Novel spatiotemporal patterns of epithelial tumor invasion in Drosophila discs large egg chambers

Authors

  • Scott Goode,

    Corresponding author
    1. Department of Pathology, Baylor College of Medicine, Houston, Texas
    2. Department of Molecular and Cellular Biology, Department of Molecular and Human Genetics, Program in Developmental Biology, and Program in Cellular and Molecular Biology, Baylor College of Medicine, Houston, Texas
    • Department of Pathology, Baylor College of Medicine, One Baylor Plaza, Houston, TX 77030
    Search for more papers by this author
  • Jun Wei,

    1. Department of Pathology, Baylor College of Medicine, Houston, Texas
    Search for more papers by this author
  • Sirish Kishore

    1. Department of Pathology, Baylor College of Medicine, Houston, Texas
    Search for more papers by this author

Abstract

Loss of Discslarge (Dlg) in early Drosophila egg chambers causes invasion of tumor follicle cells from the anterior epithelium, a pattern that resembles developmental border cell migration during mid-oogenesis. Here, we have analyzed novel spatial and temporal patterns of dlg invasion. Even though Dlg is ubiquitously expressed in all follicle cells, invasions are biased at the anterior and posterior termini. The patterns of invasion correlate with both a higher rate of follicle cell proliferation and with a greater frequency of loss of epithelial polarity at the termini compared with central regions of the egg chamber. Nonetheless, the average number of cells that invade per invasion event from terminal vs. central regions is approximately equal. Of interest, patterns of dlg invasion appear to coincide with boundaries established by proto-oncogene signals responsible for anterior–posterior patterning. The Drosophila egg chamber may thus be a useful model for exploring how epithelial tumor invasion might be a neomorphogenetic process organized by signals essential for developmental pattern formation. Developmental Dynamics 232:855–864, 2005. © 2005 Wiley-Liss, Inc.

INTRODUCTION

Over 90% of cancers are carcinomas (Alberts et al., 1994). Carcinomas evolve in renewal epithelial tissues, such as skin, colon, breast, and ovary. The unique organ-specific shapes and functions of these epithelial tissues are determined by processes that direct developmental patterning, processes that are also essential for maintaining epithelial organization throughout life, yet little is known about how developmental patterning influences carcinogenesis. The possibility that carcinogenesis is a neomorphogenetic process governed in part by rules that organize signals for developmental pattern formation has received relatively little attention, perhaps because many tumor suppressor genes and oncogenes encode ubiquitously expressed proteins (Teh et al., 1999). However, significant connections between developmental pattern and carcinogenesis seem likely, given that cancer is often triggered by the reactivation of signaling pathways down-regulated after use in embryonic development (Heiser and Hebrok, 2004). For example, epithelial growth factor (EGF) receptor, Wnt/APC/catenin, transforming growth factor-beta (TGFβ)/Smad, and Hedgehog/Patched/Smoothened pathways play crucial roles in both development and carcinogenesis. These signaling pathways regulate more than proliferation, they instruct cells of their relative position, which directs their differentiation and morphogenetic behaviors as a function of spatial location within an epithelium. This information may also direct neomorphogenetic patterns that govern tumor progression, invasion, and metastasis. For example, the tissue distribution and intensity of nuclear-localized β-catenin predicts epitheliomesenchymal and tubular branching patterns of cellular ingression that occur during sea urchin gastrulation. In human adenocarcinomas, the intratumor distribution and intensity of nuclear β-catenin also correlates with epitheliomesenchymal and tubular branching patterns of tumor invasion and, thus, resembles the patterns seen during normal sea urchin development (Kirchner and Brabletz, 2000). Because translocation of β-catenin to the nucleus is crucial for transcriptional activation, these studies raise the possibility that invasive carcinoma cells at least partially return to transcriptional profiles that direct embryonic migrations (Kirchner and Brabletz, 2000).

We are developing the Drosophila egg chamber as a simple in vivo model in which to directly compare developmental cell migrations to tumor cell invasion within the same tissue. Loss of Discslarge (Dlg) during oogenesis causes follicle cells to acquire key hallmarks of invasive cancer cells, including overproliferation, change of shape, increased motility, and cellular invasion (Goode and Perrimon, 1997; Szafranski and Goode, 2004). Indeed, Dlg function may be conserved because two human homologs, hDlg and ZO-1, are also strongly implicated in oncogenesis (Matsumine et al., 1996; Hoover et al., 1998; Gardiol et al., 2002). Furthermore, several Dlg homologs are expressed in the mouse ovary, and hDlg4 is lost in a murine model of invasive ovarian epithelial cancer (Huang et al., 2003).

