Awakening adult neural stem cells: NOX signalling as a positive regulator of the quiescence-to-proliferation transition in the Xenopus retina

The intracellular redox state recently emerged as an intrinsic regulatory mechanism for adult and embryonic stem cell self-renewal, proliferation and differentiation (Bigarella et al., 2014; Sinenko et al., 2021). Among molecular species that contribute to the cellular redox status are reactive oxygen species (ROS), such as the superoxide ion (O₂^.−), hydrogen peroxide (H₂O₂), or the highly reactive hydroxyl radical (^.OH). ROS are mainly produced in cells through the mitochondrial electron transport chain or the activity of NADPH oxidase complexes (NOX1-5 and DUOX I-II). Under normal physiological conditions, ROS levels are tightly regulated by different enzymes, including superoxide dismutases, catalases, glutathione reductases/peroxidases, thioredoxins, thioredoxin reductases or peroxiredoxins (Marengo et al., 2016). Oxidative stress occurs when ROS generation outpaces this redox machinery. The resulting accumulation of high levels of oxygen free radicals and radical-derived species leads to premature ageing, cell death or cancer through oxidative damage of cellular components, including DNA. Despite being harmful, ROS also constitute essential modulators of cell behaviours. At basal levels, these molecules are indeed involved in reversible reduction-oxidation processes known to modulate key cellular signalling pathways (Bigarella et al., 2014; Marengo et al., 2016).

Deciphering how stem cells maintain their redox homeostasis and how ROS in turn impact stem cell activity is highly relevant in different fields from regenerative medicine to cancer therapy. With regard to oxidative stress protection, it is known that cancer stem cells exhibit an intracellular redox balance distinct from that of differentiated cancer cells, which makes them resistant to ROS-mediated cell killing (Kim et al., 2019). The molecular basis of such resistance is not well understood, but it is likely shared by embryonic and adult stem cells, and has even been proposed as an innate stemness feature (Guo et al., 2010; Madhavan et al., 2006). Among the prevailing assumptions to explain the lower vulnerability of stem cells to oxidative mutagenesis are stronger antioxidant defence mechanisms, their location in hypoxic niches and/or a glycolytic metabolism occurring even in the presence of oxygen (known as the Warburg effect; Vander Heiden et al., 2009). These features presumably contribute to the low ROS content reported in embryonic pluripotent stem cells or adult stem cells, such as haematopoietic or mesenchymal stem cells (Bigarella et al., 2014; Sinenko et al., 2021). In these cell types and their respective progeny, a gradual increase of ROS levels occurs along the differentiation process. This scheme might, however, not apply to all cycling cell populations. Neural stem cells of the mammalian brain were indeed found to exhibit higher ROS levels compared with progenitors in ex vivo assays (Adusumilli et al., 2021; Le Belle et al., 2011). Besides, some studies suggest that ROS levels may vary depending on stem cell functional states (quiescent, primed or activated). Here again, data from the literature do not support a unique redox rule for all stem cell types. In the haematopoietic system, low endogenous ROS content is associated with a greater quiescence (Jang and Sharkis, 2007). In contrast, a study combining single-cell transcriptomic and ROS content labelling recently highlighted that quiescent neural stem cells from the dentate gyrus exhibit higher ROS levels compared with activated ones (Adusumilli et al., 2021). Of note, in these studies the use of different sensors with differential sensitivities for particular ROS species might explain these apparent discrepancies.

From a functional point of view, fluctuations of ROS above or below their physiological levels have been shown to affect a variety of processes, including stem cell reprogramming (Zhou et al., 2016), self-renewal, proliferation and fate (Bigarella et al., 2014; Prozorovski et al., 2015; Rampon et al., 2018). However, somehow conflicting data have been reported. For instance, increased ROS levels trigger a dramatic decline of haematopoietic stem cell number, associated with enhanced cycling and premature differentiation (Prieto-Bermejo et al., 2018; Suda et al., 2011). In contrast, high ROS levels are likely required for proper self-renewal of neural stem cells in the mammalian subventricular zone (Le Belle et al., 2011). Besides, despite an ever-increasing number of studies regarding ROS involvement in stem cell biology, few addressed their function in vivo and most of them relate to the regeneration field (Love et al., 2013; Gauron et al., 2013; Ferreira et al., 2016; Zhang et al., 2016; Hameed et al. 2015; Tao et al., 2016). We here sought to investigate redox status and ROS function in vivo, in adult neural stem cells of the Xenopus retina.

In contrast to the mammalian situation, the amphibian retina harbours continuously active retinal stem cells (RSCs), which sustain tissue growth throughout the animal's life and can be further recruited following damage to regenerate lost cells (Hidalgo et al., 2014; Miyake and Araki, 2014). The RSC niche is located in a region called the ciliary marginal zone (CMZ), the spatial organization of which is well defined. Slow cycling multipotent stem cells reside in the most peripheral part of the CMZ, followed more centrally by transit amplifying progenitors and then by their postmitotic progeny (Perron et al., 1998). Albadri and collaborators recently reported that H₂O₂ production is dynamically regulated during retinogenesis and within the zebrafish CMZ. Focusing on progenitor cells, they further showed that 9-HSA, a by-product of lipid peroxidation, regulates the switch of these cells from proliferation to differentiation (Albadri et al., 2019). The specific role of ROS in RSCs was, however, not addressed in this study. Here, we first describe that RSCs are endowed with a peculiar redox status characterized by higher ROS levels compared with retinal progenitors. This is associated with the specific expression of several redox genes. Second, our functional analyses reveal a crucial role of NOX as a homeostatic regulator of the proliferation rate within the post-embryonic stem cell pool, which limits the ratio of quiescent cells, while likely fine-tuning the mean division speed within the proliferative cohort. We further demonstrate that this function is mediated by modulation of the balance between the canonical Wnt pathway and Hedgehog signalling.

