were identified GFAP + S100bcells with stellar morphology and multiple processes.
For quantification that required analysis of the phenotypic morphology and/or overlap of multiple markers (including BrdU, CldU, IdU, and Ki67), 20 mm thick optical sections were scanned with confocal microscope (Leica SP5, Leica SP8 or a Zeiss LSM 710). Three-dimensional reconstructions and orthogonal views were obtained using Zen 2012 SP1 software (Zeiss, Thornwood, NY) or LAS AF Lite (Leica Microsystems, Buffalo Grove, IL). Cells that were located in the uppermost focal plane were excluded from the quantification to avoid overestimation. Total counts from 13 to 14 sections were multiplied by five for the total number of cells of interest in one hemisphere and then by two to get the total number of cells for both dentate gyri. The proportion of BrdU, CldU, IdU, and Ki67 positive cells among a certain cell type was calculated by dividing the total number of positive cells to previously calculated number of total cells of a particular genotype or to the total number of tdTomato + clones of a particular cell type in lineage analysis.
Clusters were defined as groups of Ki67 + cells located around the NSC, where individual cells were separated by 20 mm from their nearest neighbor. Cells beyond this limit were counted as separate clusters. Clusters were counted in every fifth section of the dentate gyrus.

Quantification of NICD1 signal
Following NICD1 immunofluorescence staining, multiple sections from each genotype were imaged (N = 3 for 2 weeks post TMX injection in iJag1 fl/fl and N = 4 for control and iLfng fl/fl mice). 5 mm thick Z-stacks images from both CA1 region and a viewpoint that covers the whole SGZ region of the section were scanned with 2048 Â 2048 pixel resolution (Zeiss710 LSM confocal microscope). Maximum intensity projections were converted to Tiff images in ZEN 2012 SP1-black edition 64 bit (Zeiss, Thornwood, NY). Regions of interest for individual NSCs (at least 20 cells per section) were drawn around in the tdTomato + NSC cell body located in the SGZ. Regions of interest in the whole SGZ were drawn as 15 mm thick stripes covering the whole SGZ in the section. To compensate the section-to-section differences due to staining irregularities, NICD1 immunofluorescence intensity of neurons in CA1 region (where Lfng-CreERT2 is not active) was used as control (75 mm X 50 mm rectangle). Blue channel fluorescence intensities (corresponding to Cy5 signal in TSA amplified NICD1 immunostaining) were measured by Histo tool in ZEN lite 2012-blue edition and intensities were normalized according to CA1 measurements.

Quantification of venus/NICD1 signal overlap
Following NICD1 immunofluorescence staining, two different sections from CBF:H2b-Venus animals (N = 3) were imaged. 15 mm thick Z-stacks images with 2048x2048 pixel resolution were scanned with Zeiss710 LSM confocal microscope. At least 200 cells were counted from SGZ and GZ of each section to evaluate the overlap between Venus and NICD1 signals in the same cell.

Detection of apoptotic cells
For activated caspase-3 staining, paraffin embedded sections (8 mm thick) from control, iLfng fl/fl , and iJag1 fl/fl animals two weeks post TMX injection were treated with antigen retrieval solution (Tris-EDTA pH 9.0), followed by primary antibody treatment (anti act-casp3 antibody; Abcam, Cambridge, MA) and DAB staining (Abcam, Cambridge, MA) according to manufacturer's protocol. Due to high yield clearance of apoptotic cells from SGZ (Abiega et al., 2016;Li et al., 2017;Sierra et al., 2010), we also used ApopTag Peroxidase in situ apoptosis detection kit (EMD Millipore, Billerica, MA). Bright field images were acquired by Zeiss Axio Imager M2 microscope and the total number of apoptotic cells located in 30 mm of SGZ was counted.

Electroconvulsive shock (ECS) and voluntary physical exercise
ECS experiments were performed using Ugo Basile 57800 Unit (Varese-Italy). Bilateral ECS was administered via moistened pads on ear clips using pulse generator in 3-month-old male Lfng-eGFP mice (N = 4, frequency, 50 Hz, shock duration 0.5 s, pulse width 0.5msec, current 50mA) at the same time each day for four consecutive days. BrdU (150 mg/kg) was injected at the fourth day, 2 hr after the last ECS treatment. Sham animals were exposed to the same procedure but did not receive a shock. Mice were sacrificed 24 hr following the BrdU injection and brains were processed for immunostaining with anti-BrdU antibody, as described above.
For voluntary physical exercise, two animals were placed in a cage (3 cages in total, N = 6) with a running wheel and their physical activity was monitored by Actimetrics system (Wilmette, IL) for a week. Control animals were housed in the same type of cage but with a locked running wheel. BrdU (150 mg/kg) was injected at the seventh day. Mice were sacrificed 24 hr later and brains were processed for immunostaining with anti-BrdU antibody, as described above.
Temozolomide, tamoxifen, DTX, BrdU, CldU, and IdU injections Temozolomide (Sigma-Aldrich, St. Louis, MO, 25 mg/kg in DMSO, PBS) was administered as four intraperitoneal injections. Control mice were injected with a vehicle. Four hours after the last injection, the mice were perfused transcardially and processed as described above. Diphtheria Toxin from Corynebacterium diphtheriae (Sigma-Aldrich, St. Louis, MO, 16 mg/kg in PBS) was administered as four intraperitoneal injections. Tamoxifen (200 mg suspended in 10 ml of 1:9 ethanol:corn oil mixture) solution was administered either at high (200 mg/kg) or low (120 mg/kg) dose. Control mice were injected with ethanol:corn oil mixture only. BrdU (150 mg/kg) was administered as described in schemes of the corresponding figures. CldU (85 mg/kg) and IdU (115 mg/kg) were administered in equimolar concentrations. BrdU and CldU were dissolved in sterile saline. IdU was dissolved in sterile saline solution that contained 2% of 0.2N NaOH.

Statistics
Statistical analysis was performed using GraphPad Prism 7.0 (GraphPad RRID:SCR_002798). The sample size was determined based on our previous publications (Sierra et al., 2010(Sierra et al., , 2015 and published data from other groups. Experiments involving two groups were compared using un-paired Student t-test. Experiments involving more than two groups with one variable were compared by One-Way Analysis of Variance (ANOVA), followed by Tukey HSD post-hoc test analysis for pairwise comparisons. Types of test were indicated in the figure legends. Exact p values were indicated in the main text next to sample size. Significance was defined as p<0.05. Data are shown as mean± SEM *p<0.05, **p<0.001, ***p<0.0001 denoted the corresponding significance levels in all graphs.