Control of photoreceptor cell morphology, planar polarity and epithelial integrity during Drosophila eye development

Many developmental processes involve conversion of a monolayer epithelium into a differentiated tissue with numerous cell types and significant morphological complexity. As the epithelium differentiates, there are coordinated changes in cell fate, cell morphology, cell polarity and the relative positions of cells. The construction of a complex tissue from an undifferentiated epithelium is exemplified by the development of the Drosophila compound eye, which comprises an almost perfect array of roughly 750 unit simple eyes known as ommatidia (Wolff and Ready, 1993). Each ommatidium includes a central core of eight photoreceptor cells referred to as retinula or R cells (R1-6 are known as the ‘outer’ R cells while R7 and R8 are referred to as the ‘central’ R cells). The adult eye develops from an epithelial monolayer known as the imaginal eye disc. The R cell precursors are first specified in the late third instar larval disc, when a dorsoventral groove, called the morphogenetic furrow, sweeps across the disc from the posterior to the anterior and leaves organized clusters of R precursor cells in its wake (Ready et al., 1976; Tomlinson, 1985; Tomlinson and Ready, 1987). During larval and pupal stages, a plethora of gene products regulate cell division, cell specification, relative cell shape and movement, as well as programmed cell death in the developing retina (Cagan and Ready, 1989a; Wolff and Ready, 1991).

One of the developmental events initiated shortly after specification of clusters of presumptive R cells is establishment of planar polarity in the imaginal disc epithelium. In a process regulated by so-called ‘tissue polarity’ genes, each of the R-cell precursor clusters senses the direction of the furrow progression and receives polarity information from its more mature neighbors (reviewed by Mlodzik, 1999). The R-cell cluster rotates through 90°, so that the R3/R4 pair, which initially faced the furrow, comes to face the dorsal or ventral pole of the disc. This rotation occurs in opposite directions in the dorsal and ventral halves of the disc, creating an axis of global mirror image symmetry known as the equator. Each cluster also loses its initial bilateral symmetry and becomes chiral as the R4 cell loses contact with the R8 cell and moves to a more equatorial position. The Jun kinase (JNK) signaling pathway has been shown to play a prominent role in creating the distinction between R3 and R4 cell fates during the establishment of planar polarity in the eye (Weber et al., 2000). When an activated form of the JUN transcription factor is expressed in the R3 and R4 precursor cells, it gives rise to a symmetrical, a misrotated or a chirally reversed ommatidium (Weber et al., 2000).

During pupal stages, the apical faces of the R cells undergo a stereotypical series of interactions that are associated with the expansion of the apical surface downwards to meet the cone cell plate at the floor of the ommatidium (Longley and Ready, 1995; Sang and Ready, 2002). The apical surface of each R cell is mechanically anchored to the cone cell plate prior to further deepening of the retina (Cagan and Ready, 1989a; Longley and Ready, 1995). As terminal differentiation occurs, a photosensitive organelle – the rhabdomere – forms at the apical surface of each R cell and eventually spreads down the entire cell length. The apical faces of the interacting R cells, which are enriched for F-actin, Armadillo, DE-cadherin and Crumbs, are dynamically rearranged during photoreceptor morphogenesis (Sang and Ready, 2002). It is not clearly understood what molecules direct the longitudinal expansion of the apical surface or of its rhabdomere, although recent evidence suggests a role for eyes closed, which encodes the p47 co-factor of a p97 ATPase that is implicated in membrane fusion, in regulation of photoreceptor morphogenesis (Sang and Ready, 2002).

The hindsight (hnt) gene (peb – FlyBase) encodes a nuclear zinc-finger protein with all of the hallmarks of a transcription factor (Yip et al., 1997). In the embryo, hnt function is required for germband retraction (Lamka and Lipshitz, 1999; Yip et al., 1997), dorsal closure (Reed et al., 2001) and tracheal morphogenesis (Wilk et al., 2000). Expression of the hnt gene is required in the amnioserosa to downregulate JNK signaling in that tissue and to create a sharp distinction between the epidermal and amnioserosal tissue (Reed et al., 2001). In hnt mutants, the expression of the JUN target genes, decapentaplegic (dpp) and puckered (puc), persists inappropriately in the amnioserosa, leading to failure of dorsal closure (Reed et al., 2001). The leading edge cells of the epidermis neither undergo appropriate cell shape changes nor accumulate apicolateral F-actin or phosphotyrosine-containing proteins characteristic of the focal complex seen in wild-type embryos at the boundary between the leading edge and the amnioserosa (Reed et al., 2001). In addition to a role in regulating JNK signaling during dorsal closure, HNT function is necessary for the maintenance of tracheal epithelia (Wilk et al., 2000). In hnt mutants the tracheal system initially develops normally but, by stage 15, the epithelium loses its integrity and forms sacs and vesicles. The mutants also lack apical extracellular structures known as taenidia, suggesting that terminal differentiation of the tracheal cells is defective (Wilk et al., 2000).