Using a ratiometric sensor transgenic line, Albadri and collaborators recently showed that the embryonic zebrafish retina exhibits higher H₂O₂ concentrations in the proliferating peripheral margin (which includes the forming CMZ), compared with the differentiating central part of the tissue (Albadri et al., 2019). To assess whether this observation holds true in the Xenopus post-embryonic retina, we adapted a protocol set up by Owusu-Ansah et al. (2008) to image ROS levels in vivo, using the fluorescent sensor dihydroethidium (DHE). We first tested DHE specificity by treating embryos 2 h with rotenone (ROT), an inhibitor of the mitochondrial electron transport chain complex I, known to increase O₂^.− production (Li et al., 2003) (Fig. 1A). As expected, ROT-treated embryos exhibited a significant increase in fluorescence intensity, suggesting that DHE staining indeed reflects intracellular ROS levels in living Xenopus embryos (Fig. 1B,C). We next analysed DHE fluorescence on retinal sections at the end of embryogenesis (Fig. S1A). In line with zebrafish results, we found a significantly higher fluorescent signal within the CMZ compared with the differentiated neural retina (Fig. S1B,C). Of note, intensity within the CMZ could be enhanced by a 2-h ROT treatment, confirming the specificity of the staining (Fig. S1B,C). Further quantitative analysis in these embryos revealed no difference in labelling between stem cells and progenitors (Fig. S1C). At post-embryonic stages, however, a peripheral-to-central gradient of fluorescence was clearly observed with a maximal intensity at the tip of the CMZ, where stem cells reside (Fig. 1D-F). These data show that, similarly to the zebrafish situation (Albadri et al., 2019), ROS levels are dynamically regulated within the Xenopus retina, both temporally (embryonic versus tadpole stages) and spatially (peripheral versus neural retina). They further highlight that post-embryonic RSCs exhibit a particularly high ROS content compared with progenitors or differentiated neurons.

We next wondered whether such high ROS content might be associated with the presence of a specific redox machinery. In this purpose, we first cloned several genes found in Xenopus databases, encoding superoxide dismutases 1-3 (sod1/2/3), catalase 2 (cat2), glutathione peroxidases 1/4/7 (gpx1/4/7) and peroxiredoxins 1-6 (prdx1-6). Superoxide dismutases are responsible for O₂^.− reduction into H₂O₂, whereas catalases, glutathione peroxidases and peroxiredoxins are involved in hydroperoxide detoxification. To our knowledge, expression patterns of cat2, sod1/2/3 and gpx1/4/7 genes have not yet been described in developing Xenopus laevis embryos. Those of Prdx genes have been published, but retinal sections were not analysed (Shafer et al., 2011). Interestingly, none of the tested genes exhibited a uniform expression profile, but instead their mRNA proved to be enriched in specific developing tissues and organs (Fig. S2A-D). Of note, several were strongly expressed in regions with high ROS levels (such as the somites or the pronephros), as inferred from DHE staining (Fig. 1B). In the eye, gpx1 showed a lens-specific expression, no obvious signal could be detected for gpx7, and prdx5 was found to be broadly expressed in the retina (Fig. S2C,D). All the others showed clear enrichment at the tip of the CMZ, as inferred from the ring-shaped labelling around the lens. Except for prdx2, sod1 and sod2, this was associated with staining of the midbrain-hindbrain boundary (Fig. 2A), an expression pattern typical of RSC markers (Parain et al., 2012). Further analysis on retinal sections did not allow the sod2 and sod3 expression profiles to be assessed as the signal was undetectable. However, it confirmed that all other genes were expressed in the stem cell-containing region (at high levels for prdx1/2/3/6 and cat2; Fig. 2B; at very low levels for sod1, gpx4 and prdx4; Fig. S2F). Their expansion within the CMZ significantly differed from one to another (Fig. 2C). Three distinct profiles could be distinguished: a homogeneous expression in the CMZ (prdx2), an enriched expression at the CMZ tip extending further at lower levels (prdx1), or an expression strictly limited to the CMZ tip (encompassing stem cells and probably young progenitors; cat2 and prdx3/6). These results support the idea of distinct and tightly regulated redox toolkits within the different CMZ domains and highlight cat2, prdx3 and prdx6 as specific and highly expressed RSC markers.

