Symmetry breaking and fate divergence during lateral inhibition in Drosophila Free

During development, stereotyped patterns of cell fates can be produced via self-organization guided by positional cues (Green and Sharpe, 2015; Schweisguth and Corson, 2019). Lateral inhibition is a conserved self-organized patterning process whereby equipotent cells inhibit each other from adopting the same fate via inhibitory cell-cell interactions mediated by Notch receptor signaling (Bray, 2006; Henrique and Schweisguth, 2019; Kopan and Ilagan, 2009; Simpson, 1990). In the absence of guiding cues, lateral inhibition produces ‘salt-and-pepper’ patterns of cell fates, but more elaborate patterns can be generated when a prepattern of Notch activation serves as the initial condition for lateral inhibition (Corson et al., 2017). Classic studies in Drosophila have indicated that stochastic fate decisions rely on the amplification of random fluctuations in Notch and/or proneural activity via an intercellular negative-feedback loop (Fig. 1A) (Greenwald and Rubin, 1992; Heitzler and Simpson, 1991; Heitzler et al., 1996; Simpson, 1997). In addition, mathematical modeling showed that an intercellular feedback loop can produce salt-and-pepper patterns when the negative feedback is strong enough to amplify fluctuating differences between adjacent cells (Collier et al., 1996). In Drosophila sensory organ formation, this feedback loop has four key elements (Fig. 1A′): the receptor Notch, its ligand Delta (Heitzler and Simpson, 1991), the E(spl)-HLH factors encoded by the direct transcriptional targets of Notch (Bailey and Posakony, 1995; Castro et al., 2005; Delidakis et al., 2014; Heitzler et al., 1996; Lecourtois and Schweisguth, 1995), and the transcription factors Achaete (Ac) and Sc (Campuzano and Modolell, 1992; Cubas et al., 1991; Skeath and Carroll, 1991). Ac and Sc specify the fate of the sensory organ precursor cells (SOPs) and are antagonized by Notch/E(spl)-HLH. They also positively regulate the signaling activity of Delta, possibly via an indirect mechanism (Chanet et al., 2009; Pitsouli and Delidakis, 2005). While the regulatory logic and general relevance of this feedback loop are well established (Apelqvist et al., 1999; Castro et al., 2005; Hellström et al., 2007; Sánchez-Iranzo et al., 2022; Simpson, 1997), the spatial and temporal dynamics of both Notch signaling and proneural activity during lateral inhibition are still largely unexplored. Indeed, although it is often thought that a phase a Notch-based competition relying on mutual inhibition precedes a phase of fate selection (Barad et al., 2011), Notch signaling has not been quantitatively studied in real time during the lateral inhibition process of SOP specification. Likewise, the time-course expression of Ac and Sc, and the dynamics of fate acquisition has mostly been studied in fixed samples in Drosophila (Corson et al., 2017; Couturier et al., 2019; Cubas et al., 1991; Huang et al., 1991; Skeath and Carroll, 1991; Troost et al., 2015; Usui and Kimura, 1993). These studies indicated that groups of cells first accumulated intermediate levels of proneural factors before one presumptive SOP was detected on the basis of increased levels of Ac and Sc, and that SOP selection was accompanied by the downregulation of Ac and Sc in the surrounding cells that were inhibited by Notch. However, when, and how SOPs emerge from proneural cluster cells have not been studied by real time imaging.

Here, we studied the dynamics of cell fate decision during lateral inhibition by live imaging a functional knock-in GFP-tagged version of the proneural factor Sc, GFP-SC (Corson et al., 2017). We focused our analysis on the selection of SOPs in the developing abdomen because high-resolution imaging can be easily performed around the time of SOP specification (Davis et al., 2022). To describe and study the process of SOP selection, we studied fate symmetry breaking (SB). The latter refers to the transition point when one cell, the future SOP, starts to stably accumulate a higher level of GFP-Sc relative to its immediate neighbors. Quantitative image analysis allowed us to identify SB and measure the rate of fate divergence during lateral inhibition. We found that fate selection occurred early and was not preceded by a detectable phase of mutual inhibition, and that cell-to-cell variations in Sc levels at SB promoted fate divergence, thus supporting the idea that SB involves an intercellular feedback loop.

