A combination of Notch signaling, preferential adhesion and endocytosis induces a slow mode of cell intercalation in the Drosophila retina

ABSTRACT Movement of epithelial cells in a tissue occurs through neighbor exchange and drives tissue shape changes. It requires intercellular junction remodeling, a process typically powered by the contractile actomyosin cytoskeleton. This has been investigated mainly in homogeneous epithelia, where intercalation takes minutes. However, in some tissues, intercalation involves different cell types and can take hours. Whether slow and fast intercalation share the same mechanisms remains to be examined. To address this issue, we used the fly eye, where the cone cells exchange neighbors over ∼10 h to shape the lens. We uncovered three pathways regulating this slow mode of cell intercalation. First, we found a limited requirement for MyosinII. In this case, mathematical modeling predicts an adhesion-dominant intercalation mechanism. Genetic experiments support this prediction, revealing a role for adhesion through the Nephrin proteins Roughest and Hibris. Second, we found that cone cell intercalation is regulated by the Notch pathway. Third, we show that endocytosis is required for membrane removal and Notch activation. Taken together, our work indicates that adhesion, endocytosis and Notch can direct slow cell intercalation during tissue morphogenesis.


Epithelial cells are polarized along the apical (top)-basal (bottom) axis and assemble into tissues via their lateral Adherens Junctions (AJ). Loss and creation of
AJs between cells can shape tissues by re-arranging the relative position of cells within the plane of the epithelium. For example, in the Drosophila germband, polarized steps of AJ loss and creation promote tissue elongation by inducing cell intercalation along the anterior-posterior axis of the embryo (Bertet et al., 2004;Blankenship et al., 2006;Tetley et al., 2016;Zallen and Wieschaus, 2004). Similar regulations take place between mesodermal cells in zebrafish to promote convergent extension (Yin et al., 2008), or during renal tube development in Xenopus for example (Lienkamp et al., 2012). In these relatively homogeneous tissues, intercalation between groups of four cells takes place over minutes.
Thus, MyoII accumulates preferentially at the AJs that shrink to drive intercalation. In this tissue, cell intercalation also requires endocytosis to shed membrane as an AJ is eliminated (Levayer et al., 2011). In other tissues such as the fly notum and wing disc, where cell intercalation also occurs over minutes, neighbor exchange is stochastic and reversible. Here, while MyoII is required for intercalation, its excessive accumulation at the AJ inhibits this intercalation (Curran et al., 2017). Furthermore, in the dorsal branch of the fly trachea, where cells intercalate to form tubes, MyoII is largely dispensable. In this case, intercalation between branch-cells is induced by forces that are generated by the migrating tip-cell, which pulls on the branch cells (Ochoa-Espinosa et al., 2017). Importantly, these examples indicate that there is no simple relationship between MyoII and intercalation.
The fly ommatidium, which is the basic visual unit of the insect compound eye, presents an interesting departure point from all these tissues in that it includes a deterministic step of neighbor exchange between four cells, that unfolds over approximately 10 hours and that is very reproducible. This slow intercalation occurs during lens formation, between four epithelial-like cells, called the cone cells.
Whether this type of slow intercalation is governed by the same mechanisms that underpin faster intercalation has not been investigated in detail. In the ommatidium, the four core cone cells are surrounded by two large primary pigment cells, themselves surrounded by a complement of narrow secondary and tertiary pigment cellscollectively referred to as Interommatidial cells (Cagan, 2009;Cagan and Ready, 1989;Charlton-Perkins et al., 2017;Ready et al., 1976;Wolff., 1993) ( Figure   1A). As they are specified, each one of the cone and pigment cells finds their location in the 2D plane of the lens through highly regulated steps of neighbor exchange (Cagan and Ready, 1989;Larson et al., 2008). These steps of neighbor exchange are controlled by preferential adhesion, whereby adhesion is favored between the primary pigment cells and the interommatidial precursors and is minimized amongst interommatidial cells. This preferential adhesion relies on the Neph/Nephrin-like immunoglobulin adhesion protein family. Hibris (Hbs; Nephrin-like) is expressed in the primary pigment cells and binds to Roughest (Rst; Neph-1), which is expressed in the interommatidial precursors (Bao and Cagan, 2005). In addition, Hbs functions in the four core cone cells to regulate their intercalation (Grillo-Hill and Wolff, 2009), suggesting preferential adhesion might be involved in cone cell intercalation. However so far, no requirement for Rst has been found in these cells.
In addition, Ncadherin (Ncad) has also been shown to play a role in regulating cone cell intercalation through establishing a planar polarized pattern of interfacial tension within the cone cell quartet. In this pattern, a higher tension, generated by MyoII, is found at the interface between the cone and primary pigment cells, when compared to the AJs between the cone cells where MyoII tension is limited (Chan et al., 2017).
However, how this MyoII pattern contributes to regulating intercalation in these cells and how it relates to preferential adhesion through Hbs is not clear. In studying this issue, we found very little requirement for RhoA-MyoII during cone cell intercalation.
Instead, our results indicate a preeminent role for the conserved Notch signaling pathways, adhesion through both Hbs and Rst, and endocytosis. Altogether our results suggest that preferential adhesion between cells is a principal mechanism of slow intercalation.