Previous work suggested that, during oogenesis, dlg tumor cell invasion resembles developmental migration of a small cluster of six to seven follicle cells called border cells. This finding allows direct comparison of normal migration to tumor invasion in the same tissue. Border cells exit the postmitotic follicular epithelium during mid-oogenesis and migrate through the center of the egg chamber to the front of the oocyte (Figs. 1, 3A–D). dlg cell invasion resembles border cell migration in that the invasive cells undergo an epithelial-mesenchyme–like transition, delaminate from the anterior epithelium, invade through the center of the egg chamber, and migrate to the oocyte, but dlg cells initiate this process much earlier than border cells and typically invade as streams of cells (Goode and Perrimon, 1997). Furthermore, dlg tumor cells do not express border cell markers, indicating they have not adopted a border cell fate (Goode and Perrimon, 1997). Two additional cell populations, the stretch cells and centripetal cells, also complete migration during mid-oogenesis (Fig. 1), but the potential relationship of dlg invasions to these cell migrations has not been explored. Furthermore, whether dlg invasion is influenced by signals governing spatial patterning within the early follicular epithelium remains to be investigated.

Figure 1.

Morphogenesis and cell migrations in the Drosophila follicular epithelium. The model is a synthesis of work cited below and in the text. Follicle cells pass through three sequential behaviors during egg chamber maturation. From stages 1–6, follicle cells proliferate to surround the growing (but not proliferating) germ cells; during stages 7–8, they differentiate; at stage 9, follicle cells migrate to the oocyte (Goode et al., 1992). The precise timing of the signaling events are not known but are thought to occur in the suggested order. Anterior–posterior patterning is determined by Jak-Stat signals activated at the termini of the epithelium (1; Xi et al., 2003). Mirror is expressed between the terminal Jak-Stat domains in the middle of the epithelium (2; Jordan et al., 2000). Anterior terminal follicle cells are distinguished from posterior follicle cells when transforming growth factor-alpha (TGFα) signals emanate from the oocyte and activate EgfR in posterior follicle cells (3; Gonzalez-Reyes et al., 1995; Roth et al., 1995; Gonzalez-Reyes and St. Johnston, 1998). At stages 6–7, Jak-Stat signals are expressed at higher levels in anterior cells in a graded pattern, leading to the differentiation of three migratory follicle cells types (4; Gonzalez-Reyes and St. Johnston, 1998; Xi et al., 2003). Border cells undergo an epithelial-mesenchyme–like transition, then migrate through the germ cells to the oocyte, then migrate dorsally in response to TGFα signals emanating from the oocyte nucleus (Duchek and Rørth, 2001). Stretch cells become squamous as they migrate over the nurse cells. Centripetal cells elongate and migrate between the nurse cells and oocyte after the epithelium completes its migration to the oocyte.

Of interest, some of the same signaling pathways that are used to establish spatial pattern in the early follicular epithelium also specify and direct cell migrations during mid-oogenesis. Early in oogenesis, complimentary patterns of Jak-Stat signaling and Mirror expression partition the follicular epithelium into terminal and main-body follicle cells (Gonzalez-Reyes and St. Johnston, 1998; Jordan et al., 2000; Xi et al., 2003). By default, terminal cells have an anterior fate until mid-oogenesis, when follicle cells in the posterior terminus of the egg chamber switch to posterior fate upon activation of Drosophila Egf-receptor (EgfR) signals by gurken TGFα emanating from the oocyte (Gonzalez-Reyes et al., 1995; Roth et al., 1995; Gonzalez-Reyes and St. Johnston, 1998). Subsequently, increased and graded Jak-Stat signals in anterior follicle cells causes them to be determined as three migratory cell types that later undergo unique shape changes as they migrate (Fig. 1). The cells are determined in a radial pattern, starting with (1) border cells at the pole, which undergo an epithelial-mesenchymal–like transition and migrate between the germ cells to the oocyte (Silver and Montell, 2001; Beccari et al., 2002); (2) stretch cells, which are adjacent to the border cells and become squamous and migrate on the outside of the egg chamber to encapsulate the nurse cells; and (3) centripetal cells, which are adjacent to the stretch cells and become elongated and migrate between the oocyte and nurse cells to encapsulate the oocyte (Fig. 1; Gonzalez-Reyes and St. Johnston, 1998). After their determination by Jak-Stat signals, the border cells are guided in their migration by a combination of platelet-derived growth factor (PDGF) and EgfR signals (Duchek and Rørth, 2001). Thus, both the Jak-Stat and TGFα-EgfR pathways are used early for early follicle cell patterning, then used subsequently for patterning border cell determination and migration, respectively.