We describe the role of HNT during eye development. We show that the HNT protein is expressed in all of the photoreceptor cell precursors in developing eye discs. Examination of a hnt temperature-sensitive allele and analyses of chimeras carrying patches of cells homozygous for lethal hnt alleles suggest that there are at least five functions of HNT during eye development. First, HNT is necessary immediately posterior to the morphogenetic furrow for the correct morphology and arrangement of cells in the ommatidial precluster (which contains the R8, R2, R5, R3 and R4 precursor cells), as well as for the specification of the R2, R5, R3 and R4 precursor cells. Second, HNT function is necessary in the outer R cells for the establishment of correct chirality and rotation of the ommatidia. This tissue polarity function may involve downregulation of the JNK signaling cascade by HNT. Third, HNT function is required to maintain ommatidial structure and integrity during the mid-pupal stage. Fourth, HNT function is required in each R cell to accomplish apical-basal extension of the cell and its rhabdomere as the adult retina is assembled in the pupal period. Finally, HNT function is required for the separation of the rhabdomeres of neighboring cells that occurs during the late pupal period. Thus, HNT possesses both eye-specific functions (precluster patterning) and general functions (regulation of the F-actin-based cytoskeleton, regulation of JNK-mediated processes, control of tissue integrity) in development. We speculate that regulation of a single molecular pathway involving cell-cell or cell-extracellular matrix adhesion, or communication, underlies all of these functions.

The following fly stocks were used in this study: hnt^XO01, hnt^EH704a, hnt^XE81, hnt^peb (Yip et al., 1997), hnt¹¹⁴² (Wilk et al., 2000), jun¹ (Kockel et al., 1997), jun^76–19 (Hou et al., 1997), bsk¹ and bsk² (Sluss et al., 1996), rho^PX81 (Freeman et al., 1992), dpp-lacZ BS3.0 (Blackman et al., 1991), and N^ts1 (Cagan and Ready, 1989b; Shellenbarger and Mohler, 1975). The FRT line w¹¹¹⁸ P{w^+mC=piM}5A P{w^+mC=piM}10D P{ry^+t7.2=neoFRT}18A and the FLP recombinase stock w¹¹¹⁸; MKRS, P{ry^+t7.2=hsFLP}86E/TM6B, Tb¹ were obtained from the Bloomington Drosophila Stock Center.

Clones were induced in the eye according to standard protocols (Harrison and Perrimon, 1993; Xu and Rubin, 1993). First instar larvae were heat-shocked for 1.5 hours in a waterbath at 38.5°C. Larval eye discs were dissected from third instar larvae and double antibody stained with the antibody of interest and α-HNT, the latter enabling identification of clones. Scars were scored in the adult or pupa and these were found at a frequency of approximately 50% among the class of flies that had the genotype required for a FLP-induced event.

Antibody staining of eye discs was carried out using standard protocols (Wolff, 2000). Primary antibodies used were: mouse α-HNT monoclonal (27B8 1G9 at 1:25 dilution) (Yip et al., 1997); rabbit α-β-Galactosidase (α-β-gal; 1:1000; Cappel); mouse α-Phosphotyrosine monoclonal (α-PY; 1:500, Upstate Biotechnology), mouse α-Boss monoclonal (at 1:1000 dilution) (Van Vactor et al., 1991); rabbit α-Spalt (at 1:500 dilution) (Barrio et al., 1999), mouse α-Glass monoclonal (at 1:5 dilution) (Moses and Rubin, 1991). All secondary antibodies were used at a dilution of 1:250 and were obtained from Jackson Immunoresearch: these were horseradish peroxidase (HRP) conjugated goat α-mouse antibody, HRP-conjugated goat α-rabbit antibody, FITC-conjugated goat α-mouse antibody and TRITC-conjugated goat α-mouse antibody. Visualization of F-actin using FITC-conjugated phalloidin (Sigma) was according to Wolff (Wolff, 2000). X-Gal staining was carried out as described elsewhere (Wolff, 2000). For double staining with the α-HNT antibody, the discs were first fixed in 4% formaldehyde in 1×phosphate-buffered saline (PBS) on ice. The X-Gal staining was then carried out overnight. Afterwards, the discs were washed thoroughly and then placed in primary α-HNT antibody. The rest of the protocol was carried out as for antibody staining. Cell death was detected using an Apoptag S1760 direct detection kit (Intergen). Standard methods were used to prepare histological sections of adult eyes (Tomlinson and Ready, 1987). A Zeiss Axioplan microscope was used for light microscopy and images were captured with a Spot digital camera (Diagnostic Instruments). A Zeiss Axiovert 100 microscope with LSM510 software was used for laser confocal microscopy. Images were processed and displayed using Photoshop software (Adobe). Samples were prepared for scanning electron microscopy (SEM) using standard methods (Ready et al., 1976). A Hitachi S570 microscope was used to view the eyes.

For the Notch experiment, N^ts1 larvae were shifted to 32°C for 24 hours as described elsewhere (Cagan and Ready, 1989b). Third instar larval discs were then dissected and stained with the α-HNT antibody. For temperature shift experiments, 1 hour hnt^peb egg collections were raised at either 18°C or 29°C and then shifted either up or down at 12 hour intervals. Once adult flies had eclosed, their eyes were scored for external roughness under a dissecting microscope.