We next examined the expression profile of NOX genes (Fig. 2D-F, Fig. S2E). No nox3 orthologue could be found in Xenopus databases and we thus focused on nox1, 2, 4 and 5. All of them were found expressed in both the embryonic and post-embryonic eye, as assessed by semi-quantitative PCR (Fig. 2E). In situ hybridization analysis did not reveal any eye staining for nox2 (Fig. S2E). Such discrepancy with the PCR result suggests that this gene might be expressed broadly but at low levels in the retina, as reported in zebrafish (Weaver et al., 2016). In contrast, a diffuse labelling was observed in the eye for nox1 and nox5, whereas nox4 showed a very specific expression in a thin domain surrounding the lens (Fig. 2F). Further examination on retinal sections revealed nox4 staining in lens progenitor cells lining the CMZ and confirmed diffuse expression of nox1 and nox5 in the neural retina, with higher levels in the CMZ (Fig. 2F). Thus, RSCs do express at least nox1 and nox5, albeit without particular enrichment compared with progenitor cells of the CMZ.

The specific expression of several redox genes in RSCs suggests a tight control of redox homeostasis in these cells, thereby raising the question of ROS signalling functions with respect to stemness features. To address this issue, we first set up optimal conditions to reduce ROS production in vivo, using two pharmacological NOX inhibitors, diphenyleneiodonium (DPI) and apocynin (APO), and the mitochondria-targeted ROS scavenger mitotempo (MITO). Working concentrations were chosen based on previous reports in Xenopus or zebrafish (Love et al., 2013; Peterman et al., 2015) and did not result in any obvious developmental defects (data not shown). To assess drug specificity and efficiency, tadpoles were treated for 16 h and then stained with either DHE or with the ROS sensor CM-H₂DCFDA (Fig. S3A). As expected, fluorescence of both sensors was lower in whole embryos following exposure to the pharmacological compounds (Fig. S3B-E). Further quantification of DHE signal on retinal sections confirmed a significantly decreased fluorescence intensity within the CMZ (Fig. S4).

We then investigated the impact of lowering mitochondrial or NOX-dependent ROS production on CMZ cell proliferation. Tadpoles were treated with either MITO or APO for 16 h and then subjected to a 5-ethynyl-20-deoxyuridine (EdU) incorporation assay (Fig. 3A). We next determined the percentage of EdU-positive cells within the whole CMZ (mainly composed of progenitors) or in its most peripheral part, which includes quiescent and self-renewing stem cells (Tang et al., 2017; Wan et al., 2016). MITO treatment did not affect EdU incorporation in either progenitors or RSCs (Fig. 3B,C). This suggests that CMZ cell proliferation is not highly dependent on mitochondrial ROS levels. APO treatment, however, generated a different result. Although the proportion of EdU-labelled cells was unchanged among progenitors, it was significantly reduced within RSCs compared with the control (Fig. 3D,E). DPI treatment generated the same phenotype (Fig. S5), suggesting distinct sensitivities of RSCs and progenitors to lowered NOX-derived ROS production.

In order to assess further the absence of proliferative defects among progenitors, we examined the consequence of a longer APO treatment on the CMZ surface (as an indication of total cell number) and neuronal production. Of note, we first confirmed through DHE staining that APO exposure for extended time periods still resulted in decreased ROS levels (Fig. S6A,B). Quantification of CMZ area following a 3-day APO treatment did not reveal any significant difference compared with the control (Fig. S6C,D). In addition, an EdU pulse-chase experiment showed that the total number of newborn cells produced by the CMZ over a period of a week was similar in APO-treated and control retinas (Fig. S6E-G). Among these newborn cells, the proportion of photoreceptors, interneurons and retinal ganglion cells was also found unchanged (Fig. S6H). This suggests that reducing NOX signalling does not impair CMZ neurogenic activity, at least in the examined time window.

We next aimed to confirm the effect of NOX inhibition by genetic means. To this end, we took advantage of a previously described translation-blocking morpholino directed against cyba mRNA (Love et al., 2013). The cyba gene encodes a regulatory subunit (also called p22^Phox) of NOX complexes 1-4 (Vermot et al., 2021) and possibly also of the NOX5 complex (Marzaioli et al., 2017). We found it to be expressed in both the post-embryonic and embryonic eye, as assessed by quantitative real-time PCR (qPCR) or in situ hybridization (Fig. S7A,B). Analysis on retinal sections revealed an enrichment of cyba expression within the CMZ and in lens progenitors (Fig. S7B). Before functional analyses, we verified that the cyba morpholino was able to inhibit the translation of a Flag-tagged cyba reporter construct (Fig. S7C,E). We next injected control or cyba morpholinos at the two-cell stage and analysed EdU incorporation on retinal sections at the end of embryonic retinogenesis. Of note, cyba knockdown frequently led to developmental defects affecting the ventral portion of the retina. We thus limited our quantification to the dorsal part of the CMZ. As observed with APO or DPI treatment, the proportion of EdU-labelled cells among progenitors was unchanged in cyba morphant retinas compared with control retinas (Fig. S7G,H). The percentage of EdU-labelled RSCs was, however, significantly decreased (Fig. 3F-H). Importantly, the phenotype specificity was verified by a rescue experiment, in which the cyba morpholino was co-injected with a morpholino-insensitive Myc-tagged cyba mRNA (Fig. 3F-H, Fig. S7D,F). These data, confirming those obtained using pharmacological compounds, firmly demonstrate the requirement of NOX signalling for proper RSC proliferation. It also highlights again the greater resilience of progenitors to reduced NOX-dependent ROS production.

