Calcium waves facilitate and coordinate the contraction of endfeet actin stress fibers in Drosophila interommatidial cells

Tissue construction during development often requires a specific subset of its constituent cells to undergo orchestrated shape change. To accomplish this, tensile forces from actomyosin networks, which are crucial for maintaining and regulating cell shapes, need to be spatially and temporally coordinated in a selective cell population (Guillot and Lecuit, 2013; Paluch and Heisenberg, 2009; Heisenberg and Bellaïche, 2013). Although the mechanism behind the actomyosin contraction in vitro and in individual cells has been extensively studied, how these forces are coordinated across a cell field during tissue formation is not well understood.

To investigate the mechanisms coordinating the contractile forces required for tissue morphogenesis, we have used the Drosophila compound eye, a convex epithelium with stereotypical cell composition and arrangement (Waddington, 1962). The compound eye consists of ∼800 unit eyes, termed ommatidia, each comprising a distal corneal lens secreted by four cells and two primary pigment cells positioned above eight photoreceptors elongated on the optical axis. Ommatidia are embedded in a honeycomb lattice of interommatidial cells (IOCs), pigment-secreting cells that separate and optically insulate ommatidia. At the retinal floor, IOCs flatten their endfeet to form the fenestrated membrane, an extracellular matrix (ECM)/cytoskeletal specialization that defines the proximal surface of the eye (Fig. 1). The ECM specializations include grommets, the circular openings that allow photoreceptor axons to exit the retina, and ridges, the dense linear ECM between neighboring grommets. This reiterative nature of the eye architecture implies the tensile forces for cell morphogenesis in one cluster are similarly applied in all other ommatidia across the retina, making the eye uniquely suited for analyzing how local forces are coordinated to ensure proper tissue formation.

Actomyosin-dependent contraction has been implicated in several processes during fly eye development, including apical cell constriction in the morphogenetic furrow (MF; an indentation that sweeps across the larval eye disc) (Benlali et al., 2000; Fernandes et al., 2014), pigment cell shape and photoreceptor position in the epithelium (Warner and Longmore, 2009a; Warner and Longmore, 2009b; Lee and Treisman, 2004), rhabdomere biogenesis (Baumann, 2004; Galy et al., 2011) and retinal lumen formation (Nie et al., 2014). In addition, actomyosin tensile forces participate in the reduction of retinal basement, a process that is essential for the formation of the curvature of the eye. During the latter half of pupal development, the IOC endfeet undergo a fourfold contraction (Longley and Ready, 1995), shaping the ommatidia into elongated hexagonal prisms that become packed to give the eye its panoramic convex curvature (Fig. 1B). Planar tension at the retinal floor has the additional role of aligning the rhabdomeres, photosensory waveguides, of each ommatidium along the optical axis. Rhabdomeres are sprung between the rigid outer dome of the faceted cornea and the inner retinal floor in a cage of specialized cell-cell junctions (Cagan and Ready, 1989) (Fig. 1). Tension in the convex retinal floor stretches the rhabdomere cage on the longitudinal axis, which aligns rhabdomeres, and must be balanced with growth and renewal of rhabdomeres across the retina (Raghu et al., 2009).

The reduction in the basal surface is largely powered by actin stress fibers and the associated non-muscle myosin II in IOC endfeet, as eyes deficient in zipper (zip; myosin II heavy chain) and actin regulation contain distorted retinal floors (Baumann, 2004; Cagan and Ready, 1989; Galy et al., 2011; Longley and Ready, 1995). At the mid-pupal stage, stress fibers form at the IOC endfeet and are anchored to the grommets via integrins (Longley and Ready, 1995), appearing as filaments emanating from the grommets. These stress fibers mature through the late pupal stages and generate tension, bringing attached grommets closer together. Concomitant with contraction, stress fibers transform from a looser, filamentous organization to larger, dense patches with higher levels of stress fibers abutting the neighboring grommets (partially contracted), and then to small, dense patches connecting the grommets (fully contracted). Thus, the contraction of endfeet stress fibers, manifested by a progressive reduction of their sizes, compacts the fenestrated membrane, including the ECM below. The regular organization of the ommatidia is preserved at the retinal floor, suggesting that the tensile forces for reducing the IOC endfeet are precisely applied.

To decipher the mechanism coordinating IOC endfeet contraction, we have combined live imaging with a Ca²⁺ sensor (Chen et al., 2013) and mutational analysis to investigate Ca²⁺ signaling in non-neuronal retinal cells. Ca²⁺ is a ubiquitous and versatile second messenger, impacting virtually all cell physiologies (Berridge et al., 2000). In addition to facilitating muscle contraction, Ca²⁺ signaling participates in actin remodeling and contraction in non-muscle cells. Ca²⁺-dependent regulation of non-muscle myosin II, a hexamer of two heavy chains, two regulatory light chains (MRLC), and two essential light chains, has been well characterized (Brito and Sousa, 2020; Heissler and Sellers, 2016). Myosin II activity is increased by the phosphorylation of MRLC, which can be upregulated by calmodulin-dependent myosin light chain kinase (MLCK) in response to Ca²⁺ signaling (Scholey et al., 1980; Heissler and Sellers, 2016). Thus, orchestrated Ca²⁺ signaling, such as a propagation of Ca²⁺ spikes through constituent cells, is capable of coordinating the activity of actomyosin networks in tissue morphogenesis (Homolya et al., 2000). In Drosophila egg activation, a single Ca²⁺ wave leads to actin reorganization and the formation of dynamic actin around the cell cortex (York-Andersen et al., 2015, 2020; Kaneuchi et al., 2015). In wing discs, oscillating Ca²⁺ waves influence actomyosin organization (Balaji et al., 2017) and facilitate recovery in response to mechanical injury (Brodskiy et al., 2019; Narciso et al., 2017; Restrepo and Basler, 2016). In vertebrate epithelia, Ca²⁺ propagation induces actin rearrangement in the surrounding cells to promote cell extrusion (Takeuchi et al., 2020).