To study Hindsight (HNT) protein expression we used a mouse monoclonal antibody (α-HNT) that specifically recognizes the HNT gene product as it does not stain hindsight (hnt) mutant embryos (Yip et al., 1997). In the third instar larval eye disc (Fig. 1A), HNT first accumulates in cells in the middle of the furrow partially overlapping with decapentaplegic (dpp) reporter gene expression (Fig. 1B). The most anterior edge of HNT expression appears serrated (Fig. 1C). In the posterior part of the furrow, HNT expression is found in all cells. Posterior to the furrow, HNT expression resolves to only the photoreceptor (R) cell precursors (Fig. 1D). This refinement most probably reflects a negative regulation of HNT expression in non-neuronal precursor cells behind the furrow and is consistent with HNT expression being a read-out of neuronal fate. The pattern of HNT staining in the R cell precursors prefigures the sequence of their neuronal determination (i.e. R8, R2, R5, R3 and R4 stain by row two, R1 and R6 stain by row four, and R7 stains by row six). From rows two to six, HNT expression is slightly higher in the R3 and R4 precursor cells relative to the rest of the cells within the clusters (Fig. 1D inset). This is the stage at which the R3 and R4 cells sense the signals that determine ommatidial planar polarity. HNT is also expressed in the ocellar precursor cells found in the anterior portion of the eye disc (Fig. 1A) but is not found in the cone or pigment cell precursors in either larval or pupal discs. HNT is, however, expressed in the bristle cell precursors in pupal eye discs (Fig. 1E).

HNT is also present in sensory neuron precursor cells in all other discs, including the wing disc (Fig. 1F), the leg disc (Fig. 1G) and the antennal region of the eye-antennal disc (Fig. 1A). The identity of these HNT-positive cells was confirmed by double antibody staining discs from A101 larvae, which have an enhancer trap in the neuralized gene (Boulianne et al., 1991; Huang et al., 1991), where there is β-gal expression in all sensory neuron precursor cells (data not shown).

In a Notch^ts1 mutant in which the function of the NOTCH receptor has been reduced, neural hypertrophy occurs in the discs (Cagan and Ready, 1989b). In Notch^ts1 discs, HNT protein is expressed in the supernumerary neuronal precursor cells (compare Fig. 1G with 1H). Thus, HNT expression in neuronal precursor cells is a read-out of sensory neuron specification. This experiment also demonstrates that wild-type Notch function is not required for HNT expression.

To understand the role of HNT during eye development, we undertook a phenotypic analysis of the viable hnt^pebbled (hnt^peb) mutant, which has a temperature-sensitive rough eye phenotype (Fig. 2) (Yip et al., 1997). In addition, we analyzed three lethal, antibody-null alleles of hnt: hnt^XE81 and hnt^EH704a, which behave as genetic nulls, and hnt^XO01, which behaves as a strong hypomorph (Yip et al., 1997). The lethal hnt mutations fail to complement the hnt^peb rough eye phenotype (Fig. 2C) (Yip et al., 1997). To examine the consequences of removing HNT function in the eye disc we generated patches (‘clones’) of cells homozygous for the lethal hnt alleles using the FLP/FRT recombination system (Golic, 1991).

In wild type, the dpp-lacZ reporter gene marks all cells in the furrow and is abruptly downregulated in the preclusters as they emerge posterior to the furrow (Fig. 3A) (Blackman et al., 1991). In patches mutant for the hnt^XE81 allele, the expression of the dpp-lacZ gene is normal (Fig. 3B). However, in hnt^peb (Fig. 3C) and in patches mutant for the hnt^XO01 hypomorphic allele (Fig. 3D), dpp-lacZ expression is normal in the furrow but persists posterior to the furrow in some R cells. The effect was more severe in the hnt^peb mutant than in hnt^XO01 patches. These results support a role for HNT in downregulating dpp posterior to the furrow. The absence of an effect in strong hnt mutants (i.e. hnt^XE81 and hnt^EH704a) suggests that there may be a positive feedback loop between DPP signaling and HNT; thus, the effect is only revealed when there is some residual HNT function.

Staining of hnt^peb discs and hnt patches with markers for neuronal determination showed that there is a delay in R-cell precluster formation posterior to the furrow in hnt mutants. For example, in hnt patches there is a one- to two-row delay (representing 2-4 hours) in expression of the marker recognized by α-HRP (Fig. 3E) and of Elav (data not shown) relative to the adjacent wild-type region of the epithelium.

We next examined early cluster morphology using the α-phosphotyrosine antibody (α-PY) that highlights all cell membranes, particularly those of the R cells (Fig. 3F). The α-PY antibody outlines the arc stages in row one at the posterior edge of the furrow [the normal developmental progression in cluster morphology is reviewed by Wolff and Ready (Wolff and Ready, 1991)]. In hnt mutant patches the arcs appear to form normally and at evenly spaced intervals (Fig. 3F). Within the hnt mutant tissue, the earliest defects involve cluster shape and cell number at the five cell precluster stage (Fig. 3F). In mutant patches, the definitive preclusters usually have the normal number of cells (five; R8, R2, R5, R3, R4) but, occasionally (i.e. in fewer than 5% of the preclusters), there are missing cells or additional cells with high levels of α-PY staining. More strikingly, some of the preclusters either have cells misarranged relative to one another or the cluster does not rotate properly.