Adult sensory bristles are distributed in two distinct patterns at the surface of the dorsal abdomen (Fig. 1B): large sensory bristles appear at regular interval along the compartment boundary to form a posterior row, whereas smaller bristles are distributed in a disordered array more anteriorly. Confirming earlier reports (Davis et al., 2022; Shirras and Couso, 1996), these two patterns were detected at the time of SOP emergence in the pupal abdomen: SOPs producing large sensory bristles appeared first along the posterior edge of the anterior dorsal histoblast nest (ADHN), while more anterior SOPs emerged later in the central region of the ADHN (Fig. 1C-E). Analysis of single and double mutant flies indicated that the proneural genes ac and sc were required for the determination of the abdominal SOPs, and that ac and sc were redundant for this process (Fig. S1A-D). We therefore examined the expression pattern of both Ac and Sc in fixed samples. To do so, we used a functional GFP-tagged knock-in version of Sc (Corson et al., 2017) and BAC transgenes encoding RFP- and GFP-tagged versions of Ac (Corson et al., 2017). GFP-Sc was not detected at 10 h (h) after puparium formation (APF; Fig. 1F,F′) and was first seen in a few cells at 13 h APF (Fig. 1I,I′) before being expressed along the posterior margin of the ADHN at 16 h APF (Fig. 1G,G′,J,J′). The pattern of GFP-Sc expression then progressively extended in the central region of the ADHN (18 h APF in Fig. 1H,H′,K,K′). Like GFP-Sc, RFP-Ac was detected in a few cells at 13 h APF (Fig. 1I,I′). However, in contrast with GFP-Sc, RFP-Ac was primarily detected in emerging SOPs at 16-18 h APF (Fig. 1I-K′ and Fig. S1E-F′; a similar observation was obtained with GFP-Ac, Fig. S1G,G′). Consistent with genetic redundancy, SOPs generally expressed both Ac and Sc (see the co-accumulation of GFP-Sc and RFP-Ac at 22 h APF in Fig. S1H-H″). As Ac appeared to accumulate at lower levels than Sc in the central region of the ADHN, we focused below on the dynamics of Sc accumulation during lateral inhibition.

To study the dynamics of SOP emergence, we performed live imaging using GFP-Sc. GFP-Sc is produced from the endogenous locus such that all Sc molecules produced in these pupae are GFP tagged. Staged pupae were imaged from ∼14 h APF, starting before the onset of GFP-Sc, to ∼22-24 h APF, ending after all SOPs have been specified (Fig. 2A-D′, Movie 1). In these movies, SOPs were identified as isolated cells with high GFP-Sc intensity at ∼20-22 h APF (Fig. 2D″). The intensity of the nuclear GFP-Sc signal was measured over time in (x,y,z,t,c) movies (n=3 movies). A nuclear RFP was used to segment and track individual nuclei. Image processing involved denoising and 3D segmentation of nuclei followed by 3D object filtering to eliminate bright dots corresponding to fragmented apoptotic nuclei, as well as large fluorescent structures (see Materials and Methods). We first examined the dynamics of Sc expression at the tissue-scale by projecting along the A-P axis the GFP-Sc intensity values from all cells to produce kymographs (Fig. 2E). This analysis confirmed that GFP-Sc appeared first along the posterior margin of the ADHN, then in the central domain of the ADHN. Our observations are consistent with earlier findings, suggesting that posterior cues might regulate early proneural gene expression in the ADHN (Bischoff et al., 2013; Kopp et al., 1997; Shirras and Couso, 1996). To study fate dynamics, SOPs from the central region of the ADHN were tracked in 3D, both forward (until they divide asymmetrically to produce a pIIa-pIIb cell pair) and backward (up to their birth from a dividing histoblast progenitor cell). A total of 130 SOPs were tracked. After manual curation of all tracks, normalized GFP-Sc levels were measured in tracked SOPs. The temporal expression profiles of GFP-Sc in tracked SOPs relative to all other untracked cells indicated that SOPs emerge within a ∼2 h time window centered around ∼17 h APF in the ADHN (Fig. 2F).

