Non-apoptotic caspase events and Atf3 expression underlie direct neuronal differentiation of adult neural stem cells

Stem cell (SC) fate decisions orchestrate adult organ maintenance and SC population renewal over a lifetime. In the brain and muscle, some adult SCs can directly acquire a differentiated fate in the absence of a division event under physiological conditions (Barbosa et al., 2015; Bjornson et al., 2012; Dray et al., 2015; Ismaeel et al., 2024; Mourikis et al., 2012). These events, described as ‘direct conversion’, have remained understudied, and it is largely unknown how they are controlled.

In the adult vertebrate brain, neural stem cells (NSCs) are radial glia-like progenitors (RGs), which are mostly quiescent (Chaker et al., 2016; Lampada and Taylor, 2023; Obernier and Alvarez-Buylla, 2019; Urbán et al., 2019; Yeh et al., 2023). In mouse, NSCs can generate differentiated neurons and astrocytes (Beckervordersandforth et al., 2010; Dulken et al., 2017; Zywitza et al., 2018). In zebrafish, NSCs express glial fibrillary acidic protein (gfap) and her4 genes, and co-express markers of mature astrocytes during their quiescence phase (Cosacak et al., 2019; Morizet et al., 2024). Upon division, they generate other NSCs (gfap^pos, her4^pos) and/or committed neuronal progenitors (NPs) (gfap^neg, her4^neg) that differentiate into neurons (Dirian et al., 2014; Furlan et al., 2017; Kroehne et al., 2011; Mancini et al., 2023). In both models, genetic clonal tracing and/or intravital imaging also suggests direct conversions from individual NSCs to neurons. In the mouse dentate gyrus, around 10% of clones in Nestin:Cre tracing of individual NSCs are single neurons after 1 month (Bonaguidi et al., 2011), suggesting a direct conversion, although the interpretation of these results could be confounded by cell death. In the zebrafish pallium, intravital imaging has revealed direct neuronal differentiation occurring from her4.3^pos or gfap^pos NSCs that do not express proliferation markers and/or have not divided for at least 14 days (four imaging time points (tp), 14-16 days] (Barbosa et al., 2015; Dray et al., 2015; Than-Trong et al., 2020). Longitudinal imaging in these cases shows no cell death events. Modeling clonal dynamics predicts that direct conversions represent 25% of fate decisions (Than-Trong et al., 2020), and are crucial to maintain homeostasis of the NSC population.

As part of our efforts to identify predictors and mechanisms of NSC fate decisions in vivo, we focus here on NSC direct conversions. First, exploiting the amenability of the zebrafish adult pallium to intravital NSC imaging, we quantify the morphodynamic features of direct conversion events in real time. We show that these parameters do not differ from those of NP delaminations, precluding the simple identification of morphometric predictors of direct conversions. As a second approach, we took an educated guess and challenged whether direct conversions could relate to non-apoptotic caspase events.

Caspases are site-specific proteases of the programmed cell death system. In addition, they are increasingly recognized for functions in non-apoptotic events during development and homeostasis in various tissues and organisms (reviewed by Abdul-Ghani and Megeney, 2008; Burgon and Megeney, 2018). In particular, there is a frequent association of non-apoptotic caspase events with cell fate decisions, including in some SCs (Fujita et al., 2008; Janzen et al., 2008). In the nervous system, non-apoptotic caspase recruitments modulate dendritic pruning, developmental circuit maturation and axonal pathfinding (Unsain and Barker, 2015), neuronal differentiation (Fernando et al., 2005), and neural progenitor proliferation (Colon-Plaza and Su, 2022; Kuranaga et al., 2006). Non-apoptotic caspase events involve the activation of effector Caspases 3 and 7 (activated forms noted as Cas3*, Cas7*) by proteolytic cleavage, and can be tracked using Cas3*/Cas7* sensors. For example, in Drosophila, CasExpress and CaspaseTracker liberate Gal4 upon cleavage at the Cas3*/Cas7* canonical site DEVD, to drive lineage labeling when caspase activation is not followed by cell death (Ding et al., 2016; Tang et al., 2015). These studies revealed multiple cells surviving an early caspase event under physiological conditions. These events can be seen as true death reversals (referred to as anastasis, as in Sun et al., 2017; Tang and Tang, 2018) or sign common pathways used to trigger cell remodeling in apoptosis and cellular decisions.

To assay whether non-apoptotic caspase events take place in adult NSCs, and, if so, test their relationship with NSC direct conversion, we generated an inducible, NSC-specific and stable Cre-mediated Cas3*/Cas7* sensor, Cas^CREAtlas. We find that non-apoptotic caspase events do take place in NSCs and are biased towards fate choices of direct neuronal generation. We identify the stress-induced transcription factor Atf3 as necessary for this NSC fate in vivo. Finally, we show that NSCs undergoing Cas3*/Cas7* events are further prone to neurogenic fates in response to lesion.

Using intravital imaging, NSC direct conversions were previously defined as the loss of expression of the NSC marker Tg(gfap:dTomato) accompanied with delamination from the pallial ventricular layer without visible division during the preceding four imaging time points (4 tp) (14 to 16 days) (Than-Trong et al., 2020). This time frame was chosen because, when a neurogenic division is visible in a movie, 85±10% (mean±s.d.) of neuron-fated daughter cells visibly express their fate (loss of gfap:dTomato) within the 10-12 days post-division (Dray et al., 2021; Than-Trong et al., 2020). Such events are followed by expression of the neuronal differentiation marker HuC/D and the presence of a neuronal process (Barbosa et al., 2015).