Here, we complete a detailed analysis of patterns of dlg tumor invasion to uncover their potential relationships to early epithelial patterning and subsequent patterns of cell migration. Although Dlg is expressed in all follicle cells (Goode and Perrimon, 1997), we find that invasions are biased at the termini of the egg chamber. The patterns of invasion correlate with both a higher rate of follicle cell proliferation and with a greater frequency of loss of epithelial polarity at the termini. Nonetheless, the average number of cells that invade per invasion event from terminal vs. central regions is approximately equal. The patterns of dlg invasion that we describe appear to coincide with boundaries established by Jak-Stat and TGFα-EgfR proto-oncogene signals responsible for anterior–posterior patterning in the early follicular epithelium. The Drosophila egg chamber may thus be a useful model for exploring how development of invasive epithelial tumors is influenced by signals essential for developmental pattern formation.

RESULTS

Kinetics of Invasion in dlghf/dlglv55 Egg Chambers

To analyze spatiotemporal patterns of invasion in dlg egg chambers, we first sought a dlg phenotype in which we could reproducibly produce tumors containing a given number of cells. We chose to precisely manipulate the number of invasive cells by using the dlg temperature-sensitive combination dlghf/dlglv55 (Goode and Perrimon, 1997). This approach is an effective strategy because dlghf/dlglv55 invasion is indistinguishable from dlgnull invasion (Goode and Perrimon, 1997). However, dlgnull tumors are not useful for the analyses described here, because they must be induced in follicle cell clones, but this induction produces massive, uncontrollable overgrowth. In contrast, dlghf/dlglv55 allows precise manipulation of the number of tumor cells by varying the duration of temperature shift, thus enabling us to clearly see patterns of invasion and to count numbers of cells.

We characterized the number of invasive dlghf/dlglv55 cells as a function of duration of temperature shift in 6-hr intervals (6, 12, 18, and 24 hr). On average, egg chambers pass through one stage of oogenesis every 6 hr (Spradling, 1993), so this shift regimen allows us to determine how many cells invade within a single stage, or span of ovarian stages. We observe invading cells starting as early as 6 hr after temperature shift (Fig. 2). For each stage there is a progressive increase in number of cells invading each egg chamber with time. For all but stages 1 and 2, a disproportionately large increase in the number of cells invading occurs between 18 and 24 hr after temperature shift (Fig. 2). We chose to use a 24-hr temperature shift for analysis of spatiotemporal patterns of dlg invasion because this time point readily enables us to identify the initiating site of invasion whether the tumor just started invading (Fig. 3) or had been invading for some time (Fig. 5). Second, a 24-hr temperature shift produces an adequate number of invasive cells to score cell numbers (see below).

Figure 2.

Tumor cell number increases with increasing duration of loss of dlg function. dlghf/dlglv55 animals were shifted to the restrictive temperature for the indicated amounts of time. Ovaries were then immediately dissected and prepared for analysis. The number of tumor cells per egg chamber was scored for stages 1 to 8 of oogenesis. Averaged results from 20 egg chambers per stage are shown.

Figure 3.

Border cell migration and patterns of dlg tumor cell invasion. Egg chambers are stained with phalloidin. dlg egg chambers were shifted to the restrictive temperature (25°C) for 24 hr. A: Wild-type (wt) s7 egg chamber. The oocyte (OO), nurse cells (NCs), and follicle cells (FCs) are labeled. Nurse cells are numbered by their position relative to the oocyte, designating the addresses where dlg tumor cells invade germ cells (E–I). B–D: Wild-type s9–10b egg chambers. The border cells (BCs, arrowhead) move between the nurse cells to the front of the oocyte. D: At stage 10B, the border cells and the follicular epithelium have completed their migration to the oocyte. The front of the epithelium, consisting of centripetal cells (CC), migrates between the oocyte and nurse cells (arrowheads). E: s7 dlg egg chamber with anterior invasion (arrow). F: s7 dlg egg chamber with invasion between the second and third nurse cells (arrow). G: s7 dlg egg chamber with invasion between first and second nurse cells (arrow). H: s7 dlg egg chamber with invasion between oocyte and first nurse cells (arrow). I: s8 dlg egg chamber with a tumor behind oocyte (arrow).

Spatial and Temporal Patterns of dlg Invasion

dlg mutant egg chambers that had been shifted to the restrictive temperature for 24 hr were scored to characterize spatial and temporal patterns of cell invasion. Invasion can initiate at any site where intercellular space exists between two germ cells (Fig. 3E–I). Invasions initiate during all stages of oogenesis after a 6 hr temperature shift, the time it takes on average for an egg chamber to pass through one developmental stage. More than one invasion per egg chamber was typically observed from stage 4 onward (Fig. 4), indicating the total number of invasions per egg chamber increases during later stages. These observations demonstrate that initiation of new invasion occurs during all stages of egg chamber maturation.