Glass is normally expressed in all presumptive R cells (Moses and Rubin, 1991). In eye discs from hnt^peb larvae, most precluster cells express Glass (compare Fig. 3G with 3H), indicating they have attained a photoreceptor identity. However (Fig. 3H), some of these preclusters are missing Glass-positive cells (22%, n=46), have displaced cells (30%, n=46) or have additional posterior cells (17%, n=46). The frequency of these defects in Glass staining is five- to tenfold higher than the frequency of preclusters with missing or additional cells as assayed using α-PY. A likely explanation for this difference in frequencies is that the correct number of cells is recruited into definitive preclusters in hnt mutants but that these cells have a reduced ability to achieve presumptive R cell fate.

To examine whether specification of individual R cell fates proceeds normally in hnt mutants, we used three markers that label the early R cells. These are the α-Boss antibody, which labels determined R8 cells beginning three rows posterior to the furrow (Van Vactor et al., 1991); the rhomboid enhancer trap, rho^PX81, which is expressed strongly in R8, R2 and R5, beginning two rows posterior to the furrow (Freeman et al., 1992); and the α-Spalt antibody, which labels the nuclei of the R3 and R4 precursor cells, beginning two rows posterior to the furrow (Barrio et al., 1999). In hnt patches, Boss accumulates in all R8 cells; however, there is a delay of one to two rows relative to the adjacent wild-type tissue and Boss levels are often reduced (Fig. 4A, arrowheads). These results suggest that, although the mutant R8 cell in each precluster is eventually properly determined, the Boss ligand is either not tightly localized to the apical surface of the R8 cell or there are lower amounts of it in the cell.

The rho^PX81 enhancer trap is expressed within hnt patches. However some of the clusters have only one or two nuclei that stain, rather than the predicted three (47%, n=129 clusters; Fig. 4B,C). As there are determined R8 cells present in these patches, we hypothesize that the R2 or R5 precursor cells are mis-specified, or fail to be specified, in many of the preclusters and thus do not express the rho^PX81 enhancer trap. In many clusters (74%), the nuclei are displaced relative to one another (Fig. 4B,C), consistent with a problem in cell shape or cell arrangement within the cluster (as visualized above with the α-PY antibody; Fig. 3F). In 79% of the triads, there are also defects in the rotation of the clusters (Fig. 4B,C).

The α-Spalt antibody, which labels the R3 and R4 precursor cells, also shows aberrant staining within hnt patches (Fig. 4D,E). In 57% of clusters, only one cell stains with α-Spalt antibody (n=145 clusters). When two nuclei stain, often they are at different planes relative to one another (84% versus 0% in the adjacent wild-type tissue). This result implies that the mutant R3 or R4 precursor cell is mis-shapen or aberrantly positioned in the cluster. In 67% of the clusters the cells fail to rotate correctly.

In summary, our results show that specific furrow markers persist posterior to the furrow in hnt mutants, that there is a delay in specification of preclusters as cells exit the furrow, that there are morphological defects in the preclusters, that recruitment of cells into specific R cell fates does not occur and that there are planar polarity defects in hnt mutants.

The mutant phenotypes we observed in the developing eye disc suggested that HNT plays a role in establishing planar polarity. To define this role at single cell resolution, we analyzed genetically mosaic ommatidia along the borders of hnt mutant patches in the adult eye. In wild-type ommatidia, each R cell occupies a distinctive position within the trapezoidal array of rhabdomeres (Fig. 5). The hnt mosaic ommatidia showed polarity defects, including reversed chirality (56%, n=43 defective ommatidia; see Fig. 5A classes II, V and VI), misrotation (44%; see Fig. 5A classes III-VI) and loss of R3/R4 asymmetry (30%; see Fig. 5A classes VII and VIII). The polarity defects within hnt patches are restricted to mosaic ommatidia; we did not see any cases where the genetically wild-type ommatidia adjacent to mosaic ommatidia are misoriented. Thus, HNT implements a local function in coordinating polarity establishment within an ommatidium.

By quantifying the genotypes of R cell subtypes in genetically mosaic ommatidia, we determined which R cells required HNT function for establishment of ommatidial polarity. Normal chirality and normal rotation require HNT function in the outer photoreceptor cells (R1-R6), as trapezoidal ommatidia with incorrect chirality and/or rotation always include at least one mutant outer photoreceptor cell (see Fig. 5A classes II-VI and Fig. 5B). Notably, the particular combination of mutant R cells does not predict to what extent an ommatidium will be misrotated or whether it will have reversed chirality. There is, however, a bias in the frequency with which specific R cells are mutant in the misoriented ommatidia (Table 1). For example, cases where the R3 (49%), R4 (36%) or R6 (26%) cells are mutant for hnt are somewhat over-represented among defective ommatidia. This finding implies that HNT function may be most important in the R3, R4 and R6 cells for control of ommatidial polarity, although the other outer R cells also clearly make some contribution.

