Asymmetric neurogenic commitment of retinal progenitors is regulated via the Notch endocytic pathway

During brain development, progenitor cells need to balance proliferation and differentiation in order to generate different neurons in the correct numbers and proportions. Currently, the patterns of multipotent progenitors’ division that lead to neurogenic entry and the factors that regulate them are not fully understood. We here use the zebrafish retina to address this gap, exploiting the possibilities for quantitative live-imaging. We show that early neurogenic progenitors arise from asymmetric divisions. Notch regulates this asymmetry as when inhibited, symmetric divisions producing two neurogenic progenitors occur. Surprisingly, Notch does not act through an apicobasal activity gradient as previously suggested but through asymmetric inheritance of Sara-positive endosomes. Further, the resulting neurogenic progenitors show cell biological features different from multipotent progenitors, raising the possibility that an intermediate state of progenitors exists in the retina. Our study thus opens new insights into the regulation of proliferative and differentiative events during central nervous system development.


Introduction
During central nervous system (CNS) development in vertebrates, the balance between proliferation and di erentiation is crucial to generate the di erent types of neurons in the correct numbers and proportions. This in turn ensures functionality of the neural tissue as it is the prerequisite for correct neuronal connectivity and circuit formation. Multipotent progenitors are key to maintain this balance. They rst grow in number through a phase of active proliferation (Fish et al., 2006;Homem et al., 2015) which is followed by neurogenesis onset. At neurogenesis onset, subsets of progenitors become neurogenic, consequently leave the cell cycle and di erentiate into neurons (Götz and Huttner, 2005;Agathocleous and Harris, 2009). Expectedly, impairment of the intricate balance between proliferation and di erentiation can lead to severe defects of brain development and is associated with numerous human diseases, including microcephaly and tumour formation (Fish et al., 2006;Thornton and Woods, 2009). Thus, understanding the parameters that di er between cells primed for continued proliferation versus cells that enter the neurogenic program is of utmost importance to comprehend brain development in health and disease.
Previous studies using the mammalian neocortex showed that at the onset of neurogenesis multipotent progenitors start dividing asymmetrically, generating one radial glial cell and one neuron or a basal progenitor and thereby balancing the ratio between di erentiation and proliferation at early neurogenesis stages (Chenn and McConnell, 1995;Huttner and Brand, 1997;Götz and Huttner, 2005;Shen et al., 2006;Shitamukai and Matsuzaki, 2012). Once a su cient number of progenitors is produced, nal symmetric consumptive neuronal divisions deplete the progenitor pool (Miyata et al., 2004;Noctor et al., 2004). Similarly to the mammalian cortex, asymmetric divisions have also been shown to be important for the maintenance of the progenitor pool before its nal depletion in the fore-and hindbrain of the teleost Danio rerio (zebra sh) (Alexandre et al., 2010;Dong et al., 2012). Thus, the division modes responsible for the generation of the rst neurons in di erent parts of the nervous system are relatively well explored, despite the fact that many questions are still open. What is much less understood, however, is how neurogenic progenitors rst arise from the multipotent progenitors. Especially in the retina, the tissue that upon maturation enables the perception of visual information, this question has not yet gathered su cient attention. Here, division patterns have mainly been investigated at later stages when the emergence of the seven di erent neurons that ensure retinal functionality has already commenced (Poggi et al., 2005;He et al., 2012;Kechad et al., 2012;Suzuki et al., 2013). It was shown that these late neurogenic divisions can be asymmetric, generating two neurons of different fate. However, how multipotent progenitors enter the neurogenic path in the rst place, what type of division patterns and what regulation pathways these initial fate decisions entail is not yet well explored.
Asymmetric divisions leading to the formation of neurons in the CNS have often been linked to the asymmetric inheritance of di erent cellular components. Examples of di erentially inherited components are the basal process (Saito et al., 2003;Okamoto et al., 2013), the apical domain (Bultje et al., 2009;Alexandre et al., 2010) and polarity components (Paolini et al., 2015;Zhao et al., 2020). In addition, di erent signalling pathways including Wnt (Hirabayashi et al., 2004) and very prominently Notch (Dorsky et al., 1995(Dorsky et al., , 1997Cayouette et al., 2001;Ohtsuka et al., 2001;Dong et al., 2012) have been shown to be involved. Notch for example can regulate the asymmetry of fate decisions through the asymmetric inheritance of Notch signalling endosomes, as shown in the zebra sh spinal cord and forebrain (Kressmann et al., 2015;Zhao et al., 2020). In the vertebrate retina, Notch signalling maintains progenitors in a proliferative state through lateral inhibition (Henrique et al., 1997;Jadhav et al., 2006). It was previously suggested that Notch drives neurogenic commitment of multipotent progenitors via an extracellular gradient along the apicobasal axis of the elongated neuroepithelial cells (Murciano et al., 2002;Del Bene et al., 2008). It was postulated that when nuclei travel along the apicobasal axis during the cell cycle (a phenomenon termed interkinetic nuclear migration, IKNM (Sauer, 1935;Baye and Link, 2007;Norden et al., 2009)), they are exposed to di erent Notch signalling levels, that in turn in uence the progenitors' probability to enter the neurogenic program (Murciano et al., 2002;Baye and Link, 2007;Del Bene et al., 2008). However, the division patterns of retinal progenitors that lead to neurogenic entry and their possible links to Notch signalling were not fully explored.
We here investigate the division patterns that give rise to the neurogenic progenitors, one cell cycle before the generation of post-mitotic neurons. We nd that divisions generating neurogenic progenitors are almost always asymmetric and that sister cells do not enter the same lineage. While Notch is involved in the generation of this asymmetry, it does not act via an apicobasal activity gradient. Instead, we provide evidence that Notch signalling acts through the asymmetric inheritance of Sara-positive endosomes. We further show that these rst neurogenic progenitors di er in their cell biological features from multipotent progenitors, indicating that they are an intermediate cell type en route to di erentiation.

