Stochastic cell-cycle entry and cell-state-dependent fate outputs of injury-reactivated tectal radial glia in zebrafish

Gliosis defined as reactive changes of resident glia is the primary response of the central nervous system (CNS) to trauma. The proliferation and fate controls of injury-reactivated glia are essential but remain largely unexplored. In zebrafish optic tectum, we found that stab injury drove a subset of radial glia (RG) into the cell cycle, and surprisingly, proliferative RG responding to sequential injuries of the same site were distinct but overlapping, which was in agreement with stochastic cell-cycle entry. Single-cell RNA sequencing analysis and functional assays further revealed the involvement of Notch/Delta lateral inhibition in this stochastic cell-cycle entry. Furthermore, the long-term clonal analysis showed that proliferative RG were largely gliogenic. Notch inhibition of reactive RG, not dormant and proliferative RG, resulted in an increased production of neurons, which were short-lived. Our findings gain new insights into the proliferation and fate controls of injury-reactivated CNS glia in zebrafish.


Introduction
Traumatic brain injury (TBI) is one clinically principal type of central nervous system insults (Burda and Sofroniew, 2014). Gliosis defined as reactive changes of resident macroglia (e.g., mammalian astrocytes) is a primary CNS response to TBI in mammals (Barres, 2008;Burda and Sofroniew, 2014). In mammals, gliosis undergoes three significant stages: Glial cells initially become reactive, hypertrophic, and inflammatory, with characteristic upregulation of GFAP and vimentin (Liddelow and Barres, 2017;Zamanian et al., 2012); subsequently, a subset of reactive glia reenter the cell cycle and become proliferative (Gallo and Deneen, 2014); finally, proliferative glia undergo gliogenesis, the process of glial cell production, and form structures known as glial scars (Burda and Sofroniew, 2014). Earlier studies have demonstrated both protective and detrimental roles of the gliosis in the injured CNS (Faulkner et al., 2004;Li et al., 2008;Silver and Miller, 2004;Sofroniew and Vinters, 2010;Wanner et al., 2013). For instance, the blockage of initial glia reactivation worsened the injury (Faulkner et al., 2004;Li et al., 2008;Wanner et al., 2013), whereas glia scars hindered neuronal regeneration (Silver and Miller, 2004). In the process, the proliferation and fate controls of injury-reactivated RG are essential but remain elusive in vivo.
Molecular mechanisms underlying the proliferation and fate controls of injury-reactivated RG have been examined in zebrafish for many years (Dias et al., 2012;Goldman, 2014). As to the proliferation control, Notch signaling is involved but is somehow context-dependent. For instance, in the zebrafish spinal cord, glial cells have low levels of Notch activity when they are in the dormant state, and enter the cell cycle by increased Notch activity after the injury (Dias et al., 2012). In contrast, dormant RG of zebrafish telencephalon exhibit high Notch activity and become proliferative by a rapid decrease of Notch activity (Chapouton et al., 2010). Notch signaling has also been reported to regulate fate outputs of reactivated MG in the injured zebrafish retina, that is, Notch inhibition leads to gliogenesis, whereas Notch over-activation results in the production of photoreceptor cells (Wan et al., 2012).
Zebrafish optic tectum, the higher sensory integration center, possesses a large population of RG (Galant et al., 2016;Ito et al., 2010). Unlike other brain regions where RG present as both dormant and proliferative forms at physiological conditions, tectal RG has been reported to be dormant, and are reactivated by injury to give rise to newborn neurons via Wnt signaling as well as Notch signaling Ueda et al., 2018). Interestingly, a recent study showed that tectal RG produced a significant number of glial cells (~25%) but not neurons (Lindsey et al., 2019). It is essential to resolve this inconsistency on the fate potentials of injury-reactivated tectal RG.