In mammalian retina, Ca²⁺ signaling is known to facilitate the formation of electrical synapses and mediates trophism between supporting cells and neurons (Barres, 2008; Allaman et al., 2011; Halassa and Haydon, 2010; Fields and Stevens-Graham, 2002). In addition, intercellular propagations of Ca²⁺ spikes sweep across glial cell networks, including astrocytes in the central nervous system (Cornell-Bell et al., 1990) and Müller retinal glial cells (Kurth-Nelson et al., 2009), although the roles of Ca²⁺ waves in these cells remain unresolved (Scemes and Giaume, 2006). In Drosophila retina, Ca²⁺ has crucial functions in phototransduction (Voolstra and Huber, 2020; O'Tousa, 2002) and Ca²⁺ dynamics in isolated live photoreceptors have been recorded (Asteriti et al., 2017; Hardie, 1996). Stimulus-independent Ca²⁺ oscillations have also been observed in developing photoreceptors and are thought to facilitate neuronal connections (Akin et al., 2019; Choi et al., 2021). The versatility of Ca²⁺ signaling and its known impact on actomyosin activity make it a good candidate for orchestrating endfeet stress fiber contraction, although whether Ca²⁺ signaling has a role in accessory cells in fly retina is not known.

Here, we report a novel observation: Ca²⁺ waves regularly sweep across the eye's honeycomb-like lattice of IOCs. These waves, formed during pupal development, require IP3R (Itpr), indicating that IOC Ca²⁺ waves are mediated by Ca²⁺ release from internal stores. Mutational analysis of IP3R suggests that Ca²⁺ waves contribute to the retinal floor reduction by facilitating the contraction of actin stress fibers at the IOC endfeet. Our work suggests a mechanism by which intercellular Ca²⁺ waves coordinate contractile forces for an orderly retinal floor condensation. Moreover, absence of IP3R results in a deformed fenestrated membrane ECM, indicating that contractile forces from the endfeet stress fibers shape the ECM network.

To monitor the intracellular Ca²⁺ levels in developing and adult Drosophila eyes, we expressed GCaMP6m (Chen et al., 2013) with longGMR-GAL4 (Wernet et al., 2003), hereafter referred to as lGMR>GCaMP6m. Like the more widely used GMR-GAL4 (Freeman, 1996), lGMR-GAL4 becomes active in cells posterior to the MF and remains active in most of the retinal cells at subsequent stages (except the bristle cells). We chose lGMR-GAL4 because its moderate GAL4 expression level, compared with GMR-GAL4, is less likely to cause cytotoxicity (Bollepalli et al., 2017; Kramer and Staveley, 2003), which may damage the eye architecture and indirectly perturb the cellular Ca²⁺ level. To minimize potential toxicity and artifact from high GAL4 expression, all lGMR>GCaMP6m recordings presented in subsequent sections were performed with heterozygotes.

Using this lGMR>GCaMP6m platform, we analyzed the spatiotemporal pattern of cytosolic Ca²⁺ increases in non-neuronal supporting cells, namely IOCs, primary pigment cells and cone cells. As the composition, arrangement and architecture of these accessory cells are well established (Cagan and Ready, 1989), we imaged lGMR>GCaMP6m retinas at different focal planes to monitor their respective Ca²⁺ activities.

Imaging young lGMR>GCaMP6m adult retina at the distal plane showed that Ca²⁺ waves propagate regularly across the hexagonal IOC lattice (hereafter referred to as IOC waves; Fig. 2A, Movie 1). We tracked the movement of the wave fronts by sequentially subtracting signal intensities from consecutive images of the time-lapse series. This revealed that the initiation of IOC waves was not restricted to a specific region. For instance, the IOC waves in Movie 1 could initiate from the middle (t=80 s), anterior (t=4 s, 152 s, 210 s, 266 s), anterior-dorsal (t=92 s), posterior-ventral (t=14 s) and posterior-dorsal (t=20 s) edges of this lGMR>GCaMP6m retina (Fig. 2B). Once initiated, the IOC waves moved equally in all directions across the eye field (Fig. 2C). Each IOC spiked approximately 2 s after a neighboring cell, translating to a wave speed of 4-5 μm/s. When two IOC waves collided, the wave fronts merged and advanced toward regions where the cells had yet to experience recent Ca²⁺ waves (Fig. 2D). If no such region was available, the colliding IOC waves ceased (Fig. 2D). The GCaMP6m intensity at the juncture of merging wave fronts was not greater than those before the collision, implying that one or more factors responsible for triggering IOC Ca²⁺ waves are limiting, rendering these cells unresponsive to an additional dose of wave-inducing signal.