Ath5 positive neurogenic progenitors arise from asymmetric divisions
To understand how neurogenic progenitors emerge from multipotent progenitors in the developing zebra sh retina, we asked how the earliest neurogenic progenitors arise, by symmetric or asymmetric division patterns or a mixture of the two. To follow these earliest neurogenic progenitors, we made use of the well-established fact that they express the transcription factor ath5 (atoh7) (Poggi et al., 2005;Zolessi et al., 2006) and we traced their origin by mosaically labelling cells in zebra sh embryos with an Ath5-driven reporter construct (Icha et al., 2016a). We followed the onset of the Ath5 signal using light-sheet imaging from the beginning of neurogenesis at 28 hpf (hours post fertilization) (Hu and Easter, 1999;Schmitt and Dowling, 1999;Martinez-Morales et al., 2005) as previously described (Icha et al., 2016b). To backtrack progenitors that began to express Ath5 (from here on referred to as 'Ath5 positive progenitors') ( Figure 1A, Supplementary Figure 1A), we labelled the plasma membrane and/or the nucleus of the same cells using hsp70:mKate2-ras or hsp70:H2B-RFP constructs ( Figure 1B). We backtracked sister cells of divisions giving rise to at least one Ath5 positive neurogenic progenitor and analysed their division patterns ( Figure 1C,C').
Tracing of 38 multipotent progenitor divisions from 9 embryos ( Figure 1B, Video 1) revealed that 92% of these divisions were asymmetric, generating one Ath5 positive (Ath5+) progenitor and one Ath5 negative (Ath5-) progenitor ( Figure 1C,C',D). Only 8% of Ath5+ cells were generated by symmetric division, producing two Ath5+ cells ( Figure 1C',D). Following Ath5-sister cells for another cell cycle showed that all cells divided again, but none of the daughter cells started expressing Ath5 (n=5 divisions, N=2 embryos) (Supplementary Figure 1B,C, Video 1). This suggests that Ath5-sister cells enter a di erent lineage than their Ath5+ counterpart. Together, these data provide rst direct evidence that neurogenic progenitors in the retina emerge from asymmetric divisions, indicating that cellular asymmetry plays an important role in the selection of progenitors that enter neurogenic programs.

Inhibition of Notch signalling a ects the pattern of divisions generating Ath5 positive progenitors
To understand the asymmetry of divisions producing Ath5+ progenitors at the molecular level, we set out to investigate the signalling factors involved. One promising candidate that drives asymmetric divisions in many neurogenic lineages is the Notch pathway (Bultje et al., 2009;Egger et al., 2010;Kechad et al., 2012;Kressmann et al., 2015). In the retina, Notch signalling has been shown to maintain progenitors in the proliferative state thus inhibiting neurogenesis (Bao and Cepko, 1997;Jadhav et al., 2006;Kechad et al., 2012;Maurer et al., 2014;Sato et al., 2016). We therefore tested whether and how Notch depletion in uenced the asymmetry of divisions generating Ath5+ progenitors using two established Notch inhibitors, the -secretase inhibitors LY411575 ( Figure 1E,F) and DAPT (Supplementary Figure 1D) (Ninov et al., 2012;MacDonald et al., 2015). In line with the reports that Notch signalling supresses neurogenesis and thereby Ath5 expression onset (Maurer et al., 2014), we found that both Notch inhibitors led to an increase in the number of Ath5+ cells at 36 hpf and 48 hpf ( Figure 1E,F). The increase of Ath5+ cells was accompanied by a striking reduction in tissue thickness ( Figure 1G, G') and retinal diameter ( Figure 1G, G") compared to controls at 48 hpf. This retinal size reduction indicated that indeed the balance between proliferation and di erentiation in the retina was disturbed upon Notch inhibition, leading to tissue-wide consequences.
The increase of Ath5+ progenitors observed upon Notch inhibition could arise from two not mutually exclusive scenarios: Ath5+ progenitors could emerge prematurely by asymmetric divisions that absorb multipotent progenitors too early, or an increase in symmetric divisions giving rise to two Ath5+ progenitors could occur that would result in less progenitors entering another cell cycle, thus also depleting the pool of remaining progenitors prematurely ( Figure 1C, C'). To explore these possibilities, live imaging of LY411575 treated embryos was performed ( Figure 1H). Compared to controls, Notch inhibition led to a signi cant increase in symmetric neurogenic divisions that generated two Ath5+ progenitors (48% compared to 8% in controls, Figure 1I, Video 2). Thus, Notch signalling regulates the asymmetry of the division that produces neurogenic Ath5+ progenitors. Interference with Notch and thereby this division asymmetry in uences neurogenesis patterns and tissue maturation in the developing retina.

