The beginning of pattern formation in the Drosophila compound eye: the morphogenetic furrow and the second mitotic wave

The developing Drosophila compound eye presents a striking instance of cellular pattern formation in which an unpattemed epithelium is organized into the precise cellular lattice of the adult eye. This transformation is mediated by a sequence of developmental events, which includes cell division, differentiation and death. Since these operations are accessible to complementary cellular, genetic and molecular analysis, fly eye devel- opment holds the promise of a description of pattern formation which encompasses both supracellular and subcellular mechanisms (see Tomlinson, 1988; Ready, 1989; Rubin, 1989 and Banerjee and Zipursky, 1990 for reviews). The initiation of pattern formation in the eye and its relation to a tightly controlled mitotic wave that follows closely upon it are the focus of this paper.

Pattern formation in the Drosophila eye begins in the morphogenetic furrow, a pronounced indentation in the monolayer epithelium from which the retina is derived. Anterior to the furrow cells are unpattemed; behind it, they are organized into stereotyped clusters which are the rudiments of the ommatidia, or unit eyes of the adult compound eye. In the furrow, new ommatidia are formed as cells aggregate in periodic groupings. The eye grows by accretion as new rows of ommatidia are added anterior to older ones; consequently, the furrow sweeps across the eye disc in a posterior to anterior direction. High resolution observations of the furrow described here suggest each ommatidial rudiment is organized from a progression of stereotypic early stages. Ommatidial rudiments are initiated sequentially along a growing row, consistent with roles for local interactions and temporal order in determining the spacing of ommatidia.

Pattern formation continues as new cells, bom in a mitotic wave trailing the furrow, are sequentially incorporated by maturing ommatidial rudiments. The close correlation between the initiation of pattern formation and the occurrence of the mitotic wave raises the possibility that the two events are linked. Cell-by- cell analysis of the mitotic wave together with a refined picture of cellular organization in the furrow supports this possibility.

Stocks were raised at 20°C on a standard cornmeal agar food and maintained on a 24h light cycle. White prepupae were collected and maintained for the reported interval. Time zero is the white prepupal stage. Wild type stock used was Canton Special.

Larvae were fed on a BrdU diet according to Truman and Bate, 1988. Briefly, animals were raised for at least 12 h on a diet containing BrdU (Aldrich Chem. Co.) at a concentration of 0.5 mg BrdUml⁻¹ food. In the pulse label experiment, third instar larvae were injected with 0.066 μI of a 0.1 mg ml⁻¹dilution of BrdU in Ringer’s solution and aged to approxi- mately 60 h. Animals were dissected in Ringer’s and eyes fixed in 5% formaldehyde for at least 45min. Subsequent treatment was according to Danova et al. (1988), with some modifications. Immediately after fixation, tissue was treated in 3 M HC1 for 30 min and then transferred to 0.05% pepsin (Sigma) in 2 % PBS, pH2 for 15 min. Tissue was washed in PBS with 0.3% Triton X-100 for 30 min prior to overnight incubation in Becton-Dickinson anti-BrdU antibody at 4°C. The Vector ‘Elite’ ABC kit was used as the secondary antibody. Tissue was mounted in glycerol.

Tissue was fixed in 4% formaldehyde for 15min, rinsed in 3 changes of PBS-T, and incubated in monoclonal antibody BP104 (kindly supplied by Allan Bieber), at a dilution of 1:1000 in PBS-T and 5% horse serum overnight. BP104 was visualized using either RITC for confocal microscopy or the Vector ‘Elite’ ABC kit for standard bright-field microscopy.

Locke and Huie’s (1981) lead sulfide stain was modified to visualize the apical surface of eye discs. Eyes were dissected in Ringer’s and transferred to primary fixative which had been aged at least two weeks, for 1-2 h at 4 °C. (Age of the primary fix is critical for high quality staining of eye discs but not for pupal eyes. The fix works best between 2 and 5 weeks and should be aged and stored at 4°C.) Tissue was taken through a series of washes beginning with 0.2 M sodium cacodylate buffer, pH 7.4 at room temperature, 0.1M sodium cacodylate buffer, pH5.6 and stock aspartate, pH5.4. After the final rinse, eyes were transferred to lead staining solution for 5 min and then replaced with fresh lead staining solution and incubated for an additional hour at room temperature. Tissue was washed in stock aspartate buffer and taken through two final washes of 0.1M cacodylate, pH5.4 before visualizing lead by counterstaining in 8 % (NH₄)₂S for 5–10 s. Eyes were mounted in glycerol.