Adherens junction dynamics during cone cell intercalation
Cone cell intercalation unfolds over 10hrs ( Figure 1A-C and Movie S1) and consists of the elimination of the AJ between the anterior and posterior (A/P) cells, followed by the creation of a new AJ between the polar and equatorial (Pl/Eq) cells ( Figure 1B-C). To establish the cellular dynamics associated with this step of slow intercalation, we used the Ecad::GFP transgene to monitor how the AJs evolve as the ommatidium develops. We found that as the cone cells undergo neighbor exchange, they also increase their apical area ( Figure 1D-E, Movie S1). Quantification revealed that despite the elimination of the AJ between the A/P cells, the cone cell cluster elongates along the Pl/Eq axis upon intercalation ( Figure 1F-G). Following this, the cone cell quartet widens along the A/P axis, as the AJ is created between the Pl/Eq cells. Further, examining the dynamics of growth of the AJs shared by the four cone cells, we found a cross-correlation between the shrinkage of the A/P cone cell AJ and the expansion of the remaining adjoining AJs ( Figure 1H). Thus, a mechanism might exist whereby membrane removed from the shrinking AJ is recycled to the neighboring AJs. This correlation did not hold when considering the AJs the cone cells share with the primary pigment cells ( Figure 1I) suggesting local membrane redistribution between the AJs shared by the cone cells, but not with those shared with the primary pigment cells.
Next, we considered that the primary pigment cells that surround the cone cells could influence cone cell intercalation and shape. These two pigment cells share AJs that run parallel to the Pl/Eq axis of the lens and thus, that are aligned with the shrinking cone cell AJ ( Figure 1A). As the cone cell AJ shrinks, the AJs shared by the primary pigment cells lengthen ( Figure 1C-D), as demonstrated by the negative correlation of their length (Supplementary Figure 1A). Here, we reasoned that the negative correlation could indicate that these processes are linked. To test this idea, we compared the corresponding patterns of AJ length fluctuation by cross-correlating the fluctuation of the lengths of these AJs. However, quantification of this parameter showed that they were not correlated, suggesting that length changes in the primary pigment cell AJs do not directly influence changes in length of the cone cell AJ (Supplementary Figure 1B).

Limited MyoII accumulation at the shrinking adherens junction suggests a minimal role in intercalation.
To assess whether MyoII powers AJ remodeling to induce cone cell intercalation, we analyzed its distribution and intensity over time using a fly strain where the Myosin light chain is tagged with GFP (Sqh::GFP) (Figure 2A-C). We could detect a marginal 10% MyoII enrichment at the shrinking AJ between the A/P cone cells ( Figure 2D). This is small when compared to the values of 30-300% enrichment reported for shrinking AJs in the germband (Collinet et al., 2015;Pare et al., 2014;Simoes Sde et al., 2010). Using live imaging, we could also visualize pulsatile apical-medial meshworks of MyoII in the cone cells, but unlike in the intercalating cells in the germband, we could not detect any flow dynamics toward the shrinking AJ ( Figure 2E), nor later on towards the new AJ as it is created between the Pl/Eq cone cells ( Figure 2F). Altogether, these results suggest that slow cell intercalation in the eye lens is not primarily driven by MyoII contractility.