The borders of the hnt^XE81 and hnt^XO01 clones also include R3/R3 type symmetrical ommatidia where the R3 or the R4 cell has not taken up the correct position (n=12, see Fig. 5A classes VII and VIII and Fig. 5C). In these ommatidia, either the R3 or R4 cells, or both, are mutant for hnt, and the chirality, as assayed by the R8 projection, is also sometimes incorrect. These results imply that HNT is required in the R3 and/or R4 cells to establish ommatidial asymmetry. The fact that all of the symmetrical ommatidia are of the R3/R3 type suggests that HNT function is necessary in the R3/R4 pair to determine R4 fate and location; in the absence of HNT, the R4 cell adopts a more R3-like position in the ommatidium.

Given the previously reported role of JNK signaling in establishing planar polarity in the eye (Weber et al., 2000) and the planar polarity defects in hnt mutants reported here, we tested for genetic interactions between JNK pathway mutants and hnt^peb. Two alleles of jun (jun¹, a hypomorph, and jun², a genetic null) and two alleles of basket (bsk¹ and bsk²) act as dose-sensitive suppressors of the hnt^peb photoreceptor defects, as assayed in histological sections. hnt^peb/Y; jun¹/+ eyes show 31% wild-type ommatidia (n=775) versus 17% in hnt^peb/Y control siblings (n=340), and hnt^peb/Y; jun²/+ eyes have 26% wild-type ommatidia (n=750) versus 15% in controls (n=207). hnt^peb/Y; bsk¹/+ eyes show 11% wild-type ommatidia (n=828) versus 7% in controls (n=509), and hnt^peb/Y; bsk²/+ have 11% wild-type ommatidia (n=957) versus 8% in controls (n=606). When polarity defects are scored in those ommatidia with wild-type numbers of R cells, from hnt^peb/Y; bsk¹/+ eyes, 14% show polarity defects versus 21% in hnt^peb/Y controls, whereas in hnt^peb/Y; bsk²/+ eyes, 17% show polarity defects versus 40% in hnt^peb/Y controls. The direction of these genetic interactions in the eye is consistent with a role for HNT in downregulating JNK signaling.

In summary, we have shown here that HNT function is required to establish planar polarity in the eye and have presented genetic evidence that this role may be accomplished by downregulation of JNK signaling.

In eyes from hnt^peb flies raised at the restrictive temperature, the facets in the eye are misaligned, with occasional ommatidial fusions (2.8%) and a small percentage of bristle displacements and bristle size defects (Fig. 2B, Fig. 6A). Transverse sections of eyes from hnt^peb flies reveal many photoreceptor cell abnormalities (Fig. 6C versus 6B). Although some ommatidia contain extra apical central cells (Fig. 6C, white arrowhead; 12%, n=820 ommatidia), the most common defect is missing photoreceptor cells (Fig. 6C, black arrowheads), including missing outer cells (43%), missing central cells (1%), or missing outer and central R cells (11%).

We determined the temperature-sensitive period of the conditional hnt^peb mutant and found that the crucial period for external eye roughness spanned from the mid-third instar to early pupal stages (data not shown). Flies that spent all of this crucial period at the restrictive temperature had the roughest eyes, implying that there is a continuous requirement for HNT function through these stages of development. The fact that flies spending only the early pupal stages at the restrictive temperature [from 2 hour after puparium formation (APF) onwards at 29°C] are also rough-eyed, indicates there is a role for HNT during the differentiation of the retinal epithelium during the pupal period (i.e. after the photoreceptor cells have been patterned and determined). Consistent with this interpretation is the fact that patches of cells homozygous for strong alleles of hnt form scars in adult eyes (Fig. 6D). Transverse sections through these scars show that they are devoid of R cells but still contain pigment cells (Fig. 6E). From our studies of hnt clones in the larval eye disc, it is clear that this phenotype is not due to a failure of the R cells to develop, but must be a consequence of loss of R cells from the retinal epithelium during the pupal stages. Longitudinal sections through hnt clones revealed that ommatidia in which all of the photoreceptor cells are mutant for hnt^XE81, hnt^XO01 or hnt^EH704a are unable to maintain their position in the developing epithelium and fall out of the retina onto the lamina of the optic lobe of the brain (Fig. 6F, arrowheads).

To ascertain whether the loss of R cells is a secondary consequence of increased programmed cell death we stained mutant patches for apoptotic cells. hnt patches show normal levels of cell death relative to adjacent wild-type eye tissue during late larval stages (data not shown). Thus, cells that are not specified correctly in the preclusters do not die but are reincorporated into the pool of undifferentiated cells in the eye disc. In addition, during early-to-mid pupal stages, there is no increased apoptosis in patches of hnt mutant cells relative to adjacent wild-type cells (data not shown; P5, 12.5-25 hour APF and P8, 47-57 hour APF at 25°C, were tested). These results imply that the loss of hnt mutant R cells from the retinal epithelium is not a secondary consequence of cell death.