To characterize the morphodynamic features of these events and determine whether and how they contrast with delamination and differentiation events occurring from neurogenic divisions (referred to as ‘post-division delaminations’), we exploited the intravital imaging movies acquired in 3 months post-fertilization (mpf) adults in the Casper;Tg(gfap:Zo1-mKate2);Tg(deltaA:egfp) background (Mancini et al., 2023). This dataset contains 828 NSCs filmed over 39-43 days every 2-3 days in four pallial hemispheres (from four different fish) and reveals NSC apical surfaces (Zo1-mKate2) and a neurogenic fate (deltaA expression). This dataset includes 83 delamination events, which we now classified according to the 4 tp cut-off. When no division is visible along the track, tp0 is the first imaging tp of the movie (Fig. 1A, top), and we only consider delaminations taking place ≥4 tp from the start of each movie. When a division is visible, tp0 is the first tp post-division (Fig. 1A, bottom). We also defined ‘delamination termination’ as the first tp with no identifiable Zo1-mKate2-negative apical surface, i.e. the moment of apical closure (Fig. 1A). With these settings, 54 delaminations terminated after at least 4 tp (13 days) without visible division, and 29 followed a visible division during the previous three imaging tp or less (≤13 days). In the first category, the time from tp0 to delamination termination varied between 14 and 41 days. Two examples are illustrated in Fig. 1B, displaying delamination events that occurred without detectable division during the previous 23 (example 1) and 35 (example 2) days.

We next compared the morphometric parameters of delaminations occurring >4 tp without division versus ≤3 tp post-division (Fig. 1C-E). Cells in these two categories did not differ significantly in their apical area at the tp preceding delamination termination (9.4±6.1 µm² and 9.3±4.8 µm², respectively; mean±s.d.) (Fig. 1C), nor in their expression of deltaA (100% deltaA^pos cells in both cases) (Fig. 1B). There was a moderate correlation between apical surface area at the onset of apical surface shrinkage (tp_sh) and the duration of apical shrinkage (Fig. 1D). When normalized over the duration of shrinkage, shrinkage rates appeared similar between the two categories (∼1 µm² per tp) (Fig. 1E). Together, these results characterize the quantitative and molecular features of the delamination process during adult pallial neurogenesis. They reveal its comparable dynamics irrespective of time post-division or apical surface area at the start.

The dataset does not allow NSCs and NPs to be directly recognized: NPs are gfap:Zo1-mKate^neg, but, when intermingled among NSCs, they will appear as a Zo1-mKate-delineated surface, making them tractable. To estimate the percentage of NSCs in our dataset, we considered the two possible NP configurations: (1) NP clusters (e.g. an NP doublet generated by the symmetric neurogenic division of an NSC or an NP, which will appear as a single apical surface surrounded by ZO1-mKate), and (2) isolated NPs surrounded by NSCs. NP clusters can be quantified using immunohistochemistry (IHC) for Zo1 (also known as Tjp1), applied on specimen fixed at the end of the film, in comparison with mKate: NSCs are Zo1 IHC^pos and gfap:Zo1-mKate^pos, while NPs are Zo1 IHC^pos and gfap:Zo1-mKate^neg. Thus, we counted the proportion of Zo1-mKate surfaces that corresponded to several adjacent NPs (Fig. S1A). This proportion increased as the Zo1-mKate surface area decreased (Fig. S1B), as expected from NPs having smaller apical surface areas than NSCs (medians at 100 µm² versus 20 µm², respectively) (Mancini et al., 2023). With an apical surface area ≤30 µm² at tp_sh (80/83 events, 96%) (see below and Fig. 1D), 21.4% of Zo1-mKate surfaces were NP clusters (Fig. S1B). Next, to count isolated NPs, an NSC marker was necessary. We used fixed Tg(gfap:gfp) pallia immunostained for GFP and Zo1; NSCs are gfap:GFP^pos and Zo1 IHC^pos, while NPs are gfap:GFP^neg and Zo1 IHC^pos (Fig. S1C). With an apical surface area ≤30 µm², 29.3% of surfaces were isolated NPs (Fig. S1D). Together, approximately half of the Zo1-mKate surfaces of 30 µm² or less that would be recorded in intravital imaging were NSCs (Fig. S1E). Thus, the morphometric features measured above (Fig. 1) likely also apply to NSC direct conversions. Together, this analysis also shows that NSC direct conversions cannot be predicted using morphodynamic parameters alone.

In search for predictive parameters of NSC direct conversions, we next considered non-apoptotic Caspase events (Fig. 2A). To track non-apoptotic proteolytic events at the DEVD Cas3*/Cas7* target site (Chéreau et al., 2003; Talanian et al., 1997), we designed a heritable lineage tracer of DEVD cleavage in NSCs. The driver line, Tg(her4:mCD8-DEVD-V5-Cre), produces a Cre recombinase (fused to a V5 tag) tethered to the plasma membrane via a DEVD site, expressed under the her4.3 regulatory elements (Yeo et al., 2007) (Fig. 2B). This promoter fragment drives expression in the same cells as the gfap regulatory elements, i.e. NSCs (Than-Trong et al., 2020). When Tg(her4:mCD8-DEVD-V5-Cre) fish are crossed into the Tg(βact:lox-stop-lox-hmg2bmCherry) reporter (Wang et al., 2011) (double background referred to as Cas^CREAtlas), non-apoptotic cleavage events at the DEVD site in her4-expressing cells should trigger Cre-mediated recombination and the permanent expression of Hmg2bmCherry in all progeny cells. A Tg(her4:mCD8-GSGC-V5-Cre) driver, immune to Cas3*/Cas7*, was used as control.

