Author response: Activation of a neural stem cell transcriptional program in parenchymal astrocytes Article Swipe

Article Figures and data Abstract eLife digest Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Adult neural stem cells, located in discrete brain regions, generate new neurons throughout life. These stem cells are specialized astrocytes, but astrocytes in other brain regions do not generate neurons under physiological conditions. After stroke, however, striatal astrocytes undergo neurogenesis in mice, triggered by decreased Notch signaling. We used single-cell RNA sequencing to characterize neurogenesis by Notch-depleted striatal astrocytes in vivo. Striatal astrocytes were located upstream of neural stem cells in the neuronal lineage. As astrocytes initiated neurogenesis, they became transcriptionally very similar to subventricular zone stem cells, progressing through a near-identical neurogenic program. Surprisingly, in the non-neurogenic cortex, Notch-depleted astrocytes also initiated neurogenesis. Yet, these cortical astrocytes, and many striatal ones, stalled before entering transit-amplifying divisions. Infusion of epidermal growth factor enabled stalled striatal astrocytes to resume neurogenesis. We conclude that parenchymal astrocytes are latent neural stem cells and that targeted interventions can guide them through their neuronal differentiation. eLife digest Regenerative medicine aims to help the body replace damaged or worn-out tissues, often by kick-starting its own intrinsic repair mechanisms. However, the brain cannot easily repair itself, and therefore poses a much greater challenge. This is because nerve cells or neurons, which underpin learning, memory, and many other abilities, are also the brain's greatest weakness when it comes to tissue repair. In most parts of the adult brain, neurons are never replaced after they die. This means that damage to brain tissue – for example, after a stroke – can have severe and long-lasting consequences. Neural stem cells are one type of brain cell that can turn into new neurons if needed, but they are only found in a few parts of the brain and cannot fix damage elsewhere. More recent work in mice has shown that astrocytes, a common type of support cell in the brain that help keep neurons healthy, could also generate new neurons following a stroke. However, the ability was restricted to small numbers of astrocytes in a specific part of the brain. Here, Magnusson et al. set out to determine the molecular mechanisms behind this regenerative process and why it is unique to certain astrocytes. The researchers used a technique called single-cell RNA sequencing to analyze the genetic activity within individual mouse astrocytes that had been exposed to conditions mimicking a stroke. This method revealed which genes are switched on or off, thus generating a profile of gene activity for each astrocyte analyzed. This experiment showed that the profiles of astrocytes that had started to produce neurons were in fact nearly identical to neural stem cells. Even the astrocytes that could not generate neurons took the first few steps towards this genetic state; however, they stalled early in the process. Treating the brains of mice withepidermal growth factor, a powerful molecular signal that stimulates cell growth, kick-started nerve cell production in a subset of these cells – showing that at least some of the non-regenerative astrocytes could be stimulated to make neurons if given the right treatment. The results of this study shed new light on how some astrocytes in the brain gain the ability to form new neurons. In the future, this knowledge could help identify a source of replacement cells to regenerate the injured brain. Introduction Neurogenesis is extremely limited in the adult brain. In most mammals, specialized astrocytes in the subventricular zone (SVZ) and hippocampal dentate gyrus (DG) are stem cells and generate new neurons continuously, but apart from that, the brain's ability to replace lost neurons is very limited. However, in response to an experimental stroke or excitotoxic lesion, some astrocytes in the mouse striatum can generate neurons (Magnusson et al., 2014; Nato et al., 2015). This neurogenic response is regulated by Notch signaling and can be activated even in the uninjured mouse striatum by deleting the Notch-mediating transcription factor Rbpj (Magnusson et al., 2014). Striatal astrocytes undergo neurogenesis by passing through a transit-amplifying cell stage. But it is not known whether these astrocytes become bona fide neural stem cells. If they do, this could have