Inter-plane feedback coordinates cell morphogenesis and maintains 3D tissue organization in the Drosophila pupal retina

The spatial arrangement of cells within an epithelium is crucial to the final form and function of the tissue. During development, genetically controlled terminal differentiation programs produce the specialized cytoskeletal structures, cell–cell junctional adhesions and cell-extracellular matrix (ECM) contacts unique to each cell type. In turn, the resulting cell shapes, structures and connections introduce specific packing constraints that influence final organ form. Although progress has been made in describing the acquisition of tissue form in simple epithelia with relatively homogeneous cell composition, how complex tissues with diverse cell fates, shapes and physical properties spatially coordinate extensive morphogenetic remodeling to maintain robust organization remains poorly understood (Collinet and Lecuit, 2021).

Previous studies have shown that coordinated cell shape changes driven by subcellular cytoskeletal and junction remodeling produce tissue-level morphology. A well-studied example is apical constriction, where supracellular networks physically couple cell apices across a tissue plane to drive numerous morphogenetic processes, including epithelial folding, bending, invagination and closure (Martin and Goldstein, 2014; Perez-Vale and Peifer, 2020). Subsequent consideration of cells as 3D units has emphasized how remodeling of cellular structure along basal or lateral planes or of the ECM can promote morphogenetic change (Daley and Yamada, 2013; Gracia et al., 2019; Harmansa et al., 2023; Roellig et al., 2022; Sui et al., 2018). Temporal sequences of independent planar changes also coordinate 3D change. Notable examples include the sequential activation of actomyosin contractility along different planes to organize the successive patterns of apical and lateral contraction required for endoderm invagination in the ascidian embryo (Sherrard et al., 2010) and lumen morphogenesis in the C. elegans vulva (Yang et al., 2017). In all these examples, the apical-lateral-basal organization inherent to polarized epithelial tissues provides an intuitive physical conduit for driving 3D cellular and tissue-level morphogenetic change. Despite the established importance of supracellular networks in providing mechanical coupling within individual planes, whether and how remodeling processes interact across different planes to produce specific 3D cellular and tissue scale morphologies remains to be explored.

The stereotyped pseudostratified epithelial architecture of the Drosophila compound eye makes it an attractive model with which approach this question. The fly retina is a complex epithelial organ whose form and function depend on the precise organization of highly specialized and uniquely shaped cell types (Fig. 1A; (Charlton-Perkins and Cook, 2010; Ready et al., 1976; Wolff and Ready, 1993). Clusters of eight photoreceptor neurons occupy the central core of each ommatidial unit, with their photosensitive rhabdomeres defining the longitudinal optical axis of the epithelium. Directly above each cluster, an apical assembly of four cone and two primary pigment cells produces the lens that will focus incoming light onto the underlying photoreceptors. A hexagonal lattice of secondary and tertiary interommatidial pigment cells (IOPCs) surrounds each photoreceptor cluster. Through their cortical cytoskeletal-enriched junctional domains and basal cell–cell/ECM junctional contacts, the IOPCs provide in-plane connections across the apical and basal planes of the retinal field.

Pupal retinal development can be separated into two phases: a pattern establishment phase and a tissue elongation phase during which pattern is maintained (Fig. 1B) (Cagan and Ready, 1989; Johnson, 2021; Ready et al., 1976). During the patterning phase, cone cell and IOPC rearrangements produce the precise hexagonal lattice pattern, while photoreceptor apical domain involution and anchorage to the cone cells align the optical axis relative to the apical and basal surfaces. During the elongation phase, cell–cell contacts and overall tissue organization are maintained, while the photoreceptor and IOPC terminal differentiation programs elaborate functional specialized structures and the epithelium elongates fourfold. Although the emergence of two-dimensional (2D) planar pattern and the individual morphogenetic events in different retinal cell types during the early patterning phase have been well described (Bao and Cagan, 2005; Baumann, 2004; Cagan and Ready, 1989; Hayashi and Carthew, 2004; Hilgenfeldt et al., 2008; Johnson, 2021; Kafer et al., 2007; Longley and Ready, 1995; Pellikka et al., 2002; Pham et al., 2008; Ready, 2002; Ready and Chang, 2021; Signore et al., 2018), how the subsequent cellular morphogenetic changes are regulated, coordinated and integrated across different tissue planes to maintain 3D retinal organization and integrity during retinal elongation has not been explored.

We chose the cytoplasmic tyrosine kinase and cytoskeletal regulator Abelson (Abl) as a tool to examine the impact of modulating retinal cell shapes and structures on 3D tissue organization. Mechanistic studies have shown how Abl modulates cytoskeletal remodeling and junctional dynamics that control cell morphology in a wide range of epithelial tissues. For example, in the early embryonic epithelium, Abl helps to coordinate apical constriction during mesoderm invagination and to produce the cell shape changes that drive convergent extension during germband elongation (Fox and Peifer, 2007; Jodoin and Martin, 2016; Tamada et al., 2012; Yu and Zallen, 2020). Inhibition of Enabled (Ena)-mediated linear F-actin assembly is a key aspect of Abl function in many contexts (Comer et al., 1998; Forsthoefel et al., 2005; Fox and Peifer, 2007; Gates et al., 2007; Gertler et al., 1995; Grevengoed et al., 2001; Kannan et al., 2014, 2017; Lin et al., 2009; Rogers et al., 2021). Best studied is the embryonic central nervous system, where, in response to different axon guidance cues, Abl modulates the balance between linear and branched F-actin by regulating the activity not only of Enabled, but also of the WAVE/SCAR complex (Forsthoefel et al., 2005; Gertler et al., 1995; Kannan et al., 2017; Liebl et al., 2000; Wills et al., 1999). In the retinal epithelium, previous phenotypic analyses showed that Abl is required for multiple aspects of photoreceptor terminal differentiation and that its loss disrupts ommatidial organization and retinal pattern (Bennett and Hoffmann, 1992; Henkemeyer et al., 1987, 1990; Kannan et al., 2014; Singh et al., 2010; Xiong and Rebay, 2011; Xiong et al., 2013).