To validate the approach, we first analyzed the expression of the her4:mCD8-DEVD-V5-Cre transgene and the inducibility and selectivity of Cas^CREAtlas tracing in the larval pallium at 6 days post-fertilization (dpf). IHC in the Tg(her4:mCD8-DEVD-V5-Cre);Tg(gfap:eGFP) background showed that V5-Cre expression is faithfully restricted to RGs (Fig. 2C). Next, we tested for Hmg2bmCherry^pos cells in the pallium at 6 dpf. Hmg2bmCherry^pos cells were rare in control pallia (on average, five per pallium). Their presence may reflect a low affinity of Cas3*/Cas7* for non-canonical sites, or the cleavage of GSGC by another protease at low frequency. In contrast, Hmg2bmCherry^pos cells were easily detectable and significantly more numerous in Cas^CREAtlas pallia (on average, 20 per pallium) (Fig. 2B′,D-F). We also tested whether Cas^CREAtlas was responsive to activation of the caspase cascade. To achieve this, 4 dpf larvae were transiently subjected to the apoptosis inducer camptothecin (CPT) (Ikegami et al., 1999) injected into the hindbrain ventricle, then chased until 6 dpf. CPT significantly increased the number of Hmg2bmCherry^pos cells in Cas^CREAtlas larvae compared to control injections (Fig. 2G-J).

The zebrafish pallium follows an outside-in neuronal generation pattern, without tangential neuronal migration (Furlan et al., 2017). We used these properties to interpret the pattern of Hmg2bmCherry cells in Cas^CREAtlas animals (Fig. 2K and see below). At juvenile and adult stages, neurogenesis follows the sequence NSCs (her4^pos or gfap^pos; Sox2^pos)>NPs (her4^neg and gfap^neg; Sox2^pos)>neurons (Than-Trong et al., 2020). At 1.5 mpf, few Hmg2bmCherry clones were seen in control pallia (Fig. 2M,M′). Based on their depth and position, they result from early non-specific activation events, as described at 6 dpf, generating rare but expanded clones. The Hmg2bmCherry pattern of Cas^CREAtlas pallia at 1.5 mpf was very different, with numerous small groups of labeled cells, some located in superficial layers (Fig. 2L,L′). IHC for Sox2, which also labels some recently born neurons, was used as a landmark and revealed deeply located (old) Sox2-positive neurons as well as more superficial (recent) events, with staining of freshly born neurons, NSCs or NPs (Fig. 2L1,L2). This pattern reflects Cre recombination events that occurred at different time points (from older to more recent) in her4-expressing cells.

Cas3/Cas7 activation in her4^pos cells may, or not, correlate with specific NSC behaviors. To address this, we compared NSC fate in Cas^CREAtlas clones with her4-mediated NSC fate tracing in the juvenile to young adult pallium (Fig. 3A). We established a temporal landmark across pallial depth, using a bromodeoxyuridine (BrdU) pulse at 1 mpf, to label neurons born at that stage. Thus, Cas^CREAtlas events having occurred between 1 mpf and the stage of analysis (2 mpf) generated Hmg2bmCherry^pos cells located above the BrdU landmark (Fig. 3B-B″). These clones were categorized and quantified. In parallel, clonal recombination events were induced with 4-hydroxytamoxifen (4-OHT) in Tg(her4:ERT2CreERT2); Tg(βact:lox-stop-lox-eGFP) animals just prior to the BrdU pulse, and the resulting clones were categorized and quantified at 2 mpf as well.

We considered that Hmg2bmCherry^pos cells belong to the same clones when they were separated by less than a two-cell diameter distance (Furlan et al., 2017; Webb et al., 2009), and recorded the composition of clones in 3D in whole-mount pallia using IHC for gfap:eGFP (NSCs). GFP^neg cells include NPs and neurons and, for simplicity, were labeled as neurons (Fig. 3C, Fig. S2A-A″). Unbiased clonal labeling generates a variety of clone types (Fig. 3D, Fig. S2B). ‘NSC-only’ clones are composed of single NSCs (remaining quiescent since labeling) or NSC doublets (from an amplifying NSC/NSC division) (red in Fig. S2A″). ‘Neuron-only’ clones include single neurons (presumably from a direct differentiation event), neuron doublets (from a direct differentiation event following an asymmetric neurogenic division, or from a symmetric neurogenic division), and groups of three neurons or more (from a direct differentiation or a symmetric neurogenic division following one or more neurogenic divisions) (respectively yellow, blue and orange in Fig. S2A″). Finally, ‘attached’ clones (magenta in Fig. S2A″) are made of an NSC and one or more neighboring neurons. This NSC was engaged in (a) recent neurogenic event(s), and attached clones correspond to neurogenically active NSCs. These fates together are in qualitative and quantitative agreement with the different division modes and fates reported in previous work (Than-Trong et al., 2020). On average, 50 Cas^CREAtlas events occurred during 4 weeks within a hemipallium (2000-2500 her4^pos NSCs). They were associated with most clone types; however, compared to her4:ERT2CreERT2-driven fates, attached clones were virtually absent in Cas^CREAtlas events, while the proportion of neuron-only clones was increased (Fig. 3D). Among the latter, the large majority (>75%) was made of single neurons (Fig. S2B). Thus, non-apoptotic Cas3*/Cas7* events in pallial NSCs correlate with a bias in the fate of neurogenically active NSCs towards lineage termination by neuronal differentiation.