We then wondered whether the reduced number of EdU-labelled RSCs observed upon NOX inhibition might reflect their depletion. To address this question, we first assessed cell death following APO or DPI treatment. TUNEL assay (data not shown) or labelling against cleaved caspase-3 did not reveal any significant increase of apoptotic cells in either the CMZ or the neural retina upon 16 h of drug exposure (Fig. S8A-C). After a 3-day APO treatment, a higher (although still limited) number of dying cells was found among treated progenitors and differentiated neurons compared with control ones, showing some toxicity of the drug. However, no apoptosis was detected within the RSC niche (Fig. S8D-F). Therefore, inhibiting NOX signalling does not impair RSC survival.

We next analysed the expression of several RSC markers [myc, prdx3, hes1, hes4 and yap (yap1)] by whole-mount in situ hybridization after a 16-h APO or DPI treatment (Fig. 4A). We also included the examination of atoh7, a progenitor marker. None of them showed altered staining in toto, reinforcing the view that progenitor cell number is not affected by reduced ROS levels, and suggesting that the RSC pool is conserved despite its lower proliferation rate (Fig. 4B). Confirming this result, no change in expression was observed for hes1, hes4 and yap following quantification on retinal sections (Fig. 4C,D) or through qPCR on eye extracts (Fig. S8G,H). To reinforce these data, we next extended APO exposure to 3 days, hypothesizing that a potential depletion of stem cells might take longer. Compared with the shorter treatment, this regimen did not aggravate the decreased EdU incorporation observed in RSCs (Fig. 4E,F and compare with Fig. 3E). In addition, qPCR analysis revealed no decrease of RSC marker expression. yap was still properly expressed, and levels of hes1, and to a lesser extent hes4, were even found to be increased (Fig. 4G,H). Taken together, these data suggest that NOX inhibition does not affect RSC maintenance.

Interestingly, the transcriptional repressor Hes1 is known to actively maintain neural stem cell quiescence in the mouse brain (Marinopoulou et al., 2021; Sueda et al., 2019). The combination of increased hes1 expression and decreased EdU incorporation observed upon APO treatment led us to hypothesize that NOX inhibition might trigger RSC quiescence. To address this question, tadpoles were pre-treated with APO for 24 h and then continuously exposed to the drug in the presence of EdU for a period of 2, 3 or 7 days (Fig. 5A). Such a protocol allows all cycling cells to be labelled as soon as the EdU pulse duration exceeds the total cell cycle duration, thereby giving an evaluation of the growth fraction (proportion of proliferating cells in a given population). The percentage of EdU-labelled RSCs was found to be stable in both control and APO-treated retinas at the three time points, showing that the plateau was already reached after a 2-day pulse. However, the labelling index was systematically and significantly reduced following APO treatment (Fig. 5B,C). Such a decreased growth fraction reinforces the idea that inhibiting NOX signalling pushes RSCs into quiescence. Altogether, these data strongly suggest that NOX signalling is required to limit the proportion of quiescent RSCs within the CMZ.

Because a subset of APO-treated RSCs still proliferate, we next examined their cell cycle kinetics by assaying cumulative labelling following 1-72 h of EdU exposure (Fig. 6A). This well-established technique allows total cell cycle (T_C) and S-phase (T_S) durations to be determined (Locker and Perron, 2019). Here again, the labelling index of APO-treated RSCs was consistently lower compared with that of controls, in both the linear part of the curve and after the plateau was reached (Fig. 6B). Confirming the previous results (Fig. 5C), the growth fraction, estimated using the Nowakowski Excel sheet (Nowakowski et al., 1989), dropped from 60% to 33% (Fig. 6C). In addition, we found that the duration of G2+M+G1 (T_C−T_S) was also decreased following APO treatment, as inferred from the time point at which the labelling index reached the plateau (Fig. 6B). Calculation of T_S and T_C revealed that both parameters were reduced compared with the control situation (−63% for T_C and −52% for T_S; Fig. 6C). These data thus show that APO-treated RSCs that remain proliferative exhibit enhanced cell cycle speed compared with the control pool of proliferative cells.

In order to dig into the molecular mechanisms underlying NOX-dependent control of RSC proliferation, we assayed whether the canonical Wnt pathway might be impacted by NOX inhibition. We considered it a good candidate as we previously showed its requirement for RSC proliferative activity (Borday et al., 2012; Denayer et al., 2008) and because it was reported to be positively modulated by ROS in other cellular contexts (Rampon et al., 2018). In order to examine Wnt activity, we took advantage of a Xenopus tropicalis transgenic reporter line, in which GFP expression is driven by a synthetic promoter harbouring seven optimal binding sequences for LEF-1/TCF (Fig. 7A; Tran and Vleminckx, 2014; Borday et al., 2018). We first aimed at verifying in this species that NOX inhibition yields the same phenotype than that observed in Xenopus laevis. We found indeed that APO treatment decreased DHE fluorescence within the CMZ and reduced EdU incorporation in RSCs, leaving progenitors unaffected (Fig. S9A-E). We next exposed transgenic embryos to APO or DPI for 16 h and analysed GFP expression on retinal sections. Compared with controls, treated individuals exhibited a significantly reduced GFP staining within the CMZ. Quantification within the RSC niche revealed a similar decreased intensity (Fig. 7B-D, Fig. S9F-H). To support these data, we then assessed Wnt activity following cyba knockdown. As observed with pharmacological treatments, morphant retinas displayed lower levels of GFP expression compared with control ones (Fig. 7E-G). Finally, as a complementary approach to evaluate Wnt activity, we examined the expression of ccnd1 (encoding cyclin D1), an established Wnt transcriptional target gene in different tissues (Shtutman et al., 1999), including the Xenopus CMZ (Borday et al., 2012). Although ccnd1 levels were similar to controls following a 16-h APO treatment, a significant downregulation was observed after 3 days (Fig. S9I,J). Altogether, these data demonstrate that the Wnt pathway is positively modulated by NOX within the CMZ.