In this study, we set out to investigate the mechanism controlling injury responses of tectal RG in vivo. We found that stab injury drove a subset of tectal RG into the cell cycle. Surprisingly, proliferative tectal RG responding to the sequential injuries at the same injury site were distinct but overlapping. Quantitative analysis showed the probability of proliferative RG responding to both sequential stab injuries could be well explained by a model incorporating stochastic cell-cycle entry at the fixed probability of~25%. Single-cell RNA-seq and functional analysis revealed this stochastic cell-cycle entry was dependent on Notch/Delta lateral inhibition. The clonal analysis showed that proliferative tectal RG underwent gliogenesis. Interestingly, post-injury notch inhibition drove reactive RG into eLife digest The brain contains networks of cells known as neurons that rapidly relay information from one place to another. Other brain cells called glial cells perform several roles to support and protect the neurons including holding them in position and supplying them with oxygen and other nutrients.
Damage to the brain as a result of physical injuries is one of the leading causes of death and disability in people worldwide. Brain injuries generally stimulate glial cells to enter a "reactive" state to help repair the damage. However, some glial cells may start to divide and produce more glial cells instead, leading to scar-like structures in the brain that hinder the repair process.
To investigate why brain injuries trigger some glial cells to divide, Yu and He systematically examined glial cells in the part of the zebrafish brain that handles vision, known as the optic tectum. The experiments showed that a physical injury stimulated some of the glial cells to divide. Repeated injuries to the same part of the brain did not always stimulate the same glial cells to divide, suggesting that this process happens in random cells.
Further experiments revealed that molecules involved in a signaling pathway known as Notch signaling were released from some brain cells and inhibited neighboring glial cells from dividing to make new glial cells. Unexpectedly, inhibiting Notch signaling after a brain injury caused some of the glial cells that were in the reactive state to divide to produce neurons instead of glial cells.
Understanding how the brain responds to injury may help researchers develop new therapies that may benefit human patients in future. The next steps following on from this work will be to find out whether glial cells in humans and other mammals work in the same way as glial cells in zebrafish.  Figure 1. Injury reactivates dormant RG to proliferate and divide. (A-B 3 ) Tg(gfap:GFP) (green), GS (red) and PCNA (gray) immunofluorescences show that PCNA + proliferative cells (gray cells) are restricted to the TPZ (white arrow in (A)) and very few radial glia (RG) (1.4 ± 0.2%, n = 5, mean ± SEM, gray cells, white arrowheads in (B-B 3 )) is PCNA + . (B-B 3 ) The high-magnification images of the boxed area (white box) in (A). (C) Schematic representation of stab injury assay. A 30G needle is stabbed into the central-dorsal region of the right hemisphere of zebrafish optic tectum. The red asterisk and yellow arrowhead indicates the injury site. RG (green cells) at the bottom of PGZ underneath the injury site are analyzed. (D-G) Tg(gfap:GFP) (green) and PCNA (red) immunofluorescences show that injury induces the proliferation of RG (GFP + /PCNA + , yellow cells) underneath the injury site at 3 days postinjury (dpi). (F and G) The high-magnification images of boxed areas in (D) and (E), respectively. (H) The design of Cre-loxP transgenic fish lines used for clonal analysis of individual tectal RG. Fish expressing mCherryT2ACreER T2 controlled by the her4.1-promotor are crossed to red-to-green reporter fish controlled by the hsp70l promoter. In Tg(her4.1:mCherry-CreER T2 Âhsp70l:DsRed2(floxed)EGFP) double transgenic fish, EGFP expression is specifically induced in her4.1-expressing RG and their progeny by TAM applications and heat shocks. (I) Experimental time course of Cre-loxP-based clonal analysis experiments shown in (J-O 1 ). Double transgenic fish are administrated