The characteristics of Ca²⁺ spikes within these IOC waves were remarkably consistent (Fig. 2E). In Movie 1, all 273 cells (204 secondary and 69 tertiary pigment cells) exhibited regular calcium oscillations (average peak number was 4.8±0.4, n=273) during the 5-min recording. These reproducible Ca²⁺ spikes rose sharply to a maximum in approximately 6.8±2.2 s (T_on, i.e. time from baseline to peak; n=1309 peaks) with a slower return to baseline (25.1±3.8 s; T_off, i.e. time from peak to baseline). The interval time, measured as the duration from one peak to the subsequent peak, was 68.6±12.4 s. The peak intensities of Ca²⁺ spikes experienced by individual cells showed no apparent dampening over time. No noticeable difference in Ca²⁺ spike characteristics was seen between secondary and tertiary pigment cells (Fig. S1B) or between cells from different regions (Fig. S1C). Pair-wise Pearson correlation analysis performed on the numeric series generated from GCaMP6m intensity over time indicated that the cells were most related to their adjacent cells, further confirming the wave-like nature of these Ca²⁺ spikes (Fig. S1D). To demonstrate that these descriptions are representative, we present the analysis of IOC waves (one still image shown in Fig. 5D) from an independently acquired lGMR>GCaMP6m recording in Fig. S2 (n=474 cells).

To determine whether other accessory cells contribute to IOC waves, we imaged young lGMR>GCaMP6m retinas more proximally, where Ca²⁺ increases in IOC waves, although not easily attributable to individual cells, were still recognizable (Fig. 3A). In this plane, GCaMP6m signals in the nuclei of primary pigment cells, the two crescent-shaped cells situated apically between the IOCs and cone cells, were particularly noticeable, possibly owing to quenching of the GCaMP6m fluorescence in the cytoplasm by the pigment granules (Fig. 3A). From two independent recordings (Movies 2 and 3), GCaMP6m signals in the primary pigment cells increased synchronously with IOC wave passages (Fig. 3A,B, arrows; Fig. S3A,B), demonstrating that the IOC wave network includes the primary pigment cells. In support of this, the characteristics of Ca²⁺ spikes in primary pigment cells resembled those observed in the secondary and tertiary pigment cells (Fig. 3C, Fig. S3C).

In contrast, cone cells, the four cells situated centrally atop the photoreceptors, exhibited Ca²⁺ activities that were independent of the IOC waves (Fig. 4A, Movie 4). Cone cell Ca²⁺ spikes were readily seen without an adjacent IOC wave, and, conversely, many cone cells were silent when IOC waves passed by (two such examples are indicated by arrows at 28 s). Whereas all secondary and tertiary pigment cells participated in propagating the IOC waves, not all cone cells were active during the 5-min recording (Fig. 4B). Of the 141 clusters analyzed, in only 51 (35.7%) did all four cells exhibit Ca²⁺ transients (Fig. 4C), and tabulation of the distribution indicated that 124 of the 564 (22.0%) cone cells were inactive. Moreover, the cone cell Ca²⁺ spike characteristics were noticeably different from those of IOC waves (Fig. 4D). For instance, from this recording, Ca²⁺ spikes (n=280) from 39 randomly selected IOC cells had a T_on of 12.03±8.37 s and a T_off of 22.18±8.45 s. In contrast, cone cell Ca²⁺ spikes (n=6179) had a shorter duration, with a T_on of 2.36±0.34 s and a T_off of 3.50±1.04 s. These Ca²⁺ spike differences, along with the lack of apparent coordination, demonstrate that IOC waves and cone cell Ca²⁺ spikes are separate events.

Interestingly, cone cells within a given cluster showed coordinated Ca²⁺ spikes. One example is shown in Fig. 4E, in which a cone cell Ca²⁺ spike initiated at t=86 s (arrow) and swept through the adjacent two cells in successive time frames. This suggests that cone cells in each cluster form their own mini wave network, independently of the IOC waves and other clusters. In support of this, pair-wise Pearson analysis of GCaMP6m signal over time with cone cells of the same cluster showed higher correlations than with those from different clusters (Fig. 4F).

To understand how these Ca²⁺ wave networks are constructed, we characterized the onset and pattern of GCaMP6m signals in developing retinas. Ca²⁺ waves were observed in ey>GCaMP6m eye discs behind and ahead of the MF (Fig. S4; Movie 5). The eye disc is highly dynamic with multiple developmental events compressed in a strong temporal gradient (Roignant and Treisman, 2009; Tsachaki and Sprecher, 2012; Wolff and Ready, 1991). Accordingly, we focused on Ca²⁺ activities in the latter half of pupal development, when pattern formation is complete (Cagan and Ready, 1989) and the temporal gradient has flattened, i.e. all cells are developing on the same timeline. At the P9 to P10 stages, lGMR>GCaMP6m retina showed no detectable Ca²⁺ activity (not shown). At the P12 stage, although most of the cells were still silent, Ca²⁺ activities in isolated IOCs and cone cells began to emerge (Fig. 5A, Movie 6). Furthermore, instances of Ca²⁺ waves propagating between adjacent IOC cells were seen (Fig. 5A,B), suggesting that the formation of IOC wave network had commenced. Imaging the same eye 1 day later showed widespread IOC waves and cone cell blinking (Fig. 5C, Movie 7), suggesting that the construction of the entire IOC wave network took place within this time.