The changing cell patterns of the Drosophila eye can be visualized at the apical surface of the eye disc using stains that highlight cell outlines. Cobalt-sulfide-stained third instar eye discs show lines of cells buckling at intervals to form preclusters, ommatidial rudiments containing the postmitotic photoreceptors, R8, R2, R5, R3 and R4 (Tomlinson and Ready, 1987). The early events leading to formation of the preclusters have been difficult to resolve in these preparations since the sooty quality of the cobalt sulfide stain obscures the small apical profiles of the furrow.

Lead sulfide stain provides an unusually fine resol- ution of apical cell outlines (Fig. 1). Rows of preclusters parallel to the furrow are the future vertical columns of ommatidia in the adult eye. The beginnings of ommatidial rotation can be seen several rows behind the furrow: ommatidia at the top of the figure have begun to rotate counterclockwise, while at the bottom of the figure they have rotated clockwise. The line along which these antisymmetrical rotations meet defines the equator which is oriented perpendicular to the furrow, dividing the eye disc into dorsal and ventral halves. (Equator is shown schematically in Fig. 8).

As unpattemed cells approach the furrow, they shorten and their apical surfaces constrict 15-fold from approximately 9 μm² ahead of the furrow to about 0.6 μm² in the furrow. Pattern formation begins in the furrow, where cells have reached their maximum state of constriction.

Rows of ommatidia emerge from the furrow along a maturational gradient; the most mature member of a row fies at or near the equator and progressively younger forms are encountered with each lateral step. Fig. 2 shows a video microscope tracing of a lead sulfide stained eye disc in which rows have been distinguished by color. The leading (purple) row is in the process of emerging from the furrow. At its growing, polar end lies the youngest ommatidial rudiment of this row, an open arc of cells. Its more mature equatorial neighbor is a closing arc. In the most mature member shown, the arc has closed.

Since a growing ommatidial row is not completed before the next row is initiated, the furrow contains the growing ends of several ommatidial rows. In Fig. 2, the growing end of the anteriormost (purple) row is nested under the next posterior (blue) row. Further laterally (not shown) the growing end of this (blue) row is nested under the following (green) row (see also Fig. 8). Mitotic labeling experiments (see below) suggest that between 3 and 5 rows emerge simultaneously from the furrow.

A progression of changing cell associations leading up to the five cell precluster can be reconstructed from a survey of early periodic forms seen in many furrows. The maturational gradient along a furrow, which delays the emergence of adjacent clusters within a row by about half an hour (see Fig. 6 legend), together with the approximately two hour difference between rows, is particularly helpful in ordering transitional forms. The following is a suggested progression through these intermediate forms, reconstructed using high magnifi- cation video microscopy of lead sulfide stained furrows. Fig. 3 is a schematic summary while Figs 4A, B and C show videomicrographs of prototypical forms of early pattern formation.

The first recognizable, periodic form is a rosette of cells consisting of a 4-5 cell core surrounded by a ring of 10-15 cells (Figs 3A, 4A). Rosettes transform into arcs as 7–9 cells on the posterior limb of the ring enlarge and become more closely associated with their side-to-side neighbors to form a rearward-pointing curve (Figs 3B, 4B). Cells on the anterior limb of the rosette merge back into the unpattemed population of cells in the furrow. Arcs then zipper shut from apex to base, beginning with the closure of R2 and R5 behind R8 (Figs 3C, 3D, 4C arrowhead). As arcs close, their cells take on elongated profiles. Typically, 7 cells zipper shut: the precluster five, plus Ml and M2 (Tomlinson et al. 1987). Core cells exit or are expelled anteriorly (Figs 3D, 3E, 3F, 4C arrows), and come to lie on the posterior face of the arc forming directly anteriorly. Additional cells at each end of the arc are displaced to either side of the core. It is not known if the core and ring of the rosette organize simultaneously from an initially equivalent group, or if the core is specified first with the ring recruited subsequently.

