A brief inhibition of Notch signaling activates qNSCs without depleting them in favor of cycling NSCs.

(A and B) Experimental scheme (A) of the treatment of 3mpf adult zebrafish with DMSO or LY for 12 or 24 hours, followed by whole-mount immunohistochemistry (IHC) and in situ hybridization (RNAscope) and cell quantifications. Rostral and caudal areas of the pallial ventricular zone were analyzed, as depicted on the brain cartoon (B) (dorsal view: A, anterior; P, posterior; ob, olfactory bulb; ot, optic tectum). (C to E) Representative images of the pallial ventricular zone (rostral areas) after treatment with DMSO (24 hours) (C) or LY for 12 hours (D) or 24 hours (E). (C’) to (E’) are higher magnifications of the areas boxed in (C) to (E). Whole-mount dorsal (apical) views, anterior left. Zo1 and Sox2 immunostainings (white) are used to identify progenitor cells with apical ventricular contact, Pcna (red) is used to label and count proliferating cells (quantifications in F), and ascl1a RNAScope (magenta) is used to read decreased Notch activity and transition toward preactivation. ascl1a is expressed at higher levels after LY treatment and higher after 24 than after 12 hours of treatment. See also fig. S1F. Scale bars, 8 μm. (F) Quantification of the proportion of cycling cells (Pcna^pos) after each treatment in the rostral or caudal part of the dorsal pallium. P values are the result of a Kruskal-Wallis test. Violin plots are built from bootstrapped random sampling of the measured proportions to estimate the distribution. (G) Experimental scheme to generate the scRNAseq dataset under LY treatment. The DMSO control dataset was reported in (16). (H) Low-dimensional embedding of quiescent NSCs (qNSCs) in the control scRNAseq dataset colored by cluster [after (16)]. (I) Low-dimensional embedding of qNSCs in the scRNAseq dataset of LY-treated fish, colored by cluster, matching colors to those from the control dataset based on inferred equivalence.

Data integration highlights Notch inhibition–induced molecular changes in NSCs.

(A) Low-dimensional embedding of qNSCs after integration of LY-treated and control datasets (Fig. 1, H and I, and fig. S2A). Cells are colored on the basis of the refined cluster annotations derived from integrated analysis; cluster q1 resolves into six clusters (1a to 1f), cluster q4 into two clusters (4a and 4b), and a previously overlooked cluster 8 is identified. (B) Violin plot comparing the expression of ascl1a and ccnd1 across all clusters for control and treated cells (clusters colored as in A). y axis: number of reads normalized for sequencing depth. (C) Depiction of the over or under-representation of distinct subpopulations of qNSCs between control and treated datasets (clusters are color coded as in Fig. 2A). The further a curve is shifted away from the central line, the more a given cluster is enriched (to the right) or depleted (to the left) in the LY-treated dataset. (D) Low-dimensional embedding of cycling cells after integration of LY-treated and control datasets. Cells are colored on the basis of the refined cluster annotations derived from integrated analysis. Cluster 1 in green represents cycling NSCs and clusters 2 to 6 represent IPCs progressively more advanced in the cell cycle. (E) Violin plot comparing the G₁/S and G₂/M scores across all clusters of cycling cells for control and LY-treated cells (clusters colored as in D). Absolute values do not reflect direct expression of genes and comparison must rely on relative values of each score. (F) Depiction of the over or under-representation of distinct subpopulations of cycling cells between control and LY-treated datasets (clusters are color coded as in D). Interpretation as in (C).

Zebrafish astrocyte–like cells respond to Notch inhibition and injury.

(A) Expression levels of indicated genes in qNSCs from DMSO- or LY-treated datasets projected on the integrated embedding of qNSCs. igfbp2a and timp4.3 mark astrocyte-like cells and ascl1a marks preactivated cells and the lack of Notch activity. q1e and q1f are circled. In the LY dataset, but not in control, many cells coexpress astrocyte-like NSC markers and ascl1a, notably in clusters q1e and q1f. (B) Expression levels of ascl1a or ccnd1 in timp4.3^pos cells from DMSO- or LY-treated datasets. (C) High magnification of the pallial ventricular surface in controls (top) or upon a 24-hour LY treatment (bottom) showing expression of timp4.1 (yellow), ascl1a (purple), and ccnd1 (green) (coexpression of ascl1a and ccnd1 appears in cyan, left merged views) revealed by smFISH together with immunohistochemistry for Zo1 (white). Examples of cells with high levels of expression of timp4.3, ascl1a and ccnd1 upon LY are indicated (yellow arrowheads). (D and D’) High magnifications of the pallial ventricular surface close to a lesion (dorsal whole-mount view), processed for immunohistochemistry (ZO1, white; Pcna, red) and smFISH (timp4.3, cyan). Proliferating cells (Pcna^pos) are more numerous neighboring the lesion. (D) all channels; (D’) Zo1 and timp4.3 only. Scale bar, 10 μm. (E) Quantification of timp4.3 expression (number of dots per cell) in Pcna^neg and Pcna^pos cells neighboring the lesion (“lesion proximity”) or far away (“control area”). P values were calculated with a Wilcoxon signed-rank test to avoid inflating sensitivity with a bootstrapped test. Control area: n = 3 brains, 92 cells; lesion proximity: n = 3 brains, 149 cells. timp4.3 levels close to the lesion are decreased, with a significant decrease in Pcna^neg cells. In contrast, Pcna^pos cells tend to express timp4.3 at higher levels close to the lesion than in control cells, but this difference is not statistically significant.