To determine whether these Ca²⁺ events deteriorate with age, we analyzed lGMR>GCaMP6m retinas from older flies. Young retina, as mentioned above, showed regular IOC waves and cone cell blinking (Fig. 5D). Whereas cone cell Ca²⁺ spikes ceased about 5 days after eclosion, IOC waves were still present in 10- and 21-day-old lGMR>GCaMP6m retinas (Fig. 5E,F). However, compared with the waves in younger retina, IOC waves in older flies appeared to be irregular, evidenced by the decrease in peak number and the large variation in spike intervals (Fig. 5G).

Because the pigment and cone cells surround the photoreceptors, we asked whether the above-mentioned Ca²⁺ events require phototransduction. Prior to each recording session, dark-adapted lGMR>GCaMP6m retina was illuminated with 647 nm light pulses to inactivate the metarhodopsin in photoreceptors. Nonetheless, IOC waves were often present immediately at the start of recording, suggesting that waves were spontaneous and ongoing in the dark. Still, we frequently observed a surge in the overall eye brightness with multiple waves merging to cover the whole retina within the first 30 s following illumination, suggestive of initial GCaMP6m signal being influenced by light. To test definitively whether the occurrence of IOC waves requires phototransduction, we monitored lGMR>GCaMP6m signals in retina mutant for norpA, the Drosophila phospholipase C β homolog known for its role in acting downstream of the rhodopsin (Bloomquist et al., 1988). We reasoned that if the initiation or propagation of IOC waves requires phototransduction, GCaMP6m signals should be absent or noticeably reduced in norpA. Although eyes mutant for norpA³⁶, a complete loss-of-function allele, displayed no response to light stimulation (Pearn et al., 1996), robust IOC waves were seen in norpA³⁶ (Fig. 6A, Movie 8) and the Ca²⁺ transient profiles were similar to those observed in wild type (Fig. 6B). These results demonstrate that IOC waves are independent of phototransduction.

Based on the fast speed (4-5 μm/s) of IOC waves (Jaffe, 2008), we speculated that the source for the IOC Ca²⁺ waves is IP3R-regulated Ca²⁺ release from the endoplasmic reticulum (ER) stores. To test this, we monitored GCaMP6m signals in retinal cells mutant for l(3)itpr^90B.0, a complete loss-of-function allele of IP3R (Venkatesh and Hasan, 1997). As IP3R is an essential gene, mosaic eyes containing clones of homozygous l(3)itpr^90B.0 cells, marked by the absence of a membrane-associated mRFP (Chang et al., 2002), were generated by FLP-induced recombination (Fig. 6C, Movie 9). Robust IOC waves were seen in wild-type cells (mRFP⁺), but did not traverse into l(3)itpr^90B.0 mutant territory (Fig. 6D), indicating that IP3R function is required for wave propagation. Moreover, no GCaMP6m flash was observed in l(3)itpr^90B.0 cells, suggesting that IOC wave initiation, like propagation, requires IP3R function.

This mosaic eye contained two wild-type patches partitioned by an IP3R⁻ clone, allowing us to investigate the behavior of IOC waves from two separate populations in the same retina. As shown in Fig. 6E, IOC waves in the larger clone had more Ca²⁺ spikes (4.0±1.2, n=210) than the waves in the small clone near the dorsal posterior region (2.4±0.9, n=21). In addition, the Ca²⁺ spikes from these two clones occurred at different time points, indicating that the IOC waves could arise independently in separate cell populations.

At this focal plane, although the identity of specific cone cells experiencing Ca²⁺ spikes cannot be ascertained (because they were out of focus), wild-type clusters with cone cell Ca²⁺ activities were still discernible (Fig. 6D, arrowheads). In comparison, such clusters were completely absent in IP3R⁻ cone cells (Fig. 6F), demonstrating that both IOC waves and cone cell Ca²⁺ activities require internal Ca²⁺ release.

To understand the functional relevance of IOC waves, we looked for morphological defects associated with IP3R⁻ IOCs. As removal of IP3R function has no effect on the retinal cell fate (Acharya et al., 1997; Raghu et al., 2000), we suspected that the IP3R⁻ mutation perturbs specific subcellular structures in IOCs. To investigate whether the IP3R⁻ mutation affects the actin stress fibers at the endfeet, adult l(3)itpr^90B.0 mosaic eyes were stained with phalloidin to label the filamentous actin. These eyes were also rendered pigmentless with cn and bw mutations to eliminate interference from the pigment granules during imaging. In wild-type clones, dense patches of fully contracted stress fibers surrounded and interconnected the grommets (Fig. 7A-C, arrows). In contrast, stress fibers in l(3)itpr^90B.0 mutant cells, marked by the absence of ubi-mRFP^nls, exhibited various degrees of abnormalities. In some IP3R⁻ cells, especially those in smaller clones (asterisk marks an example in Fig. 7E), the stress fibers retained the patched morphology although their sizes were enlarged. In others, the defect was more severe, as the stress fibers, although still anchored to the grommets, appeared frayed, stretched and unbundled (Fig. 7C, arrowheads).