In this study, we have investigated how cellular morphogenetic events are individually controlled and effectively communicated between two different retinal cell types as the late pupal retina elaborates and maintains its precise 3D organization. Our approach was to combine genetic perturbation of Abl function with single-cell resolution fixed and live time-lapse imaging to examine retinal cell shapes and tissue-scale patterns. First, we used global depletion of Abl to characterize its contributions to the cytoskeletal specializations each cell type elaborates. In contrast to a wild-type retina, where correctly elaborated and aligned photoreceptor and IOPC cytoskeletal domains confine cell shape and organization, loss of Abl disrupted photoreceptor and IOPC terminal differentiation, resulting in a tissue whose heterogenous cellular shapes, structures and intercellular connections were insufficient to maintain retinal integrity. Second, we probed how IOPCs and photoreceptors interact by restoring Abl to each individual cell type in an otherwise abl mutant background. Strikingly, these experiments uncovered a feedback interaction between the two cell types that enabled rescue of either photoreceptor or IOPC terminal differentiation to induce morphogenetic remodeling of the other cell type and thus restore retinal 3D organization and tissue integrity. Together our results suggest that Abl plays important cell-autonomous and non-autonomous roles in establishing and maintaining the network of cytoskeletal and junctional adhesions that together provide dynamic 3D organization to the pupal retinal epithelium.

We were interested in how the distinctive terminal differentiation programs of the photoreceptors and IOPCs are coordinated to ensure robust 3D tissue organization during morphogenesis. As a framework for considering this question, we briefly summarize how the apical, basal and longitudinal networks of actin-based cytoskeletal structures and junctional adhesions established during the patterning phase are subsequently remodeled to produce the specialized structures of a mature wild-type retina (Fig. 1B) (Longley and Ready, 1995; Ready, 2002).

First, the apical network is defined by the hexagonally arranged IOPC apical junctional domains. After its establishment by 50% p.d., the apical network is stably maintained with minimal change (Bao and Cagan, 2005; Cagan and Ready, 1989; Hayashi and Carthew, 2004; Signore et al., 2018). Second, at the basal epithelial plane, IOPCs refine their cell-ECM contacts by anchoring to integrin-reinforced rings, or grommets, at the center of each ommatidium, and by tiling the entire retinal floor with a radial pattern of contractile actomyosin “feet”; during the elongation phase, coordinated contraction of the basal network compacts the retinal floor (Baumann, 2004; Longley and Ready, 1995; Ready and Chang, 2021, 2023). Also referred to as the “fenestrated membrane”, this specialized contractile network supports and separates the retina from the brain. Third, the longitudinal network is defined by rhabdomeric precursors consisting of the photoreceptor involuted apical membranes and associated cytoskeletal-junctional complexes. These specialized domains bridge the apical and basal tissue planes at the centerpoint of each ommatidium through junctional connections with the cone cell apical caps and basal feet. This unique coupling is established by 50% pupal development (p.d.) and then maintained as the rhabdomeres mature and expand during the elongation phase (Cagan and Ready, 1989). Because rhabdomeres anchor basally to the cone cell feet, which like the IOPC feet are anchored to the grommets, these rings mark a hub of junctional attachments that physically connect the longitudinal and basal networks, and that couple the basal network to the ECM. Photoreceptors do not contribute directly to the 2D apical and basal patterns because their longitudinal anchor points are located, respectively, below and above the apical and basal network planes. Conversely, IOPCs contribute minimally to the longitudinal network because their thin cellular projections occupy little tissue space and do not make substantial junctional connections along the longitudinal axis.

Using this framework to explore the cellular relationships that underlie tissue-level organization, we examined the consequences of Abl loss to the patterning within these three networks. We compared wild-type and abl^null retinas along each plane at 50% p.d., when the cellular structures that define each axis are first fully aligned and connected, and then at 75 and 100% p.d., to assess pattern maintenance as the longitudinal and basal networks remodel to produce the final adult form. We used F-actin to highlight the specialized cytoskeletal structures that give retinal cells their distinct shapes and E-cadherin (Ecad) to mark the adherens junction connections that organize them into ommatidial units.

Focusing first on the apical network, variable apical cell profiles, sizes and cell-cell contacts precluded regular hexagonal packing in abl mutant retinas (compare Fig. 1F-H with Fig. 1C-E; Fig. S1A). We used secondary IOPC length, the number of cone cell apical profiles and the ommatidial shape parameter as quantitative metrics of disrupted apical network pattern. At the cell scale, loss of abl resulted in significant reductions in secondary IOPC length (Fig. 1I) and in the average number of cone cell apical profiles per ommatidium (Fig. 1J). At the ommatidial scale, a significantly increased and more variable polygonal shape parameter reported the loss of regular hexagonal packing (Fig. 1K). Qualitatively, as seen in the representative images shown in Fig. 1F-H, tissue-level apical pattern appeared to improve over time, with the regular lattice-like gridwork of ommatidia more obvious at later stages than at 50% p.d.. Because we could not track and image individual pupae over the 2-day developmental window, this apparent ‘improvement’ could simply reflect a failure of abl null animals with more severe phenotypes to survive beyond 50% p.d.. Alternatively, continued optimization of cell-cell apical contacts during the elongation phase might enable modest recovery of pattern. Live imaging over long time scales coupled with analysis of tissue-level apical pattern will be needed to explore this further.