Figure 4.

Number of invasive tumors organized by stage and site of invasion. Each graph represents a different stage(s). Invasion sites are shown in the egg chamber illustration (see also Fig. 3). The data are from dlghf/dlglv55 egg chambers that were shifted to the restrictive temperature for 24 hr. The number of invasions per site was divided by the number of egg chambers scored. In stages 2–5, most invasions occur from the anterior pole. A new pattern emerges by stage 6, in which invasions from both termini are approximately equal (A, O-NC, and O). The sum of bars present in each graph (Σ) denotes the average number of invasions per egg chamber.

During stages 2–5 of oogenesis, follicle cells appear to preferentially invade from the anterior epithelium, compared with central or posterior regions (Fig. 4). From stage 6 onward, invasive tumors are more frequent in both anterior and posterior follicle cells, compared with central regions (Fig. 4). Many posterior tumor follicle cells delaminate into the space between the oocyte and epithelium without overt invasion (Fig. 3I). The least common pattern of invasion is initiated from central, main-body follicle cells (Fig. 4).

Invading Cells Are Directed to the Oocyte

To determine where dlg cells migrate, we analyzed patterns of invasive tumor streams (Fig. 5). For streams of tumor cells with one end attached to the anterior epithelium, the streams always have a pattern consistent with taking the shortest path to the oocyte, down the center of the egg chamber (Fig. 5A). The streams of cells never bend between nurse cells 1 and 2 or 2 and 3. This pattern resembles border cell migration. Likewise, streams of tumor cells with one end attached to the epithelium between nurse cells 1 and 2 or 2 and 3 have a pattern consistent with them taking the shortest path to the oocyte, such that the stream bends at the center of the egg chamber toward the oocyte but does not bend toward the anterior or does not continue in a straight line across to the other side of the egg chamber (Fig. 5B,C). In addition, for streams with one end attached to the epithelium between nurse cell 1 and the oocyte, the stream always continues across the oocyte, and does not project anterior (Fig. 5D). This last pattern resembles normal centripetal cell migration (Figs. 1, 3D). Furthermore, for tumors attached to the posterior epithelium, the cells accumulate adjacent to the oocyte and do not extend into anterior regions of the egg chamber (Fig. 3I). In summary, invasions initiate from any region of the follicular epithelium abutting the intercellular space between nurse cells. All of the patterns of projection of the invasive streams are consistent with them taking the shortest available path to the oocyte but not to any other cells of the egg chamber (n > 10,000 egg chambers). We conclude that dlg cell invasion is directed to the oocyte. Thus, germ cell patterning before differentiation of migratory follicle cells (Fig. 1) can direct invasion in patterns resembling border and centripetal cell migration. Additionally, the invariant pattern of projection of tumor streams toward the oocyte demonstrates that dlg tumors truly migrate, rather than merely expanding into germ cells by means of proliferation. A mechanism involving proliferation alone would be expected to have a distinct pattern in which the tumor cells fill the space between nurse cells in all possible orientations relative to the oocyte.

Figure 5.

All tumor invasions migrate to the oocyte (OO). dlghf/dlglv55 egg chambers were shifted to the restrictive temperature (25°C) for 24 hr, stained with phalloidin, then visualized by laser scanning confocal microscopy. A–D: Tumors that have initiated invasion from four sites are shown: anterior (A), between the second and third nurse cells (B), between the first and second nurse cells (C), and between the first nurse cells and oocyte (D). The invariant patterns of tumor cell migration to the oocyte (see text), independent of the initiating site of invasion, indicate that all tumor cells migrate to the ooctye.

Sometimes tumor clusters delaminate from the epithelium much like border cells (Fig. 6). This type is the least common pattern of invasion. Starting as early as stages 4–5, isolated clusters are observed both in the middle of the egg chamber and contacting the oocyte (Fig. 6). These clusters represent less than 5% of total invasion events. These observations suggest that, as for dlg tumor streams, detached clusters migrate to the oocyte. Together with the switch of the cells comprising these tumor clusters from an epithelial to a mesenchymal-like morphology, these observations suggest a partial functional homology between invasive dlg tumor streams and border cell invasion.

Figure 6.

Location of tumor cell clusters that have delaminated. dlg egg chambers were shifted to the restrictive temperature (25°C) for 24 hr, stained with phalloidin, then visualized by laser scanning confocal microscopy. A: s7 dlg egg chamber with a tumor cluster in the center of the egg chamber (arrow). B: s7 dlg egg chamber with a tumor cell cluster deforming the oocyte (arrow). The graphs show the total number of tumor clusters in the center of the egg chamber or adjacent to the oocyte, divided by the number of egg chambers scored for that stage. The data are taken from the same experimental sample shown in Figure 4.