To determine whether this disruption of endfeet stress fibers affects the retinal floor area, we measured the cluster area, defined by a hexagon generated by connecting the centers of the six adjacent clusters (Fig. 7D,E). Mosaic eyes containing only wild-type clones showed no difference in the size of these hexagons, demonstrating that FLP-induced mitotic recombination had no influence on the retinal floor area (Fig. 7F). In contrast, in mosaic eyes containing IP3R⁻ clones, the retinal floor area per cluster in mutant patches was ∼40% larger than their sibling wild-type clones. This perturbation of area size in IP3R⁻ clones is specific to the retinal floor, as the apical area (the cornea level) in IP3R⁻ clones remained unaffected.

The aberrant stress fibers in adult IP3R⁻ cells could be caused by a block in the progression of stress fiber contraction during pupal development or an inability to maintain the contracted fibers in adults. To distinguish between these possibilities, we monitored stress fiber morphology in pupal IP3R⁻ mosaic eyes. If the role of IP3R is limited to maintaining contracted stress fibers in adults, stress fibers in IP3R⁻ mutant pupal eyes should appear normal. In a l(3)itpr^90B.0 mosaic eye at ∼75% pupal development, the endfeet stress fibers in wild-type cells were partially contracted, appearing as two discrete patches next to adjacent grommets (Fig. 7G-I, double arrows). In comparison, stress fibers in IP3R⁻ mutant cells were frayed and unbundled (Fig. 7G-I, arrowheads), indicating that IP3R is required for the progression of endfeet stress fiber contraction. It is noteworthy that the stage at which this phenotype is observed coincides well with the onset of IOC waves.

To investigate the effect of stress fiber deficits on the underlying ECM, we used Vkg::GFP (Kelso et al., 2004), which fluorescently tags endogenous Drosophila collagen IV α2 (viking) (Yasothornsrikul et al., 1997), a ubiquitous, network-forming basement membrane component (Kefalides, 1973; Brown et al., 2017; Fidler et al., 2017). In wild-type adult retina, basement membrane Vkg::GFP is organized in grommets and ridges (Fig. 1B,C). Compared with wild type, Vkg::GFP signal intensity in IP3R⁻ clones was reduced (Fig. 8A-D). Mean intensity measurement of Vkg::GFP showed a 32% reduction in the abovementioned hexagons in IP3R⁻ clones (wild type n=2; IP3R⁻ n=2; number of hexagons). These phenotypes were particularly noticeable in a mosaic cluster (Fig. 8, boxed area), in which the Vkg::GFP intensity was lower on the IP3R⁻ side with abnormal floor area and stress fibers. Western blot analysis of whole-head lysates suggested that this reduction in Vkg::GFP intensity was not caused by a decrease in Vkg::GFP expression (Fig. S5). This selective area expansion on the IP3R⁻ side also resulted in an apparent shift of grommet position towards the IP3R⁺ side. Lastly, the grommet area in IP3R⁻ clones was reduced by 27% (Fig. 8E), suggesting that the stress fiber contraction provides a stretch to enlarge the grommet size. Taken together, these observations demonstrate that tensile forces from the stress fibers shape the basement membrane in connection with endfeet contraction (Fig. 8F).

To understand how IOC waves regulate the endfeet stress fiber contraction, we stained l(3)itpr^90B.0 mosaic eyes with a rabbit polyclonal antibody against phosphorylated MRLC (p-MLC, Ser-19). Ser-19 phosphorylation in MRLC, a target of MLCK in response to Ca²⁺ signaling, can augment the activity of non-muscle myosin II (Scholey et al., 1980; Heissler and Sellers, 2016). The p-MLC antibody detected signals in both wild-type and IP3R⁻ cells at the retinal floor, although p-MLC signal intensity was reduced by 37% in IP3R⁻ cells (Fig. 9). This result suggests that IOC waves can impact stress fiber contraction by increasing the level of MRLC phosphorylation.

Using lGMR>GCaMP6m, we have discovered the presence of intercellular Ca²⁺ waves that spontaneously and regularly sweep across the honeycomb lattice of IOC cells and the primary pigment cells. We provide evidence that this IOC-specific communication is insulated from the cone cell Ca²⁺ spikes and phototransduction but requires IP3R function. The IOC waves are formed during pupal development and persist with age. As intercellular propagation of Ca²⁺ increases have been described in glia-glia networks in mammalian retina (Kurth-Nelson et al., 2009), our demonstration shows that communication with Ca²⁺ signaling is likely a fundamental and evolutionarily conserved feature of retinal accessory cells.