Turning to the basal network, at 50% p.d., F-actin localization outlined a recognizable radial pattern of IOPC feet in abl^null retinas (compare Fig. 1F′ with Fig. 1C′; images shown in Fig. 1C′-H″ are of the same retinas and same ommatidia shown in Fig. 1C-H, although because of basal contraction, more ommatidia are included in the field of view at 100% p.d.). However, secondary IOPC foot length was reduced, the size of the central Integrin-marked rings at the base of each ommatidium was increased, and both were more variable than in wild type (compare Fig. 1F″ with Fig. 1C″,L,O), indicating perturbation of overall pattern. At later time points, although F-actin rich stress fibers formed and contracted, basal network pattern was obviously disrupted (compare Fig. 1G′,H′ with Fig. 1D′,E′). In wild-type 75% p.d. retinas, the intensity of F-actin accumulation in the developing stress fibers forms a gradient with highest levels emanating from the central ring followed by a near-linear decay to the midpoint of the IOPC foot (Fig. 1D′,M). Across the tissue field, this sets up a regular pattern of radially arrayed stress fibers that interconnect the central rings (Fig. 1D′). In contrast, the developing stress fibers in abl^null retinas had aberrant F-actin profiles with a sharp accumulation at the point of contact with the central ring followed by an abrupt decay to a low level across the foot to the midpoint (Fig. 1G′,M). At 100% p.d., F-actin distribution was more uniform across the abl null IOPC feet compared with wild type, where the high-to-low gradient was maintained (Fig. 1E′,H′,N).

The disruption in F-actin pattern in abl null retinas was matched by loss of the Integrin enrichment around the central rings seen in wild type (compare Fig. 1G″,H″ with Fig. 1D″,E″). Central ring size became increasingly variable (Fig. 1O), suggesting a progressive disruption in tissue pattern as the basal network contracts. To assess contraction, we measured the surface area of mature 100% p.d. basal stress fibers and found a reduction in abl^null retinas relative to wild type (Fig. 1P). This suggests that Abl limits stress fiber contractility either directly by modulating F-actin dynamics in the IOPC feet or indirectly by organizing/stabilizing the cell-cell and cell-ECM connections that support the fenestrated membrane.

The disorganization and variability within the apical and basal networks predicted disruption of the longitudinal network structures that bridge the two planes. We examined the cytoskeletal-junctional structures that normally provide this connection and found abl loss caused a strong disruption (Fig. 2A-D). At 50% p.d., the photoreceptor specialized apical domains (rhabdomere precursors) align along the longitudinal axis and span the full tissue depth with anchor points at the apical cone cell caps and basal feet; an E-cadherin-marked adherens junction belt surrounds the F-actin-rich cytoskeletal domains and connects neighboring photoreceptors within each ommatidium (Fig. 2A). In contrast, and consistent with our earlier study (Xiong and Rebay, 2011), in abl mutant retinas, the photoreceptor apical domains were no longer aligned correctly along the longitudinal axis, their apical anchor points had dropped more basally and the adherens junction belt appeared reduced (Fig. 2B,E,F). By 100% p.d., the well-aligned bundles of long rod-like rhabdomeres seen in wild type (Fig. 2C) were not detected in abl null ommatidia (Fig. 2D). Instead, a tangled mass of Ecad-marked apical membrane beneath the fenestrated membrane (Fig. 2D) suggested the collapse of abl mutant photoreceptors through the mis-patterned retinal floor. Retinal depth was also reduced (compare Fig. 2D with Fig. 2C,G).

We confirmed this ‘falling’ phenotype by comparing the position of photoreceptor cell nuclei in wild-type versus abl^null 100% p.d. retina-brain complexes (Fig. 2H-J). In wild type, photoreceptor nuclei (with the exception of R8 cell nuclei that reside basally) cluster in a tight row immediately below the retinal surface (Fig. 2I), whereas in abl^null retina-brain complexes, photoreceptor nuclei were found beneath the retinal floor (Fig. 2J). Despite the aberrant position of these cells, lineage tracing confirmed their photoreceptor origin and identity (Fig. S1B-E′). These results confirmed our previous report of progressive photoreceptor ‘loss’ from the retina but refuted the dedifferentiation mechanism we had proposed (Xiong et al., 2013). Instead, our analysis suggests that the cellular defects associated with Abl loss perturb the 3D organization and integrity of the epithelium, with photoreceptor ‘falling’ a component of the cell and tissue scale disruptions.

To examine further the changes in retinal cell shapes and relative positions during the photoreceptor falling process, we labeled cell membranes by injecting a fluorescent membrane-associated dye and then performed time-lapse live imaging (Fig. 2K,L; Fig. S2A-D; Movies 1 and 2). In wild type, completion of the early patterning phase resulted in stereotyped and stable cell shapes, positions and contacts (Fig. 2K; Fig. S2A,B). Although the intensity of the injected dye precluded examination of apical network relationships, lateral views highlighted the distinctive interlocking shapes of the photoreceptors and IOPCs (Fig. 2K). Photoreceptor cell bodies occupy the bulk of the tissue space, except toward the basal plane, where they narrow to accommodate the IOPC bodies. Bundles of photoreceptor apical membrane (green asterisks in Fig. 2K), precursors to the rhabdomeres, define the longitudinal axis of each ommatidium. IOPCs maintain connection with the apical epithelial plane via thin processes that separate neighboring photoreceptor clusters and fill up the inter-ommatidial space. Despite the lack of junctional connections between photoreceptors and IOPCs, each appears to support and constrain the shape and position of the other.