We next studied the dynamics of cell fate acquisition by comparing the dynamics of Sc accumulation in each presumptive SOP relative to its immediate neighbors, defined here as the six closest nuclei, in a three-dimensional space, after manual correction of segmentation errors (Fig. 3A,A′; in the absence of cortical markers, we assumed but could not demonstrate that these neighboring nuclei marked cells that were in direct contact with the SOP). This number of neighbors was chosen because we found that SOPs had 5.8±1.1 apical neighbors (n=35 SOPs) between 17 and 18 h APF in GFP-tagged E-cadherin (Ecad-GFP) movies (see Materials and Methods and Fig. 5 below). As these nuclei were not tracked, different nuclei could participate over time to the cluster of cells associated with a given SOP. A representative example of the temporal profiles of GFP-Sc levels in a single SOP and in its six neighbors is shown (Fig. 3B, Movie 2). In this example, the presumptive SOP and its surrounding histoblasts showed a weak and slowly increasing GFP-Sc signal from t=3.6 h onwards until the presumptive SOP showed a rapid increase in GFP-Sc accumulation at ∼4.9 h (Fig. 3B; this experimental time corresponds to ∼17 h APF). As GFP-Sc levels increased sharply in the SOP, GFP-Sc levels remained relatively constant in non-selected histoblasts for more than 1 h before decreasing. This temporal profile suggested that cells underwent binary fate decisions around the time when GFP-Sc started to increase in the presumptive SOP. To study fate divergence more quantitatively, we introduced a fate difference index (FDI) that compares the relative level of Sc in the presumptive SOP and in its closest neighbors. This FDI was defined as the ratio of the difference over the sum of the GFP-Sc intensities measured in the SOP nucleus and in the six closest nuclei. This allowed us to detect when the SOP becomes different from its neighbors (Fig. 3C). To study the onset of fate divergence, corresponding to fate SB, we defined a time point when the FDI value reached and remained above a given threshold. To set the value of this threshold value, we measured FDI values over time from groups of randomly sampled non-SOP cells. This analysis indicated that background FDI values associated with cell-to-cell differences in GFP-Sc levels in non-SOP cells were below 0.005 (Fig. S2). We therefore chose this threshold value to define SB. This minimal FDI value was found to be superior to the fluctuations observed before the sharp increase in GFP-Sc levels, and we found that FDI values consistently increased after SB for almost all tracks (Fig. 3C). We also used the FDI to quantitatively evaluate fate divergence, using the rate of change (ROC) of the FDI as a proxy for the rate of fate divergence, or speed of fate decision.

To go beyond this example and evaluate cell fate dynamics more systematically, we used SB as a reference time point to re-align all tracks (n=130 SOPs). This analysis revealed that the onset of fate divergence was observed at low GFP-Sc levels, before presumptive SOPs accumulate high levels of GFP-Sc (Fig. 3D). It further showed that the FDI increased at a relatively constant rate after SB, reaching a plateau 3-4 h after SB (Fig. 3E). However, we did not observe a phase during which many proneural cluster cells would express intermediate levels of Sc before fate selection, which might correspond to the proposed phase of mutual inhibition (Cubas et al., 1991; Muskavitch, 1994). In addition, we did not observe a rapid decrease of proneural expression in non-selected cells after SB. Instead, the level of GFP-Sc remained relatively constant in non-selected cells for ∼2 h after SB. Thus, our analysis of the temporal dynamics of cell fate acquisition in the pupal abdomen showed that fate SB takes place at low Sc levels, soon after the onset of proneural gene expression.

We next studied the heterogeneity of Sc, or cell-to-cell variations in Sc levels, defined here as the coefficient of variation (standard deviation over the mean) of GFP-Sc intensities measured in the presumptive SOP and its six neighbors. Analysis of heterogeneity over time showed that it began increasing ∼1 h before SB and remained high in non-selected histoblasts for about 2 h before decreasing down to initial levels (Fig. 4A). Interestingly, the speed of SOP emergence, as determined using the ROC of the FDI, positively correlated with heterogeneity of Sc at SB (Fig. 4B). These observations raised the possibility that heterogeneous Sc levels in proneural clusters might promote fate decisions. To test this idea, we experimentally increased the heterogeneity of Sc by creating marked clones of cells expressing a RNAi against GFP (Fig. 4C-C″ and Movie 3; loss of sc activity did not prevent SOP formation, Fig. S1I,I′). Clonal analysis confirmed that cells with reduced proneural activity proliferated normally but were less likely to become selected as SOPs (Heitzler et al., 1996) (Fig. S1J,K). We then studied by live imaging the dynamics of GFP-Sc in these genetically mosaic pupae. After segmentation of nuclei marked by nuclear RFP, we tracked the wild-type SOPs located along the clone boundary (n=165 SOPs, from 12 movies) and identified the six closest SOP neighbors at each time point (Fig. 4D,D′). We then plotted the FDI over time (excluding RNAi-expressing cells to measure fate competition between wild-type cells) and determined SB for each cluster. We also determined the mean number of RNAi-expressing cells around the time of SB. As expected, the heterogeneity of Sc measured at SB using all cluster cells, thus including the RNAi-expressing cells, increased with the mean number of RNAi-expressing cells in the cluster (Fig. 4E; at least when fewer than half of the cells, n≤3.5, are expressing the RNAi construct). To test whether increased heterogeneity of Sc in this experimental condition promoted fate divergence, we plotted the ROC of the FDI at SB as a proxy for fate divergence. This analysis showed that increasing the heterogeneity of Sc positively correlated with an increased rate of fate divergence (Fig. 4F). Consistent with this, FDI values increased faster in heterogenous clusters (Fig. 4G). These results showed that cell-to-cell variations in Sc levels promotes fate divergence during lateral inhibition, thus providing experimental support to the negative-feedback loop model that was proposed to amplify small differences and speed up fate divergence (Collier et al., 1996; Heitzler and Simpson, 1991; Heitzler et al., 1996).