A machine learning approach predicts putative regulators of resistance to Notch signaling blockade in deeply quiescent pallial NSCs of cluster q5.

(A) Levels of expression of major Notch targets represented in the low-dimensional embedding of qNSCs from the LY-treated dataset. Cells belonging to cluster q5 are circled and show higher and almost exclusive expression of these genes. (B) Barplots depicting the weight associated with putative regulator-target relationships inferred through gradient boosting and automatic filtering. The height of the bar is proportional to the contribution of the indicated transcription factor’s expression to the recovery of the pattern of expression of the target (her4.1, her8a, her9, and hey1). When transcription factors were anticorrelated with their putative target we assigned a negative sign to the link. Bars are color coded to highlight specific genes and gene families. (C and D) Levels of expression of the genes encoding the best candidates regulators of q5 resistance to activation, id1 (C) and nr2f1b (D), represented in the low-dimensional embedding of qNSCs from the control (left) and LY-treated (right) datasets.

nr2f1b expression is associated with resistance to Notch inhibition.

(A) Confocal images of caudal areas in the dorsal pallium (see Fig. 1B) of fish treated for 48 hours with either DMSO or LY, immunostained for Zo1 (white, apical junctions) and Pcna (red, proliferation) and processed for smFISH for nr2f1b (cyan). Scale bars, 5 μm. (B) Quantification of the proportion of cycling cells as a function of nr2f1b expression after 48 hours of LY treatment. nr2f1b^pos cells are less likely to be cycling than nr2f1b^neg cells. Reported P value is derived from a χ² test. Violin plots are built from bootstrapped random sampling of the measured proportions to estimate the distribution but these estimated distributions are not used for statistical testing. DMSO: n = 3 brains, 1263 cells; LY: n = 3 brains, 1378 cells; nr2f1b^neg cells have zero nr2f1b mRNA dots. (C) Difference in response rate (see definition of response rate in Material and Methods) between nr2f1b^pos and nr2f1b^neg cells highlighting that nr2f1b^pos cells are less likely to respond to Notch inhibition. The histogram represents bootstrapped values to estimate a null distribution of the difference in response rates between nr2f1b^pos and nr2f1b^neg cells via Monte Carlo simulation. The dotted red vertical line represents the observed difference in response rate. Fifty bootstrapped simulations with 1000 samples each were conducted, the P value represented here corresponds to the maximum proportion of simulated values across all 50 bootstraps that were inferior to the observed difference.

Endogenous nr2f1b expression is necessary for resistance to Notch signaling inhibition.

(A) Schematic of the experiment. After morpholino (MO) intracranial injection and electroporation, fish were allowed to rest for 3 days, then treated with LY for 2 days and euthanized for quantification of the proportion of cycling NSCs. (B) Example images of the pallial surface in fish electroporated with a control MO (left) and an nr2f1b-specific MO (right) followed by 48 hours of LY treatment (dorsal whole-mount views). The brains were processed for immunohistochemistry for glutamine synthase (cyan, GS, labeling NSCs) and Pcna (red, proliferation), and the lissamine-tagged MOs (magenta) were imaged directly. Arrows point to electroporated NSCs. Scale bars, 10 μm. (C) Quantification of the proportion of proliferating NSCs as a function of the electroporated MO. NSCs electroporated with the nr2f1b MO are significantly more likely to proliferate after LY treatment than when electroporated with the control MO. Reported P value is derived from a χ² test. Violin plots are built from bootstrapped random sampling of the measured proportions to estimate the distribution. Control MO: n = 4 brains, 123 cells; nr2f1b MO: n = 5 brains, 581 cells.

nr2f1b is sufficient to promote resistance to Notch pathway inhibition.

(A) Schematic of the experiment. After plasmid electroporations and a 3-day rest (dpe, days postelectroporation), fish were treated with LY for 1 day and euthanized for quantification of ascl1a expression. (B) Schematic of the approach to quantify ascl1a molecules (magenta dots) in electroporated cells (green). NSC cell bodies can differ from the shape of their apical membrane, precluding reliable automated quantification of signal with Zo1. However, the slight leakiness of nlsGFP fills the cell body, which can be segmented. Spots falling into the circumscribed volume are assigned to the cell in an unbiased way by detecting an elbow in the plot of pixel intensities in the ascl1a channel. (C) Example of a cell with a complex morphology in which ascl1a dots were semi-automatically segmented. The first two images on top and the first image on the bottom show dorsal views of how the cell body extends beyond the Zo1^pos apical membrane and is surrounded in ascl1a dots. The third image on top shows an orthogonal view from the parenchyma displaying the complex morphology of the cell and the position of numerous ascl1a dots around it. The two last images on the bottom show the electroporated cells with only segmented ascl1a dots highlighted. Scale bars, 5 μm. (D) Example images of electroporated cells with pCMV:nr2f1b-P2A-nlsGFP or pCMV:nlsGFP after 24 hours of LY treatment. Arrows point to electroporated NSCs. Scale bars, 10 μm. (E and F) Violin plots of the number of ascl1a dots per cell in cells electroporated with pCMV:nr2f1b-P2A-nlsGFP or pCMV:nlsGFP across the pallium (E) and when separating the rostral and caudal pallial domains (F) after the experimental scheme in (A). Statistics: unpaired two-sample Wilcoxon test; n = 3 brains per condition, 201 cells in total. Overall: P value = 2.2 × 10⁻¹⁶; anterior: P value = 2.205 × 10⁻¹¹; posterior: P value = 2.771 × 10⁻¹¹.