To identify the molecular events involved in promoting NSC direct differentiation, we searched among known mediators of non-apoptotic caspase events (Sun et al., 2017; Tang et al., 2017). Many genes upregulated at an early stage of anastasis appeared to be expressed in pallial NSCs under physiological conditions, as revealed in our previously generated single-cell RNA-sequencing (scRNAseq) dataset (Morizet et al., 2024) (Fig. S3A). We focused on atf3, one of the top genes induced in mammalian cells, expression of which was detectable in a low number of quiescent NSCs (Fig. 4A). Sparse expression was confirmed using whole-mount in situ hybridization (ISH), revealing scattered atf3^pos cells across the pallial surface (Fig. 4B), with some enrichment in the posterior and dorsomedial pallium. At small scale, the atf3 pattern was not exactly identical in these areas in the two hemispheres. To address the location and morphology of atf3^pos cells, we used the Tg(gfap:eGFP) background and fluorescent chromogenic ISH. Cross-sections or horizontal optical sections showed that atf3^pos;GFP^pos cells have a delaminating profile, with their nuclei often located in deeper positions than the majority of NSCs (Fig. 4C1-D4). We also compared atf3 and Cas3. Of the two casp3a and -b duplicates, only casp3a expression was detected in scRNAseq (Fig. S3A), and only 5% of casp3a^pos cells were atf3^pos in this dataset, equivalent to the overall proportion of atf3^pos cells among all NSCs (Morizet et al., 2024). In situ, we only detected rare instances of co-expression of atf3 and casp3a (using RNAScope Hiplex whole-mount ISH) (Fig. S3B) or atf3 and Cas3* (revealed by IHC) under physiological conditions. Atf3 and Cas3* may be involved in distinct pathways, or at successive time points (or with very short windows of temporal overlap) in the same pathway, in adult pallial NSCs in situ. Finally, two atf3 transcripts were recovered in the adult pallium, corresponding to alternative splicing events predicted to encode long (Atf3-L) and short (Atf3-S) protein isoforms that differ in their N terminus (Fig. 4E). Our ISH used the atf3-L probe and does not distinguish between these isoforms.

Next, we combined gain- and loss-of-function experiments in adult pallial NSCs in vivo to test whether Atf3 impacts NSC fate. The ubiquitous pCMV:nlsGFP construct, electroporated after injection into the cerebral ventricle, highlights different cell fates on a short time scale [2 days post-electroporation (dpe)]: a large majority of ventricular cells (with a radial morphology and expressing Sox2, likely NSCs), and a minority of delaminating cells (with basally displaced nucleus and ventricular attachment, likely NPs) and of Sox2^neg parenchymal cells (interpreted as neurons) (Fig. 5A-C′). When electroporated under the same conditions, pCMV:atf3-S-nlsGFP had no effect on cell fate (Fig. 5E), while pCMV:atf3-L-nlsGFP significantly increased the proportion of neurons at the expense of NSCs (Fig. 5D). When electroporated into the Cas^CREAtlas background, pCMV:atf3-L-nlsGFP also correlated with Hmg2bmCherry expression in a minority of GFP^pos cells at 7 dpe (Fig. 5F-F″, white arrows). This may not result from chance, but the low occurrence of these co-expression events is puzzling. Cas3* may be induced only transiently or at levels too low to generate efficient levels of Cre recombinase. Alternatively, Atf3 may require a specific context to trigger Cas3* in adult NSCs. It is noteworthy that the hierarchical position of Atf3 relative to Cas3* has been reported to vary (see Discussion).

Next, to determine whether Atf3 is necessary for direct neuronal differentiation downstream of or in parallel to Cas3*, we tested the effect of blocking Atf3-L function in vivo in the Cas^CREAtlas double-transgenic context. We designed a vivo-morpholino (MO) directed against the exon2–intron2 boundary of atf3, predicted to generate a truncated Atf3-L protein devoid of its DNA-binding and leucine-zipper domains (Fig. 5G). This prediction was validated in embryos (Fig. S3C). This MO should not affect the production of the Atf3-S isoform, the start codon of which is located 3′ to the MO position. The atf3 vivo-MO, or a control vivo-MO, were injected into the cerebral ventricle of Cas^CREAtlas adults, and Cas^CREAtlas fates were analyzed after 7 days (Fig. 5H-J). Hmg2bmCherry^pos clones were categorized when located within the first one or two cell rows below the ventricular surface (corresponding to the z-position of NSC progeny cells generated over a short time scale). We found that the atf3 vivo-MO induced a significant increase in the proportion of clones composed of NSCs only, at the expense of the generation of single neurons (Fig. 5K). These results together indicate that Atf3-L expression is necessary for lineage bias, either downstream of or in parallel to physiological non-apoptotic Cas3* events.