This model also predicted that GFP-Sc dynamics in presumptive SOPs should be locally influenced by Sc dynamics in neighboring cells. Interestingly, late-emerging SOPs are more likely to be surrounded by one or several cells that have low proneural activity due to their inhibition by SOPs that have emerged earlier. We therefore predicted that late-resolved clusters should be more heterogenous, and that late-emerging SOPs should progress more rapidly towards an SOP fate in wild-type pupae. In support of these predictions, we found that both the heterogeneity of Sc and the ROC of the FDI positively correlated with the developmental time of SB (Fig. 4H,I). Thus, late-emerging SOPs appeared to progress more quickly towards the SOP fate. Together, our data support the view that initial differences in proneural gene expression become amplified to generate stable cell fates and that cell-to-cell heterogeneity in Sc levels underlies stochastic fate choice in the abdomen.

We next addressed whether the initial variations of Sc accumulation were random or resulted from some intrinsic bias associated with relative cell-cell contact size. Indeed, earlier modeling suggested that differences in apical cell area may serve as a possible source of bias for Notch-based decisions (Shaya et al., 2017). To test this possibility, we measured apical cell area of abdominal cells in Ecad-GFP pupae that were imaged live. In these movies, SOPs were identified using a nuclear GFP expressed under the control of a SOP-specific enhancer (pneur-nlsGFP) (Fig. 5A-B′, Movie 4). After segmentation of apical cell junctions, SOPs were tracked back in time (n=22 SOPs, from 3 movies). Each SOP, together with the untracked cells in direct contact with the SOP, defined a cluster. To test whether apical cell area differs between SOPs and its direct neighbors at the time of fate decision, one would need to determine SB and register in time the different clusters. Unfortunately, SB could not be determined in these movies. However, the timing of SB correlated temporally with the division of the SOP in our GFP-Sc nuclear RFP movies and we found that SOPs divided 4.7±1.1 h after SB (Fig. 5C; no correlation was found between SB and the previous division that generates the future SOP). We therefore registered in time all clusters using the SOP division as a reference timepoint (t=0 h in Fig. 5D). For each cluster, the apical area of the presumptive SOP was compared to the mean area of its neighboring cells. This analysis revealed no relative difference in apical area around the predicted time of SB (Fig. 5D). Thus, apical cell shape did not appear to bias Notch-mediated fate decision in this developmental context.

While studying Sc dynamics, we observed several clusters in which one of the six SOP neighbors showed a high level of GFP-Sc after SB (n= 16 clusters, out of 130), indicating that a presumptive SOP formed very close to the tracked SOP (Fig. 6A-C′). However, in the absence of a cortical marker, it was unclear whether these pairs of SOPs were in direct contact at SB. To address this, we performed live imaging of GFP-Sc in pupae expressing an Ecad-mKate marker. This analysis identified 20 pairs of SOPs based on high GFP-Sc expression (Fig. 6D-E′ and Movie 5; n= 399 SOPs from nine movies). Tracking back the apical cell area of these GFP-Sc cells and inferring the time of SB using the SOP division as a time reference (Fig. 5C), we found that most pairs of SOPs (n=18 pairs, out of 20) were in direct contact around SB, i.e. before they could be identified as SOPs. Additionally, we observed that these paired SOPs separated from each other at 1±1.7 h (n=13/18) after SB, before their division. Thus, lateral inhibition in the pupal abdomen failed to single out SOPs in ∼10% of the clusters, and subsequent pattern refinement primarily involved cellular rearrangements after fate specification, and not fate switching.

As SOP fate acquisition is associated with changes in acto-myosin dynamics (Couturier et al., 2017) that might be associated with changes in cell-cell adhesion, we wondered whether differential adhesion might cause a specific loss of these SOP-SOP contacts. To test this hypothesis, we segmented and tracked all cells of a GFP-Sc Ecad-mKate movie (Movie 6) and measured the rate of neighbor exchanges in both SOPs and non-SOP cells (Fig. 6F-H′). This analysis showed that all cells changed neighbors at a similar rate (Fig. 6I). Thus, global tissue fluidity, rather than differential adhesion, appeared to cause the separation of paired SOPs. We therefore conclude that initial patterning defects were corrected via cellular rearrangements associated with global tissue fluidity.