We then investigated whether the decreased Wnt activity induced by NOX inhibition might be causal to the observed RSC proliferation defects. To this end, we undertook a rescue experiment consisting of forcing Wnt activity in APO-treated embryos (Fig. 8A). Canonical Wnt signalling activation was achieved by overexpressing an inducible and constitutively active form of TCF3 (TCF3-VP16GR; Borday et al., 2012), a transcriptional effector acting at the nuclear endpoint of the pathway. At the chosen injected low dose, TCF3-VP16GR by itself did not modify the proportion of EdU-labelled RSCs compared with the control situation. However, it efficiently rescued the decreased EdU incorporation observed upon APO treatment (Fig. 8B,C). This highlights that the effect of NOX on RSC proliferation is mediated, at least in part, by modulation of Wnt signalling.

We previously showed that CMZ cell proliferation is tightly regulated by the antagonistic interplay between Wnt and Hedgehog signalling (Borday et al., 2012). As both pathways inhibit each other, we next addressed whether Hedgehog activity might also be impacted upon NOX inhibition. We thus assayed expression of the Hedgehog target gene patched 1. We found it to be upregulated, as inferred from qPCR analysis (Fig. 8D,E). NOX-dependent decrease of Wnt signalling is thus likely accompanied by enhanced Hedgehog activity. We then assessed whether decreasing Hedgehog activity might rescue the RSC proliferative defect induced by NOX inhibition. To this end, embryos were treated for 24 h with APO and cyclopamine, an antagonist of the Hedgehog receptor Smoothened (Fig. 8F). In these conditions, the percentage of EdU-positive RSCs indeed returned to the basal level observed in control embryos (Fig. 8G).

Altogether, these data reveal that the Wnt/Hedgehog balance is regulated downstream of NOX, and strongly suggest that modulation of these pathways accounts for the effects of ROS signalling on RSC proliferative activity.

Understanding how neural stem cells regulate their proliferative behaviour is a major issue for both regenerative medicine and cancer therapy. In this context, the amphibian retina model holds great advantage as it contains, like the fish, active neural stem cells, and is accessible to in vivo investigations. We here sought to decipher how the redox state might impact RSC activity. We first report that these cells exhibit high levels of ROS and express a specific set of redox genes. Our functional investigations, using both pharmacological inhibitors and morpholino-mediated cyba knockdown, further reveal that NOX signalling takes central stage in preventing quiescence of RSCs, whereas mitochondrial ROS production does not seem to alter their proliferative behaviour. Finally, we propose that NOX effects are mediated through regulation of the Wnt/Hedgehog signalling balance (Fig. 9). Altogether, our study highlights NOX signalling as a previously unappreciated player of the regulatory network that controls neural stem cell proliferative activity within the post-embryonic neurogenic niche of the retina.

Several studies have described that the route of neural stem cells towards terminal differentiation during development is accompanied by an increase in ROS levels (due to a shift from glycolytic to oxidative metabolism), an enhancement of NOX activity, and an overall weakening of the antioxidant capacity (Olguín-Albuerne and Morán, 2018). In contrast to these data, Albadri and collaborators recently revealed in zebrafish that ROS levels are higher in the proliferating retinal margin compared with the differentiated neural retina (Albadri et al., 2019). We here confirm such a peripheral-to-central gradient and further highlight that post-embryonic RSCs exhibit higher ROS content compared with neural progenitors. This is highly reminiscent of the adult mouse brain situation, where cells exhibiting the highest ROS level were found to correspond to self-renewing multipotent neural progenitors with phenotypic characteristics of neural stem cells (Adusumilli et al., 2021; Le Belle et al., 2011). These findings and ours suggest that such an oxidative status constitutes a specific and conserved feature of adult neural stem cells. Associated with the observed high ROS content, we also found that several genes involved in hydroperoxide detoxification are highly enriched or specifically expressed in RSCs. This is consistent with the general idea that stem cells exhibit stronger antioxidant defence. However, several antioxidant proteins, such as peroxiredoxins or thioredoxins, not only serve as guardians against oxidative stress but also, once oxidized by ROS, as mediators of redox signalling. Peroxiredoxin 6, for instance, is a multitasking enzyme that modulates several pathways (Arevalo and Vázquez-Medina, 2018) and was recently proposed to promote stem-like properties in small-cell lung cancer (Xu et al., 2019). The specific expression of several peroxiredoxin genes in RSCs (in particular prdx3 and prdx6) thus raises the question of their functions with regard to stemness properties.