Ath5 positive progenitors generally translocate more basally than Ath5 negative sister cells but a spatial bias is not observed at the cell population level
The selection of the sister cell that enters the neurogenic program could be regulated by di erential exposure to the extracellular environment that hosts di erent fate determinants as shown for other systems (Yamashita and Fuller, 2005;Knoblich, 2008). It was previously reported that Notch RNA is distributed in a graded manner along the cell's apicobasal axis with more Notch RNA detected towards apical positions in the retinal neuroepithelium (Del Bene et al., 2008). This nding led to the idea that distinct Notch signaling milieus exist along the apicobasal axis of the retinal neuroepithelium and that these could in uence neurogenic outcome due to nuclear movement along this axis during IKNM (Murciano et al., 2002;Baye and Link, 2007;Del Bene et al., 2008). Based on these studies, we asked whether the observed cell division asymmetry that generates Ath5+ progenitors ( Figure 1C) could be explained by di erential translocation depths between Ath5+ and Ath5-sister cells. To this end, we compared trajectories of Ath5+ and Ath5-sister cells after division ( Figure 2A). We found that indeed one sister cell soma always translocated more basally than the other ( Figure 2B), with an average di erence of 10.11 ± 7.292 m between the maximal basal positions (n=35 divisions, N=9 embryos) ( Figure 2C). In 78% of sister cells, the more basally translocating sister cell started expressing Ath5 (27/35 divisions) ( Figure 2D) while in 22% of divisions this pattern was reversed and it was the Ath5-cell that positioned the soma more basally ( Figure 2E). In the latter case however, the di erence between maximum basal position of sister cells was on average less prominent than in the reverse scenario (di erence of 6.048 ± 3.559 m when Ath5-soma is most basal compared to 11.31 ± 7.712 m when Ath5+ soma is most basal) ( Figure 2F).
To probe whether a basal positioning threshold could be de ned that predicts the onset of Ath5 expression, all trajectories were pooled and their overlap was analysed ( Figure 2G, H). To our surprise, this analysis showed a signi cant overlap of trajectories and of maximum basal positions for Ath5+ and Ath5-somas without clearly distinct Ath5+ and Ath5-zones ( Figure 2G,H,I, Supplementary Figure 2A). The measurement of the average depth occupied by somas of Ath5+ and Ath5-cells (integral mean, Supplementary Figure 2B) further con rmed that the two populations did not occupy distinguishable basal positions. These results show that while basal translocation di ers between sister cells for which the Ath5+ cell usually occupied more basal positions, overall basal position was not a clear readout for Ath5 expression onset as Ath5+ progenitors were found along the apicobasal axis.

Asymmetric neurogenic commitment is independent of maximum basal position
The large overlap between basal positioning of Ath5+ and Ath5-sister cell soma argued against a direct link between basal translocation and neurogenic commitment. To show this more directly, we eliminated the di erence of basal translocation between sister cells by equalizing their translocation pattern after apical division. To achieve this, we used a heat shock inducible dominant negative version of dynactin (Norden et al., 2009), as dynactin downregulation has been shown to induce increased basal translocation of retinal progenitor cell soma (Del Bene et al., 2008;Norden et al., 2009) (Figure 2J). Heat shock expression of the hsp70:DNdynactin-mKate2 construct led to the translocation of both sister cell somas to greater and almost identical basal positions compared to controls ( Figure 2K, Video 3, Supplementary Figure 2E). However, despite the fact that both sister cell somas reached very basal positions, only one became Ath5+ while the other remained Ath5-(n=10, N=5) ( Figure 2L, Video 3).
These results show that the asymmetry of neurogenic commitment is not linked to di erential somal translocation of sister cells along the apicobasal axis.