A failure of retinal cells to assemble properly results in a roughening of the normally smooth array of ommati- dia. Rough eyes can result from a number of defects occurring any time between the earliest stages of pattern formation in the furrow to 55 h of pupal life, when pattern formation is complete. In scabrous eyes, early defects are apparent in the furrow (Baker et al. 1990; Mlodzik et al. 19906), while in roughest eyes, a failure to eliminate cells by cell death near the end of pattern formation results in a scrambled lattice (Wolff and Ready, 1991).

An examination of sea eye discs using lead sulfide reveals pattern anomalies prior to formation of the definitive precluster. Monster cores containing 12 or more cells occur frequently in sea furrows (Fig. 4D, arrow). Along the posterior of these hypertrophied cores are elongated arcs. The closure of these long arcs results in monster preclusters. Long lines of cells are not uncommon in sea discs (Fig. 4E, arrowhead), and multiple closures within these lines may give rise to abnormally closely spaced preclusters (Fig. 4E, ar- rows). Occasional fused ommatidia can arise when adjacent preclusters do not separate (not shown).

The irregular spacing of preclusters in the first rows behind sea furrows becomes more uniform toward the back of the disc. In wild-type eyes, as photoreceptors mature, they enlarge and their nuclei migrate to stereotyped positions (Tomlinson, 1985) leading to the tight packing of clusters in square arrays. Since the majority of sea ommatidia contain normal photorecep- tor numbers, a similar type of packing may lead to this increased order. Subsequent development of sea ommatidia proceeds essentially normally, with the sequential incorporation of cone and pigment cells (Fig-5).

In the morphogenetic furrow of roughest, a rough eye mutant in which cell death does not occur normally (Wolff and Ready, 1991), spacing of arcs along the posterior of the furrow appears normal (Fig. 4F).

Previous mitotic labeling studies using [³H]thymidine autoradiography have demonstrated that two waves of cell division provide a continuous supply of new cells in the developing third instar eye disc (Ready et al. 1976; Campos-Ortega, 1980). The first wave occurs as a broad band ahead of the furrow, in the region of undifferen- tiated and unpattemed epithelium. The second wave follows the furrow and is separated from the first wave by about 8-9 ommatidial rows. The five cells of the precluster, R8, R2, R5, R3 and R4, are derived from the pool of cells generated in the first wave whereas all remaining cell types can be labeled in the second wave. The future photoreceptors R1 and R6 are the first to be recruited into a growing cluster from this pool, followed by R7 and then by the four cone cells (Tomlinson and Ready, 1988).

Technical limitations of [³H]thymidine autoradiogra- phy make it difficult to reconstruct large and small scale patterns of cell division in the second mitotic wave. The exact identity of all dividing cells in the wave can be mapped using BrdU as a mitotic label and scoring labeled cells in pupal eye whole mounts. Pupal eye whole mounts of animals fed or injected with the mitotic label BrdU show that all non-precluster cells divide and form a band between two adjacent rows of forming preclusters.

All retinal cells, except for the precluster five, which are bom ahead of the furrow, are bom in the second mitotic wave. In midpupal eyes of animals fed a BrdU- containing diet for 12–24 h as third instar larvae, preclusters appear as a negative image against a lawn of stained nuclei (not shown). On rare occasions, and always in association with pattern flaws, unlabeled gaps are present in the meshwork of stained nuclei. Cell outlines are typically not apparent at these sites, suggesting cells are absent rather than unlabeled. Among hundreds of ommatidia examined, only a single instance of non-labeling was found in an apparently normal tertiary cell.

Once a precluster has formed, between 16 and 17 cells are needed to complete an ommatidium; this number includes the remaining photoreceptors, RI, R7 and R6, the primary, secondary and tertiary pigment cells, the cone cells, the bristle mother cells and the 2–3 cells that will die at the close of pattern formation (Cagan .and Ready, 1989; Wolff and Ready, 1991). At the back of lead-sulfide-stained third instar eye discs, well after divisions are complete, 13–15 cells per photoreceptor cluster are present on the apical surface. In addition to this number, three more cells, R1, R7 and R6, belong to each cluster but become occluded from the apical surface when the cone cells close over the photoreceptors. In lead-sulfide-stained eye discs, before the second mitotic wave divisions occur, approximately 8 to 9 cells surround each precluster. The division of these cells in the second mitotic wave would suffice to generate the required 16–17 cells.

