Neural stem cell quiescence is actively maintained by the epigenome

Summary

Homeostasis of the nervous system is maintained by a population of resident neural stem cells (NSCs) retained in a state of reversible cell-cycle arrest called quiescence. Quiescent NSCs can resume proliferation in response to different physiological stimuli. Reactivation requires changes in gene expression, much of which is regulated at the epigenomic level. We mapped epigenomic changes in NSC chromatin during stem cell quiescence and reactivation in Drosophila in vivo. Contrary to expectations, chromatin accessibility is increased in quiescent NSCs. Surprisingly, genes crucial for cell-cycle progression are repressed while remaining within permissive H3K36me3-bound euchromatin. At the same time, genes necessary for cell-cell communication are derepressed by eviction of histone H1 and transition to an SWI/SNF-enriched active state. Our results reveal global expansion of accessible chromatin in quiescent NSCs without concomitant transcriptional activation. Strikingly, this process reverses upon reactivation, indicating that opening of chromatin is a quiescence-specific event.

Graphical abstract.

Introduction

Many adult tissues retain a population of quiescent stem cells to mediate repair and maintain tissue homeostasis in response to the changing physiology. Quiescence is an actively maintained state of cell-cycle arrest associated with differential transcription, epigenomic state, and metabolism.¹ In the adult mammalian nervous system, quiescent neural stem cells (NSCs) are found in the subventricular zone of the lateral ventricles, the dentate gyrus of the hippocampus, and the median eminence of the hypothalamus, where they contribute to neurogenesis in response to cues such as changing diet, exercise, pregnancy, or injury.^2–5

Quiescence is widely thought to be associated with chromatin condensation and therefore transcriptional and metabolic inactivity.^6,7 However, the quiescent state needs to be actively maintained, as quiescent cells have to retain their self-renewal and differentiation capabilities through cell cycle exit and resume proliferation in response to triggers while in communication with their niche.^7,8 A careful examination of quiescent stem cell models reveals a variety of chromatin changes dependent on the tissue of origin.⁹ In muscle stem cells, quiescence is characterized by increased H4K20me3-mediated repression.¹⁰ Quiescent lymphocytes and hair follicle stem cells, on the other hand, have lower levels of both active and repressive histone methylation.^11–13 Consequently, the quiescent chromatin signature is tissue specific and dependent on the developmental context,⁹ and therefore, studying the chromatin of quiescent cells in their native environment is crucial. The sparseness of NSCs in the mammalian brain and the lack of well-defined genetic markers lead to difficulty in profiling the genome of mammalian NSCs while still in their niche. As a result, studies investigating chromatin changes in quiescent and active NSCs have focused on chromatin accessibility^14,15 instead of the full repertoire of chromatin proteins and histone modifications. As such, little is known about the specific epigenetic modifiers that might regulate mammalian NSC quiescence and reactivation in vivo.

In contrast to the mammalian nervous system, NSCs in Drosophila melanogaster are easily identifiable in vivo by well-defined genetic markers and the stereotypic developmental timeline of quiescence induction and reactivation. Drosophila NSCs share many features with their mammalian counterparts, including similar morphology, reliance on oxidative phosphorylation during proliferation, homologous quiescence factors such as cyclin-dependent kinase inhibitors, and Akt-dependent reactivation triggered by a dietary signal.^16–20 Drosophila NSCs in the embryonic and larval brain serve as a well-studied and more easily accessible natural quiescence model.^21,22 However, the understanding of how quiescent and active NSC gene signatures are established and maintained at the chromatin level and whether epigenomic pathways instruct quiescence induction and reactivation remains lacking.

The fundamental unit of chromatin is the nucleosome, which consists of a covalently modified histone octamer and the 147 base pairs (bp) of DNA wrapped around it.²³ Classically, chromatin has been classified into transcriptionally active, generich euchromatin and silent, gene-poor heterochromatin. However, it is clear that these two broad categories can be further separated based on associated histone modifications, chromatin-binding proteins, and function.^24–26 In Drosophila melanogaster, the chromatin landscape was classified into five principal states based on the DNA-binding profiling of 53 chromatin proteins.²⁷ Euchromatin was classified into two distinct types: chromatin at highly transcribed, developmentally regulated genes associated with an SWI/SNF remodeler, Brahma (Brm; also known as “red”), and chromatin enriched in H3K36me3, present at housekeeping loci (“yellow”).²⁷ Heterochromatin was separated into constitutive, H3K9me3-bound regions found mostly at pericentromeres and telomeres (“green”); facultative, H3K27me3- and Polycomb (Pc) complex-bound heterochromatin present at developmentally regulated genes (“blue”); and finally, lamin-associated regions enriched in histone H1 that lack histone modifications and cover most silent genes (“black”).²⁷ The balance between different types of euchromatin and heterochromatin is regulated by histone-modifying enzymes that deposit methylation or acetylation at distinct amino acid residues of histone proteins.^28–30 Histone modifications and chromatin proteins ultimately regulate the spacing of nucleosomes and the formation of condensed structures, which affects the accessibility of DNA for transcription.³¹

Here, we investigated the chromatin changes that occur in NSCs during quiescence induction and reactivation. We generated a comprehensive genomic dataset of nine chromatin-binding proteins and histone modifications in three developmental conditions: proliferating NSCs prior to cell cycle exit, quiescent NSCs (qNSCs), and reactivated NSCs. Surprisingly, we found that in qNSCs, accessible chromatin is increased at signaling genes as a result of remodeling and deposition of active histone modifications. Strikingly, cell-cycle-related loci are retained within permissive, H3K36me3-bound euchromatin but are transcribed at lower levels during quiescence. Overall, our results show that quiescence in NSCs is a process that requires active chromatin remodeling, including increased accessibility and expansion of active chromatin states without, necessarily, concomitant transcription. This may enable quiescent cells to be poised for rapid reactivation.