Notch pathway components are uniformly distributed along the apicobasal axis of the retinal neuroepithelium
Our results so far argue against a graded e ect of Notch signalling on the asymmetry of divisions producing Ath5+ progenitors. Thus, we probed whether the previously reported Notch gradient at the RNA level (Del Bene et al., 2008) was also found at the level of the signalling proteins.
We performed immunostaining against two components of the Notch pathway, the receptor Notch1b and the ligand DeltaC. We found that both Notch pathway components were uniformly distributed along the apicobasal axis of the retinal neuroepithelium ( Figure 3A,A',A" and Figure 3B,B', Supplementary Figure 3). We further analysed the distribution of Notch signalling activity using two previously established reporter lines: Tg(her4.1:mcherry-CreERT2), a marker for the gene her4, a direct Notch target (Takke et al., 1999), and Tg(Tp1bglob:H2BmCherry), a marker for a secondary Notch target (Souilhol et al., 2006). This analysis con rmed that Ath5-cells that display active Notch can occupy diverse positions along the apicobasal axis of the neuroepithelium ( Figure 3C,D and Video 4). Interestingly, we noticed that Ath5+ progenitors never express Tp1 or her4 ( Figure 3C,D, arrowheads), while Ath5-cells express both markers. Thus Ath5+ and Ath5-sister cells experience di erent levels of Notch signalling activation, independently of the basal position of cell soma.
Overall, these results combined with the fact that asymmetric Ath5 expression onset persisted even when both sister cell somas translocated very basally, argue that Notch signalling does not act via an apicobasal signalling gradient for setting up the asymmetry of divisions leading to neurogenic commitment.

Basal somal translocation of Ath5 positive progenitors depends on a combination of basal process inheritance and stabilized microtubules
While we show that somal positioning along the apicobasal axis does not in uence the asymmetry of divisions that produce Ath5+ progenitors, we nevertheless found di erences in basal translocation potential between Ath5+ and Ath5-sister cells. In the majority of cases the cell that later starts expressing Ath5 (Ath5+ progenitor) translocated more basally ( Figure 2D). We thus asked what other factors could facilitate this more basal translocation pattern of Ath5+ progenitors.
To this end, we compared basal translocation kinetics of Ath5+ and Ath5-progenitor soma and performed mean square displacement (MSD) analysis. This analysis revealed that Ath5+ basal somal translocation is more directed than that of Ath5-soma ( Figure 4B), a nding con rmed by directionality ratio analysis (Supplementary Figure 2C). Furthermore, Ath5+ sister cell soma showed a higher average velocity during basal translocation (Supplementary Figure 2D), indicating that Ath5+ progenitors translocate more e ciently towards the basal side.
It was previously shown that the inheritance of the basal process allows progenitors and emerging neurons to translocate soma and nucleus more e ciently to the basal side in neuroepithelia of the neocortex and the retina (Saito et al., 2003;Icha et al., 2016a). We thus asked whether basal process inheritance was correlated with e cient basal displacement of Ath5+ soma ( Figure 4A). Indeed, in 15 out of 16 cell pairs analysed, the cell that inherited the basal process initially translocated its soma more basally ( Figure 4C). However, only in 10/16 cases (60%), that cell also became the Ath5+ progenitor ( Figure 4D). Thus, it was possible that additional factors were responsible for the more e cient basal somal translocation of Ath5+ cells. It was previously shown that stabilized microtubules emanating from the apically positioned centrosome, facilitate somal translocation of emerging retinal ganglion cell, the rst neurons born in the retina (Icha et al., 2016a). To test whether stabilized microtubules are also present in Ath5+ progenitors, we stained for acetylated tubulin in Tg(ath5:gap-GFP/RFP) embryos (Zolessi et al., 2006) at stages before and during Ath5+ progenitors emergence, 24 hpf, 28 hpf, 32 hpf and 36 hpf ( Figure 4E). This revealed that at 24 hpf, the developmental stage prior to neurogenesis onset, acetylated tubulin was seen only in the primary cilium of progenitor cells ( Figure 4E, top left). However, once neurogenesis started, acetylated tubulin was also observed in the apical process of progenitors cells ( Figure 4E  To examine whether these stabilized microtubules were directly involved in the basal somal translocation of neurogenic progenitors, we backtracked Ath5+ progenitors overexpressing the microtubuledestabilizing protein Stathmin-1 using a hsp70:Stathmin1-mKate2 construct ( Figure 4F, Video 5) (Icha et al., 2016a). Soma of Ath5+ progenitors expressing Stathmin-1 did not translocate as basally as soma of control cells ( Figure 4I,K). In contrast, the basal somal translocation of Ath5-progenitors was not a ected ( Figure 4H,J), indicating that only progenitors that enter the neurogenic state use stabilised microtubules for basal somal translocation.
This experiment further con rmed that Ath5 expression onset is independent of somal positioning along the apicobasal axis. While both sister cells remained at very apical positions after division, only one cell became an Ath5+ progenitor ( Figure 4G, 10/10 divisions). This is the opposite scenario to the dominant negative dynactin condition in which cell nuclei translocated more basally ( Figure 4J,K), but the asymmetry of Ath5 expression onset persisted in both independently of sister cell basal or apical positions ( Figure 2L, Video 3, Video 5). Together, these two results strengthen our argument that neurogenic commitment occurs independently of basal somal translocation in asymmetric divisions.
Overall, these experiments showed that the greater basal somal translocation of Ath5+ progenitors is driven by a combination of basal process inheritance and the presence of apically enriched stabilized microtubules, as seen in other examples of neuronal basal somal translocation in CNS development (Yuasa et al., 1996;Umeshima et al., 2007;Cooper, 2013;Sakakibara et al., 2014;Icha et al., 2016a).