We used the adult eye mosaics described above to ask which R cells require HNT function to maintain the integrity of the retinal epithelium during pupal development. An examination of 94 patches led to the identification of 71 genetically mosaic ommatidia. Typically, in wild-type controls, one medium-sized patch yields about 10 mosaic ommatidia; we would therefore have expected over 900 mosaic ommatidia in our experiment. The more than tenfold under-representation of mixed ommatidia along the borders of hnt clones suggests that many of them formed but subsequently lost their integrity. By quantifying the genotypes of R-cell subtypes in surviving mixed-genotype ommatidia, we were able to establish which R cells required HNT function in order for an ommatidium to be wild type with respect to photoreceptor number and internal arrangement (n=59, Fig. 5A classes I-VI). Our analyses show that there must be at least five hnt⁺ R cells – of any subtype – within the ommatidium for maintenance of its integrity in the retinal epithelium.

Beyond its role in maintenance of epithelial integrity, HNT function is also required for normal rhabdomere structure. In a wild-type adult eye, the R1-R6 cell bodies and their rhabdomeres extend over the full apical-to-basal extent of the ommatidial column (schematized in Fig. 7A). In eyes from hnt^peb flies some of the photoreceptor cells (4.2%) have enlarged rhabdomeres when viewed in individual histological sections. Such an apparent enlargement could result from failure in apical-to-basal extension of the cell and its rhabdomere. In order to study morphological defects in individual hnt mutant adult R cells, we made serial sections through genetically mosaic ommatidia containing cells homozygous for hnt alleles. Mutant rhabdomeres fail to extend the normal length of the ommatidial column, indicating that hnt cells often have shortened cell bodies (Fig. 7A, left side). This phenotype is R cell autonomous, as all R cells with defective rhabdomeres are also mutant for hnt. During the late pupal period (110 hour APF), the rhabdomeres of the individual R cells within an ommatidium separate from each other to give the canonical trapezoidal pattern found in the adult (Cagan and Ready, 1989a). An additional defect found in serial sections of hnt genetic mosaic ommatidia is failure of rhabdomere separation (Fig. 7A, right side): rhabdomere separation is often normal in the apical region of the ommatidium but, as one proceeds basally, one can observe failure of the rhadomere of a hnt mutant cell to separate from its wild-type neighbor (Fig. 7A, right side).

To investigate the developmental basis for the morphological defects in individual R cells, as well as the loss of R cells from the retinal epithelium, we stained larval and pupal discs with phalloidin-FITC, which binds to and labels the polymerized form of actin (filamentous actin or F-actin). F-actin is concentrated at the apical tip of the cells in preclusters in wild-type larval discs (Fig. 7B). Later, F-actin is present in the elongating rhabdomeres in pupal discs (Figs. 7C,D) (Longley and Ready, 1995; Sang and Ready, 2002). When patches of hnt^XE81, hnt^EH704a or hnt^XO01 cells in larval discs are stained with phalloidin, the F-actin localization is depleted and/or more diffusely localized, particularly in the more posterior regions of the mutant part of the disc (Fig. 7B). This disruption of F-actin distribution is probably not due to a late change in the apical constriction of clusters as α-PY antibody-stained discs look fairly normal with respect to these more posterior clusters (see Fig. 3F). Depletion of apical F-actin is also evident in hnt patches in early pupal discs (stage P5; data not shown). By later pupal stages (P8/9, 47-69 APF at 25°C) the R cells from ommatidia that lack all hnt function have dropped out of the epithelium (Fig. 7C,D). Ommatidia mosaic for hnt function have a lower concentration of F-actin in their R cells and it is found more basally than in their wild-type neighbors (arrows, Fig. 7C,D). This defect in actin polymerization or assembly may be a developmental precursor to the rhabdomere phenotype observed for mutant photoreceptor cells in the adult eye (Fig. 7A).

The earliest defects we observed in hnt mutant eye tissue occur during the formation of the definitive five-cell ommatidial preclusters. Specifically, we have shown that there is a 4 hour delay in specification of the ommatidial preclusters relative to wild type, the shape of individual R cells within these preclusters is often abnormal, and the relative orientation of the presumptive R cells within a precluster is disrupted. There are two plausible explanations for the observed defects: a failure in cell cycle synchronization or abnormal cellular morphology. First, it has been shown that lack of cell cycle synchronization in the furrow can lead to a one to two column delay in R cell precursor determination (e.g. in roughex mutants) (Thomas et al., 1994). Furthermore, a precluster phenotype in which nuclei are inappropriately positioned in the apicobasal plane relative to one another, is seen in mutants disrupted for cell cycle synchronization in the furrow (e.g. thick vein, saxophone, schnurri) (Penton et al., 1997). Thus, it is possible that failure of cell cycle synchronization underlies the hnt mutant phenotypes described here. Alternatively, the primary defect in hnt mutants might be the abnormal morphology of the R cells. For example, the act up mutant, which alters the apical morphology of the precluster cells, shows premature dpp expression, premature R cell differentiation and uncoordinated neuronal determination (Benlali et al., 2000). We have shown here that dpp expression persists posterior to the furrow in eye discs of hypomorphic hnt mutants. This phenotype is distinct from that described for act up or, to our knowledge, any other mutant. It is consistent, however, with our previous analyses demonstrating a role for HNT in downregulation of dpp expression in the amnioserosa (Reed et al., 2001).