Results

qNSCs gain accessible genomic regions

To investigate chromatin dynamics in NSCs in vivo, we used Targeted DNA adenine methyltransferase identification (DamID) to profile protein DNA binding without cell isolation, immunoprecipitation, or fixation.³² Techniques that rely on cell sorting, such as chromatin immunoprecipitation (ChIP), require high cell numbers per sample and highly specific antibodies.³³ In contrast, Targeted DamID is ideal for profiling genomic binding in vivo and when cell sorting can be challenging, such as for small numbers of NSCs.^32,33 Targeted DamID is designed to express low levels of E. coli Dam methyltransferase fused to a protein of interest in a tissue-specific manner using the GAL4 system.³² Dam is guided to specific genomic locations by the protein of interest and methylates the adenine in GATC sequences near its binding sites. This gives rise to methyl groups marking the sites of binding, which can be isolated using methylation-sensitive restriction enzymes. Binding data are normalized against the binding of untethered Dam, which methylates adenines at nucleosome-free GATC regions, yielding cell-specific accessibility profiles similar to assay for transposase-accessible chromatin using sequencing (ATAC-seq) or formaldehyde-assisted isolation of regulatory elements and sequencing (FAIRE-seq)³⁴ (Figure 1A).

Figure 1. Quiescent neural stem cells gain accessible regions.

(A) A diagram describing isolation of accessible genomic regions using Targeted DamID.

(B) Timeline of Dam expression for capture of pre-quiescent (embryonic proliferating NSCs [epNSCs]), quiescent NSCs (qNSCs), and post-quiescent (larval proliferating NSCs [lpNSCs]).

(C) Number of significantly differentially accessible peaks (false discovery rate [FDR] < 0.05) between epNSCs and qNSCs (left) and the volcano showing the fold change and −log₁₀ FDR for the peaks (right).

(D) Number of significantly differentially accessible peaks (FDR < 0.05) between qNSCs and lpNSCs (left) and the volcano showing the fold change and −log₁₀ FDR for the peaks (right).

(E) Feature annotation of accessible sites gained in qNSCs.

(F) Gene Ontology calls of accessible peaks gained in qNSCs.

To investigate the changes in chromatin accessibility during quiescence and reactivation in NSCs, we drove expression of Dam at different time points with an NSC-specific GAL4 driver (worniu-GAL4). First, we profiled NSCs during the embryonic wave of neurogenesis (embryonic proliferating NSCs [epNSCs]), from midway through embryogenesis until later embryonic stages (stage 9 through stage 16; according to Campus Ortega and Hartenstein staging³⁵; Figure 1B). The majority of NSCs exit the cell cycle by the end of embryogenesis (stage 17) and begin reactivation during the first larval instar.³⁶ Next, to profile qNSCs, we used a temperature-sensitive repressor of the GAL4 protein, GAL80 (GAL80^ts),³⁷ to limit Dam expression to late embryogenesis and early larval first instar (Figure 1B). GAL80^ts represses GAL4 at the permissive temperature of 18°C but is inactive at the restrictive temperature of 29°C, enabling inducible GAL4-driven expression.^38,39 Finally, we used GAL80^ts to restrict Dam expression to late first instar larvae (24 h after hatching) to profile larval proliferating NSCs (lpNSCs) (Figure 1B).

We used differential binding analysis of Dam-enriched fragments to recover differentially accessible genomic peaks. When filtered by the false discovery rate (FDR) (FDR < 0.01) to retain only significantly different fragments, we observed higher numbers of accessible peaks in qNSCs in comparison to epNSCs or lpNSCs (Figures 1C and 1D). We assayed the genomic distribution of differentially accessible regions and found that chromatin was decondensed upon quiescence induction at both genic and intergenic regions (Figure 1E). This indicated that qNSCs undergo global expansion of accessible chromatin. This process reverses upon reactivation, indicating that the opening of chromatin is a quiescence-specific event (Figure 1D). Genes with a higher number of accessible peaks in qNSCs are associated with Gene Ontology calls related to cell-to-cell signaling, such as the G-protein-coupled receptor (GPCR) signaling pathway and ion transport (Figure 1F; example tracks are provided in Figure S1). Therefore, contrary to expectations, we found that qNSCs decondense chromatin at both genic and intergenic regions (Figure S1).

NSC chromatin can be classified into seven functional states

Having established that qNSCs gain accessible regions, we investigated the epigenomic changes that enable NSCs to remodel their chromatin through quiescence entry and reactivation. We performed a series of Targeted DamID experiments profiling chromatin modifiers and histone modifications in NSCs. Using the time points defined in Figure 1B, we expressed a selection of tools that recognize a variety of histone modifications specific to both euchromatin and heterochromatin (Figures 2A, S2, and S3). We used well-established constructs expressing Dam fused to Drosophila chromatin proteins with known roles and binding patterns: the SWI/SNF factor Brm, histone H1, heterochromatin protein 1a (HP1a), Pc, and RNA polymerase II (RNA Pol II)^32,40,41 (Figures 2A, S2, and S3). We also expressed Dam fused to transgenic engineered chromatin reader domains specific against histone modifications, TAF3-phd (H3K4me3)^42,43 and Dnmt3a-pwwp (H3K36me3),^43,44 as well as Dam fused to nanobodies recognizing H4K20me1^43,45 and H3K9ac.^43,46

Figure 2. Neural stem cell chromatin can be separated into 7 functional states.

(A) A diagram showing the repertoire of DamID tools used for hidden Markov modeling of chromatin-state data.

(B) Emission probabilities of the 8-state concatenated chromatin model, i.e., how likely a given signal is to be present within a given state.

(C) Chromatin accessibility of given states in the three conditions, measured using counts per million (CPM) of Dam-only libraries.

(D) Genomic feature distribution of regions annotated as a given chromatin state.

(E) Genomic tracks of chromatin-state annotation and associated Dam fusion average signal at four loci in embryonic NSCs. The Dam fusion signal is measured in Dam-fusion/Dam-only ratios, and accessibility is measured as read coverage. Inscuteable (insc), a gene expressed in NSCs, is marked by H3K4me3 at the promoter (active state 1) and by RNA Pol II at the promoter and throughout the gene body together with H3K36me3, H4K20me1, and H3K9ac (active states 3 and 4). Brahma and H3K9ac are present upstream of the promoter (active state 2). Deadpan (dpn) is a highly expressed NSC marker. As such, it is mostly annotated by state 1, with high levels of active histone modifications and RNA Pol II throughout the locus. Deformed (Dfd) is a Hox gene repressed by Polycomb in epNSCs and labeled as repressive state 1 accordingly. Finally, acetylcholine nicotinic receptor 6 alpha (nAChRalpha6) is not expressed in proliferating NSCs and is mostly marked by histone H1 and labeled as repressive state 2.