far-reaching implications for regenerative medicine. Astrocytes make up a large fraction of all brain cells (10–20% in mice) (Sun et al., 2017) and are distributed throughout the central nervous system. They would thus represent a very abundant source of potential neural stem cells that might be recruited for therapeutic purposes. Although certain injuries and Rbpj deletion can both trigger neurogenesis by astrocytes, it almost exclusively does so in the striatum. And even within the striatum, primarily the astrocytes in the medial striatum readily activate neurogenic properties (Figure 1a). This suggests that neurogenic parenchymal astrocytes either occupy an environmental niche favorable to neurogenesis or that only they have an inherent neurogenic capacity. In order to recruit astrocytes for therapeutic neurogenesis, a first step is to understand the mechanisms underlying this process. If these mechanisms are understood, they could potentially be targeted to induce localized therapeutic neurogenesis throughout the central nervous system. Figure 1 with 1 supplement see all Download asset Open asset Neurogenesis by striatal astrocytes can be reconstructed using single-cell RNA sequencing. (a) Deletion of the gene encoding the Notch-mediating transcription factor Rbpj activates a latent neurogenic program in striatal astrocytes (Magnusson et al., 2014). Nuclei of Dcx+ neuroblasts are indicated by red dots. Not all striatal astrocytes undergo neurogenesis, shown by the restricted distribution of Dcx+ neuroblasts and the fact that many recombined astrocytes (gray) remain even 2 months after Rbpj deletion. (b) We performed single-cell RNA sequencing using two protocols. For the AAV-Cre dataset, we deleted Rbpj exclusively in striatal astrocytes using a Cre-expressing AAV and sequenced the transcriptomes of recombined cells five weeks later. (c) Dimensionality reduction using UMAP captures the progression from astrocytes, through proliferating transit-amplifying cells, to neuroblasts. Panel (d) shows markers for the different maturation stages. Here, we generated two separate single-cell RNA sequencing datasets to study neurogenesis by parenchymal astrocytes in mice. We found that, at the transcriptional level, Rbpj-deficient striatal astrocytes became highly similar to SVZ neural stem cells and that their neurogenic program was nearly identical to that of stem cells in the SVZ, but not the DG. Interestingly, astrocytes in the non-neurogenic somatosensory cortex also initiated a neurogenic program in response to Rbpj deletion, but all stalled prior to entering transit-amplifying divisions and failed to generate neuroblasts. In the striatum, too, many astrocytes halted their development prior to entering transit-amplifying divisions. We found that stalled striatal astrocytes could be pushed into transit-amplifying divisions and neurogenesis by an injection of epidermal growth factor (EGF), indicating that it is possible to overcome roadblocks in the astrocyte neurogenic program through targeted manipulations. Taken together, we conclude that parenchymal astrocytes are latent neural stem cells. We posit that their intrinsic neurogenic potential is limited by a non-permissive environment. Recruiting these very abundant latent stem cells for localized therapeutic neurogenesis may be possible but is likely to require precise interventions that guide them through their neurogenic program. Results Transcriptome-based reconstruction of neurogenesis by striatal astrocytes To understand the cellular mechanisms underlying neurogenesis by parenchymal astrocytes, we decided to perform single-cell RNA sequencing of striatal astrocytes undergoing neurogenesis in vivo. To this end, we generated two separate single-cell RNA sequencing datasets, each with complementary strengths. For the first dataset, which we call the Cx30-CreER dataset, we used Connexin-30 (Cx30; official gene symbol Gjb6)-CreER mice to delete Rbpj and activate heritable tdTomato or YFP expression in astrocytes throughout the brain. This method targeted up to 100% of astrocytes in the striatum (tdTomato expression in 92 ± 10% of glutamine synthetase+, S100+ cells; mean ± S.D.; n = 6 mice). These mice also target neural stem cells in the SVZ, but no other brain cell types (Magnusson et al., 2014). Rbpj deletion in healthy mice was chosen as the means by which to activate the neurogenic program of astrocytes. That is because (1) it constitutes a single, precisely timed trigger throughout the brain, (2) it mimics the endogenous stimulus of reduced Notch signaling by which stroke induces neurogenesis by