Cells spend a relatively brief time rounded up at the apical surface during mitosis. Of the approximately 250 mitoses occurring every two hours (8–9 divisions for each of 32 ommatidia per row) an average of 15 cells were seen at the apical surface between rows 2 and 8 in five cobalt-sulfide-stained eye discs (row 1 is defined by prominent arcs).

Cell divisions in the second mitotic wave parallel the emergence of clusters from the furrow: they begin near the center of a row and spread laterally. This pattern is seen in midpupal eyes of animals injected with BrdU during the third instar, in which segments of three to five rows are labeled in a step-like pattern (Figs 6, 7). The most anterior rows are labeled centrally, while label shifts laterally in more posterior rows. With each step back, fewer ommatidia within a row are labeled, suggesting ommatidial initiation may slow as it ap- proaches the periphery.

BrdU-labeled cells are distributed approximately symmetrically across the equator. In 13 out of 17 pupal eyes scored, the number of clusters labeled on either side of equator in the anterior, leading row differed by no more than two. In the remaining eyes, between 3 and 8 more labeled clusters were found on one side of the equator.

The relationship between the center-out maturation- al gradient of the furrow and cell division in the mitotic wave is schematized in Fig. 8. Panel A shows a snapshot of a third instar eye disc in which 6 rows are emerging from the furrow. Trailing the newly forming clusters, a band of cells divides in the second mitotic wave (asterisks). Subsequent development, including the completion of the ommatidial rows and the conversion of the disc’s square array into the hexagonal lattice of the mature eye, carries the band of dividing cells into the bow shown in panel B.

In the BrdU injections used in these experiments, label may be available for approximately 1h. Since a new row is added every two hours, the labeling of only a single row (except where growing ends overlap) indicates BrdU is available for less than two hours. In a late third instar eye disc, the division of 8–9 cells for each of the 30–32 preclusters in a row would be expected to generate between 480 and 576 cells every 2h. However, in three injected animals about half this number (245, 255 and 294) of cells were labeled (Fig. 7), suggesting BrdU is available for half this time.

Cells bom in the second mitotic wave contribute to the posterior face of one row and the anterior face of its posterior neighbor, so that cells simultaneously in S phase lie between forming rows (Fig. 7). This is most notable in the labeling of the primary pigment cells: labeled posterior primaries of one row face labeled anterior primaries in the following row. This ‘in- between’ pattern is also evident in the labeling of photoreceptors RI, R7 and R6 in the anteriormost row. In these ommatidia, the posterior primary pigment cell is usually labeled, but the anterior primary is virtually never labeled. The posterior primary pigment cells and photoreceptors RI, R7 and R6 are recruited from cells lying to the posterior of the precluster, but the photoreceptors are carried equatorially by the omma- tidial rotation that occurs in third instar eye discs (Tomlinson and Ready, 1987).

BP104, an antibody raised against the neuron-specific form of neuroglian (Hortsch et al. 1990), is detectable beginning approximately 6 rows behind the morpho- genetic furrow. The leading row of BP104-expressing clusters shows heavy staining centrally and trails off on the lateral ends (Fig. 9), reflecting the center out progression of ommatidial initiation. Unlike neural antigen 22C10, which is expressed by R8 before R2 and R5, BP104 expression begins at the same or nearly the same time in photoreceptors R8, R2 and R5.

In addition to the photoreceptors, BP104 is also expressed by the cone cells at the back of the third instar eye disc (Fig. 10). Hortsch et al. (1990) also report BP104 expression in non-neural support cells of embryonic peripheral sense organs. In a gain of function sevenless mutant, the expression of BP104 by the cone cells has been taken as evidence that these cells have adopted a neural identity (Basler et al. 1991).

Events in the morphogenetic furrow lay the foundation for all subsequent eye development. Of these changes, the most notable is the transition from an unpattemed epithelium into an array of periodic clusters. This occurs in the furrow where cells reach their greatest contraction in the apical-basal dimension, as well as the maximal constriction of their apical surfaces. No causal link has been established between these cellular events and the onset of pattern formation in the furrow, but the coincidence is provocative. If apical membrane receptors are potentiated when concentrated in massed arrays, the approximately fifteen-fold reduction in apical surface area could facilitate cell communication by ‘corralling’ these receptors into high density. A heightened level of communication might be critical for parceling the epithelium into periodic multicellular domains.