We used machine learning to create chromatin-state annotations based on the genomic binding profiles of these nine Dam fusions. We fitted a concatenated hidden Markov model for the three NSC conditions to our datasets with ChromHMM⁴⁷ (explained in detail in Figures S4A and S4B). We identified four active and three repressive chromatin states in the 8-state model (Figures 2B and 2C). The model classified chromatin in accordance with those previously obtained by state modeling in different Drosophila tissues and cells.^{26,27,40,41,48} To maintain continuity with previous annotations,^27,40,41,48 in particular Filion et al.,²⁷ we used a similar color labeling code to denote different chromatin states (Figure 2E).

Previous chromatin-state annotations in Drosophila have separated two types of euchromatin: trithorax (Trx)-associated red regions and H3K36me3-associated yellow regions.^26,27,40,41 In this work, active state 1 (red) is characterized by high levels of H3K4me3, RNA Pol II, and other active histone modifications and is found mostly around the transcriptional start sites (TSSs) and at short, highly expressed genes, such as the NSC gene deadpan (Figures 2B, 2D, and 2E). This state is most similar to the red chromatin in Filion et al.²⁷ and state 1 (red) in Kharchenko et al.²⁶ Active state 2 (pink) is characterized by high levels of Brm, RNA Pol II, and other active chromatin marks. Brm is a Drosophila ATPase-dependent chromatin remodeler, homologous to the SWI/SNF factor,⁴⁹ that can modulate H3K27ac via its association with histone acetyltransferase CBP/nej and H3K27me3 demethylase ubiquitously transcribed tetratricopeptide repeat, X chromosome (Utx).⁵⁰ Many regions annotated as active state 2 are found at introns and distal intergenic regions (Figures 2D and 2E). As active state 2 is associated with H3K27ac and present in non-coding or intronic regions, it is suggestive of active enhancers²⁶ and transcribed genes with large 5' introns, similar to state 3 (brown) in Kharchenko et al.²⁶

Active states 3 and 4 (orange and yellow) exhibit characteristics of transcriptional elongation at gene bodies, such as H3K36me3 and H4K20me1^51,52 (Figure 2B). The identity of active states 3 and 4 is supported by their presence in exons (Figures 2D and 2E). Active state 1 is often followed by active state 4 in the genome, reflecting their localization at promoters and active gene bodies, respectively (Figures 2E and S4C). Both states reflect state 2 (pink) from Kharchenko et al.²⁶ classification and yellow chromatin from Filion et al.²⁷ Indeed, they are mostly present at short genes lacking long intronic regions (Figures 2D and 2E). Importantly, chromatin accessibility is highest in states 1 and 2, at the regions associated with regulatory elements (Figure 2C).

The remaining states are characterized by lower chromatin accessibility and the presence of chromatin proteins that suggest repressive chromatin (Figures 2B and 2C). We observe a clear separation of Pc-bound regions (blue chromatin in previous studies^27,40,41 and gray in Kharchenko et al.²⁶) and HP1-related heterochromatin (green in previous studies^27,40,41 and blue in Kharchenko et al.²⁶), representing repressive states 1 and 3, respectively (Figures 2B, 2C, 2E, and S4C). Repressive state 2 is reminiscent of the black or “silent” chromatin reported previously in Drosophila^26,27,40,41 or the “naive” chromatin in mammalian cell lines²⁴ and consists of repressive regions with fewer histone modifications and high histone H1 levels (Figure 2B). Finally, we identified an “empty” state devoid of any signal that occurs due to data processing steps allowing to generate a high signal-to-noise ratio at the expense of the amount of genomic coverage (see STAR Methods). These regions were named “unannotated” and excluded from most further analyses (Figure 2B).

qNSCs gain active chromatin domains

Having assigned identities to the chromatin states, we assayed the genomic frequency of each state in NSC genomes during embryonic proliferation (epNSCs), quiescence (qNSCs), and reactivation (lpNSCs). Remarkably, we found that the fraction of active states 1 and 2 (red and pink) increased in quiescence, confirming the results we obtained investigating chromatin accessibility alone (Figure 3A).

Figure 3. Quiescent NSCs gain domains marked by active chromatin states.

(A) Chromatin-state frequency at the whole genome in the 3 tested conditions.

(B) Chromatin-state frequency at state-annotated promoters (defined as regions 500 bp upstream and 200 bp downstream of the TSS) in the 3 tested conditions.

(C) Number of significantly differentially bound H3K4me3 (TAF3) peaks (FDR < 0.05) between epNSCs and qNSCs (left) and the volcano showing the fold change and −log₁₀ false discovery rate (FDR) for the peaks (right).

(D) Alluvial plot for gene-based chromatin-state transitions for all Drosophila genes. Genes that remain in unannotated chromatin in all 3 conditions are filtered out. Transitions assayed in later figures are highlighted.

In particular, the genomic frequency of active state 1 increased during quiescence, followed by a reciprocal decrease in reactivated NSCs. This suggested that a group of promoters or genes becomes activated only in qNSCs (Figure 3A). To confirm that the increase of active state 1 is specifically associated with the opening of promoters, we compared the chromatin-state frequency at TSS-associated regions (Figure 3B). We defined the putative Drosophila promoter region as 500 bp upstream to 200 bp downstream of the TSS and annotated each TSS region with a chromatin state. We found that the active state 1 increase in quiescence is more pronounced at TSS-associated regions. To validate this finding using analysis independent of the chromatin-state model, we assayed the number of H3K4me3-bound peaks during quiescence induction with differential binding analysis and found, accordingly, that the number of H3K4me3-bound regions was higher in qNSCs (Figure 3C). Overall, we showed that chromatin remodeling that leads to a higher number of accessible regions in qNSCs is due to a gain of active states 1 and 2 associated with promoter and enhancer regions. This result was striking, as chromatin was shown to condense and gain repressive histone methylation in other models of quiescence.⁹

qNSCs maintain permissive chromatin at cell cycle genes

We were surprised to find increased open chromatin during quiescence, given that chromatin accessibility is generally thought to represent areas of active transcription. We asked if specific chromatin-state transitions were associated with proliferation genes and with genes that are transcribed in qNSCs. To better reflect chromatin transitions at the single-gene level, we developed a new bioinformatic tool (feat_annot) to annotate each Drosophila locus with a chromatin state (Figure S4D). Although a single gene is likely marked by multiple states, we used the chromatin state with the highest coverage at given loci for gene annotation to reduce the dimensionality of the data. We used this annotation to plot the frequency of chromatinstate transitions at all Drosophila genes and found unique chromatin transitions that occur between active embryonic NSCs and qNSCs, within both euchromatin and heterochromatin (Figure 3D).