astrocytes (Magnusson et al., 2014), and (3) it does not induce potentially confounding cellular processes like reactive gliosis or cell death. Striatal tdTomato+ cells were isolated by flow cytometry from mice homozygous for the Rbpj null allele 2, 4 and 8 weeks after tamoxifen administration, and from control mice with intact Rbpj, 3 days after tamoxifen administration (hereafter called ground-state astrocytes) (Figure 2—figure supplement 1a–d). Sequencing libraries were prepared using Smart-seq2 (Picelli et al., 2013). In addition to parenchymal astrocytes, Cx30-mediated recombination also targets SVZ neural stem cells, whose progeny might be sorted along with striatal astrocytes and, thus, confound our results. We therefore generated a second single-cell RNA sequencing dataset, for which labeling of cells was restricted to the striatum. Striatal astrocytes were selectively recombined via injection of an adeno-associated virus (AAV) expressing Cre under the control of the astrocyte-specific GFAP promoter into the striatum of Rbpjfl/fl; R26-loxP-Stop-loxP-tdTomato mice (Figure 1b). We isolated tdTomato+ cells 5 weeks after virus injection, a time point at which Ascl1+ astrocytes, transit-amplifying cells and neuroblasts are all present (Magnusson et al., 2014). We sequenced these cells using 10X Chromium by 10X Genomics. Despite the GFAP promoter, the AAV-Cre dataset also contained oligodendrocytes and some microglia (Figure 1—figure supplement 1). These were excluded from the final dataset based on our previous finding that they are not involved in a neurogenic program (Magnusson et al., 2014). The AAV-Cre dataset contained a somewhat higher proportion of transit-amplifying cells than the Cx30-CreER dataset (Figure 2—figure supplement 1f), likely because the AAV-Cre dataset only contained cells from 5 weeks after Rbpj-KO, near the peak of transit-amplifying divisions (Magnusson et al., 2014). We first asked whether we could reconstruct striatal astrocyte neurogenesis computationally using the AAV-Cre dataset. Using Uniform Manifold Approximation and Projection (UMAP), we found that the sequenced cells indeed included Rbpj-deficient astrocytes (Cx30+), activated astrocytes and transit-amplifying cells (Egfr+, Ascl1+, Mki67+, Dcx+) and neuroblasts (Dcx+), up until the migratory neuroblast stage (Nav3+) (Figure 1c–d). We conclude that it is possible to use single-cell RNA sequencing to study the transcriptional mechanisms underlying the astrocyte neurogenic program. Neurogenesis by astrocytes is dominated by early transcriptional changes associated with metabolism and gene expression We were interested in the transcriptional changes that take place as astrocytes first initiate the neurogenic program. For this question, we primarily used the Cx30-CreER dataset, which included astrocytes both at ground state and at 2, 4 and 8 weeks after Rbpj deletion (Figure 2a). Furthermore, the Cx30-CreER mice allowed us to study the neurogenic process in the uninjured brain, without the potentially confounding effects of a needle injection. Two weeks after Rbpj deletion, astrocytes had upregulated genes mostly related to transcription, translation and metabolism (Figure 2b, S1g, Supplementary file 1). To study in detail how gene expression changed over the course of neurogenesis, we next reconstructed the differentiation trajectory of these astrocytes computationally using Monocle (Trapnell and Cacchiarelli, 2014; Figure 2—figure supplement 2; Methods). This analysis captured the progression from astrocyte to neuroblast in pseudotime (Figure 2c). The pseudotemporal axis reflected neurogenesis, as revealed by plotting along pseudotime the expression of genes that are dynamically induced in canonical neurogenesis (Figure 2d). This analysis confirmed the previous finding that neurogenesis by parenchymal astrocytes involves transit-amplifying divisions (Magnusson et al., 2014; Niu et al., 2013), and does not occur through direct transdifferentiation, a process during which intermediate stages of differentiation are skipped (Briggs et al., 2017). Figure 2 with 3 supplements see all Download asset Open asset Neurogenesis by astrocytes is dominated by early transcriptional changes associated with metabolism and gene expression. (a) We prepared a second single-cell RNA sequencing dataset using SmartSeq2. This dataset consists of cells isolated from Cx30-CreER; Rbpjfl/fl mice at 2, 4 and 8 weeks after Rbpj deletion, and ground-state astrocytes with intact Rbpj. (b) Two weeks after Rbpj deletion, astrocytes have upregulated genes mainly associated with metabolism and gene