Asymmetric inheritance of Sara-positive endosomes a ects the asymmetry of neurogenic commitment
We conclude that the greater basal translocation of progenitors that start expressing Ath5 was driven by a combination of basal process inheritance and a stabilized microtubule network. However, these ndings do not yet explain how Notch signalling is involved in the asymmetry of neurogenic commitment independently of a Notch activity gradient. It was shown that Notch signalling components can segregate asymmetrically during division of the mother cell in other systems (Le Borgne and Schweisguth, 2003;Dong et al., 2012;Kechad et al., 2012;Coumailleau et al., 2009), which leads to the inheritance of di erent levels of Notch by the two daughter cells. In particular, Sara-positive endosomes (Sara endosomes) that carry active Notch signalling via the endocytosis of Delta receptors and Notch ligands were shown to be involved in asymmetric neurogenic fate decisions in the Drosophila sensory organ precursor and the zebra sh spinal cord (Coumailleau et al., 2009;Kressmann et al., 2015).
To test whether Sara-endosomes could also be involved in divisions generating Ath5 positive progenitors, we assessed whether Sara endosomes are present in retinal progenitors at stages at which Ath5+ progenitors emerge. Embryos were injected with mRFP-Sara mRNA and stained for Notch1b and DeltaC. We found that these components indeed co-localise in retinal progenitors ( Figure 5A,A' and Figure 5B,B'). To understand whether Sara-endosomes are asymmetrically distributed between progenitors upon division as suggested elsewhere (Coumailleau et al., 2009;Kressmann et al., 2015), we injected mRFP-Sara RNA into membrane labelled Tg(actb1:HRAS-EGFP) embryos. At all developmental stages analysed, 24 hpf, 28 hpf and 36 hpf, asymmetric distribution of Sara-endosomes was observed in dividing cells (Figure 5C). Cells mainly contained 1 to 3 endosomes and rarely more than that (Supplementary Figure 4A, n = 25 dividing cells, N = 7 embryos). In most cases one of the two sister cells inherited all endosomes (Supplementary Figure 4B, suggesting that the majority of Sara endosome related Notch signalling is dispatched to only one sister cell. Light-sheet live imaging con rmed asymmetric distribution of Sara endosomes during multipotent retinal progenitor divisions ( Figure 5D, D').
These experiments indicated that Notch signalling is asymmetrically partitioned through Sara-positive endosomes. Thus, Sara endosomes inheritance could mediate asymmetric neurogenic commitment in the retina. Our nding that Ath5+ progenitors never displayed active Notch signalling ( Figure 3C,D) is in agreement with this notion. Combined, these ndings argue that the depletion of Sara endosomes is linked to Ath5+ fate. We propose that Sara endosomes are mostly inherited by the Ath5-cell, which retains high Notch signalling and thus stays in the proliferative state (see working model in Figure 5G and cell in Supplementary Figure 4D, Video 6). To further test the involvement of Sara endosomes in asymmetric neurogenic commitment, we depleted the Sara protein using an established Sara splice morpholino (Kressmann et al., 2015). We found that the number of Ath5+ cells was greatly reduced compared to controls upon Sara inhibition, leading to the presence of fewer neurons ( Figure 5E, Supplementary Figure 4C). At the same time, the number of Tp1-positive cells increased ( Figure 5F), suggesting an increase in active Notch signalling in both daughter cells as postulated by our working model ( Figure 5G). This is the opposite scenario to what we observed upon Notch signalling inhibition which lead to the generation of more neurogenic progenitors through an increase in symmetric neurogenic divisions ( Figure 1E,H,I). In contrast, depletion of Sara led to more symmetric proliferative divisions generating two Ath5-cells.
Overall, these results suggest that Sara-mediated intracellular Notch signalling is necessary to create asymmetries in Notch activity among progenitors, thereby selecting the progenitors that enter the neurogenic path.