We focused on a group of 335 genes that transitioned from active state 1 (equivalent to red euchromatin according to the classification by Filion et al.²⁷) to H3K36me3-associated active state 4 (yellow euchromatin) during quiescence induction (Figures 4A and 4B). As active state 4 (yellow) is associated with lower RNA Pol II levels and chromatin accessibility (Figure 2B and 2C), we speculated that this gene group would be downregulated during quiescence. Strikingly, we found that most genes within this category are necessary for the progression of the S and M phases of the cell cycle based on Gene Ontology calls, including DNA replication, organization of the mitotic spindle, and chromatid separation (Figure 4C). States 3 and 4 have lower chromatin accessibility and RNA Pol II levels than state 1 (Figure 2C), which indicates that these loci are within permissive, but not highly active, chromatin states.

Figure 4. Quiescent NSCs maintain euchromatin at cell cycle genes.

(A) Alluvial plot for genes undergoing a transition from active state 1 to state 4 during quiescence induction (335 genes).

(B) mRNA expression levels and chromatin-state annotation at three example loci: origin recognition complex subunit 5 (Orc5), aurora A (aurA), and structural maintenance of chromosomes 2 (SMC2).

(C) Gene Ontology (biological process Slim) of genes shown in (A).

(D) UCell expression score of gene set in (A). All gene expression results are based on the scRNA-seq dataset from Gherghina et al.⁵³

To assess transcription levels at these loci, we analyzed our single-cell RNA sequencing (scRNA-seq) dataset of the embryonic and larval brains.⁵³ Several genes necessary for proliferation (aurora A, Orc5, and SMC2 are shown) were maintained in active states in qNSCs (active states 1, 3, and 4 [red, orange, and yellow]) but were not transcribed, as indicated by the scRNA-seq of qNSCs and lpNSCs performed previously by our group⁵³ (Figure 4B). Finally, we calculated the expression score (UCell score⁵⁴) for genes moving between active states 1 and 4 in qNSCs and lpNSCs. We found that cell cycle genes that transition from active state 4 to active state 1 during reactivation increase their expression only in lpNSCs (Figure 4D). This indicates that several genes required for the M and S phases of the cell cycle are within permissive chromatin during quiescence but are not highly transcribed.

qNSCs derepress signaling genes

The transition of a large group of genes from repressive H1 chromatin (repressive 2) in epNSCs to active euchromatin associated with Brm (active 2) in qNSCs suggested a group of genes that is specifically derepressed during quiescence (Figure 3D). We assayed the putative “quiescence-expressed” group of loci characterized by a transition out of repressive state 2 (black) to active state 2 (pink) (Figure 5A). This group consisted of over 700 genes with “neuronal-like” functions, such as synaptic and neurotransmitter signaling, ion transport across membranes, cell adhesion, and microtubule remodeling, based on Gene Ontology calls (Figure 5B).

Figure 5. Quiescent NSCs derepress neuronal signaling genes.

(A) Alluvial plot for genes undergoing a transition from repressive state 6 to active state 2 during quiescence induction (740 genes).

(B) Gene Ontology (biological process Slim) of genes shown in (A).

(C) UCell expression score of gene set in (A). All gene expression results are based on the scRNA-seq dataset from Gherghina et al.⁵³

(D) Signal ratio of the TAF3-Dam fusion (reflecting H3K4me3) in the gene group shown in (A) in embryonic proliferating NSCs (epNSCs), quiescent NSCs (qNSCs), and larval proliferating NSCs (lpNSCs).

We found that many of these genes transition back to repressive state 2 (black) in lpNSCs, indicating that they become repressed during reactivation (Figure 5A). Importantly, this suggests that transcription of neuronal genes only occurs in qNSCs and is not a feature of NSCs in general. We validated the expression of genes undergoing the repressive state 2-to-active state 2 transition in qNSCs and reactivated NSCs (lpNSCs) using scRNA-seq⁵³ and found that these genes are transcribed in quiescent cells but not in reactivated cells (Figure 5C). Accordingly, we have assayed the presence of H3K4me3 in this group of genes and found that the peak of the H3K4me3 signal is highly enriched at the TSS in qNSCs in comparison to epNSCs or reactivated NSCs (Figure 5D).

To assess whether H3K4 methylation is necessary for NSC reactivation, we performed NSC-specific RNAi knockdowns for components of Trx Drosophila complex of proteins associated with Set-like (dCOMPASS-like) and Trx-related (Trr) dCOM-PASS-like complexes using the wor-GAL4 driver. We found that histone methyltransferases trx, which is responsible for the deposition of H3K4me2/3 at developmentally regulated promoters, and trr, required for H3K4 monomethylation at promoters and enhancers, are necessary for the timely exit of NSCs from quiescence (Figures S5A and S5B). The knockdown of both led to a delay in reactivation as measured by the mitosis marker phospho-histone H3 (pH3) and the presence of the Worniu protein, which is only present at reactivation, in contrast to worniu mRNA¹⁹ (Figures S5A and S5B). Likewise, Utx H3K27me3 demethylase, which is a member of the Trr-related dCOMPASS complex, is also required for proper timing of reactivation (Figure S5C). In contrast, we found that depletion of brm does not affect reactivation, indicating that even though Brm marks active chromatin at genes upregulated in qNSCs, its binding is either a consequence of the chromatin state at these regions or there is redundancy with other ATP-dependent remodeling complexes (Figure S5D). In conclusion, we found that chromatin remodeling in qNSCs enables them to undergo reversible cell-cycle arrest while inducing the expression of genes necessary for signaling and extension of the basal projection.