expression. (c) Dimensionality reduction using Monocle organizes astrocytes and their progeny along pseudotime (see Figure 2—figure supplement 2 for cell state definitions). (d) Pseudotime captures the neurogenic trajectory of astrocytes, as shown by plotting the expression of classical neurogenesis genes along this axis. (e) Gene clustering and GO analysis shows distinctive dynamics of functional gene groups with example genes shown in (f). Genes involved in lipid and carbohydrate metabolism are upregulated initially but downregulated as cells enter transit-amplifying divisions. Genes involved in transcription and translation are expressed at steadily increasing levels. Accordingly, the number of genes detected per cell increases initially (g) – an increase not explained by a mere increase in cell size (h). A gene cluster with remarkable dynamics contains genes involved in synapse organization (e), presumably important for both astrocytes and neurons. (i) All these findings are supported by a pseudotemporal analysis of the AAV-Cre dataset. The AAV-Cre dataset, however, only contains cells from the 5-week time point and thus does not show changes that happen early in the neurogenic program. (TA cells, transit-amplifying cells). In order to better understand the transcriptional changes that take place as striatal astrocytes initiate neurogenesis, we used Monocle's gene clustering algorithm, which clusters genes whose expression levels vary similarly along pseudotime (Figure 2e). These gene clusters were then characterized using the gene ontology (GO) tool DAVID (Huang et al., 2009). This approach enabled us to see which biological processes changed as differentiation proceeded (Supplementary file 2). We found that, as astrocytes initiated their neurogenic program, genes associated with metabolism, particularly lipid (e.g. Fabp5, Fasn, Elovl5) and carbohydrate metabolism (e.g. Gapdh, Aldoa, Aldoc) were at first upregulated but dramatically downregulated as cells entered transit-amplifying divisions (Figure 2e–f, [Cluster 1]). This suggested that the neurogenic program requires high metabolic activity to prepare for transit-amplifying divisions. In addition to these metabolic genes, the expression of genes involved in the regulation of transcription (e.g. mRNA splicing genes Snrpb, Srsf5, Srsf3) and translation (e.g. ribosomal subunits Rpl17, Rpl23, Rps11) progressively increased throughout the neurogenic process (Figure 2e–f, [Cluster 2]). Accordingly, we found that the number of genes detected per cell increased in the initial stages of neurogenesis (Figure 2g): pre-division astrocytes expressed 1.4 times as many genes as ground-state astrocytes (p=10−12; 95% C.I. 1.3–1.5; Materials and methods). In theory, such an increase in the number of detected genes could be an artifact caused by a simultaneous increase in cell size; however, no cell size increase was observed during this early phase of neurogenesis, as measured by flow cytometer forward scatter (Figure 2h). The changes to metabolism and gene expression activity described above were the two dominating transcriptional changes that occurred early in the neurogenic process of astrocytes. At the peak of metabolic gene expression, astrocytes initiated transit-amplifying divisions (Figure 2e; Materials and methods). Accordingly, they upregulated genes associated with cell division (e.g. Ccnd2, Ccng2, Cdk4; Figure 2e–f, [Cluster 3]). Finally, as they exited transit-amplifying divisions, cells upregulated classical marker genes of neuroblasts (e.g. Cd24a, Dcx, Dlx1/2, Tubb3 [βIII tubulin]; Figure 2e–f, [Cluster 3]), indicative of early neuronal commitment and differentiation. Interestingly, one group of genes showed a remarkable pattern of initial downregulation followed by upregulation as cells developed into neuroblasts (Figure 2e–f [Cluster 4], S4a). This gene group consisted mainly of genes encoding proteins localized to synapses (e.g. Cplx2, Ctbp2, Shank1, Chrm4, Bdnf). This suggested that synapse maintenance was compromised very early as astrocytes initiated neurogenesis, but also that some of the same genes are important for synapse maintenance in both astrocytes and neurons. The gene expression changes described above were confirmed in the AAV-Cre dataset, using a differential expression analysis along pseudotime (Figure 2i). But because the AAV-Cre dataset only contained cells isolated 5 weeks after Rbpj deletion, this dataset did not contain ground-state astrocytes and could not capture changes that happened early in the neurogenic program. Astrocytes gradually lose astrocyte features as they enter the