Cas3/Cas7 activation and Atf3-L appear, together, to be linked with the specific NSC fate choice of direct neuronal differentiation under physiological conditions. To address further the relevance of this regulatory process, we first tested whether experimental stimulation of Cas3/Cas7 activity in adult fish could modify NSC fate. CPT was used as an inducer. Incubation in CPT triggered Cas3* induction and was followed by NP death at larval stages [as revealed by the Tg(ubi:secA5-mVenus) reporter, whereby Annexin 5-mVenus expression marks cell death; van Ham et al., 2010] (Fig. S4A-B1). In striking contrast, pallial Cas3* cells were seen instead to delaminate when CPT was injected into the cerebral ventricle in adult animals (Fig. 6A-C1′). To track their fate longer term, CPT was applied to Cas^CREAtlas adults and clone types were assessed (Fig. 6D-G, Fig. S4C,D). We found a proportional increase of neuron-only clones (yellow in Fig. 6F′), mimicking the fate observed for physiological Cas^CREAtlas events in the adult pallium (Fig. 6G). Neuron-only clones were, in large majority (>70%), composed of single neurons (Fig. S4D). Over this short time frame, attached clones were virtually absent and the phenotype was observed at the expense of NSC clones (red in Fig. 6F′).

Finally, we tested whether Cas3*/Cas7* events were recruited under challenges that impact NSC state or fate. Quiescent NSCs are radiation resistant, possibly via an efficient mechanism of DNA repair (Barazzuol et al., 2019; Hellström et al., 2009; Mineyeva et al., 2019). However, radiation-induced differentiation has also been described (Konirova et al., 2019; Schneider et al., 2013). We used X-ray irradiation of live adults (3 mpf) to support this observation and test whether radiation resistance could also be accompanied by a NSC fate change in vivo. Short (1-h) treatment with a low radiation dose (5 Gy) induced γH2AX-positive foci in NSC/NP nuclei, indicative of double-strand DNA breaks and the recruitment of the repair machinery (Fig. 7A-B′). This process was complete by 2 h post-treatment (Fig. S5A-C″) with no visible effect on NSC fate and was observed until very high irradiation doses. At 40 Gy, a low number of cells located very close to the ventricular surface became Cas3*^pos after a 24-h chase (Fig. 7C-D1). These cells displayed a delaminating profile with a cell body partly displaced into the parenchyma (Fig. 7D2,D3) but sometimes keeping a ventricular attachment (Fig. 7D ₂-_2′) and were Sox2^neg. Considering that Cas3* induction may be rapid and transient (Andrews et al., 2016), we also conducted a minimal time series (9 h versus 24 h chase) with a 80 Gy dose, and observed a similar quantitative and qualitative phenotype at both time points (Fig. S5D-E′). Although we did not ascertain cell survival at later chase times, these results suggest that, upon irradiation at high dose, a few Cas3* events are induced and correlate with the first steps of neuronal commitment. Low to moderate irradiation schemes, however, lead to repair and do not recruit Cas3*.

Mechanical lesions applied to the adult pallium induce NSC division for neuronal repair (Baumgart et al., 2012; Kishimoto et al., 2012; Kroehne et al., 2011; März et al., 2011), accompanied with a partial fate shift towards symmetric neurogenic divisions (Barbosa et al., 2015). We tested whether atf3^pos or Cas3*/Cas7* events could also be involved in lesion response. atf3 expression appeared to be induced around the lesioned ventricular zone starting at 1 day post-lesion (dpl) and very prominently at 3 dpl (Fig. 7E-H). It was also massively downregulated in the contralateral hemisphere at 3 dpl, a phenomenon not reported yet for other lesion-responsive genes (Fig. 7H, arrowhead). Like under physiological conditions (Fig. 4C1-D4), atf3^pos cells in lesioned pallia displayed a delaminating morphology (Fig. 7I-J′). We next used Cas^CREAtlas to track the fate of Cas3*/Cas7* events in a lesioned context. At 7 dpl, Cas^CREAtlas fates recorded in lesioned hemispheres were biased towards neurogenic fates at the expense of NSCs, compared to Cas^CREAtlas fates in non-lesioned brains (Fig. 7K). Among these neurogenic fates, in lesioned, like in unlesioned brains, the vast majority (>70%) corresponded to single neurons (Fig. 7L). We conclude that, within 7 dpl, the neurogenic fate of NSCs that experience a Cas3* event is further enhanced, without change in neurogenic mode. We have not attempted to link Atf3 and Cas3* in this context.

In this work, we focus on adult NSC direct neuronal conversion and bring together several observations in link with this fate. First, we provide a quantitative, morphometric and molecular characterization of delamination/neuronal differentiation events in the NSC/NP population of the adult pallium. Second, we demonstrate that non-apoptotic Cas3*/Cas7* events occur in NSCs during development and homeostasis of the zebrafish pallium and, at adult stage, are preferentially associated with NSC fate choices of lineage termination by neuronal differentiation, among which the generation of single neurons is predominant. Upon parenchymal lesion or when induced by CPT, Cas3*/Cas7* events in NSCs are also further biased towards lineage termination and neuronal differentiation. Finally, we show that Atf3 is sufficient to trigger the direct generation of single neurons from NSCs and is necessary, downstream of or in parallel to Cas3*/Cas7*, for the realization of this fate. Together, our results provide a first molecular insight, with lineage and functional tracking in vivo, into NSC direct neuronal conversion. This fate decision balances fate choices to preserve homeostasis of the pallial NSC population over a lifetime (Than-Trong et al., 2020). Adding such a regulatory level beyond NSC division fate choices could add flexibility to the control of NSC numbers and neuronal production.

The Tg(gfap:Zo1-mKate);Tg(deltaA:egfp) intravital imaging dataset includes NSCs and NPs that cannot be distinguished, but we estimate (Fig. S1) that around half of the recorded events are NSCs, and show that NSC conversions follow a classical delamination process. The observed significant duration of this process overall may permit important regulations of cell state towards neurogenesis during this event (Baek et al., 2018). The fact that delaminating NSCs do not undergo apoptosis and convert into neurons, not directly assessed here, is supported by arguments from other works: apoptosis is not observed in the adult pallium under normal conditions (Barbosa et al., 2015; Webb et al., 2009), neurons are the sole parenchymal fate of pallial NSCs (Furlan et al., 2017; März et al., 2010), and delaminating NSCs express neuronal markers (Barbosa et al., 2015).