Discussion

The balance between NSC proliferation and quiescence, as well as the proper order of reactivation events, is crucial to maintain tissue homeostasis. Previous studies have shown that quiescent cells are characterized by the accumulation of heterochromatin, low metabolic rates, and low transcription levels.^10,55–58 Here, we show that one of the main chromatin transitions in Drosophila qNSCs is an increase in chromatin accessibility and frequency of active chromatin states. This remodeling is the result of a transition from H1-bound chromatin (repressive state 2) to a Brmmarked state (active state 2) during quiescence induction. At the same time, proliferation genes remain in permissive euchromatin (Figure 6A).

Figure 6. Quiescent NSCs undergo expansive chromatin remodeling.

(A) Model of chromatin remodeling in NSCs. During quiescence induction, NSCs exhibit increased chromatin accessibility due to the derepression of neuronal genes, which transition from state 6 (H1-bound) chromatin to active (Brahma-bound) euchromatin associated with higher transcription. As NSCs reactivate, they begin to slowly transition back to H1-bound repressive state 2 chromatin and lower their transcription levels, although some genes have transitioned back to repressive chromatin by late first instar. At the same time, several genes necessary for proliferation (specifically M and S phases) transition from H3K4me3-marked active state 1 associated with high expression levels to permissive euchromatin characterized by H3K36me3 (active state 4) and lower transcription. As NSCs reactivate, proliferation genes become upregulated due to a transition back to active state 1.

(B) Example genomic tracks of two neuronal genes that undergo the repressive state 2-to-active state 2 transition during quiescence induction: beaten path IIIa (beat-IIIa) and Ca²⁺-channel protein α₁ subunit D (Ca-alpha1D).

(C) Example genomic tracks of two proliferating genes that undergo the active state 1-to-active state 4 transition during quiescence induction: tumbleweed (tum) and polo kinase (polo).

As an example, genes necessary for axonal fasciculation (beaten path IIIa [beat-IIIa]) and calcium signaling (Ca²⁺-channel protein α₁ subunit D [Ca-alpha1D]) move from repressive state 2 to active state 1 in qNSCs. By late first instar, some neuronal genes have moved back to repressive state 2 (Figure 6B). At the same time, genes necessary for cell cycle progression, such as the polo kinase (polo) and Rho GTPase tumbleweed (tum), move from active state 1 chromatin to active state 4 permissive euchromatin marked by H3K36me3 (Figure 6C). This transition is associated with reduced transcription and is reversed upon reactivation, when NSCs start transcribing genes needed for replication and mitosis. We also observed small changes in the frequency of active states between embryonic and larval active NSCs that most likely arise from differences in the spatiotemporal patterning between the embryonic and larval CNSs.

Quiescent stem cells re-enter the cell cycle in response to specific stimuli. Reactivation can occur within hours, such as in response to dietary signals in Drosophila NSCs. To ensure a rapid response, it might be most effective for proliferation genes to remain within permissive chromatin and boost their transcription once NSCs are reactivated. A similar process has been reported in quiescent myoblasts, where the cyclinA2 locus is found in permissive chromatin, marked by both H3K4me3 and H3K9me2,⁵⁹ but not transcribed. Permissive euchromatin at cyclinA2 inhibited the deposition of H3K27me3, and therefore hyper-repression of the locus that would block or slow down cell cycle re-entry upon reactivation.⁵⁹

qNSCs reactivate within a specific sequence to maintain the progenitor pool and build appropriate neural circuits. Interestingly, NSCs change morphology during quiescence induction from large, round cells to small cells that extend a basal process toward the neuropile of the larval CNS.⁶⁰ This suggests that genes with functions related to building the basal process and cell signaling, including microtubule-binding proteins, ion channels, neurotransmitter receptors, and synaptic proteins, are briefly derepressed during quiescence but cease transcription upon reactivation. We previously found that Drosophila qNSCs transiently transcribe neuronal genes that encode synapse-associated proteins and ion channels needed for depolarization.⁵³ Disruption of membrane polarization led to an abnormal order of NSC reactivation.⁵³ Here, we have shown that genes identified previously by scRNA-seq⁵³ are part of a larger group of loci that undergo a chromatin transition from H1 histone-marked repressive state 2 (black) heterochromatin to active state 2 bound by Brm (pink). During reactivation, NSCs retract the basal process, increase their size, and adopt a rounded shape.^53,60 Therefore, the downregulation of microtubule binding and synaptic genes during reactivation could be necessary for the disassembly of the qNSC projection. In fact, we also previously showed that over 48 h after reactivation, genes with signaling and neuronal functions are still repressed by H1 black chromatin⁴⁰ (here named repressive 2).

These neuronal signaling genes also gained H3K4me3 at their promoters to briefly become derepressed during quiescence. Consequently, we showed that depletion of COMPASS complex components trx, trr, and Utx is sufficient to disturb the timing of NSC reactivation, suggesting that deposition of H3K4 methylation is necessary to induce or maintain this derepression. The presence of Brm in this gene group suggests that quiescence induction is accompanied by the mobilization of nucleosomes and the eviction of histone H1. These processes can be induced by sequence-specific pioneering transcription factors that bind to nucleosomes.⁶¹ Crucially, this can occur prior to SWI/SNF recruitment to the locus.⁶¹ Here, we found that depletion of a single chromatin remodeler, Brm, is not sufficient to affect reactivation timing. However, we speculate that sequence-specific transcription factors recognizing motifs that we discovered in promoters of the genes that become derepressed in qNSCs (e.g., the GAGA factor [GAF]) could potentially induce the repressive state 2-to-active state 2 transition (Figures S6A and S6B). Disruption of a combination of chromatin remodelers or sequence-specific transcription factors could also impact the proper order of reactivation of NSCs by preventing expression of the signaling machinery.