The spread of ommatidial organization across the eye disc bears a striking resemblance to the growth of periodic patterns in other epithelia, notably the well- studied feather placode pattern in the chick (Dhouailly, 1984). Feather rudiments emerge as local cellular condensations as a wave sweeps from back to front and from a symmetry midline out. In both systems, the wave of periodic rudiment initiation does not appear to be propagated by inductive contact; rudiments appear on schedule in unpatterned epithelium even when it is dissected away from the oncoming pattern wave (Sengel, 1975; Lebovitz and Ready, 1986). Nothing is known of the apparently intrinsic property that organ- izes the temporal progress of the wave.

The observation that ommatidia emerge sequentially as roughly equidistant rosettes of cells is compatible with a number of models, but several mutant pheno- types suggest a form of lateral inhibition may contribute to ommatidial spacing. Mutations in Notch, which codes for a transmembrane protein implicated in numerous cell-cell interactions, notably lateral inhi- bition (Campos-Ortega, 1988), DER, the Drosophila homolog of the epidermal growth factor receptor (Baker and Rubin, 1989), and scabrous, a secreted fibrinogen-related protein (Baker et al. 1990), all disrupt normal ommatidial spacing. The observation that normal spacing is determined well before the differentiation of R8 and that monster cores are found in sea eye discs, suggests sea may be important for periodicity from the earliest stages. This would be consistent with the observed expression of sea in clusters of cells ahead of the preclusters (Baker et al. 1990).

Ommatidial development shows interesting parallels to the generation of sensory organs in the developing Drosophila peripheral nervous system (Ghysen and Dambly-Chaudiere, 1989). In the PNS, a sensory mother cell is designated within a ‘proneural cluster’; in the eye, the precluster is selected from the arc. In normal eyes, approximately 4 cells at the ends of each arc are ultimately not incorporated into the precluster. In seven-up mutant ommatidia, two ‘surplus’ cells of the line, Ml and M2 have been suggested to contribute extra cells which differentiate as photoreceptors (Mlod- zik et al. 1990a).

Cell division in the postfurrow mitotic wave could be either a triggered event, for example induced by contact with a forming precluster, or an intrinsic behavior of epithelial cells not selected to become members of the precluster. A cell-by-cell analysis of divisions in the second mitotic wave favors an inductive control. A consistent observation in eyes of animals injected with BrdU is that cells between adjacent rows are mitotically active at about the same time. One scenario that could account for such a pattern is that divisions are triggered by contact with differentiating cells. Cells contacting the early-differentiating R8, R2 and R5 (purple cells, Fig. 3) will be triggered to divide before those contacting the later-differentiating R3 and R4 (yellow cells, Fig. 3). Where two ommatidial rows meet, the ‘delayed’, R3/R4 faces of preclusters in the posterior row alternate with the ‘advanced’ R8/R2/R5 faces of ommatidia in the next row. If the developmental delay between the formation of one row and the next approximates the time lag between advanced and delayed sides of the precluster, a signal triggering divisions would be presented from alternating sides to a band of cells lying between ommatidial rows. Cells bom in the ensuing divisions would contribute to their respective ommatidia on complementary faces (Fig. 11).

This scenario does not adequately account for the division of cells not contacting a precluster. It is difficult to determine the precise number of such cells in the dynamic cellular environment of the furrow, but in lead-sulfide-stained discs, for every precluster, there are about one or two apical profiles that do not contact a developing photoreceptor. It is possible that these cells have made transient, inductive contacts with preclusters, or have made them below the apical surface, but these alternatives have not been explored.

A new retinal floor, the fenestrated membrane, forms behind the mitotic wave (Cagan and Ready, 1989). Except for the photoreceptors, all retinal cells stand on this new basement membrane which lies above and parallel to the original basement membrane. As cells divide in the wave, they release their anchorage to the original basement membrane, divide at the apical surface and re-extend their basal feet to the new plane. Cell division seems an economical way to step up to a new plane; the formation of the fenestrated membrane could be complicated if some non-precluster cells did not release their original anchorage and divide. Precluster photoreceptors maintain their footing on the original basement membrane and extend growth cones posteriorly on this surface.

We thank Dr Nancy Bonini for helpful comments on the manuscript, and Dr Allan Bieber for BP104 antibody. This work was supported by USPHS grant no. ROI AG09302.