We next examined the dynamics of Notch activity during lateral inhibition. To do so, we used a destabilized GFP (deGFP) expressed downstream of a Notch-Responsive Element (NRE-deGFP) (He et al., 2019). A strong NRE-deGFP signal was detected in the posterior region of the ADHN at ∼14 h APF (Fig. 7A, Movie 7). We further noticed that the posterior-most ADHN cells showed a lower NRE-deGFP signal, indicative of lower Notch activity. Thus, Notch signaling may provide a negative template to define where the first SOPs will emerge, as seen in the notum (Corson et al., 2017). A strong NRE-deGFP signal was also observed in the PDHN (Fig. 7A). This initial pattern of Notch activity was detected before the onset of Ac and Sc expression. Consistent with this, it did not depend on Ac and/or Sc (Fig. S3A,C). Live imaging revealed that this pattern gradually evolved to produce a salt-and-pepper pattern extending over the central region of the ADHN, with presumptive SOPs emerging as non-fluorescent cells (Fig. 7B,C; Movie 7). Averaging the NRE-deGFP signal over the posterior part of the ADHN (outlined in green in Fig. 7A,B) showed that the NRE-deGFP signal decreased to reach a minimum at t=7 h before increasing again (Fig. 7D). This decrease in signal intensity implied that the degradation rate of deGFP was faster than its synthesis rate. Of note, detection of a NRE-deGFP signal does not necessarily imply that Notch signaling is active, as the deGFP protein has a measured half-life of ∼2 h in fly cells (He et al., 2019). Therefore, a twofold decrease in fluorescence over a 2 h period would suggest that Notch signaling is switched off (but was active at an earlier time point). To interpret the temporal dynamics of the NRE-deGFP signal in terms of instantaneous Notch activity, we examined the ROC over time of this signal (Fig. 7E). This analysis showed that Notch activity decreased to reach a minimum at t=3.5 h (corresponding to ∼17.5 h APF; Fig. 7E), implying that the signaling activity of Notch was minimal at this stage. Of note, this minimum value (0.157) corresponds to a signal decay rate of 4.4 h (at ∼22°C), which is close to the half-life of deGFP measured in cultured cells (∼2 h, at 25°C). This suggested that deGFP synthesis, and hence Notch signaling, must be very low at that stage. Notch signaling activity then increased in both the posterior ADHN region (Fig. 7D,E) and in the central ADHN area (Fig. 7B). In summary, Notch signaling appeared to be low before proneural gene expression and SOP selection, while it increased as SOP emerged. Additionally, Notch activity remained low in sc^10-1 ac³ mutants (Fig. S1B,D), further indicating that increased Notch activity was associated with the emergence of SOPs.

We confirmed these observations by studying the expression of the Notch target E(spl)HLH-m3 gene. Using a tagged version of endogenous E(spl)m3, GFP-m3, we found that E(spl)m3 was expressed at 13 h APF in the posterior cells of the ADHN, as well as in the PDHN (Fig. S3F,F′). Expression of E(spl)m3 extended to the central ADHN domain as SOPs emerged at 17-18 h APF (Fig. 7I-I‴, Fig. S3E-G″). Analysis of an mCherry-tagged E(spl)m3 showed that GFP-Sc was detected before E(spl)m3 in the central domain of the ADHN [Fig. S3E-E″; other E(spl)-HLH factors showed a similar spatial-temporal dynamics, see Fig. S3H-J′]. Thus, E(spl)m3 was detected after the onset of Sc expression and upon, but not before, SOP emergence. This indicated that SOP selection in the abdominal epidermis took place at low Notch signaling levels.

We next studied Notch dynamics at single cell resolution using NRE-deGFP. To do so, we used the same approach as the one used for GFP-Sc, restricting our analysis to the neurogenic region of the ADHN where no initial NRE-deGFP expression was detected. All nuclei were segmented using a nuclear RFP and presumptive SOPs, which were identified as NRE-deGFP low cells (Fig. 7C), were backtracked (n=21 SOPs, from two movies). To compare the expression of NRE-deGFP in each SOP relative to its closest neighbors, we calculated a FDI and identified a point of SB in Notch activity (SBN; this nomenclature was chosen to distinguish SBN from SB defined by proneural dynamics). We then registered all clusters relative to SBN (defined as t=0) and plotted the temporal profile of NRE-deGFP expression (Fig. 7F). This showed that SBN occurred at low Notch activity levels and that non-SOP cells showed strong Notch signaling activity only after SBN (Fig. 7G; note that FDI values were negative and decreasing because NRE-deGFP was expressed in non-SOP cells). Thus, this quantitative approach did not reveal a phase of strong and reciprocal Notch signaling during which proneural cluster cells would have similar levels of Notch activity before SOP emergence.