Although ROS involvement in neurogenesis is well-documented (Bórquez et al., 2016; Prozorovski et al., 2015; Wilson et al., 2018), how variations in their production mediate successive steps of proliferation and differentiation is still poorly understood (Bórquez et al., 2016). In addition, only a few studies have so far addressed the question of ROS functions in adult neural stem cells specifically. A great wealth of in vivo data indeed relates to embryonic precursors or, when conducted at the adult stage, do not truly distinguish bona fide stem cells from their more committed proliferative progeny (globally referring to neural stem/progenitor cells when examining for instance Sox2- or Nestin-positive cells). Both the subventricular zone and the dentate gyrus of the adult mammalian brain exhibit high expression levels of the NADPH oxidase complex NOX2. Using knockout mice, three different teams reported the requirement of NOX2-derived ROS to maintain normal proliferation and neuronal production in these neurogenic regions (Dickinson et al., 2011; Le Belle et al., 2011; Nayernia et al., 2017). In these studies, however, whether and how the loss of NOX2 affects neural stem cell proliferative behaviour in vivo remains unclear. Only ex vivo data, using neurosphere assays, support the idea that ROS might be instrumental in neural stem cell self-renewal (Le Belle et al., 2011; Paik et al., 2009). Taking advantage of the spatial organization of the CMZ, which allows for a topological distinction between stem cells and amplifying progenitors, we here identified a function of NOX in maintaining a proper ratio of activated neural stem cells within the CMZ. We currently do not know which NOX proteins are involved. NOX1 and NOX5 might be candidates as they are expressed in RSCs. Alternatively, a non-cell-autonomous function of NOX4 is possible, because it is expressed in lens precursor cells, in the vicinity of RSCs. Of note, we postulate that the RSC proliferative defects resulting from NOX inhibition are due to the consequent decrease of ROS concentration. However, blocking NOX activity may also increase NADPH levels and thereby alter the activity of the pentose phosphate pathway (the main producer of NADPH) through negative retroaction. Whether this might contribute to the observed phenotype remains to be determined.

A functional link between low ROS levels and quiescence in haematopoietic stem cells has long been known (Kohli and Passegué, 2014; Prieto-Bermejo et al., 2018). Current knowledge points to mitochondria, rather than NOX enzymes, as the main source of ROS driving the transition of these cells to a proliferative state, as they escape their hypoxic niche and switch to a more energetic oxidative metabolism. In the developing brain, Khacho and collaborators recently revealed that neural stem cells, although relying on glycolysis, possess functional elongated mitochondria, whereas committed progenitors exhibit fragmented ones. They further show that alteration of mitochondrial fusion/fission dynamics impairs the balance between self-renewal and commitment in a ROS-dependent manner (Khacho et al., 2016). Whether mitochondrial ROS production also contributes to modulate the proportion of proliferatively active neural stem cells is, however, still unknown. Our own data suggest that RSCs are not highly sensitive to decreased mitochondrial ROS production, with regard to their proliferative ability. We do not exclude the possibility, however, that longer MITO treatment duration or higher concentration of the drug might provoke their dormancy and/or affect their maintenance or fate.

In addition to the increased rate of quiescence, we also, surprisingly, found that the mean division speed was enhanced upon NOX inhibition. A reasonable scenario to explain this phenotype might rely on a possible heterogeneity among RSCs in terms of cell cycle kinetics. In this respect, lowering NOX activity might push the more slowly proliferating cells to enter quiescence, leaving faster cells still active, and thus biasing the mean speed measure towards a higher value. Alternatively, NOX-derived ROS might act as a brake slowing down RSC divisions.

Digging into the mechanisms that underlie NOX impact on RSC proliferation, we found that NOX blockade results in reduced canonical Wnt signalling and enhanced Hedgehog activity. Furthermore, the defective RSC proliferation could be successfully rescued by forcing Wnt activation or decreasing Hedgehog activity. These data thus raise two main questions: how does NOX regulate these pathways and how does their regulation impact RSC proliferative activity? We previously showed that both pathways are active in RSCs and reciprocally inhibit each other activity (Borday et al., 2012). NOX signalling could thus directly or indirectly modulate one, the other, or both. Of note, NOX-derived ROS have already been reported to modulate Wnt and Hedgehog signalling in diverse biological contexts (e.g. Gauron et al., 2016; Love et al., 2013; Meda et al., 2016; Thauvin et al., 2022a,b). In addition, Albadri and collaborators recently reported that 9-HSA, a known end product of lipid peroxidation, enhances the expression of Wnt target genes within the CMZ (Albadri et al., 2019). However, little is known about the molecular mechanisms underlying ROS-dependent regulation of Wnt and Hedgehog signalling. Only the Wnt pathway has proved to be directly activatable by H₂O₂in vitro through the oxidation of nucleoredoxin, a small redox-sensitive protein, which interacts with the adaptor protein Dishevelled (Funato and Miki, 2010). Whether this might hold true in RSCs as well remains to be determined. Besides, it is possible that NOX-mediated modulation of the Wnt/Hedgehog balance might in turn regulate ROS levels in RSCs. Indeed, Hedgehog was recently described as a positive modulator of H₂O₂ levels in vitro (Thauvin et al., 2022b), as well as in vivo during zebrafish appendage regeneration (Meda et al., 2016; Thauvin et al., 2022a). It would thus be worth assessing the occurrence of such a regulatory loop in RSCs.