Our analysis further suggests that Cas^CREAtlas captures a measurable fraction of these events, which are characterized by the expression of a specific molecular pathway involving Cas3*/Cas7* activation and Atf3. Several arguments support this conclusion. First, the direct conversion events recorded by intravital imaging (a few dozen events in ∼800 NSCs in 40 days) and the number of Cas^CREAtlas clones (37-46 events in ∼1000 NSCs in 56 days), the majority of which are single neurons, are of the same order or magnitude. Second, Cas^CREAtlas activation is tracked from NSCs and generates persisting clones (Fig. 3). Third, CPT treatment reveals that induction of the apoptosis cascade is sufficient to trigger NSC delamination (Fig. 6), while Atf3 blockade impairs the expression of the direct neuronal conversion fate (Fig. 5G-K), together suggesting that a cascade involving Cas3* and/or Cas7* and Atf3 actually drives the neuronal conversion fate. Obviously, intravital imaging of Cas3*/Cas7*/Atf3 events would be useful to connect these molecular events directly to our morphometric description, and this important aspect is currently lacking. This is, however, not possible to achieve with our current tools, notably because Hmg2B-mCherry needs several days to be directly visible by fluorescence, which would bypass the initial steps of the fate process. An NSC transcriptomic state possibly linked with direct conversions was recently proposed based on in silico analyses of scRNAseq data (Mitic et al., 2024). This interpretation remains to be validated with lineage and functional assays in vivo, but testing its relationship with Cas3*/Cas7* and Atf3 could be interesting.

The fact that non-apoptotic Cas3*/Cas7* events are physiological components of NSC population fates is an important finding: such events do occur in NSCs of the adult pallium and bias NSC fates towards lineage termination, including, in a large part, the direct generation of neurons (Fig. 3). To our knowledge, this is the first demonstration of non-apoptotic Cas3*/Cas7* events associated with a specific SC fate choice in vivo. Our work also provides information on the effectors or facilitators of this process. The cleavage site of the Cas^CREAtlas construct is also recognized by Cas7*. casp7 is not expressed in the larval brain (Spead et al., 2018), arguing for a Cas3*-specific cleavage at larval stages (Fig. 2). However, both casp3a and casp7 are transcribed in quiescent NSCs (Fig. S3A). We observe a correlation between Cas3* IHC and Cas^CREAtlas events in the number and spatial distribution in response to CPT (Fig. 6), strongly suggesting that Cas^CREAtlas reads, in part, Cas3*. Finally, when Atf3 function is abrogated, Cas^CREAtlas is still induced in NSCs, but its associated fate bias is abolished (Fig. 5K), suggesting that Atf3 is either a mediator of non-apoptotic Cas3*/Cas7* events or a parallel and converging actor. There remain, however, several missing steps in our molecular understanding, among which a key issue is the link between Cas3*/Cas7* and Atf3. The apparent non-correlation of casp3a and atf3 transcripts under physiological conditions could mean that these two factors are functionally unrelated, or that they act in temporally distinct windows along the non-apoptotic process studied here. It remains that non-apoptotic Cas3*/Cas7* and Atf3 functionally interact, even if this does not prove their action along a single pathway. The relationship between non-apoptotic Cas3*/Cas7* and Atf3 is also not a simple hierarchy: overexpressed Atf3 can induce Cas^CREAtlas (although such experiments can mimic the activity of related factors) (Fig. 5F), while loss-of-function assays reveal that Atf3 is also necessary downstream or in parallel of Cas3* (Fig. 5K). Such dual positioning of Atf3 is reminiscent of previous observations; Atf3 has been described both upstream and downstream of Cas3* (Lu et al., 2007; Syed et al., 2005). Solving the relationship between Atf3 and Cas3*/Cas7* in adult NSCs will be an important next step to deepen our work, but this may require a more sensitive Cas3*/Cas7* readout, detecting Atf3 protein, or tracing Atf3^pos events over time. Conducting the same comparison under challenge (such as CPT, irradiation or lesion), and epistasis analyses between Cas^CreAtlas events and Atf3 expression in these contexts, could provide further information. However, with our tools fish survival proved to be technically challenging. Atf3 targets in adult NSCs also remain to be identified, as well as Cas3*/Cas7* effectors, given that blocking Atf3 does not fully abolish direct neuronal differentiation (Fig. 5K). We also do not identify the functionally relevant Cas3*/Cas7* substrates that will bias NSC fate. Intense DNA damage, which can be caused by Caspase activation (upon cleavage of the caspase-activated DNase inhibitor) and trigger cell differentiation, e.g. in muscle (Larsen et al., 2010), is here efficiently repaired in adult pallial NSCs (Fig. 7A-D3). Finally, Cas^CREAtlas is also induced in NSCs that follow other endogenous fates, such as symmetric NSC-NSC divisions (Fig. 3D). Thus, Cas3*/Cas7* alone is not sufficient to encode the direct differentiation fate in NSCs and a further level of regulation must exist.