with TAM for three consecutive days (black dots) before the injury. EdU is injected to the injured fish to label the newborn cells for six consecutive days (red dots). Fish (21 to 24 dpi) are heat-shocked to induce EGFP expression in recombined cells and their progeny. (J-K 3 ) Representative RG-derived clone (EGFP + /EdU + , white arrows) underneath the injury site at 8 dpi. (K-K 3 ) The high-magnification images of the boxed area in (J). Two EGFP + /EdU + (white arrowheads) cells and an EGFP + radial process (open white arrowhead in (J)) are found underneath the injury site in this clone. (L-N 2 ) Representative 1 cell (L-L 2 ), 2 cells (M-M 2 ) and 3 cells clones (N-N 2 ) derived from single Figure 1 continued on next page Injury-reactivated tectal RG enter the cell cycle stochastically

gfap:GFP GS PCNA DAPI
To investigate the cell-cycle entry of tectal RG, we examined PCNA expression in tectal RG of Tg (1016tuba1a:GFP), a transgenic line used as the reporter for retinal MG reactivation after the injury (Fausett and Goldman, 2006). Under physiological conditions, weak GFP signals were present in tectal RG (Figure 2A and F). At 1 dpi, robust GFP signals were already observed together with the upregulation of Vimentin, a hallmark of glial reactivation at the early stage of gliosis ( . From 4 to 7 dpi, the number of PCNA + RG gradually dropped back to the same level observed before the injury ( Figure 2K and Figure 2-figure supplement 1C). At 3 and 5 dpi, we found some robust GFP signals and PCNA + cells at the injury site and in the region underneath ( Figure 2C and D).The GFP signals were likely due to hypertrophic responses of RG's processes and other cells at the injury site ( Figure 2D We further measured the spatial relation between reactive RG (with robust GFP + signal) and proliferative RG (GFP + /PCNA + ) ( Figure 2L). In coronal sections, reactive RG were primarily distributed in an area with a width of 186 ± 4 mm (n = 7, mean ± SEM) underneath of the injury sites ('Reactive Zone') while the majority of proliferative RG (88 ± 3%, n = 7, mean ± SEM) were located in an area with the width of 76 ± 5 mm (n = 7, mean ± SEM) underneath of the injury site ('Proliferative Zone') ( Figure 2L and M). The small variation of the width of both zones indicated the high reproducibility of stab injury outcomes ( Figure 2M).
Although the injury reactivated all RG underneath the injury site, only a subset of them (~25%, n = 8) became proliferative ( Figure 2H and Figure 2-figure supplement 1C). It raised an immediate question as to whether the proliferation of a subset of RG was due to stochastic cell-cycle entry or the presence of distinct RG subpopulations that respond differentially to the stab injury. To test this, we designed a sequential stab injury experiment. We examined the responses of reactive tectal RG to two sequential stab injuries performed at the same physical site ( Figure 2N). The first injury was introduced followed by BrdU pluses for six consecutive days to label proliferative RG responding to the first injury, and the second injury was introduced at 12 dpi followed by EdU pluses for six consecutive days to mark proliferative RG responding to the second injury ( Figure 2N). Finally, the fish were sacrificed, and coronal sections were stained for BrdU, EdU, radial glial marker BLBP, and neuronal marker HuC/D at 23 dpi ( Figure 2O-O 5 ). We found that although the number of proliferative RG induced by the first and the second injury showed no significant difference (the first injury: 84.4 ± 15.0 cells, n = 8; the second injury: 83 ± 9.4 cells, n = 8; mean ± SEM; p>0.05; Figure 2P), two sets of proliferative RG were distinct with some degree of overlapping (23 ± 6, n = 8, mean ± SEM; Figure 2O-O 5 and Figure 2P). More importantly, the proportion of overlapping RG (those reactivated after both injuries) was statistically indistinguishable from the multiplication of the reactivation probabilities of either injury, which suggested that individual reactive RG entered the   Figure 2-figure supplement 1. Figure 2 continued on next page cell cycle in the stochastic manner (prediction: 7.1 ± 1.9%, n = 8; experiment: 6.8 ± 1.7%, n = 8; mean ± SEM; p>0.05; Figure 2Q).