From a functional point of view, we previously showed that Hedgehog signalling has the ability to increase retinal precursor cell cycle kinetics (Locker et al., 2006). Although we did not examine its effect on RSCs specifically, its increased activity upon NOX inhibition might account for the apparent enhanced cell cycle speed observed. It might, however, also contribute to trigger RSC quiescence. In line with this idea, Daynac and collaborators examined the consequences of sustained Hedgehog signalling in adult neural stem cells of the subventricular zone through conditional deletion of its receptor Patched. They found a phenotype very similar to that observed in this study, with quiescent stem cells accumulating and the remaining active ones cycling faster (Daynac et al., 2016). In contrast to Hedgehog activation effects, increased canonical Wnt signalling (at least to a moderate level) proved sufficient to promote reactivation of hippocampal quiescent neural stem cells in vitro (Austin et al., 2021). The same was observed upon short-term in vivo clonal analyses of neural stem cells in which the sfrp3 gene (which encodes a secreted Wnt inhibitor) had been knocked out (Jang et al., 2013). Finally, our own data demonstrated a crucial role of Wnt signalling in the maintenance of CMZ cell proliferation (Denayer et al., 2008). Altogether, we thus propose that NOX activity is instrumental in controlling the rate of quiescent versus activated RSCs, and possibly their cell cycle speed, through its opposed effects on Wnt and Hedgehog activities.

All animal care and experiments were conducted in accordance with NeuroPSI guidelines, under the institutional licences C91-471-102 and A91272108. The study protocols were approved by the animal care committee CEEA 59 (Paris, France) and by the Direction Départementale de la Protection des Populations (APAFIS 21474-2019071210549691v2 and 32589-2021072719047904v4; Courcouronnes, France).

Xenopus laevis embryos and tadpoles were obtained by in vitro fertilization, staged according to Nieuwkoop and Faber (Nieuwkoop and Faber, 1994) and raised at 18-21°C in 0.1× Modified Barth's Saline (MBS). The Wnt reporter transgenic Xenopus tropicalis line (carrying a pbin7Lef-dGFP transgene) was previously described and validated (Borday et al., 2018; Tran and Vleminckx, 2014). Transgenic embryos were obtained by crossing a wild-type male with a homozygous transgenic female. They were grown in 0.05× Marc's Modified Ringer solution at 23°C.

The ROS sensor DHE is a reduced form of ethidium that is rapidly taken up by live cells and emits red fluorescence upon oxidation (Thermo Fisher Scientific). It was resuspended in 100 µl anhydrous DMSO to a 30 mM stock solution. The staining protocol was adapted from Owusu-Ansah and collaborators (Owusu-Ansah et al., 2008). Briefly, embryos/tadpoles were incubated in 0.1× MBS solution supplemented with a freshly made DHE solution (5-20 µM) for 10 min in complete darkness and then washed three times, 5 min per wash, in 0.1× MBS. If previously treated with oxidant or antioxidant compounds, tadpoles were washed with 0.1× MBS solution supplemented with the drug they had been incubated in prior to DHE staining. The sensor 5-(and-6)-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate acetyl (CM-H₂DCFDA; Thermo Fisher Scientific) is a chloromethyl derivative of H₂DCFDA, oxidation of which yields a green fluorescent adduct that is trapped inside the cell. It was reconstituted in DMSO to a 2 mM stock solution. Tadpoles were incubated in darkness for 2 h in 0.1× MBS containing 5 µM CM-H₂DCFDA and then transferred for 30 min in 0.1× MBS to activate the probe. In toto imaging was performed immediately after, following tadpole anaesthesia in a 0.005% benzocaine solution (Sigma-Aldrich). MITO stock solution (3 mM; Sigma-Aldrich) was obtained by powder resuspension in dH₂O and stored at −20°C. APO (100 mM; Abcam), DPI (600 µM; Sigma-Aldrich) and ROT (2 mM; Abcam) stock solutions were systematically made freshly following reconstitution in DMSO. These drugs were then added into the embryo rearing medium for various durations, as indicated, and at the following final concentrations: 20 µM MITO, 200 µM APO, 2-4 µM DPI, 1-2 µM ROT. Cyclopamine (50 µM; LC Laboratories) was applied to the tadpole culture medium for 24 h. The different solvents of the drugs were used as negative controls.

Redox gene sequences were searched in Xenbase (https://www.xenbase.org/xenbase/). Some orthologues could not be found in the database (e.g. nox3) or failed to be cloned (e.g. cat1). cDNA encoding catalase 2, superoxide dismutases 1/2/3, glutathione peroxidases 1/4/7, peroxiredoxins 1-6 and NOX2/4 were amplified by PCR and inserted into a pCR®II-TOPO® vector through conventional TA-cloning procedures. nox1/5 sequences (300 bp) were synthetized by Eurofins Genomics (Germany) and cloned into a pBluescript KS plasmid. The list of PCR primers used and nox1/5 sequences are provided in Table S1. Plasmids used for antisense probe synthesis against hes1, hes4 and atoh7 (El Yakoubi et al., 2012), yap and c-myc (Cabochette et al., 2015), as well as the pCS2-TCF3-VP16GR plasmid (Borday et al., 2012) have been described previously. pCS2+ plasmids encoding Myc- or Flag-tagged cyba (pCS2+-Myc-cyba and pCS2+-cyba-Flag) were kindly provided by Enrique Amaya (Love et al., 2013). cyba translation-blocking antisense morpholino oligonucleotides (GeneTools) were previously described and validated (Love et al., 2013). Of note, they perfectly matched with both the Xenopus laevis and Xenopus tropicalis target sequence. A standard morpholino was used as a negative control (CCTCTTACCTCAGTTACAATTTATA).