Modeling cone cell intercalation predicts a predominant role for adhesion.
The small increased Sqh::GFP accumulation at the A/P AJ compared to the adjoining AJs could regulate shrinkage of this AJ, however the small magnitude could mean that this MyoII activity is not sufficient to drive AJ shrinkage, and that instead adhesion between cells might dominate this process. It is also possible that extrinsic regulations are involved, for example from the primary pigment cells that surround the cone cells. To distinguish between these possibilities and assess a potential role for the primary pigment cells during cone cell intercalation, we built a computational vertex model (Farhadifar et al., 2007) of the ommatidium allowing us to vary the relative levels of contractility and adhesion at the cells' AJs.
Vertex models depend on modelling tension at each junction which is a combination of contractility from MyoII and counteracting adhesion forces. Therefore, to set up the vertex model, we quantified the intensity of MyoII at all AJs, and surveyed the principal adhesion molecules that have been shown to be involved in mediating intercellular adhesion in the lens. This includes Ecadherin and Ncad, as well as Rst and Hbs (Bao and Cagan, 2005;Carthew, 2005;Chan et al., 2017) ( Figure 3A).
Intensity values for each of these adhesion molecules were normalized and summed to give a pan-adhesion parameter, relative to the expression levels on the nonshrinking cone cell AJs (Supplementary table 1  We then used the model to explore how cone cell arrangement is affected when varying adhesion and/or contractility through modeling. The tension ranges were obtained by varying the averaging weight by 0, 30, 50, 70 or 100 percent to reflect the varying ratio of contribution of adhesion (cyan) and MyoII contractility (magenta) to tension at the AJ ( Figure 3D, y-axis). In addition to varying the tension at the AJs, we simulated a range of cytosolic MyoII contractility levels in the primary pigment cells, whereby contractility in these cells applies tensile forces on the cone cell cluster (Blackie et al., 2020) ( Figure 3C, dashed lines; Figure 3D, x-axis).
Considering an ommatidium in a stable post-intercalation topology (green zone in Figure 3D heatmap, Figure 3E''), a reduction in AJ adhesion (moving down Y-axis) causes the topology to switch out of post-intercalation topology to the four-way vertex (yellow zone, Figure 3D,E') and even back to the pre-intercalation topology (red zone, Figure 3D,E). Whereas a reduction in cone cell AJ MyoII contractility (moving up Y-axis) coupled with a reduction in primary pigment cell cytoplasmic MyoII (moving left on X-axis), ommatidium can remain in the stable post-intercalation state even for a significant amount of MyoII reduction (diagonal green zone, Figure 3D).
Thus, while a wide range of contractility/adhesion can support stable cone cell intercalation, our model suggests it can be achieved through adhesion, with minimal input form MyoII. Our model also suggest that stable cone cell intercalation might be mechanically influenced by the primary pigment cells ( Figure 3D).

A limited role for actomyosin contractility during cone cell intercalation
Our vertex model suggests that there are a range of values of MyoII, adhesion and PPC contractility where intercalation is achieved (green stable zone, Figure 3D).
These range from adhesion dominating with little role for MyoII-dependent contractility (upper left region, Figure 3D), to MyoII dominating but only with strong contractile forces from the PPCs to stabilize intercalation (lower right region, Figure   3D). To test which of these is true for the retina, we manipulated MyoII and made predictions from our model for how the tissue would behave for each region. To test these suggestions, we perturbed MyoII expression and activity specifically in the cone cells. A prediction from our model is that a reduction in MyoII levels in the cone cells should not push the cone cells back into a pre-intercalation state ( Figure 3D a shift up the y axis stays in the 'green' post-intercalation zone).
Consistent with this, we found that expression of the dominant negative version of the MyoII heavy chain, Zipper DN ::YFP (Barros et al., 2003) had a minimal impact on cone cell intercalation, with less than 5% of the ommatidia examined showing a failure to shrink the A/P AJ ( Figure 4A-B). Similarly, expressing a shRNAi (IR) against the heavy chain of MyoII (Zipper; zip) in the cone cells interfered with intercalation in less than 10% of the cases ( Figure 4C). In these experiments, we cannot measure how much MyoII activity is inhibited and it is therefore possible we are underestimating the role of MyoII in cone cell intercalation. For this reason and to further probe the contractile actomyosin cytoskeleton, we also examined requirement for the MyoII activator RhoA. Expressing a dominant negative RhoA transgene (RhoA N19 ) or shRNAi targeting this small GTPase in cone cells, led to significant relaxation of these cells, which strongly suggests inhibition of the contractile actomyosin cytoskeleton ( Figure 4D-E). However, this did not affect cone cell intercalation. Therefore, taken together, our experiments indicate a minimal requirement for the contractile actomyosin cytoskeleton during cone cell intercalation.
Next, to assess the second prediction of our vertex model, that if MyoII contractility dominates at the cone cell AJs, then stable intercalation would require pulling tension from the primary pigment cells, we expressed the shRNAi against RhoA specifically in the primary pigment cells ( Figure 4F). We found this did not block cone cell intercalation. However, quantification of the length of the AJ that forms between the polar and equatorial cone cells after intercalation showed that this AJ was shorter