Quantitative live imaging of the proneural factor Sc allowed us to detect fate SB and monitor fate divergence during lateral inhibition in the pupal abdomen. This showed that SB takes place at low level of Sc. After SB, the accumulation of Sc in presumptive SOPs was not associated with a concomitant decrease of Sc in all non-SOP cells. Instead, the mean level of Sc in the cells surrounding the emerging SOP remained relatively constant, whereas heterogeneity of Sc increased, implying that signal-receiving cells were asynchronously excluded, as seen earlier in the pupal notum (Corson et al., 2017; Couturier et al., 2019). We also found that cell-to-cell variations in Sc levels increased before SB. These variations positively correlated with fate divergence. Moreover, an experimental increase of GFP-Sc heterogeneity resulted in a faster rate of fate divergence. These results confirmed that initial differences in Sc levels become amplified via an intercellular feedback loop during lateral inhibition. Although the notion that cell-to-cell variability promotes fate divergence and increases patterning speed is not novel and applies to other juxtacrine signaling systems (Rudge and Burrage, 2008), our observations are nevertheless important, as they provide experimental support for the intercellular feedback loop model (Collier et al., 1996; Heitzler and Simpson, 1991). Our data also showed that the initial bias in Sc expression might play a significant role in SB. Of course, cell-to-cell variations in Ac (not studied here) might have a similar effect. The origin of the initial heterogeneity of Sc is unclear. In principle, cell-to-cell variations in Notch activity might provide a negative bias for the onset of Sc. Consistent with this, the heterogeneity of clusters producing late-emerging SOPs was high, possibly due to local Notch activation associated with the emergence of earlier SOPs in the vicinity. We were, however, unable to test whether Notch activity anti-correlates with Sc levels before SB due to the low levels of Sc and E(spl)-HLH at this stage (note also that RFP-Ac and GFP-m3 were not detectable by live imaging, possibly due to low expression and fast turn-over of GFP-m3 relative to the maturation time of GFP). Other sources of heterogeneity may also be considered. For example, in mouse intestinal organoids, cell-to-cell differences in mechanics have been involved in initiating a Delta/Notch signaling that drives SB (Serra et al., 2019). Although a potential role for mechanics in regulating fate decision remains to be studied in the abdomen, our analysis of apical area indicated that the size of the cell-cell contact did not provide a clear bias for fate decision in this tissue. In other systems, non-genetic heterogeneities can result from cell-to-cell differences in the capacity of producing and maintaining a stable pool of active protein levels (Colman-Lerner et al., 2005). Future studies will address the basis of the early heterogeneity in Sc (and Ac) accumulation.

The formation of the SOP pattern is thought to involve a phase of competition, or mutual inhibition, followed by a phase of fate selection during which selected SOPs inhibit their neighbors, and then a phase of pattern refinement during which patterning errors are corrected (Barad et al., 2011; Cohen et al., 2010; Simpson, 1990). Although mutual inhibition might be dispensable in contexts where binary fate decisions are strongly biased, as during neuroblast selection (Simpson, 1997), this phase is thought to precede the selection of individual cells in tissues where stochastic fate decisions produce salt-and-pepper patterns (Barad et al., 2011; Campuzano and Modolell, 1992; Cubas et al., 1991; Muskavitch, 1994). Contrary to this view, no prolonged phase of intermediate Sc expression or Notch activity was detected before fate SB in the abdomen. GFP-Sc expression was very low before SB in the central region of the ADHN. Likewise, live imaging of a Notch transcriptional reporter showed that Notch activity was minimal before SBN, and that its upregulation dependent on the activity of Ac and/or Sc. Additionally, activation of the Notch target gene E(spl)m3-HLH was detected around the time of SOP emergence. Thus, fate selection appeared to operate at low levels of both proneural and Notch activities in the abdomen. We therefore suggest that mutual inhibition can only be transient in the abdomen. It will be of interest to further examine the significance of mutual inhibition in other systems using quantitative live approaches.

Our study also revealed that lateral inhibition did not faithfully single out SOPs in the pupal abdomen. Indeed, pairs of presumptive SOPs were unexpectedly found in direct cell-cell contact at the time of SB in ∼10% of the clusters. Thus, our observation suggested that the feedback loop may not be robust enough to faithfully impose an alternative fate choice and that SOPs can rapidly become deaf to inhibitory signals produced by their direct SOP neighbors, possibly due to strong cis-inhibition of Notch (del Álamo et al., 2011). Pattern refinement was proposed to rely on fate switching, cell elimination and/or cell rearrangement (Barad et al., 2011; Cohen et al., 2010). In the pupal notum of Drosophila, pairs of emerging SOPs were proposed to resolve through fate reversal (Cohen et al., 2010). These conclusions, however, may need to be confirmed, as the reporter used to identify SOPs was later shown to be also expressed in non-SOP cells (Corson et al., 2017). Here, we found that cell-cell rearrangements led to the separation of these SOP pairs after SB and that pattern refinement relied on global tissue fluidity, not fate reversal. Interestingly, SOP patterning in the notum occurs in a field of non-dividing cells (Corson et al., 2017), whereas cells proliferate as they are patterned by Notch in the abdomen. Whether the mode of pattern refinement, based on cell-cell rearrangements and/or fate reversal, correlates with cell proliferation and/or tissue fluidity remains to be examined.