Discussion
We here used long-term live imaging to explore the emergence of neurogenic progenitors in the vertebrate retina. Our study reveals that early neurogenic commitment arises via asymmetric divisions of multipotent retinal progenitors that generate one Ath5+ neurogenic progenitor and one Ath5-progenitor that will enter a di erent lineage. This asymmetry is regulated by Notch signalling and is important to ensure the balance between proliferating cells and cells entering di erentiation routes at early stages of neurogenesis. Interference with the division asymmetry leads to premature or delayed di erentiation. We provide evidences that Notch does not act through an apicobasal activity gradient as previously proposed but via the endocytic pathway. Figure 5G summarises our main ndings and presents a working model.
Our quantitative imaging approach showed that progenitor divisions that generate the earliest retinal neurogenic progenitors, Ath5 positive progenitors, are reproducibly asymmetric. This approach allowed us to investigate whether and how the asymmetry is in uenced by Notch signalling. Our nding that Notch inhibition leads to more symmetric neurogenic divisions producing two neurogenic progenitors explains the increase in the total number of neurons at the linage level, adding to previous ndings that probed this issue at the tissue level (Austin et al., 1995;Ahmad et al., 1997). We show that this increase in symmetric divisions leads to the premature exhaustion of multipotent progenitors.

of 24
In line with our ndings, it has been shown earlier that Notch is involved in asymmetric divisions that generate two di erent neurons or one neuron and one proliferative cell in other neural systems including the vertebrate spinal cord and mammalian neocortex (Mizutani et al., 2007;Nelson et al., 2013;Kressmann et al., 2015). For the early neurogenic progenitor divisions in the retina investigated here, it was speculated that Notch acts via a signalling gradient along the apicobasal axis of the neuroepithelium (Murciano et al., 2002;Del Bene et al., 2008). Two studies proposed that more basally translocating progenitors are more likely to become neurogenic, generating two neurons after the next division (Murciano et al., 2002;Baye and Link, 2007). This led to the hypothesis that Notch signalling acts mainly at the apical side of the neuroepithelium where it suppresses neurogenic fate. However, due to technical challenges at the time, these analyses could only take a small number of unrelated cells into consideration. We here analysed large cohorts of sister cells and tracked their basal translocation patterns and indeed found that the more basally translocating sister cell is more likely to enter the Ath5+ lineage. However, when the entirety of Ath5+ cells was taken into account, a signi cant overlap of basal positioning of Ath5+ and Ath5-cohorts was observed. In addition, the asymmetry of neurogenic commitment persisted when the translocation position of sister cells was levelled towards either the basal or the apical side. Together, this shows that somal position along the apicobasal axis of the retinal neuroepithelium does not in uence asymmetric neurogenic commitment.
Instead, we present evidence that it is the asymmetric partitioning of Sara endosomes that drives the asymmetry of retinal progenitors divisions to enter the neurogenic pathway, as shown before in the Drosophila sensory organ precursor and the zebra sh spinal cord (Coumailleau et al., 2009;Kressmann et al., 2015). Ideally, we would have directly correlated Sara-positive endosomes inheritance with neurogenic or proliferative fate occurrence in numerous cells. However, this correlation could not be systematically quanti ed over several instances of asymmetric divisions in our system, due to its complexity and long timeframes between division and fate readout. Nevertheless, we could follow one division in which Sara endosomes where asymmetrically inherited and correlated the inheritance with sister cell fates. As expected, the cell inheriting all endosomes becomes the Ath5-progenitor and thus remains proliferative (Supplementary Figure 4D, Video 6) while the cell without Sara-endosomes starts Ath5 expression and thereby enters the neurogenic lineage (see working model Figure 5G and Supplementary Figure 4D, Video 6). This outcome is coherent with the fact that Ath5+ cells do not show active Notch signaling. These observations, combined with the nding that asymmetric inheritance is observed at stages at which asymmetric Ath5 decisions occur and that Sara knockdown impairs the balance between proliferation and neurogenesis in the Ath5 lineage, strongly argue that Notch signaling via Sara-positive endosomes is important for lineage asymmetry in the retina. The asymmetric distribution of Sara-positive endosomes most likely leads to the downregulation of Notch in one sister cell. This explains our observation that upon Sara knockdown more progenitors express active Notch signaling, as the Sara protein guides the asymmetric dispatch of these endosomes (Coumailleau et al., 2009). It should be noted that our systematic analysis not only enhances the understanding of Notch signalling in this particular neurogenesis event, but also shows the importance of quantitative approaches in developmental systems. Only due to the analysis of large cohorts of sister cells combined with a backtracking approach could we achieve the depth of data needed to exclude the role of a possible Notch gradient which led to the investigation of the Notch endocytic pathway. The advance of imaging techniques, image analysis algorithms and the possibility of performing backtracking studies in the retina are exciting emerging tools to follow similar questions in future studies on brain development. Such sophisticated studies will in turn promote our understanding of reproducibility and/or plasticity of lineage decisions at diverse developmental stages in di erent tissues of the central nervous system.
Despite the fact that basal position is not directly linked to Ath5+ fate commitment, we nd that the majority of progenitors entering the Ath5 lineage nevertheless translocate their soma more basally than their Ath5-sister. This observation argues that the more basal translocation of neurogenic Ath5+ progenitors is a consequence of neurogenic commitment, not the cause for it. We nd that basal translocation of Ath5+ progenitors is linked to apically localized stabilized microtubules and basal process inheritance. This is remarkably similar to what has been shown for somal translocation events of retinal ganglion cells (Icha, Kunath, et al., 2016). In contrast, this is di erent to nuclear and thereby somal translocation during IKNM in multipotent progenitors, which has been shown to be caused by passive nuclear displacements (Norden et al., 2009;Kosodo et al., 2011;Leung et al., 2011). Thus, an interesting interpretation of these ndings could be that Ath5+ progenitors constitute an intermediate cell type placed between multipotent progenitors and di erentiating neurons at the level of cellular arrangements including cytoskeletal components. This is reminiscent of intermediate progenitors found in the mammalian neocortex (Haubensak et al., 2004;Miyata et al., 2004;Noctor et al., 2004). To our knowledge such intermediate progenitors have not yet been described in the retina, meaning that our nding opens the possibility of exciting future studies of this cell type. It will be interesting to probe how cytoskeletal changes could be linked to cell identity changes in the developing retina and beyond. Such studies will open new avenues to understand the biology of fate decisions downstream of transcription pro les towards the cell biological scale.
Asymmetric progenitor divisions are an important mean to maintain the balance between proliferation and di erentiation in developing organs. To understand the regulation of these types of divisions and thereby embryogenesis as a whole, their investigation needs to span scales from the molecular to the cellular to the tissue level. Furthermore, investigations need to combine knowledge of cell type speci c transcription factor pro les, the signalling factors involved in guiding fate decisions and the cell biological consequences that di erentiation entails. The emerging possibilities of quantitative in vivo studies mentioned above in combination with the advent of single-cell sequencing data that is becoming more and more extensive (Wagner and Klein, 2020;Xu et al., 2020), opens new perspectives for an understanding of these phenomena at a previously unimaginable level. This, in combination with the continued use of the zebra sh as a model for quantitative developmental biology, has the potential to reveal general principles that can then be extended to other, often less accessible systems.