Rst and Hbs regulate cone cell intercalation.
A likely process that might induce cone cell intercalation is intercellular adhesion.
Amongst the adhesion molecules that might play a role in these cells are Rst and Hbs. In the case of the cone cells, and with analogy to how these two factors promote preferential adhesion between interommatidial cells ( To examine this possibility, we used shRNAi to decrease the expression of hbs. When expressed in either the Eq and Pl cells, this led to limited defects in intercalation with a minority (ie 10 to 15%) of clusters failing to elongate the new AJ to complete intercalation ( Figure 5H' and Figure 5I). Decreasing hbs expression in the A and P cone cells led to a worse phenotype, where up to 30% of the clusters fail to shrink the A/P cone cell AJ ( Figure 5H'' and Figure 5I). In addition, it also led to other defects in cone cell configuration, whereby the mutant cell rounded up to minimize its interface with the flanking primary pigment cell ( Figure 5H''' and Figure   5I). Altogether, these results confirm previous work (Grillo-Hill and Wolff, 2009).
Interestingly, decreasing the expression of hbs in all four cone cells had very little effect on their intercalation, suggesting differential Hbs expression amongst these four cells is required. These results and the pattern of Hbs expression suggest that interaction between the cone and primary pigment cells are at play in cone cell intercalation.
To determine whether Rst is also required, we made use of the rst 6 mutant allele.
Homozygous rst 6 animals are viable, and as noted before (Grillo-Hill and Wolff, 2009), cone cell intercalation occurs normally in these animals. Based on our results that no defects in intercalation are observed when all cone cells are deficient for Hbs, we reasoned that a role for Rst might only be revealed in mosaic situations, when some cone cells are wild type. To test this idea, we recombined the rst 6 allele onto an FRT chromosome and generated somatic mutant clones. This approach showed that removing rst specifically in the Pl and Eq cone cells leads to defects in intercalation ( Figure 5J-J', quantified in Figure 5K). In these experiments, up to 40% of the ommatidia presenting both Pl and Eq rst 6 mutant cone cells failed to intercalate properly. In addition, a small proportion (less than 10%) of the clusters containing an A or P cone cell mutant for rst 6 , also show defects in intercalation. Altogether, the requirement we find for hbs and rst in the cone cells using mosaic analysis suggest that differential Hbs/Rst activity is required for cone cell intercalation.

Cone cell intercalation is regulated by the primary pigment cells.
A and P cone cells deficient for hbs expression minimize their AJ with the surrounding pigment cells suggesting a model whereby adhesion between these cells and the surrounding primary pigment cells could be regulated by the Hbs-Rst system. In this hypothesis, Rst would be required in the primary pigment cells. To test this, we examined ommatidia lacking rst in one or both primary pigment cells.
Firstly, we could confirm that decreasing hbs expression in the primary pigment cells did not affect cone cell intercalation (Grillo-Hill and Wolff, 2009) ( Figure 5L). In contrast, we found that abolishing rst expression in both primary pigment cells, affected cone cell intercalation in ~65% of the cases ( Figure 5M). In addition, 35% of the cases had cone cell configurations resembling those seen when decreasing the expression of hbs in the A and P cone cells ( Figure 5H'''). Together with the patterns of Rst::GFP and Hbs::GFP expression ( Figure 5G), these results support the model that the interface between the A and P cone cells and the surrounding pigment cells is promoted by the Hbs-Rst adhesion system, with Hbs in the cone cells engaging in trans with Rst expressed in the surrounding primary pigment cells.