In contrast, the shapes, positions and contacts of abl^null retinal cells were aberrant and irregular (Fig. 2L; Fig. S2C,D). Photoreceptor cell bodies were found mispositioned basally at or below the plane of the IOPCs, confirming our previous observations (Xiong and Rebay, 2011), and the apical membrane bundles (rhabdomere precursors) were disorganized (Fig. 2L). Occasional photoreceptors appeared to have lost contact with surrounding retinal cells, with the bulk of their cell volume beneath the IOPC cell bodies (Fig. 2L, cyan cells). IOPC cell shapes were also aberrant (compare Fig. 2L with Fig. 2K and Movie 2), disrupting the cell-cell contacts that pattern the apical and basal networks (Fig. S2C,D).

Together, these observations suggest that Abl function is broadly required in multiple cell types to establish and maintain the 3D organization of the retinal epithelium. At the cellular level, the distinct shapes and spatial arrangement of the photoreceptors and IOPCs are required for correct apical, basal and longitudinal network patterns. At the tissue level, correct 3D organization is essential to the ability of the tissue to maintain epithelial integrity during the elongation phase.

To define the cellular and subcellular contexts for Abl function, we examined Abl protein localization using an endogenously GFP-tagged allele (Nagarkar-Jaiswal et al., 2015). Previous antibody-based analyses of Abl protein localization in the developing retina emphasized its enrichment in the photoreceptor apical membranes (Bennett and Hoffmann, 1992; Xiong and Rebay, 2011), although Bennett and Hoffmann also noted low levels of Abl protein within the IOPC apical network at 25% p.d.. Using the Abl^GFP allele, we detected Abl expression not only in the photoreceptors but in all retinal cell types. Subcellularly, Abl appeared enriched in the IOPC and photoreceptor cytoskeletal and junctional domains that define the apical, basal and longitudinal networks (Fig. 3 and Fig. S3). For example, Abl overlapped with F-actin along the full length of the developing (Fig. 3A-D′ and Fig. S3A) and mature rhabdomeres (Fig. S3B). Abl was also detected in the IOPCs (Fig. 3B,B′,D,D′), where the overlap with F-actin was particularly striking in the basal feet (Fig. 3D,D′ and Fig. S3). The expression and localization of Abl in both the photoreceptors and IOPCs position it in time and space to participate in the coordinated morphogenetic changes that collectively provide retinal 3D organization.

Assembly of branched F-actin networks supports the specialized structures in both photoreceptors and IOPCs (Galy et al., 2011; Signore et al., 2018). Given that a balance between linear and branched F-actin networks is required for a cell to build specific cytoskeletal structures (Burke et al., 2014; Kannan et al., 2017; Suarez and Kovar, 2016), and that Enabled (Ena) promotes linear F-actin assembly and is inhibited by Abl in a variety of cellular contexts (Comer et al., 1998; Forsthoefel et al., 2005; Fox and Peifer, 2007; Gates et al., 2007; Gertler et al., 1995; Grevengoed et al., 2001; Kannan et al., 2014, 2017; Lin et al., 2009; Rogers et al., 2021), we asked whether the mechanisms underlying Abl function in the photoreceptors and IOPCs involved Ena.

We first confirmed Ena expression in both cell types. Ena protein was enriched in and overlapped with F-actin in the photoreceptor apical domains and was detected in the IOPC apical domains (Fig. S4A,A′). At the basal plane, Ena was prominent in bristle cells but little, if any, was detected in the IOPC feet, consistent with the minimal F-actin enrichment at this stage (Fig. S4B,B′).

We next asked whether reducing the genetic dose of ena could suppress the abl null phenotype. Using rhabdomere presence (Fig. 4A-B′) and photoreceptor nuclear position (Fig. 4C-D′) at the subapical plane as readouts, we found that heterozygosity for ena, which on its own did not perturb retinal development, partially suppressed the abl mutant phenotype. For example, in contrast to abl null ommatidia where only scattered foci of F-actin were observed (Fig. 4A), organized bundles of rhabdomeric F-actin were seen when ena dose was reduced (Fig. 4B). Similarly, Elav-positive photoreceptor nuclei were again detected at the appropriate plane upon ena reduction (compare Fig. 4D with Fig. 4C). To confirm the relevance of Abl-Ena antagonism in photoreceptors, we selectively expressed Ena dsRNA (Ena^RNAi) in the photoreceptors (using elav-Gal4) and as a control in IOPCs (using LL54-Gal4) in an abl^null background. Only photoreceptor-specific Ena^RNAi visibly improved the organization of the photoreceptor clusters (Fig. 4E,F, compare with abl^null in Fig. 2B). Confirming that IOPC-specific Ena knockdown alone did not disrupt retinal pattern and thereby mask a suppressive interaction, control retinas appeared wild type (Fig. 4G, compared with wild type in Fig. 2A).

Our results thus far suggested that Abl function in the photoreceptors and IOPCs collectively supports the elaboration of a stereotyped 3D tissue architecture that maintains retinal integrity. Returning to the photoreceptor-specific Ena knockdown experiment (Fig. 4E), we also noticed significant improvement in 50% p.d. apical network pattern (Fig. 4H-L). The observation that proper photoreceptor terminal differentiation could non-autonomously correct IOPC patterning defects raised the possibility that interactions between these two retinal cell types coordinate the overall 3D patterning process. We considered two possibilities. First, one cell type might be the primary organizer, with their shape, position and connections imposing dynamic physical constraints that influence how the terminal differentiation program of the other unfolds. Alternatively, there might be redundancy, with local interactions between photoreceptors and IOPCs mutually constraining and influencing the structures and contacts they each contribute to the 3D tissue organization. In either scenario, loss of Abl disrupts these interactions, generating variability within the apical, basal and longitudinal networks that compromises the stereotyped 3D morphogenetic program.