Zebra sh husbandry
Wild-type zebra sh were maintained at 26°C. Embryos were raised at 21°C, 28.5°C, or 32°C in E3 medium, which was changed daily. To prevent pigmentation, E3 medium was supplemented with 0.2 mM 1phenyl-2-thiourea (Sigma-Aldrich) from 8 hours post fertilization (hpf). Animals were staged in hpf according to Kimmel et al. (1995). All animal work was performed in accordance with European Union directive 2010/63/EU, as well as the German Animal Welfare Act.

DNA and RNA injections
To mosaically label cells in the zebra sh retina, DNA constructs were injected into 1 cell stage embryos. Injected volumes ranged from 0.5 to 1 nl. The amount of injected DNA ranged from 10 to 20 pg per embryo and did not exceed 40 pg even when multiple constructs were injected. See Table 1 for details. pCS2 expression vectors containing CFP-Sara or mRFP-Sara were kindly provided by Marcos Gonzáles-Gaitán and published previously (Kressmann et al., 2015). The Sara mRNA was synthesized using the Ambion mMessage mMachine kit. To visualize SARA endosomes, 100 pg of CFP-SARA or mRFP-SARA where injected into one cell stage embryos for ubiquitous expression.

Heat shock of embryos
To induce expression of the heat shock promoter (hsp70)-driven constructs, embryos were incubated at 24 or 28 hpf in water bath at di erent temperatures (see Table 1 for details). Imaging always started 3-4 hours after heat shock.

Blastomere transplantation
Donor embryos from the double transgenic line Tg(hsp70:H2B-RFP),Tg(her4:EGFP) and acceptor were dechorionated in pronase. At the high stage, cells from the animal pole of donors were transplanted into the acceptor wild type embryos. Transplanted embryos were grown at 32°C to recover for 4 hours, then moved to 28°C. The E3 medium was supplemented with 100 U of penicillin and streptomycin (Thermo Fisher Scienti c). Transplanted embryos were heat shocked at 24 hpf for 15 minutes at 39°C and imaged from 28 hpf.

Drug treatments
Notch inhibitors DAPT (Sigma-Aldrich) and LY411575 (Sigma-Aldrich) were dissolved in DMSO and used at 50 M and 10 M concentration respectively and compared to DMSO only controls. Embryos were dechorionated and treated with the drugs at 28°C in E3 medium, either in a 24 multi-well plate or the light sheet microscope chamber for live imaging experiments. All treatments started at 24 hpf. For live imaging experiments, sh were pre-incubated with the drug for 4 hours and imaging started at 28 hpf.