The evidence for specific chromatin effectors in the regulation of mammalian NSC function is limited. For instance, loss of an ATP-dependent chromatin remodeler, Chd7, in the mouse subgranular zone of the hippocampus increased the number of cycling cells, depleting the stem cell pool over time.⁶² Recent studies of mammalian NSC chromatin attributed the loss of functionality of qNSCs with age in mice to decreased chromatin accessibility, leading to a downregulation of cell adhesion genes and, ultimately, the loss of adherence of quiescent cells.¹⁴ Such a comparison is not possible in Drosophila, as developmental NSCs are thought to apoptose during pupation, suggesting that adult neurogenesis stems from a different source.⁶³ However, this indicates that both Drosophila and mammalian qNSCs may require maintained accessibility for their proper function. Similarly, in an in vitro NSC progenitor (NSPC) quiescent system, activated NSPCs were shown to gain accessibility at H3K27ac-marked enhancers found in distal intergenic regions, whereas promoters remained stably accessible during reactivation.¹⁵ Aging also led to decreased accessibility in qNSCs in this system.¹⁵ Finally, BMP-4-induced quiescence in mouse NSCs in vitro led to the activation of over 6,000 enhancers, twice as many as the number of active enhancers in proliferating NSCs,⁶⁴ indicating that it is likely that chromatin accessibility also increases in mammalian qNSCs.

Chromatin accessibility has been associated with quiescence in the context of B cell lymphocytes,¹¹ whereas decreased histone methylation (both active and repressive) characterizes hair follicle stem cells.¹³ As pluripotency is associated with open chromatin,⁶⁵ this mechanism may enable NSCs to retain their plasticity and ability to respond to different environmental stimuli. In conclusion, we found that in contrast to other quiescence systems, in which cells condense chromatin and shut down nuclear processes (such as in nutrition-deprived yeast⁶), NSCs actively remodel their chromatin to induce functional expression of signaling components and safeguard proliferation genes against hyper-repression. The profiling of active histone modifications and chromatin remodelers at those loci will allow us to elucidate the extent to which these processes are conserved in mammalian models.

Limitations of the study

First, our study is limited by the bp resolution of Targeted DamID, which is dependent on the frequency of GATC regions recognized by Dam. In flies, this resolution is limited to about 300 bp.³² As the Drosophila genome is very compact, with limited intergenic regions, this could introduce issues when annotating DamID datasets. However, with this resolution, we recovered global differences in chromatin states. Second, the NSC driver worniu-GAL4 drives expression in a small number of mushroom-body neuroblasts, which remain proliferative throughout the quiescent period.⁶⁶ The contribution of these NSCs is averaged out among the NSC population, and the differences that we observed were quiescence specific, as we used the same driver for all conditions. Third, wor-GAL4 drives NSCs that arrest in both the G₀ and G₂ phases of the cell cycle,¹⁹ introducing heterogeneity into our dataset. While the separation of G₀ and G₂ NSCs necessitates cell sorting, which could influence gene expression and the chromatin state of the NSCs, chromatin transitions that occur specifically in G₀- and G₂-arrested cells are of interest for future investigations. Finally, this study supported Targeted DamID results with previously published scRNA-seq datasets from qNSCs and larval reactivated NSCs⁵³; however, an scRNA-seq dataset of epNSCs is not available for comparison.

Resource Availability

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Andrea H. Brand (andrea.brand@nyulangone.org).

Materials availability

This study did not generate new unique reagents.

Star★Methods

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Guinea pig anti-Deadpan	Caygill and Brand67	–
Rat anti-Worniu	Abcam	ab196362
Rabbit anti-phosphohistone H3 (pSer10)	Millipore	06–570; RRID:AB_310177
Goat anti-guinea pig Alexa Fluor 568	Invitrogen	A11075; RRID:AB_2534119
Goat anti-rabbit Alexa Fluor 405	Invitrogen	A31556; RRID:AB_221605
Goat anti-rat Alexa Fluor 633	Invitrogen	A21094; RRID:AB_2535749
Chemicals, peptides, and recombinant proteins
EDTA 0.5M	Millipore	324506
Phosphate Buffer Saline (1×PBS)	–	–
RNAseA (DNAse free)	Roche	11119915001
DpnI	NEB	R0176
DpnII	NEB	R0543S
Klenow 3′–5′ exo-enzyme	NEB	M9212L
Klenow Fragment (5 U/μL)	NEB	M0210S
MyTaq HS DNA Polymerase	Bioline	BIO-21113
NEBNext High Fidelity 2 × PCR Master Mix (6.25 mL)	NEB	M0541L
PEG-8000	Sigma-Aldrich	P5413
T4 DNA Ligase	NEB	M0202S
T4 DNA polymerase (3 U/μL)	NEB	M0203S
T4 polynucleotide kinase	NEB	M0201S
Quick ligase	NEB	M2200S
1MTris pH 8.0	–	–
SpeedBead Magnetic Carboxylate	Cytiva	65152105050250
100% ethanol	–	–
Deoxynucleotide (dNTP) Solution Set	NEB	N0446S
16% Formaldehyde Solution (w/v)	ThermoFisher Scientific	28908
Triton X-100	Sigma Aldrich	T9284
Vectashield Antifade Mounting Medium	Vector Laboratories	H-1000-10
Critical commercial assays
QiaAMP DNA Micro kit	Qiagen	56304
QIAquick PCR Purification Kit	Qiagen	28104
Qubit dsDNA Quantification Assay Kits	Thermofisher	Q32854
Genomic DNA ScreenTape	Agilent	5067−5365
Genomic DNA Reagents	Agilent	5067−5366
Deposited data
Targeted DamID data	This study	Array Express: E-MTAB-15696
Experimental models: Organisms/strains
wor-GAL4	Albertson et al.68	–
tub-GAL80ts	BDSC	7019
UAS-LT3-NDam	Southall et al.32	–
UAS-LT3-NDam-RNA Pol II	Southall et al.32	–
UAS-LT3-Dam-Pc	Marshall et al.40	–
UAS-LT3-Dam-HP1a	Marshall et al.40	–
UAS-LT3-Dam-Brm	Marshall et al.40	–
UAS-LT3-Dam-H1	Marshall et al.40	–
UAS-LT3-Dam-TAF3-phd	van den Ameele.43	–
UAS-LT3-Dam-Dnmt3a-pwwp	van den Ameele.43	–
UAS-LT3-Dam-15f11	van den Ameele.43	–
UAS-LT3-Dam-19E5	van den Ameele.43	–
UAS-ta-RNAi	BDSC	33703
UAS-Utx-RNAi	BDSC	31076
UAS-trr-RNAi	BDSC	36916
UAS-brm-RNAi	BDSC	34520
UAS-mCherzy-RNAi	BDSC	35785
Software and algorithms
damidseq_pipeline	Marshall and Brand69	vR.1
damMer suite	Tang et al.70	–
quantile_norm_bedgraph.pl	Owen Marshall	–
average_tracks.pl	Owen Marshall	–
scale.pl	Owen Marshall	–
bowtie2	Langmead and Salzberg71	–
Samtools	Li et al.72	–
deeptools	Ramirez et al.73	–
MACS2	Zhang et al.74	–
ChromHMM	Ernst and Kellis47	v1.27
ChIPSeeker	Wang et al.75	–
coverage_over_states	This study	github.com/annamalkowska/Malkowska-et-al
feat_annot	This study	github.com/annamalkowska/Malkowska-et-al
Bedtools	Quinlan and Hall76	2.31.1
BEDOPS	Neph et al.77	v2.4.41
DiffBind	Stark and Brown78	3.21
PANTHER	Thomas et al.79	18.0
Seurat	Hao et al.80	5.3.0
Gmisc	–	3.0.4
tidyverse	–	2.0.0
pheatmap	–	1.0.13
rstatix	–	0.7.2
Adobe Illustrator	Adobe	27.5
Homer	Heinz et al.81	5.1
R	R-project	4.5.0
RStudio	RStudio	–
Other
Illumina HiSeq 1500	Illumina	–
Illumina NovaSeq 6000	Illumina	–