Finally, we show that Cas3*/Cas7* events, and the associated fate of direct neuronal differentiation, can be engaged under non-physiological conditions of NSC stress (e.g. ionizing radiations) or neuronal repair (e.g. mechanical lesion). In the latter situation, the biased fate towards lineage termination of Cas3*/Cas7*^pos NSCs appears to be enhanced (Fig. 7K), which may accelerate neuron generation for repair. This occurs without apparent change in the neurogenic mode of Cas^CreAtlas events, i.e. a predominant generation of single neurons (Fig. 7L). As such, it differs from the post-lesion symmetric neurogenic divisions described by Barbosa et al. (2015).

At present, we do not know what induces Cas3*/7*/Atf3 pathways physiologically. In adult mouse muscle satellite cells, direct SC differentiation is increased upon abrogation of Notch signaling or under regeneration (Bjornson et al., 2012; Ismaeel et al., 2024; Mourikis et al., 2012). Notch3 signaling is also a major gatekeeper of NSC quiescence in the adult pallium (Alunni et al., 2013; Chapouton et al., 2010) and whether abrogating this pathway or quiescence will result in mobilizing direct neuronal differentiation fates remains to be tested. Neuronal differentiation and (neuro)epithelial delamination, as undergone by delaminating NSCs in the adult pallium, also involve major cellular remodeling (Kasioulis and Storey, 2018; Kuijpers and Hoogenraad, 2011), which may exploit Caspase (Unsain and Barker, 2015) or Atf3 (Rohini et al., 2018) activities. Finally, it will also be interesting to determine whether the neurons issued from direct neuronal conversion of NSCs have specific structural features and identity.

All tools and reagents are listed in Table S1.

Wild-type (AB) and Tg(GFAP:eGFP) (Bernardos and Raymond, 2006), Tg(ubi:SecA5) (van Ham et al., 2010), Tg(her4:ERT2CreERT2) (Boniface et al., 2009), Cas^CreAtlas (see below), Tg(βactin:lox-stop-lox-hmg2B-mcherry) (Wang et al., 2011) transgenic zebrafish were used. Embryos/larvae up to 5 dpf were maintained and staged as described (Kimmel et al., 1995). Adult zebrafish were maintained using standard fish-keeping protocols by the Ethics Committee n°39 of Institut Pasteur (authorization #36936) and DDPP-2021-921 of the Direction Départementale de la Protection des Populations de Paris.

The mCD8-DEVD-V5-Cre (CDVC) construct was PCR-generated by fusing in-frame the mCD8-Diap1 region of a plasmid encoding Drosophila Casexpress DQVD (Ding et al., 2016) to a V5-CRE recombinase cassette and subcloned into the Tol2-kit vector pME-MCS to generate pME-CDVC. pME-CDVC was used to generate the transgenesis vector pTol2-HCDVC (her4:mCD8-diap1-V5-CRE-SV40pA) using the L/R recombinase reaction and the Tol2 vectors p302 and p395, and the 5′ vector N11. In the final product, the DEVD caspase site was mutagenized to GSGC to generate the control plasmid pTol2-HCDVC* by the Round-the-horn mutagenesis method (https://openwetware.org/wiki/%27Round-the-horn_site-directed_mutagenesis). Transgenic lines were made by injecting one-cell embryos with a mix containing 60 ng/µl of plasmid and 60 ng/µl of transposase capped RNA.

The atf3 cDNAs was amplified from reverse-transcribed 16 hours post-fertilization (hpf) embryo RNA and subcloned into pSCA. The atf3 RNA probe was generated from the long version of the RNA (full coding sequence). The atf3-P2A-GFP constructs were generated with the Gibson method and subcloned into the pCMV5 vector using the NEBuilder^® HiFi DNA Assembly Cloning Kit.

cDNA, extracted from 24 hpf embryos, was amplified by RT-PCR using primers atf3_FL_fwd and atf3_FL_rev and the following cycles: 98°C, 1 min; 98°C, 30 s; then 35 cycles of 98°C, 10 s; 63°C, 30 s; 72°C, 20 s; then 72°C, 2 min.

Brains were dissected in 1× PBS at 4°C, their tela choroida was manually removed and the brains were directly transferred to a 4% paraformaldehyde solution in PBS for fixation. They were fixed overnight at 4°C under permanent agitation. After four washing steps in PBS, brains were dehydrated through a 5-10 min series of 25%, 50% and 75% methanol diluted in 0.1% Tween-20 PBS solution and kept in 100% methanol at −20°C. Rehydration was performed using the same solutions, and then brains were processed for whole-mount IHC. After rehydration, the telencephala were dissected out and subjected to an antigen retrieval step using HistoVT One for 1 h at 65°C. Brains were rinsed three times for at least 10 min in a 0.1% DMSO and 0.1% Triton X-100 PBS 1× solution (PBT) and then blocked with 4% normal goat serum in PBT (blocking buffer) 4 h at room temperature on an agitator. The blocking buffer was later replaced by the primary antibody solution (diluted in blocking buffer), and the brains were kept overnight at 4°C on a rocking platform. The next day, brains were rinsed five to ten times over 24 h at room temperature with PBT and incubated in a solution of secondary antibodies diluted in PBT overnight, in the dark, and at 4°C on a rocking platform. In some instances, to be able to use the ZO1 dye-coupled antibody in the presence of another primary mouse mAB, secondary antibody free sites were blocked by incubation with 2% mouse serum in PBT for 1 h before applying the dye-coupled ZO1 antibody. After three rinses in PBT over 4 h, brains were transferred into PBS. Dissected telencephala were mounted in PBS on slides using 0.5 mm-thick holders. The slides were sealed using a glue gun.