Sc was GFP-tagged at its N-terminus in the scute^GFP knock-in line. This line was generated using CRISPR-mediated homologous recombination (HR) using 3xP3-RFP as a selection marker (Corson et al., 2017). As 3xP3-RFP gave a spotty signal in the abdominal nest, this marker, flanked by loxP sites, was removed using the Cre recombinase (BL-851). Proper excision was verified by gPCR. The rfp^- version of scute^GFP appeared to be fully functional and was used in this study. The GFP-Ac and RFP-Ac BAC transgenes encode functional versions of the ac gene (Corson et al., 2017). The E(spl)m3-HLH factor was GFP tagged at its N terminus in the GFP-m3 CRISPR knock-in line (Couturier et al., 2019), whereas the functional GFP-m8, GFP-mδ and GFP-mγ factors were encoded in transgenic BACs marked by 3xP3-RFP (Couturier et al., 2019). The Cherry-tagged version of E(spl)m3 was produced as a BAC transgene by recombineering in E. coli an E(spl)-C BAC (Chanet et al., 2009), as described for the GFP-m3 BAC transgene (Couturier et al., 2019; Venken et al., 2006). The mCherry-m3 BAC was integrated at attP site located in 99F8 (VK20 line). Injection was performed by BestGene. The sc^10-1 ac³ (BL-36541), sc^M6 (Marcellini et al., 2005) and Df(1)91B (Gibert et al., 2005) mutants were used. The activity of Notch was monitored using the NRE-deGFP line (He et al., 2019). Different mRFPnls lines were used to mark nuclei. Apical junctions were tracked using versions of Ecad intracellularly tagged with GFP (Huang et al., 2009) or mKate (Pinheiro et al., 2017). SOPs were identified using a neur-nlsGFP transgene (Aerts et al., 2010). The silencing of GFP in marked flp-out clones was achieved using a flp-out strategy in pupae carrying hs-flp, nlsRFP and GFP-Sc on the X, P{y[+t7.7] v[+t1.8]=VALIUM20-EGFP.shRNA.3}attP40 (BL-41559) on the second chromosome, and P{UAS-His3.3.mIFP-T2A-HO1}attP2 (BL-64184) and P{w[+mC]=AyGAL4}17b (BL-4413) on the third chromosome. Tubes with third instar larvae were heat-shocked (40 min, 36.5°C) in a water bath. The RNAi-mediated knockdown of GFP-Sc in the clone was studied in scute^GFP mRFPnls/hs-FLP ; UAS-shRNA(gfp)/+ ; pActin-FRT-stop-FRT-Gal4 (AyGal4)/UAS-His3.3-mIFP-T2A-HO1 pupae.

Imaging of living pupae can be performed at high spatial resolution from 13 h APF onwards, after the pupal case detaches from the pre-cuticle secreted by the developing imago, and after histolysis of abdominal muscles, preventing body movements and allowing stable imaging. Pupae were selected at ∼0 h APF: individuals entering puparium formation were selected every 45-60 min and incubated in a moistened chamber. Stage pupae were mounted for imaging by removing the pupal case using fine forceps at ∼14 h APF. Before imaging, larvae and pupae were grown at 18, 21 or 25°C, but image acquisition was carried out at 23-25°C. Slight differences in the absolute timing of SB may have resulted from these varying conditions. Spinning disk microscopy was performed using either a Leica DMRXA microscope equipped with a 40× (PL APO, N.A. 1.32 DIC M27) objective, a Yokogawa CSU-X1 spinning disk, a sCMOS Photometrics PRIM95B camera, 491/561/642 lasers and the Metamorph software, or a Nikon Ti2E microscope equipped with a 40× (N.A. 1.15 Water WD 0.6) objective, a Yokogawa CSU-W1 spinning disk, a sCMOS Photometrics PRIM95B camera and 488/561/640 lasers. Typically, we imaged GFP every 5 mins with 20% laser power for 300 ms and RFP every 2.5 min with 9% laser power for 200 ms. A z-stack of ∼40 mm was imaged with Δz=1.33 mm and 0.7 mm for GFP-Sc and Ecad-GFP movies, respectively. Reproducible dynamics of Sc were obtained over 10-16 h of imaging. All snapshot views and movies shown here correspond to maximal projection views.

The nuclear channel of the GFP-Sc nlsRFP movies was preprocessed to correct the contrast and denoised using Noise2Void (Krull et al., 2019). The 3D segmentation of nuclear boundary was performed based on a custom-made program (Corson et al., 2017). Undesired structures (dead corpses, non-epithelial cells) among the segmented objects were filtered using different criteria (object volume, object intensity, object solidity). Nuclei belonging to the gfp RNAi clone were identified using intensity-based thresholding of the His3.3-mIFP signal.