neurogenic program Throughout the initial phase of neurogenesis and up until the point where cells entered transit-amplifying divisions, astrocytes maintained normal astrocyte morphology and expression of common marker genes (e.g. Aqp4, S100b, Slc1a2 [Glt1], Glul) (Figure 2—figure supplement 3b–c, Video 1). They never showed signs of reactive gliosis (e.g. morphological changes or upregulation of Gfap), indicating that reactive gliosis is not required for neurogenesis by astrocytes in vivo. As the astrocytes entered the neurogenic program, they did eventually lose their typical branched morphology; however, they did so only gradually, and only after entering transit-amplifying divisions. In fact, they retained rudiments of astrocytic processes for several transit-amplifying cell divisions (Figure 2—figure supplement 3c, Video 1). As an additional sign that they originated from parenchymal astrocytes, many clustered transit-amplifying cells retained trace levels of the astrocyte marker protein S100β, even though they no longer expressed the S100b gene. S100β protein is found in parenchymal astrocytes but not in SVZ stem cells (Figure 2—figure supplement 3d–g). Accordingly, we did not detect any S100β protein in SVZ transit-amplifying cells (Figure 2—figure supplement 3d–g). This suggests that the low levels of S100β protein seen in striatal transit-amplifying cells were lingering remnants of their parenchymal astrocyte origin. It indicates that trace levels of S100β protein can function as a short-term lineage-tracing marker for transit-amplifying cells derived from parenchymal astrocytes. Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Gradual loss of astrocyte morphology in astrocyte-derived transit-amplifying cells. Striatal astrocytes retain their branched astrocyte morphology (see YFP signal) even after they have initiated transit-amplifying divisions in vivo. Shown are several examples of different Ascl1+ astrocytes and astrocyte-derived transit-amplifying cells at various stages of transit-amplifying divisions. EdU indicates that these cells had divided within the 2 weeks leading up to the analysis (Materials and methods). Note that astrocyte processes gradually become less branched and more retracted the longer transit-amplifying divisions proceed. A recent study in Drosophila melanogaster showed that many quiescent neural stem cells are halted in the G2 phase of the cell cycle (Otsuki and Brand, 2018). However, we analyzed EdU incorporation in dividing striatal astrocytes. We found some single astrocytes that had not yet divided but were EdU+ (Figure 2—figure supplement 3c). This suggested that these astrocytes initiated the cell cycle by synthesizing DNA and thus had not been suspended in the G2 phase. Parenchymal astrocytes are located upstream of a neural stem cell state and acquire a transcriptional profile of neural stem cells upon their activation We were interested in a direct comparison between neurogenic parenchymal astrocytes and adult neural stem cells to see how their neurogenic transcriptional programs relate to one another. A number of published single-cell RNA sequencing datasets exist for mouse neural stem cells generated by 10X Chromium or Smart-seq2. We first analyzed our AAV-Cre dataset together with a previously published 10X Chromium dataset for SVZ neural stem cells and their progeny (Zywitza et al., 2018). A UMAP analysis of these cells together with our AAV-Cre dataset revealed that the striatal astrocytes and their progeny clustered closely together with the corresponding stages of SVZ cells (Figure 3a–b). This indicated that neurogenesis in these two brain regions proceeds through the same general developmental stages. Particularly interesting was the observation that our 'Astrocyte Cluster 2' grouped closely together with SVZ neural stem cells (Figure 3b), indicating that these cells are transcriptionally similar. Figure 3 with 1 supplement see all Download asset Open asset Striatal astrocytes are upstream of SVZ neural stem cells in the neurogenic lineage. (a) We analyzed our AAV-Cre dataset together with a previously published dataset of SVZ cells. (b) All maturational stages along the striatal neurogenic trajectory cluster together with the corresponding stages in the SVZ. The AAV-Cre dataset has few ground-state astrocytes because all cells were isolated five weeks after Rbpj deletion. (c) We therefore compared our Cx30-CreER dataset with a previously published dataset of SVZ cells using Monocle's pseudotime reconstruction (d). This analysis suggests that striatal astrocytes are located upstream of SVZ stem