Our demonstration that Abl is required for both photoreceptor and IOPC terminal differentiation (Figs 1 and 2) allowed us to test the possibility of non-autonomous feedback interactions by restoring Abl function selectively to one cell type and then assessing the impact on the other abl mutant cell type and on overall retinal organization. We used the partial rescue strategy instead of a selective knockdown approach because abl^RNAi does not recapitulate abl^null phenotypes in the retina (Xiong et al., 2009). Control experiments confirmed the specificity of the genetic strategy, with the expected restriction of elav-Gal4- and LL54-Gal4-driven expression to photoreceptor and IOPCs, respectively (Fig. S5A-F′), and no leaky expression or rescue with UAS-Abl^GFP alone (Fig. S5G-L).

Focusing on 50% p.d., the timepoint marking completion of pattern establishment, we first asked how restoring Abl to the photoreceptors influenced IOPC shapes and apical and basal network patterns in an otherwise abl^null retina (elav>Abl; abl^null). As expected, expressing Abl specifically in the photoreceptors restored their morphology, with marked improvement in organization and alignment of the photoreceptor apical membrane bundles that pattern the longitudinal network (Fig. 5A-C,M,N). When we examined the IOPC response, we found significant improvement in both apical and basal network pattern (Fig. 5E-G,O-Q) These results suggest that correct photoreceptor cell morphology, and by extension an intact longitudinal network, promotes correct IOPC morphology and pattern within the apical and basal networks.

Second, in the reciprocal experiment, we examined the response of abl mutant photoreceptors to IOPC-specific restoration of Abl (LL54>Abl; abl^null). As expected, expressing Abl in the IOPCs rescued their cellular morphology and improved apical and basal network patterns (Fig. 5H,L,O-Q). When we examined the photoreceptor response in the longitudinal plane, the organization, alignment and position of the rhabdomere precursors were all improved (Fig. 5D,M,N). Time-lapse imaging further emphasized the improved 3D organization, with the characteristic shapes and relative positions of both photoreceptors and IOPCs resembling those of a wild-type retina (Fig. 5R, compare with Fig. 2K,L, Fig. S6A,B and Movie 3). Together, these partial rescue experiments suggest that local interactions between these two major retinal cell types, with each non-autonomously influencing the morphology of the other, redundantly coordinate pattern establishment across the different epithelial planes.

The non-autonomous influence of correct cellular morphology and pattern across orthogonal tissue planes suggested that interactions between the IOPCs and the photoreceptors are an integral organizing feature of the 3D epithelial structure. These interactions could be primarily passive, with cell shapes adjusting to fill available tissue space, or more active, with morphogenetic change in each cell type promoting the terminal differentiation program of the other. Given the extensive deposition of cellular material and elaboration of structure that occurs within the longitudinal and basal networks during the elongation phase, we predicted that active interactions would be needed to coordinate the completion of these two terminal differentiation programs.

To test this, we used the same cell-type-specific partial rescue strategy and assessed rhabdomere formation, fenestrated membrane contraction, tissue elongation and epithelial integrity in 100% p.d. retinas. We first asked whether selective restoration of Abl function to the photoreceptors (elav>Abl; abl^null), which supported the elaboration and elongation of a mature longitudinal network of rhabdomere bundles (Fig. 6I), was sufficient to rescue the structure and function of the basal network of contractile IOPC feet. In contrast to the disorganized pattern seen in abl^null (Fig. 6B versus Fig. 6A), in elav>Abl; abl^null partial rescue retinas, the radial alignment of F-actin bundles in the IOPC feet and the uniformity in central ring size both appeared to be improved (Fig. 6C), suggesting successful elaboration of the cytoskeletal structures and connections needed for basal network contraction. Similar results were obtained by genetically suppressing the abl^null phenotype with photoreceptor-specific Ena knockdown (elav>Ena^RNA; abl^null), whereas little or no suppression was obtained with IOPC-specific Ena knockdown (LL54>Ena^RNAi; abl^null), consistent with Abl using an Ena-dependent mechanism in the photoreceptors and an Ena-independent mechanism in the IOPC basal stress fibers (Fig. 6E-H). Measurement of retinal depth (Fig. 6K) confirmed rescue of the entire tissue-level morphogenetic program. These results suggest that photoreceptor-IOPC interactions actively induce morphogenetic remodeling in the IOPCs, and that, at the tissue level, an intact and correctly aligned longitudinal network can nonautonomously correct patterning across the orthogonal basal network.

Finally, we examined the reciprocal partial rescue experiment to determine whether elaboration of correct apical/basal network pattern was sufficient to induce remodeling and elongation of the photoreceptor rhabdomeres (LL54>Abl; abl^null). As expected, restoring Abl function to the IOPCs reduced basal network variability (Fig. 6D,G,H). Remarkably, this non-autonomously rescued the terminal differentiation program of the abl^null photoreceptors such that they elaborated well-defined rhabdomere bundles that spanned the full longitudinal axis of the epithelium (Fig. 6J) and maintained junctional attachments to the apical cone cell caps (Fig. 6J) and basal cone cell feet (Fig. S6C-E). Not only were rhabdomere structures, junctional contacts and organization restored, but their elongation was also recovered (Fig. 6K), indicating active remodeling of this highly specialized structure. We propose that interactions between photoreceptors and IOPCs, mediated across different tissue planes through a 3D network of specialized cytoskeletal structures and junctional connections, provide a mechanical feedback mechanism that actively coordinates their terminal differentiation programs (Fig. 7). Together, these interactions confer physical robustness to the stereotyped patterns that support the tissue-level morphogenetic program.