Our results obtained from mosaic analysis of hnt alleles clearly demonstrate that HNT function is necessary in all of the outer R cells in the developing eye disc, particularly in the R3, R4 and R6 precursor cells, for correct planar polarity. A role for the R3 and R4 precursor cells in establishing planar polarity in the eye has previously been proposed on the basis of analyses of the JNK pathway in this process (Weber et al., 2000). Those experiments could not rule out a role for other outer R precursor cells, which also express the JUN transcription factor (Bohmann et al., 1994; Weber et al., 2000). hindsight must now be added to the list of genes, such as strabismus (Wolff and Rubin, 1998), that regulate planar polarity and have a clearly described role in all of the outer R cells.

The fact that all of the symmetrical ommatidia along the borders of hnt clones are of the R3/R3 conformation suggests that HNT function is necessary for correct R4 fate and orientation. It has been suggested that, owing to its closer proximity to the polarizing signal from the equator, a stronger activation of the JNK pathway occurs in the R3 precursor cell (Weber et al., 2000). Activated JUN would then be responsible for the upregulation of the target gene, Delta, in the R3 precursor cell relative to the R4 precursor cell (Weber et al., 2000). As our results in the eye (this study) and results in the embryo (Reed et al., 2001) imply that HNT is necessary for downregulating JNK function, we propose that the wild-type function of HNT is to downregulate JNK activity in the R4 precursor cell. Such downregulation would enhance JNK signaling differences between the R3 and R4 cells. In the absence of the HNT gene product, JNK signaling would be inappropriately elevated in the R4 precursor cell, thereby upregulating the transcription of JNK targets such as Delta, leading that cell to behave more like a R3 precursor cell. Consistent with this model, we have found that Delta hypomorphs act as enhancers of the hnt^peb rough eye phenotype (A. T. P., R. Wilk and H. D. L., unpublished). R3/R3 symmetric clusters are observed both when the R4 cell is mutant for hnt and the R3 precursor is hnt⁺, and when the R3 cell is mutant for hnt and the R4 precursor is hnt⁺. In the latter case, the above model would lead one to expect normal R3/R4 clusters. As only R3/R3 clusters are observed, we speculate that HNT can affect the R4 precursor cell when expressed only in the neighboring R3 precursor cell (i.e. that there may be some communication feedback between these cells leading to local non-autonomy of the hnt phenotype).

The early morphological and fate determination defects and the later planar polarity defects seen in hnt eyes may be causally connected. We have shown that specification of the early outer R cell precursors (R2, R5, R3 and R4) is often disrupted in hnt tissue. It has been established that early disruption of R3 and R4 fate can perturb planar polarity (e.g. in the seven-up mutant) (Fanto et al., 1998). This result suggests that accurate interpretation of extrinsic polarity signals may require each R cell to already be properly determined. We have also shown that the relative positioning of outer R cells within a precluster is frequently deviant in hnt mutants before precluster rotation. Therefore the relative distance to the equator, or to a neighboring cluster, may also be crucial for differential reading of the polarity signal by pairs of outer cells within the cluster (R3 versus R4, R2 versus R5).

Our previous analyses have shown that that HNT function in the amnioserosa is required for the assembly and/or maintenance of focal complexes in adjacent epithelial cells along the leading edge of the dorsal ectoderm (Reed et al., 2001). We have demonstrated that these F-actin rich structures, which also accumulate high levels of α-PY reactive proteins, are required for morphogenetic events during normal dorsal closure. In wild-type eye discs, F-actin is enriched at the apical tips of presumptive R cells in ommatidial preclusters. This apical F-actin forms part of a tightly localized signaling complex enriched for receptor and ligand molecules such as Sevenless (Banerjee et al., 1987; Tomlinson et al., 1987), Boss (Van Vactor et al., 1991), Notch (Fehon et al., 1991) and Delta (Kooh et al., 1993). Here, we have shown that, in late larval eye discs, hnt mutant R cells have reduced F-actin at their apical tips. Furthermore, although the R8 precursor cell is correctly determined in hnt mutant patches, the Boss ligand is less concentrated at the apical tips of some of the hnt R8 cells. The F-actin phenotype seen in hnt patches in larval eye discs occurs late and is therefore probably not a direct cause of disruption of the apical signaling complex in the eye disc. However, the defects in F-actin accumulation may be a marker indicative of perturbed cytoarchitecture in the apical region of the mutant R cell. Such a perturbation would have serious consequences for correct R cell specification, which requires intimate contacts between adjacent R cells for proper intercellular signaling.