Notch signaling regulates cone cell intercalation.
Previous studies have shown that the expression of Hbs in the primary pigment cells is regulated by Notch (Bao, 2014). This connection and the finding that Hbs regulates cone cell intercalation prompted us to investigate the role of Notch in this process. Firstly, we examined the distribution patterns of Notch and its ligand Delta before and after intercalation using live imaging of functional GFP-tagged proteins ( Figure 6A-B) (Corson et al., 2017;Trylinski et al., 2017). At the onset of cone cell intercalation, we found that Notch was present at the apical cortex of all cone cells ( Figure 6A) whereas, confirming previous observations (Bao, 2014), we found that Dl was highly expressed in the A cell and to a lesser extent, in the P cell ( Figure 6B).
We also examined the distribution of Neuralized (Neur), a developmentally regulated E3 ubiquitin ligase that is required for Dl endocytosis and N receptor activation ( Figure 6C) (Schweisguth, 2004;Weinmaster and Fischer, 2011). While Neur was detected in all cone cells, we observed a higher level in the A and P cells. These patterns of expression suggest that high levels of Dl in the A cell activate N in the P, Pl and Eq cell. To test this suggestion, we monitored N signaling over time by measuring activated nuclear Notch, or NICD, in the cone cells of living NiGFP pupae (endogenous Notch was tagged with GFP in its intracellular domain so that nuclear GFP reveals NICD; Figure 6D and 6E) (Couturier et al., 2012). NICD was detected in the Pl and Eq cells before and after intercalation. In contrast, NICD was detected in the P cell only before intercalation, suggesting that loss of direct contact with the A cell results in loss of N receptor activation. Finally, NICD was not detected in the A cell, further suggesting that this cell is the Dl signal-sending cell within the cone cell quartet. This pattern of N activity was confirmed by examining the expression of two direct N targets, E(spl)m -HLH ( Figure 6F) and E(spl)m3-HLH (not shown) using GFP-tagged proteins. Altogether, our analysis revealed that directional signaling occurs within the cone cell quartet and that a specific change in the pattern of N activity correlates with cone cell intercalation.
To test whether N signaling contributes to cone cell intercalation, we used a to defects in their intercalation ( Figure 6G-H). Live imaging of these retinas revealed that these defects were due to failures in stabilizing the new Pl/Eq AJ, which led to reversion to the original configuration with an AJ between the A and P cells (Movie S4 and Figure 6I). In good agreement with our finding that N is active in the Eq and Pl cells throughout intercalation, examination of ommatidia mosaic for Mam DN revealed a strong requirement for transcriptional regulation downstream of N in each of these two cells ( Figure 6J-K). From these experiments, we conclude that signaling from either the A or P cell is sufficient to activate N in the Eq and Pl cells and that N activity in the Eq and Pl cone cells is critical for intercalation. This pattern of N and the requirement for Hbs in all cone cells does not suggest a simple Notch-Hbs link in this case. In addition, expressing Hbs in Pl/Eq cone cells also expressing Mam DN , did not rescue the intercalation phenotype (not shown).

Endocytosis is required at all stages of cone cell intercalation.
Our finding suggesting that membrane is recycled from the shrinking junction to the neighboring junctions ( Figure 1H) led us to hypothesize that endocytosis is involved in intercalation by promoting membrane removal during AJ shrinkage. To test this idea, we used the thermo-sensitive Shibire ts1 protein (ortholog of dynamin). Inhibiting endocytosis in the cone cells as they intercalate caused the AJs they share to become convoluted ( Figure 7A). We therefore conclude that endocytosis regulates membrane turnover during cone cell intercalation. Additionally, cone cell intercalation was blocked either during AJ shrinkage (72% of cases) or at the four-way vertex (18% of cases) ( Figure 7B-C) when assayed at a stage where in wild-type retinas roughly 100% of the ommatidia should have completed intercalation.
In order to refine this analysis, we then inhibited endocytosis for only 4hrs during each of the three stages of cone cell intercalation: AJ shrinkage, four-way vertex and AJ elongation ( Figure 7D-F). In these experiments, intercalation was stalled or reverted to an earlier stage in a significant number of cases ( Figure 7G). Time-lapse imaging confirmed that inhibition of endocytosis caused a failure in cone cell A/P AJ shrinkage (Movie S5). In agreement with our quantifications ( Figure 7G), live imaging revealed that inhibiting endocytosis at the four-way vertex stage (24%APF) led to reversions to the initial cone cell configuration ( Figure 7H). Altogether these results demonstrate that endocytosis is required for stable intercalation to proceed between the cone cells. AJ convolution is compatible with an excess of plasma membrane, which is consistent with a role for endocytosis in shedding membrane during slow intercalation.
Another aspect of cone cell intercalation that is expected to be affected when . We found that expression of Brd R simultaneously in the A and P cone cells led to defects in intercalation in 40% of the cases, while expressing Brd R in either the A or P cell rarely led to defective intercalation ( Figure 7I). Thus, next to promoting membrane removal, endocytosis is also required for N-activity in cone cell intercalation.