Organ form and function derives from the precise arrangement of different cell types with various sizes, shapes and specialized structures. In complex tissues, coordinating different cellular morphogenetic events in 3D to achieve the correct final form presents a major developmental challenge. Here, we show that the Drosophila pupal retina resolves this challenge by forming a physically coupled supracellular network of specialized cytoskeletal domains and junctional attachments that organizes 3D tissue pattern by integrating two modules of regulation. First, cell-type specific terminal differentiation programs mediated by Abl specialize the cytoskeletal domains that pattern the individual planes of the tissue. Second, cell-cell interactions underlie a tissue-intrinsic feedback relay that coordinates morphogenetic change between different retinal cell types. Together, this ensures the fidelity and integrity of retinal morphogenesis.

Late pupal retinal morphogenesis involves drastic shape changes at the cell and tissue level. In contrast to epithelia where tissue elongation and growth involves changes of junctional contacts through cell intercalation, division or motility along the axes of elongation/growth (Aigouy et al., 2010; Baena-López et al., 2005; Blankenship et al., 2006; Clarke and Martin, 2021; Dye et al., 2017, 2021; Etournay et al., 2015; Glickman et al., 2003; Irvine and Wieschaus, 1994; Mao et al., 2013; Paré and Zallen, 2020; Shindo, 2018; da Silva and Vincent, 2007; Warga and Kimmel, 1990; Wilson and Keller, 1991; Zallen and Wieschaus, 2004), the pupal retina relies on a persisting network of cell-cell contacts and cytoskeletal domains to generate and withstand the morphogenetic changes that produce adult organ form. Thus, one significant finding in this study is that photoreceptors and IOPCs coordinate their morphogenetic programs in order to reinforce stereotyped tissue organization during morphogenesis.

First, consistent with previous studies (Baumann, 2004; Longley and Ready, 1995; Ready, 2002), our 3D reconstructions highlight the concomitant elaboration of photoreceptor rhabdomeres with the elaboration and contraction of IOPC feet. As both cytoskeletal structures converge from different planes to anchor to the cone cell feet, the cellular coordination between photoreceptors and IOPCs likely stems from the physical coupling between their specialized cytoskeletal structures along orthogonal planes. Second, restoring Abl function in either the photoreceptors or IOPCs is sufficient to support the terminal differentiation program of the other abl mutant cell type. This result suggests that the non-autonomous rescue is not merely a passive restoration of cell position and shape, but also involves active remodeling of the cytoskeletal domains and structural specializations. Reiteration of local photoreceptor-IOPC interactions across the ommatidial field restores large-scale tissue order.

One way to explain this tissue-level coordination is through feedback within the cytoskeletal adhesion network architecture (Fig. 7A). Previous studies described planar supracellular networks that span multiple cell apices as central to many morphogenetic processes, including epithelial closure, elongation, invagination and folding (Kiehart et al., 2000; Martin et al., 2009; Popkova et al., 2021; Yevick et al., 2019). Our study extends this concept to 3D. We speculate that the specialized apical and basal cytoskeletal/junctional domains of the IOPCs, which are mechanically linked to longitudinally oriented cytoskeletal structures in the photoreceptors, together constitute a mechanically coupled supracellular 3D scaffold that channels cell and tissue-scale feedback (Fig. 7A,B). Through spatially organized cell-cell interactions, the cytoskeletal structure and patterning in one cell type along one plane promote cytoskeletal organization and patterning in the other cell type along the orthogonal plane. This positive-feedback system allows dynamic intercellular communication that ensures the concomitancy of their morphogenetic progress. Given the significant morphogenetic changes that unfold during the elongation phase, such inter-plane feedback may be crucial for maintaining robust retinal organization. More broadly, inter-plane mechanical coupling might be a general feature of tissue morphogenesis, taking on context-specific forms but providing analogous cell and tissue-level coordination to the acquisition of 3D form.

Two general mechanisms could underlie such inter-plane communication. One relies on the incompressibility of the cytoplasm, with competition for cell volume constraining cell shapes and coordinating 3D change (Bagnat et al., 2022; Gelbart et al., 2012; Harmansa et al., 2023; Stooke-Vaughan and Campàs, 2018). Alternatively, forces could be transmitted directly through the specialized cytoskeletal/junctional domains, with the precise geometrical arrangement affecting inter-plane communication as in an adaptive planar supracellular network (Aigouy et al., 2010; Chanet et al., 2017; Coen and Rebocho, 2016; Huang et al., 2018; Khalilgharibi et al., 2019; Mao and Baum, 2015; Rebocho et al., 2017; Shyer et al., 2013; Wyatt et al., 2020; Yevick et al., 2019).

Our study highlights how photoreceptors and IOPCs coordinate their morphogenesis to maintain the optimal organization required for vision. Although we have focused on interactions between photoreceptors and IOPCs, additional interactions, such as anchorage to the cone cells and to the basal ECM, contribute to retinal organization and integrity. The cone cells provide a hub that physically connects apical and basal planar hexagonal patterns to the orthogonally oriented rhabdomeres. The cone cell feet also plug the central rings and complete the in-plane tiling of the basal floor by making septate junction connections with the IOPC feet (Banerjee et al., 2008). Thus, concomitant remodeling of cone cell shape and junctional connections with both photoreceptors and IOPCs will further reinforce 3D tissue organization. In abl mutants, the increased and irregular ring size implicates correct cone cell feet shape and attachments as integral components of 3D retinal organization and elaboration. Identification of Neurexin IV, a septate junction component, and Pax2, a transcription factor that presumably regulates many aspects of cone cell terminal differentiation, as ‘photoreceptor falling’ mutants (Banerjee et al., 2008; Charlton-Perkins et al., 2017) further supports these ideas.