We have shown that, in the pupal retina, the concentration of F-actin in the apical tips of the R cell clusters is depleted in hnt mutant eye tissue. Still later, less F-actin accumulates in the extended rhabdomeres. F-actin and associated proteins in the apical surfaces of the photoreceptors play a key role in the initiation of rhabdomere morphogenesis (Cagan and Ready, 1989a; Longley and Ready, 1995; Sang and Ready, 2002). In addition, F actin is a component of the rhabdomere terminal web, a structure that anchors the rhabdomere membranes along the length of the differentiating R cell and prevents them from collapsing into the R-cell cytoplasm (Chang and Ready, 2000). When the function of Drac1, which regulates F-actin arrangement, is disrupted, the rhabdomere membranes ‘spool out’ into the cytoplasm (Chang and Ready, 2000). Interestingly, in addition to its role in rhabdomere morphogenesis, Drac1 is thought to signal through the JNK cascade and, like HNT, is necessary earlier on for the establishment of planar polarity in the eye (Fanto et al., 2000) and for dorsal closure of the embryo (Harden et al., 1995; Harden et al., 1999; Reed et al., 2001). The pupal photoreceptor phenotype that we observe in hnt R cells, where rhabdomeres do not extend fully in the apicobasal axis, is qualitatively different from that seen for a dominant negative Drac1 allele (Chang and Ready, 2000).

We speculate that the depletion of F-actin in hnt mutant R cells may affect morphogenetic events that precede rhabdomere terminal web maturation. For example, there may be defects in extension of the specialized membrane down the length of the R cell or in closure of cone cells over the highly constricted R cell apices (Cagan and Ready, 1989a; Longley and Ready, 1995; Sang and Ready, 2002). The morphogenetic parallels between closure of the leading edge epidermal cells over the constricted amnioserosa of the embryo and closure of the cone cells over the constricted R cell apices in the pupal eye are striking. Furthermore, both require assembly and function of F-actin-rich complexes at the boundary between the two cell types involved.

We have shown that hnt mutant R cells are unable to maintain their integrity within the retinal epithelium during retinal differentiation and morphogenesis in pupal discs. By analogy, the hnt phenotype in the tracheal system is first seen at stage 14, when the overtly normal tracheal epithelium begins to disintegrate, forming sacs and vesicles from the collapsed dorsal trunk and branches (Wilk et al., 2000). In hnt hypomorphs, a specific proportion of the cells in the amnioserosa fall out of the epithelium during dorsal closure (B. A. Reed and H. D. L., unpublished). Thus, HNT function is required for maintenance of the integrity of the epithelia in which it is expressed.

In the eye, loss of integrity occurs in clusters containing fewer than five out of eight HNT-expressing R cells. We speculate that delamination of clusters may occur because they lack a threshold level of apical F-actin required for inter-photoreceptor communication and/or adhesion. For example, if hnt mutant R cells fail to form focal contacts with the overlying cone cells, a genetically mosaic R cell cluster could slip basally, eventually falling out of the epithelium before consolidation of the fenestrated membrane during the pupal period.

In the eye, patches of tissue mutant for certain of the integrins (myospheroid and inflated) have missing R cells or R cells with shortened rhabdomeres (Zusman et al., 1993; Zusman et al., 1990). Detailed studies have shown that integrins are expressed in the cone and pigment cells and that the mutant phenotypes may trace their origin to a structural defect in the cone cell plate at the retinal floor (Longley and Ready, 1995). HNT is not expressed in cone or pigment cells and there is no gross defect in the retinal floor beneath hnt mutant eye tissue. However, as we observe an F-actin defect and it is known that extracellular engagement of adhesion molecules physically links F-actin bundles with the cell surface to provide structural integrity, further studies to examine possible requirements of HNT for extracellular matrix production or function may be revealing.

Our results clearly implicate HNT in regulation of several types of cellular events that are common to the different contexts in which it functions. These include establishment or maintenance of the morphology of individual cells within an epithelium, as well as maintenance of the integrity of the epithelium per se (this study) (Lamka and Lipshitz, 1999; Reed et al., 2001; Wilk et al., 2000). There are also shared molecular correlates of these HNT functions. In particular, HNT is required for establishment of localized F-actin- and phosphotyrosine-rich complexes in the leading edge epidermal cells, as well as in the photoreceptor cells (this study) (Reed et al., 2001). We have also presented evidence that HNT functions to regulate two JNK signaling dependent processes, planar polarity in the eye and dorsal closure of the embryo (this study) (Reed et al., 2001), possibly by downregulating JNK signaling in time and space. Whether the functions of HNT in different tissues and at different times during development derive from control of the same molecular pathway will await genetic and molecular analysis of the genes regulated by HNT.

We thank an anonymous reviewer for insightful comments on the parallels between closure of embryonic epidermal leading edge cells over the amnioserosa and cone cell closure over the R cells during retinal morphogenesis; these have been integrated into the Discussion. We also thank Dr D. Ready for helpful advice and discussions upon our initiation of this research project; Drs S. L. Zipursky, K. Moses and R. Barrio for supplying antibodies; Dr M. Mlodzik and the Bloomington Drosophila Stock Centre for providing fly stocks. We are grateful to Drs J. Brill, T. Erclik, R. McInnes, B. Reed, K. Yee and R. Wilk for helpful comments on the manuscript. A. T. P. was supported in part by a postdoctoral fellowship from the Research Training Center of the Hospital for Sick Children, M. L. L. by a postdoctoral National Research Service Award from the USPHS and M. L. R. Y. in part by a predoctoral fellowship from the Howard Hughes Medical Institute. This work was funded by a an operating grant to H. D. L. from the National Cancer Institute of Canada with funds from the Canadian Cancer Society.