Average Number of Invasive Cells per Invasion Event Is Independent of the Initiating Site of Invasion

In wild-type egg chambers, bromodeoxyuridine (BrdU) -labeling experiments indicate that the rate of cell proliferation is the same in terminal vs. central regions of the follicular epithelium (Goode and Perrimon, 1997). In dlg egg chambers, BrdU-labeling experiments, as well as comparisons of total cell numbers, indicate that there are more cells at the termini compared with central regions of the follicular epithelium (Goode and Perrimon, 1997). This finding suggests the possibility that more invasions initiate from the termini of dlg mutant egg chambers, because there is a greater pool of cells at the termini to initiate invasion. To characterize the relationship between patterns of dlg invasion and proliferation in more detail, we compared the number of cells per invasion that initiated from terminal vs. central regions of the egg chamber (Fig. 7). We scored stage 7 and 8 egg chambers because a preferential pattern of terminal invasion is clearly distinguishable by this time (Fig. 4). More invasive cells on average were observed at the termini (Fig. 7, upper graphs), in a similar pattern to invasions per egg chamber (Fig. 4). To establish the average number of cells invading per invasion site, the total number of invasive cells at each site (Fig. 7) was divided by the number of times invasion occurred at that site (Fig. 4). Surprisingly, the average number of cells per invasion was approximately equal for all sites of invasion (Fig. 7).

Figure 7.

Number of cells per invasion. Invasion sites are indicated in the egg chamber illustration. Stage 7 and 8 graphs are shown. The upper graphs show the average number of cells per egg chamber that invade at each invasion site. The largest number of invasive cells was observed at the termini (A, O-NC, and O). The lower graphs show the average number of invasive cells at each site (upper graphs), divided by the number of invasions at that site (graphs on Fig. 4). The number of invasive cells at each invasion site is approximately equal. The data were derived by scoring the same sample of egg chambers twice, first for site of invasion and subsequently for numbers of cells at each site.

We assume that the number of follicle cells that initiate invasion from terminal and central regions is on average the same, because the average number of cells per tumor is approximately the same for all invasion sites. If the number of follicle cells that initiated invasion from the termini were consistently greater, then we would expect that these tumors would also consistently have more cells compared with tumors initiating from central regions. With this assumption, our data suggest that tumor growth occurs at similar rates, independent of the initiating site of invasion. One caveat of this interpretation is that tumor size may be limited by the duration of the temperature shift or other factors, and these limiting factors may bias the measurements.

In summary, these data correlate a greater number of cells at the termini of dlg egg chambers with a higher frequency of invasion and, thus, suggest the possibility that the greater pool of cells at the termini contribute to the increased frequency of invasion. However, without more direct evidence, we cannot definitively say whether the greater number of cells at the poles is a cause of the higher invasion frequency at the poles, an effect of increased invasion, or both.

dlg Tumor Cells at the Poles of the Egg Chamber Preferentially Lose Polarity

dlg cells undergo at least three morphogenetic changes to become invasive; they overproliferate, switch polarity, and migrate into the germline (Goode and Perrimon, 1997). Thus, in addition to an increased rate of proliferation at the termini, it is also possible that a differential change in follicle cell polarity at the termini or differential invasive capacity of terminal follicle cells contributes to spatially selective invasion. To address these possibilities, we analyzed noninvasive dlg mutant combinations in which follicle cells accumulate of at the termini of the egg chamber without invading (Fig. 8A1–C1). Noninvasive dlg mutant combinations disrupt distinct Dlg domains compared with invasive mutants; thus, invasion appears to be a genetically separable from other Dlg activities (Goode and Perrimon, 1997). Like the invasive dlg mutants, the noninvasive mutants cause follicle cells to lose epithelial polarity and delaminate from the native epithelium, except that they accumulate in multiple layers (Fig. 8A1–C1). There are more follicle cells at the termini of the noninvasive mutants, correlating with increased BrdU labeling at the termini in invasive dlg mutants (Goode and Perrimon, 1997), suggesting that the extra cells are derived from selective overproliferation. The noninvasive mutants, thus, are attractive for addressing the hypothesis that terminal cells lose epithelial polarity more frequently than main-body follicle cells, because the phenotype is clearly separable from invasion.

Figure 8.