Image acquisition
Laser scanning confocal microscopy (LSCM) Fixed samples from immunostaining experiments were imaged with the Zeiss LSM 880 inverted pointscanning confocal microscope (Carl Zeiss Microscopy) using the 40x/1.2 C-Apochromat water immersion objective (Zeiss). Samples were mounted in 1% agarose in glass-bottom dishes (MatTek Corporation) and imaged at room temperature. The microscope was operated with the ZEN 2011 (black edition) software (Zeiss).

Spinning disk confocal microscopy (SDCM)
Confocal stacks for xed samples for Notch inhibition and Sara knockdown experiments were taken using an Andor SDCM system (Andor), consisting of an Andor IX 83 stand equipped with a CSU-W1 scan head (Yokogawa) with Borealis upgrade and an Andor iXon Ultra 888 Monochrome EMCCD camera. 40x/1.25 or 60x/1.3 silicone objectives were used to acquire 50-80 m z-stacks were recorded with 1 m optical sectioning. Fixed samples were mounted in 1% agarose in glass-bottom dishes (MatTek Corporation) and imaged at room temperature. The microscope was operated by Andor iQ 3.6 software.
In vivo light sheet uorescent imaging (LSFM) Light sheet imaging started between 28 hpf and 30 hpf. Embryos were manually dechorionated and mounted in glass capillaries in 1% low-melting-point agarose as previously described (Icha et al., 2016b). The sample chamber was lled with E3 medium containing 0.01% MS-222 (Sigma-Aldrich) to immobilize embryos and 0.2 mM N-phenylthiourea (Sigma-Aldrich) to prevent pigmentation. Imaging was performed on a Zeiss Lightsheet Z.1 microscope (Carl Zeiss Microscopy) equipped with two PCO.Edge 5.5 sCMOS cameras and using a 20x/1.2 Zeiss Plan-Apochromat water-immersion objective at 28.5°C. Zstacks spanning the entire retinal epithelium (70-100 m) were recorded with 1 m optical sectioning every 5 min for 15-20 hours using the double-sided illumination mode. Sara endosomes inheritance during cell division was imaged with 1 min time resolution and 0.5 m optical sectioning. The system was operated by the ZEN 2014 software (black edition).

Image analysis
Images from live imaging experiments were cropped and averaged in ZEN Black and/or Fiji (Schindelin et al., 2012) and were corrected for drift using a Fiji plugin (Manual drift correction) created by Benoit Lombardot (Scienti c Computing Facility, Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany).

Analysis of sister cell somal translocation
Ath5 progenitor divisions were identi ed and cells were backtracked to the previous cell division in order to identify cells giving rise to Ath5+ progenitors and the sister cell arising from the division. From the point after division, the translocation of sister cells was tracked in 2D in maximum projected substacks by following the center of the cell body in Fiji using the semi-automated ImageJ plugin MTrackJ (Meijering et al., 2012). MSDs and directionality ratios were calculated in the DiPer program (Gorelik and Gautreau, 2014), executed as a macro in Excel (Microsoft). To calculate the average instantaneous velocities, a custom-made scripts was written using Python 3 and can be found here.

Integral mean calculations
The integral meand was computed using Python 3 (available here) as follows: Where T is the time length of each cellular trajectory and I is the area below the trajectory curve, calculated as: Where deptht > 0.

Retinal size measurements
Retinal size was manually measured using 48 hpf retinas treated with LY411575 10 M or equal volume of DMSO, as illustrated in Figure 1G. Thickness and diameter were measured in the central z plane of the nasal, central and temporal region of the retina, using the Fiji line tool.

Notch1b and DeltaC spatial distribution
Prior to analysis, a gaussian blur of 0.5 was applied to all the images. Apical, central and basal ROIs (560 m 2 ) of the same dimensions were set for each image, dividing the length of the apicobasal axis by 3. Three central slices per embryo were analyzed in the nasal, central and temporal regions ( Figure 3A"). SpotCounter plugin for Fiji (Spot counter, Nico Stuurman, ValelabUtils package) was used to measure the number of spots in each ROIs. The number of spots in each ROIs was then normalized by the total number of spots in the 3 ROIs of a slice.

Sara-positive endosome count
Prior to analysis, a gaussian blur of 0.5 was applied to all the images. Confocal z-stack of retinal progenitors were reconstructed in 3D using the volume renderer ClearVolume (Royer et al., 2015) for Fiji. The number of Sara-positive endosomes was manually counted in dividing cells from 3D reconstructions.

Statistical analysis
All Statistical tests used are indicated in the gure legend, as well as the de nitions of error bars. All tests used were two-sided and 95% con dence intervals were considered. P values and sample sizes are noted in gure panels or gure legends. Data was analyzed using GraphPad Prism 6 or Python 3. Statistical analysis was performed using GraphPad Prism 6 and MATLAB.