In adult Drosophila eye, photoreceptors depolarize with Ca²⁺ influx in response to light stimulation (Asteriti et al., 2017). As IOC and cone cells contact the photoreceptors, it seems plausible that the occurrence of IOC waves and cone cell spikes depends, at least in part, on phototransduction. The wave surges observed near the beginnings of recordings further lend support to the notion that IOC waves are influenced by illumination. However, the occurrence of IOC waves was clearly unaffected in a norpA null mutant. The characteristics of Ca²⁺ spikes in norpA³⁶ retina were comparable to those in wild type, suggesting that the complete removal of a phospholipase C that is essential for phototransduction has no modulatory effect on IOC waves. Moreover, although stimulus-independent Ca²⁺ oscillations in developing photoreceptors have recently been reported (Akin et al., 2019), these Ca²⁺ oscillations occur too early (∼55-95 h after pupa formation) to influence the IOC waves, and their Ca²⁺ characteristics show no resemblance to those of IOC waves. These observations imply that IOC waves are insulated from photoreceptor activities. Likewise, a mechanism must exist to compartmentalize the IOC and cone cell Ca²⁺ networks. One possible mechanism may involve a restricted propagation of second messenger required for Ca²⁺ increases. For instance, Inx2 (Innexin 2), a Drosophila homolog of the gap junction component connexin, is expressed specifically in IOCs and localized to the endfeet (Richard and Hoch, 2015). As Inx2 mediates the injury-induced slow Ca²⁺ waves in wing discs (Restrepo and Basler, 2016; Balaji et al., 2017), it may act similarly in the eye to regulate a cell-specific passage of second messenger required for IOC waves.

More importantly, using mutational analysis of IP3R, we have demonstrated involvement of Ca²⁺ communication in facilitating and coordinating stress fiber contraction at IOC endfeet. During pupal development, the appearance of endfeet stress fibers transitions from arrays of filaments emanating from the grommets at mid-pupal stage to large patches at late pupal stage (Baumann, 2004; Longley and Ready, 1995). The profiles of these structures continue to consolidate and appear as small dense patches by eclosion. In IP3R⁻ adult and pupal mutant retinas, the morphology of stress fibers arrests at earlier stages. Concomitantly, the retinal floor in IP3R⁻ clones increases by ∼40%, consistent with the notion that stress fiber contraction powers the retinal floor condensation. The apical surface in IP3R⁻ clones remains unaffected, suggesting that IP3R⁻ does not disrupt actin cytoskeleton globally and its effect on endfeet stress fibers is specific. Collectively, these results indicate that IP3R has a role in facilitating the progression of stress fiber contraction at the endfeet. Interestingly, whole IP3R⁻ eyes, generated by the EGUF-GMR-hid method (Stowers and Schwarz, 1999), appear larger and rounder (Bollepalli et al., 2017), in support of our observations that IP3R is linked to the formation of retinal curvature.

Several lines of evidence suggest that IOC waves and endfeet stress fiber contractions are causally linked. First, the phenomena of IOC waves and endfeet compaction are spatially and temporally correlated. Both processes are manifested in the secondary and tertiary pigment cells, and the emergence of IOC waves overlaps with the duration of endfeet stress fiber contraction. In addition, elimination of IP3R function in the eye does not affect cell fate determination and phototransduction (Acharya et al., 1997; Raghu et al., 2000), suggesting that these defects in IOC waves and stress fiber contractions are specific. Furthermore, MRLC Ser-19 phosphorylation at the retinal floor in IP3R⁻ clones is reduced, supporting a plausible mechanism by which IOC waves regulate myosin II activity through modulation of MRLC phosphorylation. Based on these observations, we propose that the removal of IP3R function abolishes Ca²⁺ increase from internal stores, which directly disrupts the occurrence of IOC waves. The absence of cytosolic Ca²⁺ increase in IOCs then inactivates myosin II at the endfeet, thereby inhibiting stress fiber contraction and retinal floor condensation. As Ca²⁺ increases in IOCs are elicited by waves, the stress fiber contractions required for retinal floor morphogenesis are coordinated. Alternatively, cytoskeletal elements can alter the subcellular distribution of IP3R and impact Ca²⁺ signaling (Walker et al., 2002; Turvey et al., 2005; Hours and Mery, 2010), and it is possible that the interaction between the actomyosin network and IP3R in IOC endfeet contraction is more complex than we envision.

The consistency of Ca²⁺ spikes in IOC waves makes this cellular communication well suited for coordinating endfeet stress fiber contraction for floor condensation. Analysis of IOC Ca²⁺ spikes suggest the presence of a regular refractory period, which ensures that all IOCs of the same eye experience a similar number of spikes, regardless of the frequency of IOC wave initiations. Furthermore, no increase in Ca²⁺ spike amplitude was seen in cells at wave merging, suggesting that all IOCs experience Ca²⁺ spikes of similar amplitude. Indeed, we confirmed that IOC Ca²⁺ spike parameters from different regions of the eye were similar. It is possible that this uniformity of Ca²⁺ spikes in IOC waves ensure that the contractile forces are evenly applied across the retina during floor reduction. Ca²⁺ oscillations in cultured myofibroblasts have been shown to induce periodic micro-contractions of actin fibers (Castella et al., 2010), and it is likely that IOC waves have a similar effect on actin stress fibers at the pigment cell endfeet. In this scenario, regular passages of IOC waves would induce periodic and incremental contractions of the endfeet stress fibers to compact the retinal floor. Tensile forces driving the apical constriction of cells in the mesoderm furrow during gastrulation and pulling the epidermis cells to move dorsally during dorsal closure also occur in a pulsed fashion (Solon et al., 2009; Martin et al., 2009).