Single-cell RNA-seq analysis reveals cellular states representing RG reactivation and proliferation To further examine the molecular mechanism underlying this stochastic cell-cycle entry of injury-reactivated RG, we carried out single-cell RNA sequencing (scRNA-seq) analysis of tectal RG at 3 dpi, at which stage the number of proliferative tectal RG nearly reached the plateau in terms of cell number ( Figure 2K and Figure 2-figure supplement 1C). We dissected and dissociated the optic tecta of Tg(gfap:GFP) fish at 3 dpi and sorted out GFP + RG using fluorescence-activated cell sorting (FACS) for further scRNA-seq on the 10x Genomics platform ( Figure 3A and The remaining 1174 cells exhibited radial glial characteristics and were thus used for further analysis. They were segregated into five major cell clusters using t-SNE analysis ( Figure 3B). Each cell cluster had a characteristic gene expression ( Figure 3C). Cluster 1 cells (RG of dormant state, dRG) constituted the most abundant cell population with the high expression of milk-fat globule-epidermal growth factor 8a (mfge8a, Figure 3D), whose analog, mfge8, is a phagocytosis factor that maintains the pool of radial glia-like cells by controlling cellular quiescence in mice (Zhou et al., 2018). Cluster 2 cells (RG of reactive state) were characterized by their up-regulation of vimentin (vim), a hallmark of RG reactivation ( Figure 3E and Figure 2-figure supplement 1A-B 1 ). Proliferative RG were composed of RG of proliferative-S (mcm2 and pcna, cluster 3; Figure 3F and G) and proliferative-G2 states (cdk1 and nusap1, cluster 4; Figure 3H and I). Cluster 5 cells highly expressed vimentin (vim) and were likely to represent Vimentin + cells from neighboring tissues under the optic tectum in the midbrain due to possible contamination during the dissection of the optic tecta ( Figure 3E and Figure 3-figure supplement 2G-2I 1 ). We did not observe such a high expression of vim in the tectal RG ( Figure 3E). Thus, we excluded cluster 5 cells from further analysis. Cell cycle phases analysis ( Figure 3J) and pseudo-time analysis ( Figure 3K and Figure 3-figure supplement 2J) were performed and suggested the temporal order of 4 remaining cell clusters, thereafter termed as the state of dormant RG (dRG), the state of reactive RG (reactive RG), the state of proliferative-S RG and the state of proliferative-G2 RG. Next, we looked into the expression dynamics of the genes that differentially expressed across the states. mfge8a was abundant in dormant RG (cluster 1), began to decrease in reactive RG (cluster 2) and became rapidly diminished in proliferative RG (cluster 3 and 4) ( Figure 3L). Kruppel-like transcription factor 6a (klf6a), the transcription factor essential for optic axon regeneration (Veldman et al., 2010;Veldman et al., 2007), exhibited a peaked expression in reactive RG (cluster 2) ( Figure 3M), and Insulinoma-associated 1a (insm1a), encoding a transcriptional repressor that has been reported to be necessary for MG-based retina regeneration (Forbes-Osborne et al., 2013;Ramachandran et al., 2012), highly expressed in proliferative-S and -G2 RG (cluster 3 and 4)    Figure 3N). To verify their expression, we performed in situ hybridization. The results were consistent with our scRNA-seq data, mfge8a was down-regulated in injured-induced PCNA + proliferative RG at 3 dpi ( Figure 3O-P 1 ), whereas klf6a and insm1a mRNA expression increased in the 2-dpi ( Figure 3Q-R 1 ) and 3-dpi ( Figure 3S-T 1 ) optic tecta, respectively. Interestingly, the signals of klf6a ( Figure 3Q and Q 1 ) and insm1a ( Figure 3S and S 1 ) were mainly distributed in the processes of RG.

Notch/Delta expression pattern correlated with the cell-cycle entry of reactive RG
Notably, during the transition of reactive (cluster 2) and proliferative states (cluster 3 and 4), the expression of her4.1, the targeting gene of Notch signaling (Takke et al., 1999), decreased ( Figure 4A and B), whereas deltaA expression increased ( Figure 4C and D). Further correlation analysis showed that pcna and deltaA expression were correlated, while pcna and deltaA were uncorrelated with the expression of her4.1 and her4.2 ( Figure 4E). Our results suggest proliferative RG with an increase of deltaA expression and a decrease of Notch activity.