DISCUSSION
Most instances of epithelial cell intercalation studied so far have focused on rapid neighbor exchanges, occurring over periods of tens of minutes, as is the case in the fly germband (Bertet et al., 2004;Zallen and Wieschaus, 2004), or a couple hours as during kidney tubule morphogenesis and early retinal development (Lienkamp et al., 2012 ;Robertson et al., 2012). Intercalation of the cone cells in the fly is much slower as it unfolds over ~10 hours. Combining mathematical modeling and genetics experiments, we found that it involves regulations that appear more complex than the ones found to control faster intercalation (Heisenberg and Bellaiche, 2013;LeGoff and Lecuit, 2015). Our results show that cone cell intercalation requires transcription downstream of Notch in two of the four intercalating cells. We also find that cone cell intercalation shows a low requirement for the contractile MyoII pathways, and instead, relies more on adhesion, regulated by Rst and Hbs. In addition, our experiments suggest that membrane endocytosis is essential for stable intercalation by shedding plasma membrane during AJ shrinkage and by promoting N-activation.
Finally, we present evidence that cone cell intercalation involves processes extrinsic to the cone cell quartet, involving tensile force and adhesion in the neighboring cells.

The RhoA-MyoII axis is largely dispensable for cone cell intercalation
We found a small enrichment for MyoII at the shrinking cone cell AJ, suggesting that it could contribute to inducing shrinkage of the AJ over time. In addition, inhibiting the activity and expression of this motor protein led to cone cell intercalation defects in approximately 10% of the ommatidia we examined. Therefore, MyoII plays a limited role during cone cell intercalation. However, we cannot exclude the possibility that residual MyoII levels are present in our genetic perturbations, which are sufficient to promote cone cell intercalation. Our perturbations of RhoA, an upstream regulator of the actomyosin cytoskeleton, show that while RhoA inhibition in cone cells leads to a relaxation of their apical profile, intercalation is not impacted.
These experiments also support our model that MyoII is minimally required in cone cell intercalation. This conclusion is also consistent with recent work revealing that normal cone cell intercalation requires Ncad to downregulate MyoII at the AJs shared between the cone cells (Chan et al., 2017). Therefore, irrespective of potential redundancies in MyoII activation in retinal cells, this work and our present study strongly suggest that MyoII concentration/activity needs to be limited at the cone cell AJs for intercalation to proceed normally. This situation is somewhat analogous to that reported in the fly notum where lowering MyoII activity allows for tissue fluidity (i.e. neighbor exchange) (Curran et al., 2017). However, unlike in the notum, AJ remodeling is not stochastic but deterministic in the cone cells.

Adhesion regulates cone cell intercalation.
We found that both Hbs and Rst regulate cone cell intercalation. However, the general adhesion logic at play between the cone cells, and also between them and the surrounding primary pigment cells is not clear. Hbs is required in all individual cone cells, and our work argues that it plays a role in stabilizing the new AJ created between the Eq and Pl cells after intercalation. Rst in the Eq/Pl cells also contributes to regulating creation of this AJ. It is therefore possible that a Hbs/Rst interaction in trans takes place between the Eq/Pl cells to stabilize the newly created AJ.
However, Rst::GFP and Hbs::GFP show no detectable enrichment at these AJs. In fact, one of the strongest effects on cone cell intercalation is seen when rst is removed from the primary pigment cells. Rst::GFP and Hbs::GFP are detected at the AJ between these cells and the cone cells. We therefore hypothesize that Hbs-Rst interactions take place between these two cell types. In this hypothesis, we envisage that Hbs expressed in the cone cells contacts Rst expressed in the primary pigment cells. The primary pigment cells also express Hbs, which accumulates at the AJ they share with the interommatidial cells (Bao and Cagan, 2005). Our work therefore raises the possibility that the primary pigment cells express both Rst and Hbs, and that these proteins have a planar polarized distribution in this cell type. This possibility is consistent with our finding that both Rst and Hbs are strongly localised to the AJ between the two primary pigment cells.

Notch-signaling between cone cells regulates intercalation.
In addition to a role for Hbs and Rst, our work also revealed that a N-Dl code exists between the four cone cells that regulates their intercalation. In this code, Dl signals from the A or P cell whereas N is activated and required in the Eq and Pl cells. This function for N requires transcription since it is blocked by Mam DN . Previous work has established that N can induce Hbs expression (Bao, 2014). However, the pattern of N activation in the Eq/Pl cells and the requirement for Hbs in all cone cells does not suggest a Notch-Hbs pathway in this case. Thus, more work will be required to understand how N regulates intercalation and how expression of Hbs and Rst is controlled in the cone cells. We also note that Notch activation has previously been placed downstream of the Hbs-Rst system in the eye (Singh and Mlodzik, 2012).
Such regulation could also be at play during cone cell intercalation. Considering N signaling between the cone cells, it is possible that perturbing Hbs/Rst expression in these cells might affect the Dl-N interface by changing the surface contact these cells share.