The data was fed to Mastodon (https://github.com/mastodon-sc) for semi-automatic tracking of SOPs from their birth to their division (when available). The presumptive SOPs were identified as high GFP (GFP-Sc movies) or low GFP cells (NRE-deGFP movies) on a maximal z-intensity projection at late stages. The six closest neighbors of SOPs were automatically detected in 3D based on their physical distances to SOPs and errors (due to the presence of dead corpses, over-segmentations and/or non-epithelial cells) were manually corrected using a customized web application (https://gitlab.pasteur.fr/4dunit/tracked-nuclear-neighbor-analyzer). The GFP signal was normalized in 3D by dividing the value measured in a single nucleus by the mean value of the signal measured at each time point in the same channel in the entire image stack (given the low intensity of the GFP-Sc signal, this value should closely reflect the autofluorescence noise).

The GFP-Sc kymograph was computed by dividing the AP axis into continuous bins with width=5 pixels. The mean GFP-Sc of all segmented nuclei within each bin at a given time is then calculated and normalized by the mean GFP-Sc of the image stack at that time.

Images were processed using max z-intensity projection for the most apical three or four z stacks to limit the projection to Ecad signal. Probability map of cells was generated using TrainableWeka (Arganda-Carreras et al., 2017), segmented automatically and manually corrected on Tissue Analyzer (Aigouy and Prud'homme, 2022). SOPs were semi-automatically tracked on Tissue Analyzer.

For the movies of RNAi cells clones, the SOP neighbors expressing RNAi are excluded when computing the FDI.

To measure the rate of cell-cell intercalation, the segmented apical areas were tracked and corrected by Tissue Analyzer (Aigouy and Prud'homme, 2022; Aigouy et al., 2016). SOP cells were identified based on GFP-Sc intensity and linked to their corresponding apical IDs. For each cell, we identified the IDs of its apical neighbors and scored changes over time in the ID of the neighbors that are not associated with cell division. This analysis was performed around the time of SB, from t=0 to t=4.5 h. Transient changes in neighbors (persisting for less than 50 min) were not counted.

Linear least-square regression was used to estimate the linear correlation between two datasets and the slope was tested using a Wald test. Pearson correlation analysis was used to measure correlation strength (R). Mean and standard deviation were used in time-series plots. Student's unpaired t-test was used to test the mean difference between two distributions.

Dissection of staged pupae and antibody staining were performed using standard procedures. Briefly, nota were dissected from staged pupae using Vannas micro-scissors, fixed in paraformaldehyde (4% in PBS 1×, 20 min) and incubated in PBS 1× with 0.1% Triton X100 and primary antibodies for 1.5 h at room temperature. The following antibodies were used: rat anti-Ecad (DCAD2 from DHSB, 1:100), goat anti-GFP (Abcam, ab6673, 1:1000), rabbit anti-DsRed (Clontech, 632496, 1:500), guinea-pig anti-Sens (from H. Bellen, Baylor College of Medicine, TX, USA; 1:2000) (Nolo et al., 2000), mouse anti-Cut (2B10, from DSHB, 1:100) and mouse anti-Hnt (1G9, from DSHB, 1:50). Secondary antibodies were from Jackson ImmunoResearch: donkey anti-rabbit Cy3 (1:1000; 711-165-152), anti-goat Alexa 488 (1:1000; 705-545-003), anti-guinea-pig Cy5 (1:1000; 706-175-148), anti-rat Cy3 (1:1000; 712-165-150), anti-mouse Cy5 and Cy3 (1:1000; 715-175-151 and 715-165-150). After washes in PBT, nota were mounted in 4% N-propyl-galate and 80% glycerol.

Images were acquired using a confocal Zeiss LSM780 microscope with 63× (PL APO, N.A. 1.4 DIC M27) and 40× (PL APO, N.A. 1.3 DIC M27) objectives. Adult flies were imaged using a Zeiss Discovery V20 stereo-macroscope using a 1.0× (PlanApo S FWD 60 mm) objective.

We thank Stein Aerts, Laure Bally-Cuif, Yohanns Bellaiche, Hugo Bellen, Francis Corson, Jean-Michel Gibert, Li He, Yang Hong, Romain Levayer, Juan Luna, Norbert Perrimon, Jean-Yves Tinevez, Alexis Villars, Flybase, Developmental Studies Hybridoma Bank and the Bloomington Drosophila Stock Center for reagents, resources and discussion. We thank the UtechS Photonic BioImaging (Imagopole; supported by France BioImaging, ANR-10-INBS-04) for use of a spinning-disk microscope. We thank R. Levayer for critical reading.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.203165.reviewer-comments.pdf