mRNA (200 to 350 pg; synthesized with mMessage mMachine kit; Life Technologies) or 2 pmol morpholinos were injected into both blastomeres at the two-cell stage. mRNAs encoding β-Galactosidase and GFP were injected as controls and lineage tracers, respectively. Activity of the TCF3-VP16GR chimeric protein was induced by incubating the embryos for 24 h in 4 µg/ml dexamethasone (DEX, Sigma-Aldrich) from stage 37/38 to stage 41.

Total RNA from 30-40 dissected eyes per condition was isolated using the TRIzol reagent (Life Technologies) and purified with the NucleoSpin^® RNA Plus kit (Macherey-Nagel). Reverse transcription was performed using the Superscript II enzyme (Thermo Fisher Scientific). qPCR reactions were performed in triplicate using SsoFast EvaGreen Supermix (Bio-Rad) on a CFX96 thermal cycler (Bio-Rad). Results were normalized against the expression of reference genes odc1 and rpl8. Primer sequences are listed in Table S1.

Embryos/tadpoles were either injected intra-abdominally (at stage<41) or bathed (at later stages) with/in a 1 mM EdU solution (Thermo Fisher Scientific). They were then fixed after the desired time period in 4% paraformaldehyde. Paraffin-embedded embryos/tadpoles were sectioned (11 µm) on a Microm HM 340E microtome (Thermo Fisher Scientific). EdU incorporation was detected on paraffin sections using the Click-iT EdU Imaging Kit (Thermo Fisher Scientific) according to manufacturer's recommendations. Immunostaining experiments were conducted using standard procedures. Antigen retrieval was performed by boiling the sections in 10 mM sodium citrate and 0.05% Tween 20 for 9 min. Primary antibodies were incubated at 4°C overnight, and secondary antibodies at room temperature for 2 h. Cell nuclei were counterstained with Hoechst 33258 (Sigma-Aldrich). For western-blots, protein extracts were prepared using eight embryos per condition (neurula stage), which were frozen on dry ice and then resuspended and homogenized in extraction buffer (20 mM HEPES, 15 mM MgCl₂, 20 mM EDTA, 1 mM DTT), supplemented with protease inhibitors. Extracts were centrifuged 10 min at 18,000 g at 4°C and the interphase (devoid of pellets and lipids) was resuspended in migration buffer. All antibodies used are listed in Table S2.

Digoxigenin-labelled sense and antisense RNA probes were generated according to the manufacturer's instructions (DIG RNA Labelling Mix; Roche), from the corresponding linearized plasmids. Whole-mount in situ hybridization was carried out as previously described (Parain et al., 2012). Sections were either cut using a Leica VT1000S vibratome (50 μm thick) after 4% agarose embedding, or with a Microm HM 340E microtome (15 μm thick) following paraffin embedment.

Fluorescence and brightfield images of whole-mount embryos/tadpoles were captured with an AxioZoom fluorescence macroscope (Zeiss). Retinal sections were imaged with an ApoTome-equiped Axio Imager.M2 microscope and processed using ZEN (Zeiss) and Photoshop CS5 (Adobe) software. The intensity of DHE/CM-H₂DCFDA fluorescence or in situ hybridization signals was quantified with Fiji software (Schindelin et al., 2012). For DHE/CM-H₂DCFDA fluorescence, data were corrected by subtracting the mean autofluorescence intensity of unstained negative controls. For in situ hybridization signals, noise correction was performed using a blank region (devoid of specific signal) of constant surface on each section. In both cases, corrected values were then normalized to the corresponding region of interest surface. Unless otherwise specified, all quantifications on retinal sections took into account both the dorsal and ventral CMZ. Quantification of EdU-labelled cells was performed by manual counting. For each condition, six to eight sections per retina were considered and a minimum of six retinas were analysed. Stem cells were considered as the four to six most peripheral cells of the CMZ. EdU cumulative labelling experiments were analysed as previously described (Locker and Perron, 2019). Growth fraction (proportion of proliferative cells), total cell cycle length and S-phase length were determined using the Excel spreadsheet provided by Dr R. Nowakowski (Nowakowski et al., 1989). All experiments were performed at least in duplicate. Shown in figures are results from one representative experiment. In graphs, data are represented as mean±s.e.m. Statistical analyses were performed using nonparametric tests when at least one condition had n<30 measurements. For experiments with n>30 measurements in all conditions, parametric tests were only applied when Shapiro–Wilk test for normality was passed. Statistical significance is given as: *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001.

We thank Enrique Amaya for providing the tagged cyba plasmids. We are very grateful to our colleagues from TEFOR, the Paris-Saclay's aquatic zootechnical service, for hosting and supplying Xenopus.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201463.reviewer-comments.pdf