Follicle cells at the termini of dlgm35/dlgip20 egg chambers lose polarity. A1,B1,C1:dlgm35/dlgip20 egg chambers are stained with Actin and: Par6 (A1), Crumbs (B1), and DE-Cadherin (C1). dlgm35/dlgip20 follicle cells develop in a multilayered adenoma-like pattern at the termini of the egg chamber (brackets), corresponding to the region where most dlg tumors initiate invasion (Figs. 4, 7). A2,B2,C2: Actin localization. Actin is enriched on the apical surface of most follicle cells in lateral regions of the egg chamber (arrows), as in wild-type follicle cells (not shown). Actin is distributed uniformly in terminal cells that have delaminated (brackets) and in the terminal regions of the native epithelium (A1–C1 dotted outlines), indicating that these cells have lost polarity. Less frequently follicle cells in central regions of the follicular epithelium lose actin polarity (asterisks). A3: Par6 is enriched on the apical surface of most follicle cells in the lateral regions of dlgm35/dlgip20 egg chambers (arrows), as it is normally in all follicle cells of wild-type egg chambers (not shown). Par6 is reduced and diffusely distributed in dlgm35/dlgip20 tumor cells that have delaminated (brackets), in the native epithelium in terminal regions (dotted outline A1), and less frequently in follicle cells in central regions of the epithelium (asterisks), indicating that these cells have lost polarity. B3: Crumbs is enriched on the apical surface of follicle cells in lateral regions of the egg chamber (arrows), as it is normally in all follicle cells of wild-type egg chambers (not shown). Crumbs is reduced and diffusely distributed in dlgm35/dlgip20 tumor cells that have delaminated (brackets) and in the native epithelium in terminal regions (dotted outlines B1) and, less frequently, in follicle cells in central regions of the epithelium (asterisks), indicating that these cells have lost polarity. C3: DE-Cadherin is enriched at the apical–lateral adherens junctions in lateral regions of the egg chamber (arrows), as it is normally in all follicle cells of wild-type egg chambers (not shown). DE-Cadherin is reduced and diffusely distributed in dlgm35/dlgip20 tumor cells that have delaminated (brackets) and in the native epithelium in terminal regions (dotted outlines C1), indicating that these cells have lost polarity.

We analyzed follicle cell polarity proteins Actin, Par6, Crumbs, and DE-Cadherin (Tanentzapf et al., 2000) in dlgm35/dlgip20 mutants (Fig. 8). In wild-type follicles, actin is enriched in the apical membrane, whereas Par6 and Crumbs are localized specifically to the apical membrane of follicle cells. DE-Cadherin is localized in the apical–lateral adherens junction. As described in Figure 8, both follicle cells in the native epithelium and tumor follicle cells at the termini of dlgm35/dlgip20 mutants preferentially lose epithelial polarity, whereas follicle cells in more central regions of the epithelium lose polarity less frequently. Terminal follicle cells, thus, appear to be more susceptible to loss of epithelial polarity in dlg mutants. These data establish a correlation between greater invasion and greater loss of polarity at the termini of dlg mutant egg chambers, independent of a difference in invasive capacity. Thus, in addition to a higher rate of proliferation, greater invasion from the termini of dlg mutant egg chambers may also be influenced by a higher frequency of loss of epithelial polarity.

DISCUSSION

Our purpose in this study was to complete a detailed analysis of spatial–temporal patterns of dlg tumor invasion to generate hypotheses about mechanistic similarities and differences between normal and tumor cell movements. In previous studies, we had emphasized an anterior pattern of invasion, because it most closely resembles border cell migration (compare Figs. 3A–D and 3E and 5A; Goode and Perrimon, 1997). However, since our initial work, much has been learned about how the early follicular epithelium is patterned and the importance of the early pattern formation for establishing patterns of cell migration during mid-oogenesis (Fig. 1 and see Introduction section). We thus reasoned that re-evaluation of dlg invasion patterns would enable us to generate a more accurate and comprehensive understanding of potential mechanistic similarities and differences between tumor cell invasion and normal cell migration during oogenesis.

We discovered two novel patterns of dlg invasion that had not been appreciated previously. First, posterior invasions occur between the oocyte and adjacent nurse cells in a pattern that resembles normal centripetal cell migration during mid-oogenesis (compare Fig. 5D to Figs. 1 and 3D). This pattern is the second most frequent pattern of invasion (O-NC, Fig. 4). The pattern emerges in posterior follicle cells, which by stage 6 develop more tumors than anterior follicle cells (compare O and O-NC to A in Fig. 4). However, only roughly half of posterior tumor cells invade, because most delaminate into the space between the epithelium and oocyte (Fig. 3I). Because all invasions are directed toward the oocyte (Figs. 5, 6), these cells have no place to move, even though they likely have the capacity to migrate. The second novel pattern of invasion initiates from central regions of the follicular epithelium, in main-body follicle cells (Fig. 5B,C). These are the least frequent invasions (1-2NC and 2-3NC, Fig. 4).