To visualize the Notch/Delta dynamics in vivo, we employed a reporter line Tg(Tp1bglob:EGFP) (hereafter referred to as Tg(Tp1:EGFP)), in which EGFP is driven by the TP1 element, the direct target of the intracellular domain of Notch receptors (NICD) that is generated upon Notch activation (Parsons et al., 2009;Quillien et al., 2014). We performed PCNA immunostaining on the coronal sections of Tg(Tp1:EGFP) at 3 dpi ( Figure 5F-G 3 ). Interestingly, the results showed~82% (97/119 cells, n = 6 sections) of PCNA + proliferative RG had no EGFP signal, indicating low Notch activity ( Figure 4H). Notch activity and PCNA signal were mostly exclusive ( Figure 4F 1 -F 3 ). Consistently, in situ hybridization of deltaA followed by immunostaining of PCNA showed~81% (60/74 cells, n = 10 sections) of PCNA + RG expressed deltaA ( Figure 4I-4J). Our results suggest that Notch/Delta lateral inhibition may be at work.

(K) Pseudo-time developmental trajectory of identified states using
Monocle showing that the trajectory is booted from dRG cluster (cluster 1) and end at pRG-G2 cluster (cluster 4). (L-N) Violin plots of expression for genes enriched in dRG cluster (mfge8a, cluster 1), rRG cluster (klf6a, cluster 2) and pRG-S/G2 cluster (insm1a, cluster 3 and 4). (O-T 1 ) In situ hybridization showing the expression of mfge8a (O-P 1 ), klf6a (Q-R 1 ) and insm1a (Q-R 1 ) in the optic tecta after injury. The white arrowheads shown in (O and O 1 ) indicate PCNA + proliferative RG are mfge8a À , the open white arrowheads indicate klf6a (Q and Q 1 ) or insm1a (S and T 1 ) mRNA signals are located in processes of proliferative RG. White dashed lines represent the tectal ventricle boundary. t-SNE, t-stochastic neighbor embedding; RG, radial glia; PGZ, periventricular gray zone, TS, torus semicircularis. Scale bars, 30 mm. See also of a recent study , Notch inhibition was also sufficient to trigger the proliferation of tectal RG even without any injury (DMSO-treated: 6.3 ± 0.48 cells, n = 4; LY411575-treated: 26.7 ± 2.4 cells, n = 5; mean ± SEM; p>0.05; Figure 5D, E, H, I, J, Figure 5-figure supplement 1C and D), which was reminiscent of the increase of constitutively proliferative RG in the zebrafish telencephalon by Notch inhibition (Chapouton et al., 2010).  Furthermore, we took advantage of Tg(hsp70l:gal4ÂUAS:NICD-Myc) double-transgenic fish, in which a heat shock promoter drives mosaic expression of the NICD-Myc fusion protein, allowing conditional and potent over-activation of Notch signaling ( Figure 5K) . Misexpression of NICD significantly blocked the cell-cycle entry of tectal RG following stab injury, that is,~94% (31/33 cells, n = 15 sections) of NICD-overexpressed RG underneath the injury sites were PCNA À ( Figure 5L-N and Figure 5-figure supplement 1E-E 3 ). Torus semicircularis (TS) is the midbrain tissue under the PGZ of the central optic tectum, and their boundary could be unambiguously defined by DAPI staining (Figure 1-figure supplement 1A-C 2 and Figure 5-figure supplement  1F-G 3 ). We noticed that stab injury induced some cells in the TS underneath the injury site (close to the boundary of TS and PGZ) to become proliferative in some animals, which required further investigation ( Figure 5L-L 3 ). In sum, Notch inhibition mediates the stochastic cell-cycle entry of reactive RG after the injury.