Although the floor area increase in IP3R⁻ implicates IOC waves in floor condensation, the phenotypic severity (40%) is significantly less than the fourfold area reduction required (Longley and Ready, 1995), indicating that IOC wave-mediated contraction is not the sole mechanism for basal surface compaction. In support of this, the pupal retinal floor reduction commences at ∼55% pupal development, earlier than the start of IOC wave formation. Thus, the retinal floor condensation consists of at least two phases: an early IOC wave-independent phase, which contributes to the majority of basement reduction, and a late IOC wave-dependent phase. We have observed mid-pupal retina containing clusters at different stages of floor reduction (Fig. S6D-F), implying that the condensation of ommatidial basement is initiated locally in patches and that the clusters in different patches may compact at different rates. Tensile forces coordinated by IOC waves during the late phase may fine-tune the floor condensation by smoothing out these differences from the early phase.

The retinal basement membrane, the ECM plane that defines the proximal surface of the retina, is organized into two specializations: grommets and ridges. Grommets, the exit ports through which photoreceptor axons leave the retina, first form in early pupae at the completion of pattern formation (Fig. S6A-C). Ridges, dense linear ECM between adjacent grommets, emerge as stress fiber contraction draws grommets together. In wild-type pupal eyes containing a mixture of fully and partially contracted stress fibers, ridges were prominent in regions with dense stress fibers (Fig. S6D-F, arrowheads). In IP3R⁻ clones with reduced contraction, ridges were reduced. Together, these observations suggest that ridges are raised as compression deforms and buckles the ECM meshwork. Ridges assemble into a hexagonal network oriented orthogonally to the contractile stress fiber network. As Vkg (collagen IV) stiffens ECM (Walma and Yamada, 2020), the ridge network may supply elastic energy that counterbalances the contractile forces, thereby maintaining a stable retinal floor. The fly eye should be an excellent model for investigating how stress fiber contractile forces and ECM dynamics cooperate to shape tissues.

All fly crosses were carried out at 25°C in standard laboratory conditions. lGMR-GAL4 (8605), UAS-GCaMP6m (20XUAS-IVS-GCaMP6m; 42748), norpA³⁶ (9048), FRT^82B, GMR-src-mRFP (7124), FRT^82B, ubi-mRFP^nls (30555) and ey-GAL4, UAS-FLP; FRT^82B, GMR-hid/TM2 (5253) were obtained from Bloomington Stock Center (Bloomington stock numbers are indicated in parentheses). Both lGMR-GAL4 and UAS-GCaMP6m have transgenic insertions on the 2nd and 3rd chromosomes, and we have generated lGMR>GCaMP6m recombinants for respective chromosomes. Both lGMR>GCaMP6m recombinants exhibited similar Ca²⁺ waves; nonetheless, for simplicity and consistency, we used the 2nd chromosome recombinant to generate all the movies presented. The FRT^82B, l(3)itpr^90B.0 recombinant was a gift from Dr Roger Hardie (Cambridge University, UK), and the vkg::GFP (GFP^vkg-G00454) protein trap line (Morin, 2003) was a gift from Dr Wu-Min Deng (Tulane University School of Medicine, USA).

To monitor Ca²⁺ activities in IP3R mitotic clones, lGMR>GCaMP6m/CyO; FRT^82B, l(3)itpr^90B.0/TM3 males were mated with ey-FLP/CyO; FRT^82B, GMR-src-mRFP/TM3 females. IP3R clones in the adult eyes of ey-FLP/lGMR>GCaMP6m; FRT^82B, l(3)itpr^90B.0/FRT^82B, GMR-src-mRFP progeny, easily recognized by the absence of membrane-associated mRFP, were imaged as described below. To examine actin stress fiber morphology in IP3R clones, cn, bw; FRT^82B, l(3)itpr^90B.0/TM3 males were mated with ey-FLP; cn, bw; FRT^82B, ubi-mRFP^nls/TM3 females. To monitor Vkg::GFP localization in IP3R clones, cn, bw, vkg::GFP; FRT^82B, l(3)itpr^90B.0/TM3 males were mated with ey-FLP; cn, bw, vkg::GFP; FRT^82B, ubi-mRFP^nls/TM3 females. To generate IP3R⁻ Vkg::GFP-containing retinas for western blotting, cn, bw, vkg::GFP; FRT^82B, l(3)itpr^90B.0/TM3 or cn, bw, vkg::GFP; FRT^82B (control) males were mated with ey-GAL4, UAS-FLP; FRT^82B, GMR-hid/TM2 females.

For each Ca²⁺ activity recording, a live fly was glued eye-first onto a clean glass coverslip (Franceschini and Kirschfeld, 1971), maintained over an hour in the dark for adaptation, and exposed to 647 nm light pulses to shift active metarhodopsin to rhodopsin prior to imaging. Mounted flies were imaged on an inverted Nikon Eclipse TE2000 using a Nikon 40×1.3 Plan Fluor objective. Time-lapse movies were captured at one frame every 2 s, unless otherwise indicated, for 5 min using a Photometrics CoolSNAP camera controlled using Metamorph. Images were binned 2×2 and Nikon ND filters were used to attenuate excitation for typically 400 ms exposures. No decrease in signal was observed over single or repeated observations. Mounting is non-injurious to eyes and animals; when kept in a humid chamber and fed with sugar water, it was possible to image flies repeatedly over several days.