Experimental Model And Study Participant Details

The following fly stocks were used to drive expression of transgenes: worniu-GAL4,⁶⁸ tub-GAL80^ts (BDSC 7019), UAS-LT3-NDam and UAS-LT3-NDam-RNAPol II³², UAS-LT3-Dam-Pc, UAS-LT3-Dam-HP1a, UAS-LT3-Dam-Brm, UAS-LT3-Dam-H1⁴⁰, UAS-LT3-Dam-TAF3, UAS-LT3-Dam-Dmnt3a, UAS-LT3-Dam-15F11, UAS-LT3-Dam-19E5,⁴³ UAS-mCherry-RNAi (BDSC 35785), UAS-trx-RNAi (BDSC 33703), UAS-Utx-RNAi (BDSC 31076), UAS-trr-RNAi (BDSC 36916), UAS-brm-RNAi (BDSC 34520).

Method Details

Targeted DamID experimental design

To perform Targeted DamID, wor-GAL4 or wor-GAL4, tubGAL80^ts, depending on the condition were crossed to flies carrying a Dam fusion. UAS-LT3-Dam was used as a control for all experiments, and three or more replicates were performed for each condition.

For the embryonic proliferating NSC (epNSC) condition, Dam fusion lines were crossed to wor-GAL4. Flies were reared in cages at 25°C and allowed to lay eggs on apple juice plates for 1 h. The plates were kept at 25°C for 12 h and 50 μL of embryo pellet per replicate was collected into microcentrifuge tubes and frozen at −70°C.

For the quiescent NSC (qNSC) condition, Dam fusion lines were crossed to wor-GAL4, tubGAL80^ts. Flies laid eggs on apple juice plates for 1 h and the plates were then transferred to 18°C for 28 h to prevent expression of Dam with the use of the temperaturesensitive GAL80 repressor, restricting the transgene expression to after embryonic stage 17. To induce expression of Dam, the plates were shifted to 29°C for 12 h. About 300 whole larvae per replicate were collected into microcentrifuge tubes and frozen at −70°C.

For the larval proliferating NSC (lpNSC) condition, Dam fusion lines were also crossed to wor-GAL4, tubGAL80^ts. Flies laid eggs on apple juice plates for 1 h and the plates were then transferred to 18°C for 56 h, restricting transgene expression to late first larval instar (equivalent to 24 h after larval hatching, ALH, at 25°C). The plates were then shifted to 29°C for 12 h. About 200 whole larvae per replicate were collected into microcentrifuge tubes and frozen at −70°C.

Targeted DamID library processing

Targeted DamID were processed according to a published DamID-seq protocol.⁷⁰ Briefly, genomic DNA was extracted with the QIAamp DNA Micro Kit and digested with DpnI and DpnII to isolated methylated DNA. The fragments are PCR amplified and sonicated before generating a sequencing library and multiplexing with a homebrew Truseq kit. Sequencing was performed as single end 50 or 100 bp reads generated by an Illumina HiSeq 1500 or Illumina NovaSeq 6000, dependent on the dataset, at the Gurdon Institute NGS Core Facility.

RNAi experiments

Worniu-GAL4 virgins were crossed to given UAS-expressing males and reared at 25°C. Flies laid eggs on apple juice plates for 3 h at 25°C, after which the plates were moved to 29°C until hatching. Hatched larvae were transferred to food plates and placed again at 29°C for either 17 h (equivalent to 20h ALH at 25°C) or 20 h (equivalent to 24 h ALH at 25°C) before dissection. Dissected larval brains were fixed for 20 min in 4% PFA and washed 3 times with PBS +0.3% Triton (PBST) for 15 min. The samples were then incubated in primary antibodies solution at room temperature overnight. The following primary antibodies were used: guinea pig anti-Deadpan (Andrea Brand,⁶⁷ 1:5000), rat anti-Worniu antibody [5A3AD2] (ab196362, 1:100) and rabbit anti-phosphohistone H3, pSer10 (Millipore 06–570, 1:200). Samples were washed 3 times for 15 min in PBST before incubation with secondary antibodies overnight at 4°C. The following secondaries were used: Alexa Fluor 405, 488, 568, 633 (Life Technologies). Larval brains were mounted in Vectashield (Vector Laboratories) and imaged on an SP8 Leica confocal microscope.