Endocytosis and cone cell intercalation
Our work reveals a key function for membrane endocytosis during cone cell intercalation. This could be due to several reasons. Blocking endocytosis in the primary pigment cells leads to defects in Rst localization (Johnson et al., 2008), and by analogy, blocking endocytosis in the cone cells could also affect the Rst/Hbs pathway and thus intercalation. Similarly, endocytosis of Dl is required to activate the Notch pathway (Schweisguth, 2004;Weinmaster and Fischer, 2011). Indeed, blocking Dl endocytosis by expressing brd DN in the A/P cone cells interferes with intercalation. Endocytosis is also likely to be required to control the turnover of Ecad/Ncad in the cone cells and promote plasma membrane removal during AJ

shrinkage. For instance, as these cells intercalate Ecad becomes depleted from the AJ between the cone cells and Ecad overexpression in the Eq/Pl cone cells blocks
intercalation (Carthew, 2005). It is possible that endocytosis promotes this inhibition.
Consistent with this type of model, similar regulation of Ecad endocytosis takes place in the fly embryo during intercalation to shrink AJs (Levayer et al., 2011).

Extrinsic regulation of cone cell intercalation.
In addition to showing a dominant role for adhesion over MyoII in regulating cone cell intercalation, our vertex model also predicts that external tensile forces in the surrounding primary pigment cells could play a role (Fig. 3D, moving left on X-axis in the heat map). The primary pigment cells present medial meshworks of actomyosin that are contractile (Blackie et al., 2020). This contractility regulates the width of these cells' apical profiles and would be expected to pull onto the AJ these cells share with the A and P cone cell. Perturbing the actomyosin cytoskeleton in the primary pigment cells by inhibiting RhoA expression led to a decrease in the length of the cone cell central contact but did not completely block intercalation (Fig. 4E-F).
Taken together with the fact that reduction of MyoII in the cone cells doesn't inhibit intercalation but reduction of adhesion does, this suggests that ommatidia lie in the upper/left section of the predictive heatmap generated by our model, whereby this slow mode of intercalation is dominantly driven by adhesion within the cone cells.
Animals were raised at 18°C.

Inhibition of endocytosis using shibire ts1
UAS-shibire ts1 flies were crossed to prosGal4 flies and raised at 25°C until the stated time of development, then transferred to 31°C to block endocytosis for either 4hrs or overnight. Retinas were dissected at 40% APF and scored for progression of cone cell intercalation.

Antibodies and immunostaining
Pupae were staged at 25°C or 29°C to 40%APF then retinas were dissected in PBS on ice and fixed in 4% paraformaldehyde for 20mins at room temperature (RT). were acquired on a Leica SPE, Leica SP5 or Leica TCS SP8 confocal microscope. A 40x oil objective was used for imaging of whole retinas for quantification and a 63x oil objective was used for higher magnification images.

Statistical tests
Statistical tests were performed in GraphPad Prism 7. Data was compared using a Ttest (paired or unpaired as appropriate) or One-way ANOVA with Tukey's post-hoc tests.

Laser ablation
Ecad::GFP pupae were raised and mounted as for time-lapse imaging experiments.
Ablations of AJ between primary pigment cells and between interommatidial cells, were performed using a Zeiss LSM880 microscope with a Plan Apochromat 63x/NA1.4 oil objective using 740nm multiphoton excitation from a Ti-sapphire laser.
An ROI of 3x3 pixels was drawn over an AJ and ablated with 5-10% laser power at the slowest scan speed for 1 iteration. Images were acquired every 1sec after ablation. Settings were optimized by imaging sqh AX3 ;sqh::GFP; flies during ablation to ensure only the AJ-associated MyoII was removed and that the medial meshwork of MyoII remained intact. The AJs repaired after every instance of ablation, indicating that the cell was not damaged. Positions of the two adjoining vertices after ablation were manually tracked using FIJI and the distance between them calculated at each frame after ablation. Recoil velocity was calculated by a linear fit across the first frames after ablation. One-way ANOVA was performed in GraphPad Prism7 to compare stages.

Cone cell area through time
The Tissue Analyser FIJI plugin (Aigouy et al., 2010) with manual correction was used to segment the cone cells on a time-lapse (5mins/frame) of ;Ecad::GFP; retinas.
The areas of individual cone cells were then measured and averaged across four ommatidia.