For larval eye discs, tissues were dissected from crawling third instar larvae in their own hemolymph under a drop of halocarbon oil (Oil 10 S, VOLTALEF; VWR) on a coverslip. Coverslips were mounted on a microscope slide with spacers allowing full exposure of the halocarbon oil to air while imaging on the inverted Nikon microscope.

The image files were rotated to bring anterior to the right and processed with ImageJ as described as below.

Time-lapsed image stacks (2 s per frame for a 5-min recording) were separated into 151 individual png files using ImageJ (Image>Stacks>Stack to Images). A file series comprising sequential subtractions of signal intensities from consecutive png files were generated in Python (code provided upon request). Composite files with color-coded wave fronts from various time points are generated with ImageJ (Image>Color>Merge Channels). This operation was only performed on IOC recordings because the Ca²⁺ spike intervals in the cone cell movie (Movie 4) were too short.

To analyze the cellular signal intensities, the cells were labeled by hand in ImageJ with fixed-sized circular regions of interest (ROIs) (5-pixel diameter), which were imported into ROI manager. Text files containing the mean signal intensities over time were then generated with the Multi-Measure function, and the numeric series were processed by a Python code (code provided upon request). The code normalized the intensity values from each cell by subtracting the minimal value in the series and dividing by the range [i.e. F_normalized=(F−F_min)/(F_max−F_min)]. With an artificial cutoff of 0.24 (the value was empirically determined from applying the algorithm on several data sets, and F_normalized<0.24 is considered noise), the number of peaks, duration of peaks, intervals between peaks, time of onset for the peaks (T_on), and time to baseline return for the peaks (T_off) were counted.

The cone cell movie (Movie 4) had significant sample movements, which interfered with the extraction of mean intensity over time in individual cells. This issue was remedied by digitally stabilizing the image frames with the ImageJ Template-Matching plug-in. To improve the image quality, a minimum z-stack projection was used as denominator for each frame with ImageJ Ratio Plus to make a ratio movie, followed by the ImageJ Enhance Contrast function. To characterize cone cell spikes, cells were manually labeled with fixed-sized circular ROIs and mean signal intensities over time were obtained with ImageJ as described for IOCs (see ‘Quantitative analysis of IOC GCaMP6m signals’). However, because the signal intensities in cone cells were dim and often influenced by the passages of IOC waves, the cone cell signal intensities were normalized by subtracting the signal intensity from a nearby IOC.

For comparing GCaMP6m in IP3R⁻ mosaic eye, we obtained an mRFP map of the eye prior to the acquisition of a time-lapse stack of the GCaMP6m signal. A dual color movie of IP3R⁻ mosaic eye (Movie 9) was generated from 151 composite files, derived from merging individual GCaMP6 m image files with the mRFP map file in ImageJ.

For immunostaining, dissected retinas were fixed with 4% paraformaldehyde and permeabilized with PBS+0.3% Triton X-100. Retinas were incubated with rabbit anti-phospho-MLC (3671T; SER-19; Cell Signaling Technology) polyclonal antibody at 1:10 overnight at 4°C and washed four times with PBS+0.3% Triton X-100. Alexa-conjugated phalloidin 647 (A22287, Invitrogen) and Alexa Fluor 488 secondary antibody (A11034, Invitrogen) were used at 1:100. All images were acquired with Zeiss LSM 710 laser-scanning confocal microscope. Zeiss Zen 2010 and Volocity (Improvision) softwares were used to render confocal stacks for 3D reconstruction. To quantify phospho-MLC staining, six equal sized squares from wild-type and IP3R⁻ patches were randomly selected, and the mean signal intensities within these squares were determined with ImageJ.

For western blot analysis, 15 cn, bw, control, or IP3R⁻ heads, generated by EGUF method (Stowers and Schwarz, 1999), were homogenized in 50 μl 1× SDS loading buffer and boiled at 95°C for 5 min, then 8 μl samples were separated on 10% polyacrylamide gels, transferred onto Bio-Rad nitrocellulose membrane using semi-dry transfer (at 10 V for 45 min), and probed with rabbit polyclonal anti-GFP (ab6556, Abcam) or mouse monoclonal anti-β-tubulin (E7, Developmental Studies Hybridoma Bank, Iowa) antibodies at 1:1000 and 1:100, respectively. HRP-conjugated anti-rabbit (611-103-122, Rockland Immunochemicals) and anti-mouse (sc-2005, Santa Cruz Biotechnology) secondary antibodies were used at 1:10,000, and blots were developed using Luminol Chemiluminescent HRP Substrate (Thermo Scientific).

We thank Dr Roger Hardie (Cambridge University, UK) for sharing the FRT, IP3R recombinant and for discussion of IP3R phenotypes. Funding for the LSM710 was provided by a National Institutes of Health NCRR Shared Instrumentation Grant (1 S10 RR023734-01A1).

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199700.