Quantification And Statistical Analysis

Targeted DamID data processing

Fastq files were analyzed with damidseq_pipeline (vR.1),⁶⁹ using the damMer suite for parallelization and pairwise comparisons of DamID replicates.⁷⁰ Briefly, damidseq_pipeline was used to align fastq files with bowtie2⁷¹ and extend the reads toward 300bp or the first GATC site. Next, the bam files were used as input to generate normalized binding tracks in bedgraph format. DamMer was used to parallelize each Dam-Only to Dam-fusion comparison and perform quantile normalization between comparisons and averaging. The variability of the extended *.bam files was visualized using Pearson correlation performed using deeptools multiBam-Summary and plotHeatmap -pearson commands. Extended reads were also used to perform MACS2 broad peak calling for all pairwise comparisons between Dam-fusion and Dam-only replicates. Peaks were thresholded based on FDR < 10^-5, merged and filtered based on appearance in more than 50% replicates. Peaks were also called on Dam-only replicates alone, serving as a chromatin accessibility readout.

Generating hidden Markov models

Peak files were used as input to generate hidden Markov model using ChromHMM.⁴⁷ Prior to learning the final chromatin state model, we have performed extensive comparison of hidden Markov models dependent on levels of pre-processing prior to learning the model and evaluated them based on signal-to-noise ratio and putative biological identities of generated states. We have found that model based on reproducible and thresholded peaks was characterized by lower signal-to-noise ratio and more sharply defined states at the expense of the percentage of genome that is annotated. This results in the presence of an ‘Unannotated’ state with no signal as a result of biological limitations (i.e., dependent on how many chromatin proteins were profiled) and technical limitations (due to thresholding peaks based on significance, binning and binarization).

Reproducible peak files were binarized with BinarizeBed with bin size of 300bp and the ‘peaks’ option in a concatenated format (all NSCs condition used as input to learn one model together). Learn Model command was executed for each condition with a differing number of states between 5 and 12. The resulting emission probabilities of 5-12-state models were compared and the 8-state models was selected as the most relevant for further analyses.

Examination of chromatin states

Feature annotation of chromatin states bins was performed with ChiPSeeker.⁷⁵ Visualization of emission and transition probabilities derived from ChromHMM and chromatin state frequency was performed using custom R scripts. Dam-Only read coverage per chromatin state was calculated with a custom script (coverage_over.sh), which used bedtools coverage to calculate read count per million (CPM) per state and visualized it using pheatmap R package.

Chromatin state annotation

A custom script was created (feat_annot.sh) to annotate genomic features (genes and TSS regions) with the chromatin state that is most prevalent within that DNA region. Briefly, genome annotation files generated by ChromHMM were split into separate files based on the chromatin state. Bedmap was used to calculate the base percentage of a chromatin state at given feature. For the TSS regions, a region 500bp upstream and 200bp downstream was used as reference. Chromatin state with the highest base percentage was treated as the state annotation for given feature. The ‘unannotated’ state was thresholded at 75% i.e., the ‘unannotated’ state was only called when a feature had a base percentage higher than 75%, otherwise the second most prevalent state was called. These annotations were used to plot chromatin state frequency at the TSS and to perform all alluvial plots, which were generated with Transition-class from the ‘Gmisc’ R package.

Differential binding analysis

For comparison of chromatin accessibility between NSC conditions, the DiffBind R package was used.⁷⁸ Broadpeak and extended read files of the Dam-only dataset were used as input for differential accessibility analysis. For differential binding of H3K4me3 peaks, TAF3 broadpeaks and bam files were used as input, with the Dam-only bam files as control. Default settings were used for majority of commands, apart from dba.count, where bSubControl = TRUE, minOverlap = 1, fragmentSize = 0, summits = TRUE.

Gene Ontology

Gene Ontology was performed with PANTHER18.0⁷⁹ as a statistical overrepresentation test (Fisher’s Exact text with False Discovery Rate correction) of the GO biological process complete set, using whole Drosophila gene set as reference. The database was accessed used rbioapi package in R and the results were visualized with custom R scripts.

Single cell RNA-sequencing dataset analysis

A previously published scRNA-seq dataset of NSCs at embryonic stage 17 and 24h ALH⁵³ was analyzed using the Seurat suite (v4).⁸² Cells in the two datasets were filtered based on RNA count and mitcQC, merged and sctransform normalization was applied. The two datasets were integrated and clustered. NSC-specific cluster that expressed dpn and other NSC-markers was subsetted and the dataset was normalized and scaled within that cluster. Gene group expression score was calculated using UCell.⁵⁴

Motif analysis

Putative promoter regions of the ‘Repressive state 2’ to ‘Active state 2’ gene group were defined as 500 bp upstream and 200 bp downstream of the TSS and recovered from the dm6 genome. Motif discovery and comparison to known motif databases was performed with Homer (v5.1) using the findMotifsGenome.pl command with the -size given option.

Image quantification

Quantification of cell numbers based on confocal images of the RNAi experiments was performed in Fiji (v2.17.0) using CellCounter. Two-tailed Student’s t test was performed for each comparison using R package rstatix (v0.7.2). Statistical significance is indicated as follows: ns is non-significant, * is < 0.05, ** is < 0.01, *** is < 0.001, **** is < 0.0001.

Data visualization

All graphs were generated using ggplot2 (v3.4.2) or pheatmap (v1.0.13) unless stated otherwise. Average signal tracks of TAF3 (Figure 5D) were generated using SeqPlots (v1.4.1). All genomic tracks were created with pyGenomeTracks (v3.9). Figures were assembled using Adobe Illustrator (v27.5).

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116732.

Highlights.

• Chromatin becomes accessible during neural stem cell quiescence

• Chromatin modeling shows that quiescent cells gain active domains to maintain plasticity

• Permissive chromatin does not correlate with active transcription

• Cell-type-specific profiling of chromatin factors in vivo with Targeted DamID

In brief.

Malkowska et al. comprehensively mapped chromatin-binding proteins in quiescent and proliferating neural stem cells in vivo. Quiescent neural stem cells maintain plasticity by gaining accessible chromatin.

Data and code availability

• Targeted DamID datasets, including chromatin-state genome annotations as *_model.bed files, have been deposited at ArrayExpress under accession number E-MTAB-15696. scRNA-seq data have been previously reported in Gherghina et al.⁵³

• All original code is publicly available as of the date of publication at GitHub: github.com/annamalkowska/Malkowska-et-al.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.