Cone cell axes length through time
The perimeter of the cone cell cluster was traced manually using the Freehand selection tool in FIJI on every 20th frame (100mins) of time-lapse of ;Ecad::GFP; retinas. To measure A/P and Eq/Pl axes length an ellipse was fitted over the cone cells, measured over time and then expressed as a ratio. Measurements were averaged for each time point over 13 time-registered ommatidia from two independent retinas.

AJ perimeter measurements
AJ lengths were taken as the inter-vertex distance and were measured manually on each frame of time-lapse movies of ;Ecad::GFP; retinas using the line tool in FIJI.
Measurements were averaged from 13 time-registered ommatidia from two independent retinae.

Cross-correlation of AJ length
Curves of AJ length over time for the cone cells were smoothed by taking a five-point moving average. Data were detrended by taking the running difference to find the change in AJ length over time. Cross-correlation was performed in R statistical package using the ccf function. The mean cross-correlation function was calculated as the average correlation coefficient at each time lag across 13 ommatidia from two independent retinas.

Scoring cone cell intercalation
Retinas were dissected at 40% APF, fixed and stained for Arm to label the AJs.
Whole retinas were imaged with a 40x objective taking a Z-series in 0.5μm slices in two tiles. Tiles were manually aligned and combined using Align3 TP plugin in FIJI.
Ommatidia were manually scored for stage of cone cell intercalation across each

Scoring switches between stages of cone cell intercalation
For the Ecad::GFP; and ;Ecad::GFP/Ecad::GFP;UAS-Mam DN /pros-Gal4 time-lapse movies, ommatidia were scored for which stage of cone cell intercalation they were in for each frame of the movie (5min intervals). A 'switch' in a given frame is defined as being scored as at a different stage to that of the previous frame. The total number of switches was measured from 10% of retinal development (before cone cell intercalation) for roughly 12hrs (until after cone cell intercalation is completed in the wild type).

Particle Image Velocimetry (PIV)
sqh AX3 ;sqh-sqh::GFP; ommatidia were imaged on a Zeiss LSM880 microscope with a Plan Apochromat 63x/NA1.4 oil objective using airyscan detectors to increase resolution. Movies were processed by Bleach correction, Gaussian blur and registered with the Stack-reg plugin (Thevenaz et al., 1998) if needed in FIJI. PIV analysis was performed using the FIJI PIV plugin (Tseng et al., 2012), by choosing an 8x8 pixel window with a time lag 4.34 seconds. Cell contours were tracked using the Tissue Analyzer plugin (Aigouy and Le Bivic, 2016) in FIJI or manually in FIJI to segment the cone cells. The angles of each PIV vector within the cone cells over time were plotted in a polar histogram in MATLAB to show lack of overall directional flow.
Here, E(R i ) is the total energy of the system for a given set of vertex position (R i ) that the algorithm minimizes. The system is composed of N c cells (a = 1…N c ), and N v vertices (i = 1…N v ). For area conservation term (i), K α is the elasticity coefficient, A α is the current area of the cell, and A (0) α is the ideal area. For the tensile/adhesive contact energy contribution (ii), Λ ij is the line tension coefficient for the junction couple (i,j), and l ij is the length of the junction between vertices. For the simulations in this manuscript, the base tension is Λ 0 ij = 0.26, base A (0) α is 1.0 (see Supplementary   Table 1, Figure 3), and K α is 1.0. The small cone cell area is set to A (0) α and remaining cell sizes are scaled accordingly with experimental size scaling measurements (Sup. Table 1). The symmetrical side junctions in between the cone cells are selected as the base, similar to experimental intensity measurements depicted in Figure 3A-A''''. A weighted average of the adhesion and myosin intensity measurements ( Figure 3B') are used as surrogate for scaling tension values (Sup Table 1). To scale the tension contribution of adhesion to a cell-cell junction, the base tension level is scaled inversely to normalized adhesion intensity of the junction. For scaling MyoII levels, the tension is scaled directly proportional to normalized MyoII intensity. This is given in the following equation: Where S is a scaling factor for the line tension, (w) are the averaging weights of myosin and adhesion and (c) are the normalized intensities. During the postintercalation (late) phase, the primary pigment cell contact and contacts between interommatidial cells on the sides of the ommatidium were not significantly different, whereas the tension on the top & bottom interommatidial cell contacts was 51 percent higher ( Figure 3C). The laser ablations depict the true tension of a bond, and model parameters are scaled accordingly where information is available (Sup Table   1). Simulations are carried out on a setup of 91 connected ommatidia with fixed boundaries, and analysis is done on the central ommatidium.          Development: doi:10.1242/dev.197301: Supplementary information