Independent of the site from which dlg cells initiate invasion, all invasions resemble border and centripetal cell migration, in that they are directed toward the oocyte (Fig. 4). In addition, the cells have a rounded, mesenchymal-like morphology, which partially resembles the morphology of border cells. In contrast, centripetal cells switch to an elongated-epithelial cell shape as they migrate. Thus, the cell shape change associated with loss of dlg is conducive to patterns resembling either border or centripetal cell migration but most closely resembles border cell morphology. Because this morphology is conducive to a tumor invasion pattern that resembles developmental centripetal cell migration, possibly the differences in cell shape of border or centripetal cells is less important for their migration compared with the reduction of epithelial polarity associated with migration of both cell types.

The invariant movement of dlg cells to the oocyte indicates that an intrinsic pattern exists in germ cells starting very early in oogenesis, before border cell migration, that guides dlg cells to the oocyte. Our study does not address whether the intrinsic pattern is the same as that used for border cell migration. However, given what is known about expression of the molecular cues that guide border cell migration, it is plausible that the same mechanism is used to guide dlg tumor cells. The gurken locus produces an EGF/TGFα-like growth factor, expressed specifically in the oocyte starting early in oogenesis (Neuman- Silberberg and Schupbach, 1996). Furthermore, we have found that a PDGF-like growth factor (Duchek et al., 2001) is expressed in the oocyte starting early in oogenesis (S. Goode, unpublished data). Both factors are needed to guide the border cells to the oocyte (Gupta and Schupbach, 2001). Because they are both expressed starting early in oogenesis, it is possible that one or both are also responsible for guiding dlg tumor cells to the oocyte. Another possibility, not mutually exclusive from oocyte signals, is that an anterior–posterior haptotactic or adhesion gradient is expressed on the surface of nurse cells starting early in oogenesis.

Our data suggest that greater cell invasion from the termini may be caused by either a greater number of cells, or a higher frequency of loss of epithelial polarity at the termini of dlg mutant egg chambers (Fig. 8), or both of these factors. Another possibility is that increased proliferation at the termini is driven by the need to replace cells in the native epithelium as cells lose polarity and repetitively delaminate, with or without invasion. This interpretation is also supported by data shown in Figure 2, which shows that, for each stage of oogenesis, there is a progressive increase in time in the number of invasive cells per egg chamber. Thus, if delaminating or invading tumor cells exited without being replaced by proliferation, we would expect to observe gaps, holes, or thinning of the native epithelium, but this finding is rarely observed (Figs. 3, 5, 8). This interpretation suggests that overproliferation may not only contribute to greater invasion from the termini but also is a consequence of a mechanism that ensures that cells are replaced within the native epithelium after cellular delamination, with or without invasion. This interpretation is consistent with the observation that the average number of cells per invasion is roughly equivalent, independent of the spatial location from which invasion initiates (Fig. 7), suggesting that tumor growth may occur at roughly equal rates, once invasion is initiated. We propose that dlg tumors grow as invasion projects toward the oocyte, pulling cells out of the follicular epithelium, thus causing increased proliferation in the native epithelium, ensuring a continual supply of cells to build the invasive tumor stream.