The retinal floor comprises not only the IOPC basal contractile network but is also constrained by the underlying ECM; close coupling between these two components enables coordinated force transmission across the plane (Ready and Chang, 2023). IOPCs form integrin-mediated focal adhesions that anchor the contractile fibers to the underlying ECM. Concomitant with contraction of the IOPC basal network, the ECM shows signs of increasing collagen IV (Vkg-GFP) deposition and increased integrin intensity around the central rings (Fig. S2E). This suggests that mechanical coupling between the elaborating ECM layer and the IOPC basal network matches their developmental progress during elaboration of the specialized retinal floor, thereby contributing to its integrity. Consistent with this, integrins have long been known to be essential to retinal integrity (Longley and Ready, 1995). The specific contributions of different ECM components are just beginning to be explored (Ready and Chang, 2023).

In addition to the cone cell and ECM components, previous studies have identified several other cytoskeletal/junctional components or regulators whose loss compromises retinal integrity, such that photoreceptors fall below the retinal floor. The list includes: Arp2/3 and WAVE/SCAR components (Galy et al., 2011); Cofilin (Pham et al., 2008); DRac1 (Chang and Ready, 2000); Pebbled/Hindsight (Pickup et al., 2002); Mbt/Pak4 (Schneeberger and Raabe, 2003; Walther et al., 2016); Rap1 (Walther et al., 2018); and Afadin/Canoe (Matsuo et al., 1999). A few on the list have been implicated in Abl function in other developmental contexts (Grevengoed et al., 2003; Kannan and Giniger, 2017; Kannan et al., 2017; Yu and Zallen, 2020) and could provide insight into mechanisms of Abl function in the retina.

In these ‘photoreceptor falling’ mutants, it is interesting to consider the cause of the basal-ward collapse. One appealing possibility is that photoreceptor axonal connections in the brain exert a pulling force (Langen et al., 2015; Lee and Treisman, 2004) that is normally resisted by redundant structural features of the retinal epithelium. In mutants such as abl that disrupt the junctional coupling of the photoreceptor clusters, their anchorage to the overlying cone cells and/or the organization of the retinal floor, tension from the axon-brain connections may be sufficient to pull them basally through the fenestrated membrane. Analogous to how oriented cell division and mitotic nuclear movements physically displace adjacent cells and redistribute patterns of mechanical tension and adhesion in simple epithelia to impact the morphogenetic program (Bosveld et al., 2016; Kondo and Hayashi, 2013; Leen et al., 2020; Mao et al., 2013), the collapsing photoreceptor clusters will disrupt cell shapes, contacts and force balances within each ommatidium, progressively perturbing pattern across all tissue planes as they drop basally. Future study of the roles of Abl and other photoreceptor falling mutant genes in the cell-cell junctions and interactions that organize, support and provide mechanical feedback will provide further insight into how cell and tissue-scale processes collectively produce 3D organ form.

All crosses were carried out at 25°C in standard laboratory conditions. w¹¹¹⁸ (3605), abl¹ (3554), abl² (8565), Abl^mimicGFP (59761), LL54-Gal4 (5129), Elav-Gal4 (8765), ena²³ (8571), G-TRACE (2820) were from the Bloomington Stock Center. UAS-Abl-GFP (O'Donnell and Bashaw, 2013) was a gift from G. Bashaw (University of Pennsylvania, PA, USA). UAS-ena^RNAi (106484) was from the Vienna Drosophila Resource Center. vkg-GFP (CC00791) was from Flytrap (Buszczak et al., 2007). Chp^4.5-Gal4 (Mishra et al., 2013) was a gift from S. L. Zipursky (University of California Los Angeles, CA, USA). GMR-wIR^13D (Lee and Carthew, 2003) was a gift from R. Carthew (Northwestern University, IL, USA).

w¹¹¹⁸ was used as wild type in all experiments. To generate abl^null animals, abl²/TM6B males were crossed to abl¹/TM6B virgins and abl¹/abl² (non-Tubby) were selected. To generate abl² clones, abl²,FRT80B/TM6B males were crossed to ey-FLP; GFP^nls,FRT80B virgins. To reduce ena dose in abl² clones, ena²³/CyO, abl² FRT80B/TM6B males were crossed to ey-FLP; GFP^nls,FRT80B virgins. For cell type-specific rescue experiments, abl² was recombined with the UAS-Abl-GFP transgenes inserted into the 86Fb docking site (O'Donnell and Bashaw, 2013). abl²,UAS-Abl-GFP/TM6B males were crossed to LL54 (or Elav)-Gal4; abl¹/TM6B virgins. To test Abl-Ena interactions in specific cell types, UAS-ena^RNAi; abl²/TM6B males were crossed to LL54 (or Elav)-Gal4; abl¹/TM6B virgins. For lineage tracing, w,GMR-w.IR;+/+;chaoptin-Gal4, abl²/TM6B females were crossed to w; G-TRACE; abl¹/TM6B males (G-TRACE = P{w[+mC]=UAS-RedStinger}4,P{w[+mC]=UAS-FLP.D}JD1,P{w[+mC]=Ubi-p63E(FRT.STOP) Stinger}9F6/CyO). GMR-w.IR reduced pigment deposition, improving imaging.