Primary antibodies were used at a final concentration of 1:1000 for chicken anti-GFP and chicken anti BrdU, 1:500 for dye-coupled-ZO1, 1:300 for Casp3a, 1:500 for RFP, 1:200 for Sox2, ZO1 and mAb anti-GFP, and 1:100 for γH2AX. Secondary antibodies were all used at a final concentration of 1:1000.

ISH was performed as described previously (Bosco et al., 2013; Chapouton et al., 2010; Ninkovic et al., 2005) except for the additional presence of 5% dextran sulfate during the hybridization phase. For combined ISH and IHC, the ISH was developed using Fast Red. See Table S1 for details of antibodies and probes used in this study.

Micro-injections into the adult brain ventricle were performed on anaesthetized fish as described (Rothenaigner et al., 2011) except that DNA was injected at the midbrain midline to avoid damaging the pallium. vivoMOs (Gene Tools) were injected at a concentration of 0.125 mM. Micro-injections into 4 dpf larval hindbrain ventricle were performed on anesthetized larvae immobilized in 4% methyl cellulose.

For electroporation, plasmid DNA was diluted to 1 µg/µl in 0.1×PBS and injected into the ventricle. Electrodes (Tweezertrodes, 5 mm platinum) were placed on each side of the fish head. Fish were then administered two electric pulses (70 V, 50 ms width, 1000 ms space).

4-OHT treatment was performed as previously described (Mosimann et al., 2011) on her4:Ert2CreErt2, ßactin:LoxSTOPloxhmgbmCherry. Clonal recombination conditions at 1 mpf were 10 min with 0.5 µM 4-OHT as described by Than-Trong et al. (2020. They were followed by a 4 h pulse of 1 mM BrdU. Cas^CreAtlas were only treated with 1 mM BrdU for 4 h. Fish were then washed four times, transferred into fresh fish water, and grown as usual until 2 months of age.

CPT was dissolved as a 10 mM stock in DMSO, aliquoted and stored frozen until use. Just before use, an intermediate solution was prepared in DMSO and further diluted 50 times in PBS (for injections) or fish water (for incubations) to reach the working concentration.

Fish (3-5 months old) were placed into 2-l cages inside an X-ray irradiator (Gulmay CP160/10, 250 kV, 12 mA) and irradiated for the length of time needed to reach the expected dose (5 Gy= 212 s). Controls were placed for the same duration in the chamber but left untreated. Following irradiation, fish were kept in 6-l cages in a 28°C incubator for the desired length of time. Fish were fed once a day and water changed daily if needed. Fish brains were then dissected, their tela choroida was removed, and the brains were then fixed for ISH or IHC.

Images of whole-mount immunostained telencephali were acquired on a confocal microscope (LSM700 and LSM710, Zeiss) using a 20× objective or a 40× oil objective (Plan-Apochromat 40×/1.3 Oil M27) and tile images of four to eight z-stacks were stitched with ZEN2009 software. 3D renderings were generated using Imaris® software (versions 8 and 9, Bitplane). Vertical plane images were extracted when needed. Cas^CREAtlas clones were resolved manually using 3D rendering and the slice mode and highlighted in different colors. For Fig. 7A-D and Fig. S3, a nuclear mask was created in Imaris using Sox2 nuclear staining in order to quantify nuclear H2AX staining. For Fig. 4D, single planes at different depth were extracted.

For dorsal whole-mount views of the telencephalon (ISH in blue), images were taken using a Nikon macroZoom.

All experimental data were analyzed using Prism software and are expressed as mean±s.e.m. Significance was set at P<0.05. Global comparison of proportions between experimental and control conditions (Figs 3D, 5D-E, 6H, 7K,L, Figs S2B, S4C) were performed via a contingency test based on a χ² analysis. When a statistically significant difference was detected, the specific cell fate category(ies) significantly different between experimental and control datasets was determined using a Fisher's exact test with Bonferroni correction to account for multiple comparisons (corresponding to the number of categories tested). The number of fish tested is indicated in the legend for each figure; we report the number of hemispheres tested, and always only analyzed one hemisphere per fish. All counts are provided in Table S2.

We thank the ZEN team for input, Isabelle Foucher for expert assistance in particular with the generation of her4:ERT2CreERT2-driven clones, Sara Ortica for conducting ISH and IHC experiments to analyze lesioned brains and cas3/atf3 expression for the revision of this paper, Nathan Guibert for maintaining the Cas^Cre Atlas background, Emmanuel Than-Trong for his specific re-analysis of 4-OHT-induced clones after a 4-week chase (for comparison with Fig. 3D), Laure Mancini for initially producing the movies that were here re-analyzed in Fig. 1 and Fig. S1 for direct differentiation events, the Institut Pasteur Irradiator service platform (Philippe Casanova and Claire Mallet) for expert assistance with X-ray use (for Fig. 7A-D3 and Fig. S5), and Emeline Perthame and Pascal Campagne from the Institut Pasteur Bioinformatics platform for their help in the choice of tools for statistical analyses (for Figs 3D, 5D,E,K, 6G and 7K,L and Figs S2B and S4D). We thank Denise Montell for sharing constructs, and Tjakko van Ham for sharing the Tg(ubi:SecA5) line. We are also greatly indebted to Romain Levayer and his lab, to Miria Ricchetti and Shahragim Tajbakhsh for insightful discussions and suggestions, and to Romain Levayer for his critical reading of the manuscript.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.204381.reviewer-comments.pdf