Long-term tracing reveals proliferative RG are gliogenic
To examine the fate outputs of proliferative RG after the injury, we utilized the Cre-loxP system to perform the clonal analysis of single RG after stab injury and analyze clonal cell-type compositions by immunostaining of BLBP, a putative maker for RG, and HuC/D, a putative marker for neurons. Notably, the newborn cells were largely BLBP positive, indicative of RG identity ( Figure 6A-B 3 ). These results raised an immediate question as to whether injury-induced proliferative RG are gliogenic.
After the injury, we often observed a physical wound at the injury site on the surface of the optic tectum (1343 ± 315.7 mm 2 , n = 10, mean ± SEM; Figure 6-figure supplement 1C-E). More strikingly, these stab wounds remained up to 300-400 dpi (1339 ± 768.6 mm 2 , n = 7, mean ± SEM; p>0.05; Figure 6-figure supplement 1E and F-I 3 ). These wounds were surrounded by BLBP signals but without cell nuclei, suggesting that the hypertrophic processes of RG formed a glial scar-like structure surrounding the wound, and thereby blocking the repair of the wound (Figure 6-figure supplement 1A-B 3 and F-I 3 ). Our results suggest a limited regenerative capacity of the adult zebrafish optic tectum.

Discussion
How do reactive RG enter the cell cycle? Consistent with astrogliosis in injured mammalian CNS, we find that zebrafish tectal RG are undergoing the consecutive phases of glial reactivation and glial proliferation in response to the injury (Figure 2A-K) (Burda and Sofroniew, 2014). After stab injury, Figure 7 continued but not 1-3 dpi significantly increases the number of EdU + /HuC/D + newborn neurons in the injured optic tectum. In the uninjured optic tecta, Notch inhibition during both 1-3 dpi and 4-5 dpi increases the number of EdU + newborn cells, but not EdU + /HuC/D + newborn neurons (mean ± SEM, ***p<0.001, **p<0.01, *p<0.05, ns, p>0.05; one-way ANOVA followed by Tukey's HSD test). See also Figure 7-source datas 1 and 2 for quantification. (G) Proportion of EdU + /HuC/D + newborn neurons to EdU + newborn cells in (B-D). Notch inhibition during 4-5 dpi increases the proportion of the neuron production, whereas Notch inhibition during 1-3 dpi decreases the proportion (mean ± SEM, **p<0.01; ns, p>0.05; one-way ANOVA followed by Tukey's HSD test). See also Figure 7-source data 3 for quantification. (H and I) Schematics of the experimental procedure for Notch inhibition experiments shown in (J-M). After the injury, fish are treated with either DMSO or LY411575 during 4-5 dpi and are injected with EdU for three consecutive days during 1-3 dpi (H) or 4-6 dpi (I). All the fish are sacrificed and analyzed at 7 dpi. (J-M) Representative images of HuC/D (green) and EdU (red) immunofluorescences of the 7-dpi optic tecta after the treatment in (H and I). With the treatment of LY411575 during 4-5 dpi, EdU pluses during 4-6 dpi (L and M) but not 1-3 dpi (J and K) label significant more newborn neurons. White arrowheads indicate EdU + /HuC/D + newborn neurons. (N and O) Quantification of EdU + newborn cells (N) and EdU + /HuC/D + newborn neurons (O) in (J-M) (!3 replicates for each group; mean ± SEM, ***p<0.001, ns, p>0.05; two-way ANOVA followed by Tukey's HSD test). See also Figure 7-source datas 4 and 5 for quantification. (P) Proportion of EdU + /HuC/D + newborn neurons to EdU + newborn cells in (J-M). EdU pulses during 4-6 dpi significantly increase the proportion of neuron production (!3 replicates for each group; mean ± SEM, ***p<0.001; ns, p>0.05; two-way ANOVA followed by Tukey's HSD test). See also Figure 7-source data 6 for quantification. (Q) Schematic summary of the working model. Injury induces all RG underneath the injury site to become reactive. Only~25% of reactive RG enter the cell cycle and become proliferative. The cell-cycle entry of reactive RG is regulated by Notch/Delta lateral inhibition. In the injury condition, proliferative RG largely undergo gliogenesis (~3-5% newborn neurons). The resulting newborn cells could survive up to 300 dpi. In the Notch inhibition