For developmental staging, white pre-pupae (0 h APF) were selected and aged at 25°C. 50%, 75% and 100% p.d. animals for dissection were selected by time (48, 72 and 96 h APF, respectively), confirmed by morphological landmarks before dissection (Bainbridge and Bownes, 1981) and by cellular features after dissection.

50% p.d. pupal eye discs were dissected in PBS and fixed for 10 min in 4% paraformaldehyde in PBT (PBS with 0.1% Triton X-100). For 75% and 100% p.d. dissections, pupal/adult heads were pre-fixed for 20 min, retina were dissected and fixed for 10 min, washed three times in PBT, blocked in PNT (PBT+3% normal goat serum) for 1 h, incubated overnight at 4°C with primary antibodies diluted in PNT, washed three times in PBT, incubated in secondary antibodies diluted in PNT for 6 h, washed three times in PBT and mounted in 90% glycerol in 0.1 M Tris (pH 8.0) with 0.5% n-propyl gallate.

In cell type-specific rescue experiments in which UAS-Abl-GFP was ectopically expressed under the control of LL54-Gal4 or Elav-Gal4, F-actin was stained with AlexaFluor-488 Phalloidin. Although GFP and AlexaFluor-488 are both excited by the same wavelength, the UAS-Abl-GFP signal faded during our standard staining protocol and was no longer detected with the confocal settings used to image F-actin. For Fig. S3, discs were imaged after only a 2 h incubation with AlexaFluor-568 Phalloidin in order to preserve the UAS-Abl-GFP signal.

Antibodies were from the Developmental Studies Hybridoma Bank (DSHB): mouse anti-Elav 9F8A9 (1:50); mouse anti-Ecad DCAD2 (1:500), mouse anti-βPS integrin CF.6G11 (1:100) and mouse anti-Ena 5G2 (1:50). Cy3/Cy5/Alexa488/Alexa594-conjugated secondaries were diluted 1:2000 (Jackson ImmunoResearch, 715165150, 115175146, 715546150 and 715585151). AlexaFluor-488 and AlexaFluor-568 Phalloidin were diluted 1:1000 (Thermo Fisher Scientific, A12379 and A12380).

Fluorescent images were obtained using a Zeiss LSM880 confocal microscope with Airyscan. Image processing and measurements were performed using Fiji. To obtain the image of different planes, multiple z-stack slices (0.4 µm interval) that encompass the region of interest were averaged.

3D reconstructions and lateral views were made in Imaris (Bitplane). For views that include intact rhabdomeres, the 3D reconstructions were rotated and cropped to expose the lateral view of one line of aligned ommatidia.

Graphing and statistical analyses were performed in Prism7 (GraphPad). Data were plotted as mean±s.e.m. and statistical differences between conditions were determined with two-tailed unpaired t-tests or t-tests with Welch's correction.

To measure retinal depth, we chose the central region of the retina for all genotypes to avoid regional variation. We generated multiple lateral views in Imaris and used the average length of 10 ommatidia in the central region as the retinal depth. The same wild-type and abl^null control datasets were used in multiple figures to make different comparisons of the various retinal parameters.

Staged 50% p.d. pupae were selected and the puparium was dissected away to expose the distal (apical) surface of the retina. After injecting ∼0.5 μl of CellMask Deep Red Plasma Membrane Stain (Thermo Fisher Scientific) diluted 1:25 in Schneider's Drosophila medium into the head, the pupae were incubated in a humid chamber at 25°C for 4 h. Injected pupae were mounted in a glass-bottom dish such that the region of the head containing the developing retina was directly in contact with the glass in a thin layer of Halocarbon 700 oil (Halocarbon Products). Pupae were oriented and immobilized using a log of petroleum jelly and the imaging vessel was kept humid with blotting paper soaked in Schneider's Drosophila medium (method adapted from Hellerman et al., 2015). Mounted retinas were immediately imaged on a Zeiss LSM 880 laser scanning confocal microscope using the 40×oil-immersion objective on AiryScan mode. Optical slices of 1 μm were acquired through the entire apical-basal depth of the tissue every 5 min. Although we cannot rule out the possibility that our protocol stalls development or negatively impacts the tissue, in a control experiment, injected animals continued to develop and 85% (n=20) eclosed with no obvious morphological defects. Because the dye injection resulted in a huge intensity difference between apical and basal planes, Fiji Top-Hat (Spot Radius=0.75) was run on the entire 4D image stack to equalize signal and enhance cell outlines (script by G. Landini, adapted for use on hyperstacks by C. Labno: https://blog.bham.ac.uk/intellimic/g-landini-software/). The modified macro can be accessed on the University of Chicago Microscopy Core's GitHub (https://github.com/UChicago-Integrated-Light-Microscopy/ImageJ_macros).

Individual cells were annotated by hand in ImageJ using the xy-planes because of the ease in distinguishing cell outlines. After manually correcting drift, lateral projections of the resulting time-lapse images were made using ImageJ software.

We thank Ed Munro, Ellie Heckscher and Rick Fehon for experimental advice and comments on the manuscript; Saman Tabatabaee for assistance with graphics; Christine Labno (University of Chicago Integrated Light Microscopy Core Facility, which receives support from the Cancer Center Support Grant P30CA014599) and Audrey Williams for live image processing advice; past and present Rebay lab members for helpful discussions; and G. Bashaw, S. L. Zipursky, R. W. Carthew and S. Horne-Badovinac for fly stocks. Some of the text and figures in this paper formed part of Xiao Sun's PhD Thesis in the Committee on Development, Regeneration and Stem Cell Biology at the University of Chicago in 2023.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201757.reviewer-comments.pdf