cells in the neurogenic lineage, as if representing highly quiescent neural stem cells that, upon Rbpj deletion, initiate a neurogenic program very similar to that of the SVZ stem cells. (e) the expression pattern of genes Figure shows that striatal astrocytes initiate neurogenesis with a phase of metabolic gene SVZ stem cells on this metabolic a for differentiation increases in Rbpj-deficient astrocytes to the same levels as in SVZ stem cells to entering transit-amplifying divisions, striatal astrocytes are in their most stem upregulated many markers of neural stem cells is one Note the expression pattern of many of these (i) classical however, are not our AAV-Cre dataset only contained cells isolated 5 weeks after Rbpj deletion, this dataset could not how striatal astrocytes in their ground state relate to SVZ stem cells. We therefore next to our Cx30-CreER dataset and analyzed these cells together with SVZ cells from previously published Smart-seq2 dataset et al., Figure 3c). We used Monocle to perform pseudotemporal of our Cx30-CreER dataset together with the SVZ cells, using genes involved in the GO Neurogenesis and the cells the pseudotime axis (Figure Interestingly, on this neurogenic ground state astrocytes were located upstream of the SVZ stem cells, as if representing highly quiescent cells that through an initial activation phase before the SVZ stem cells (Figure We next compared the gene expression changes that occur in SVZ cells with in our Cx30-CreER dataset. We used Monocle's gene clustering analysis of only the SVZ cells, without the striatal cells. We first confirmed the previously published finding that SVZ neurogenesis is by changes in genes associated with metabolism (Figure supplement et al., 2015). however, SVZ neural stem cells to be maintained at a high of metabolic gene expression (Figure supplement striatal astrocytes first through a phase of metabolic gene upregulation to this level, seen by the (Figure and (Figure supplement metabolic gene (Figure This suggested that the neurogenic program in striatal astrocytes with an activation phase during which these astrocytes to SVZ stem cells with to for an initial activation phase was found also in gene expression As described we detected a in the number of genes detected in Rbpj-deficient astrocytes compared to ground-state astrocytes. This is similar in to a previously detected between adult SVZ stem cells and striatal astrocytes et al., This suggested that the gene expression activity of Rbpj-deficient astrocytes to that of neural stem cells. We asked if we could use our RNA sequencing data to whether the differentiation of striatal astrocytes to that of SVZ stem cells. is a of the of a It the number of possible by all proteins expressed by a It is a for a differentiation and and the underlying that a highly which in turn to a of differentiation As a of the signaling we signaling pseudotime in the SVZ dataset and that SVZ cells their signaling as activated stem cells or transit-amplifying cells (Figure supplement We found that the signaling of striatal astrocytes increased after Rbpj deletion (Figure to the levels seen in SVZ stem cells (Figure = ± pre-division astrocytes ± mean pre-division astrocytes pre-division SVZ ± This supported the that striatal astrocytes the same of differentiation as SVZ neural stem cells. before entering transit-amplifying divisions, astrocytes were in their most neural stem In addition to the described striatal astrocytes in this pre-division state had upregulated some marker genes of neural stem cells (e.g. Figure and Figure such markers were expressed by ground-state astrocytes and their RNA levels did not after Rbpj deletion (e.g. Figure supplement Interestingly, some classical neural stem cell markers were never upregulated by Rbpj-deficient astrocytes. For example, and at low levels throughout the neurogenic process (Figure showing that neurogenic astrocytes do not become transcriptionally identical to SVZ stem cells. Supplementary file 1 a of genes expressed between ground-state astrocytes and in their stem pre-division The neurogenic program of striatal astrocytes is more similar to that of SVZ stem cells than to that of stem cells In the adult brain, neural stem cells exist in two regions, the SVZ and the and produce neuronal We asked whether the neurogenic program of striatal astrocytes is more similar to that of stem cells in the SVZ or the DG. For we our AAV-Cre dataset with the SVZ dataset from et al., and previously published 10X Chromium dataset of neurogenesis et al., Figure We used our AAV-Cre dataset than the Cx30-CreER dataset for this