condition, dormant RG can become proliferative but only generate~1% of newborn neurons. However, Notch inhibition during 4-5 dpi drives reactive RG into the cell cycle, giving rise to significant more neurons (~12-20%). Interestingly, these over-produced Single-cell RNA sequencing To perform single-cell RNA sequencing (scRNA-seq), cells after FACS were loaded onto the Chromium Single Cell Chip (10x Genomics, USA) according to the manufacturer's protocol. The scRNAseq libraries were generated using the GemCode Single-Cell Instrument and Single Cell 3' Library and Gel Bead kit v2 Chip kit (10x Genomics, 120237) by following the manufacturer's protocol. Library quantification and quality assessments were performed by Qubit fluorometric assay (Invitrogen) with dsDNA High Sensitivity Assay Kit (AATI, DNF-474-0500) and the fragment analyzer with High Sensitivity Large Fragment À50 kb Analysis Kit (AATI, . The indexed library was tested for quality, and sequenced by the Illumina NovaSeq 6000 sequencer with the S2 flow cell using paired-end 150 Â 150 base pair as the sequencing mode. The sequencing depth was 60K reads per cell.

Single-cell sequencing data analysis
Single-cell FASTQ sequencing reads (Novogene) were processed, and converted to digital gene expression matrices after mapping to the zebrafish genome (Zv10) using the Cell Ranger Single Cell Software Suite (v2.1.0) provided on 10x genomics website (https://support.10xgenomics.com/singlecell-gene-expression/software/pipelines/ latest/what-is-cell-ranger). 66,817 mean reads per cell and 1325 mean genes per cell were obtained.
For further analysis, we used an analysis pipeline provided by Seurat R package (http://satijalab. org/seurat/). Firstly, the Seurat object was created to filter low-abundance genes, cell doublets and low-quality libraries (with low gene numbers and high mitochondrial transcripts). Secondly, the filtered data were normalized and used to identify highly variable genes based on expression and dispersion. Thirdly, the data were scaled, and the unwanted sources of variation were removed. Fourthly, cell clustering analyses were performed by the t-SNE projection (Figure 3-figure supplement 1B). Finally, we found out the markers for every cluster (Figure 3-figure supplement 1C). Due to possible contamination during tissue dissociation and FACS, the samples were contaminated with other types of cells from the optic tecta and other neighboring tissues. Based on the markers of each cluster, these contaminated cells were identified and removed after the initial clustering.
As many proliferative progenitors are present in the tectal proliferation zone (TPZ) (Galant et al., 2016;Ito et al., 2010), a big-area injury induced a lot of PCNA + tectal RG at 3 dpi. We obtained two pcna + clusters in the t-SNE plot (cluster 1 and 2, Figure 3- figure supplement 2A and B). However, based on experimental evidence: 1. Injury caused the obvious down-regulation of her4 PCNA + RG, whereas her4 was highly expressed in RG in TPZ (Figure 3-figure supplement 2B-F); 2. The previous study showed progenitors in TPZ were able to generate oligodendrocytes. We did not find any new-born cell derived from injury-induced PCNA + RG was oligodendrocyte, and olig2 expression was noticed in cluster 1 and 10 but not in cluster 2 (Figure 3-figure supplement 2B and C). We identified cluster 1 as the progenitors in the TPZ, and it was removed from our data. Following these step-wise filtering processes, we obtained the purified data of each sample.

Cell cycle phase analysis
To obtain the cell-cycle properties of the cells in our sample, the 'CellCycleScoring' function of Seurat was used. Briefly, each cell was scored based on its expression of G2/M and S phase marker genes. Then the numbers of cells in different cell cycle phases were counted and the ratios of individual cell cycle phases were calculated.

Gene-gene correlation analysis
The gene-gene correlation was measured according to the pairwise Pearson correlational distances. 'bioDist' R package was used to calculate these correlational distances.