In Figure 9, we relate patterns of dlg invasion to corresponding domains of follicle specification described by others. Study of the Jak-Stat, TGFα-EgfR, and Mirror pathways demonstrate that the follicular epithelium is divided along the anterior–posterior axis into two terminal and one main-body domain (Gonzalez-Reyes and St. Johnston, 1998; Jordan et al., 2000; Xi et al., 2003). These domains correspond to patterns of dlg tumor invasion (Fig. 9), with most invasions initiating from terminal follicle cells and the least from main body cells. We propose that terminal follicle cells are more susceptible to invasion, because both termini are prepatterned as anterior follicle cells by the Jak-Stat pathway, and anterior follicle cells will subsequently change shape and become migratory follicle cells (Fig. 1). The main-body follicle cells are the least likely to change shape and migrate, because they are not part of a Jak-Stat migratory prepattern and, thus, are not destined to migrate. Thus Jak-Stat and/or TGFα-EgfR pathways (Fig. 1) may directly influence the ability of dlg tumor cells to change shape and migrate, but another possibility is that other genes or gene activities expressed in a similar pattern control spatial patterns by which follicle cells are susceptible to switching polarity and, thus, the invasiveness of dlg tumor cells along the anterior–posterior axis (e.g., Kekkon-1; Goode and Perrimon, 1997). An ill-defined parameter of this proposal concerns the precise timing of action of the Jak-Stat and TGFα-EgfR pathways. Although the ligands for both of these pathways are expressed starting as early as stage 1 of oogenesis, the anterior and posterior cell fates determined by these signals may not be established until stages 5–6 (Gonzalez-Reyes et al., 1995; Xi et al., 2003). Therefore, we cannot definitively say whether the timing of these signals corresponds to the timing of the terminal patterns of invasion that we have described. Nonetheless, our data suggest that rather than resembling patterns of border cell migration (Goode and Perrimon, 1997), patterns of dlg invasion appear to more closely correspond with patterns of anterior–posterior cell determination, which are established during early oogenesis. Proto-oncogene signaling pathways that play a crucial role in tumor progression in humans establish the anterior–posterior pattern (Fig. 1). These pathways may be influencing the plasticity of terminal and main body follicle cell populations, causing the terminal dlg tumor cells to be more susceptible to loss of polarity and shape change and, thus, may play a crucial role in determining the neomorphogenetic patterns of cell invasion observed. This process is an attractive hypothesis, considering the same signals are subsequently used to specify the border cells and attract them to the oocyte. The Drosophila egg chamber, thus, may be a simple model system to compare developmental patterning and the pathology of invasive tumorigenesis, which may ultimately lead to a deeper understanding of how normal developmental programs are bypassed during cancer, and how tumorigenesis may be driven by neomorphogenetic processes.

Figure 9.

Correspondence between follicle cell patterning and patterns of invasion in dlg mutant egg chambers. The upper portion of the figure shows domains of follicle cell determination in the follicular epithelium (see Fig. 1). Below, a color-coded graph shows the suggested correspondence between these domains and patterns of follicle cell invasion in dlg mutant egg chambers. By stage 7 of oogenesis, tumor cells that delaminate adjacent to the oocyte (O) and that invade between the oocyte and nurse cells (O-NC) are the most prevalent tumors in dlg egg chambers (red). This domain corresponds with overlapping posterior Jak-Stat and TGFα-EgfR signals (red; Fig. 1), which are subsequently used to specify and guide border cell invasion, respectively. The next most prevalent domain of dlg invasion corresponds to anterior (A) follicle cells, which are subsequently determined as the three migratory populations of follicle cells (blue). The fewest follicle cell invasions occur between nurse cells 2 and 3 (2-3 NC) and nurse cells 1 and 2 (1-2 NC), corresponding to main-body follicle cells, which receive little or no Jak-Stat and TGFα-EgfR signals.

EXPERIMENTAL PROCEDURES

Genetics

The dlg mutants dlghf321 (temperature-sensitive), dlglv55, dlgip20, and dlgm35 were used in this study (Woods and Bryant, 1991). dlghf321/dlglv55 flies were grown at the permissive temperature (18°C) then shifted to the restrictive temperature (25°C) for the indicated amounts of time. dlgip20/dlgm35 flies were reared at 25°C.

Histochemistry and Imaging

Ovaries were fixed in 5% phosphate buffered saline, 4% formaldehyde solution as previously described (Szafranski and Goode, 2004). Alexa488 phalloidin (1:10; Molecular Probes) was used to stain and visualize egg chambers. Images were acquired with a Zeiss LSM510 confocal microscope, and confocal images were processed using Photoshop software (Adobe).

Scoring Tumor Cell Invasion

Egg chambers were scored based on developmental stage, the site of invasion of tumor cell clusters, and the number of cells within each tumor cluster. It was difficult to score the number of cells in the invasive streams because the cells are distributed in three-dimensions. We started in the highest focal plane that included the tumor and counted and recorded the number of cells. Concentrating sharply, we then gradually focused to the next lower focal plane, scoring newly appearing cells, while keeping track of cells that had already been scored. This method often required focusing up and down several times to be confident that we did not miss any cells or score the same cell twice. Also, sometimes we drew the tumor in a particular focal plane to keep track of the cells. We continued this procedure stepwise until we reached the lowest focal plane that included the tumor. We practiced before starting the actual experiments to ensure that we could reproducibly score the cell numbers. For tumors of ∼30 cells, we routinely recorded the same number of cells ± two to three cells. Furthermore, by using this approach, S.G. and S.K. completed two independent experiments, with similar results. The data shown in Figure 7 is the pooled data from both experiments.

Acknowledgements

We thank Kwang-Wook Choi, Haiyun Cheng, members of the Goode lab, and anonymous reviewers for critical review of the manuscript. We thank P. Szafranski for helping with confocal images in Figure 8. S.G. was funded by the NIH.

Ancillary