Genetic control of programmed cell death in the Caenorhabditis elegans hermaphrodite germline

The nematode Caenorhabditis elegans is characterized by a highly reproducible development. The patterns of cell divisions, cell migrations and cell deaths that give rise to the adult animal are nearly invariant and have been completely described. Thus, the fate of every somatic cell arising during (C)elegans development is known (Sulston and Horvitz, 1977; Kimble and Hirsh, 1979; Sulston et al., 1980, 1983).

Of the 1090 cells that are generated during development of the hermaphrodite, 131 are fated to undergo programmed cell death (apoptosis). Analysis of these deaths revealed a genetic pathway that controls the programmed death and elimination of apoptotic cells in C. elegans (reviewed by Ellis et al., 1991b; Horvitz et al., 1994). Biochemical characterization of the genes in this pathway has led to the identification of an apoptotic machinery. Briefly, the proximal cause of apoptosis in C. elegans is the activation of the caspase CED-3 (CEll Death abnormal) from the inactive zymogen (proCED-3) into the mature protease (Yuan et al., 1993; Xue et al., 1996). This activation is mediated by the Apaf-1 homolog CED-4 (Yuan and Horvitz, 1992; Zou et al., 1997). In cells that should survive, CED-3 and CED-4 pro-apoptotic activity is antagonized by the Bcl-2 family member CED-9 (Hengartner et al., 1992; Hengartner and Horvitz, 1994a). CED-9 has been proposed to prevent death by sequestering CED-4 and proCED-3 in an inactive ternary complex, the apoptosome (Hengartner, 1997). In cells fated to die, CED-9 is in turn inactivated by the pro-apoptotic BH3 domain-containing protein EGL-1 (EGg-Laying defective), likely through a direct protein-protein interaction (Conradt and Horvitz, 1998). How the apoptotic machinery is controlled such that it only becomes activated in cells that should die is not yet understood.

Each of the key C. elegans cell death genes has one or several mammalian homologs, and all the classes of protein-protein interactions described for C. elegans cell death proteins have also been detected for their mammalian counterparts (reviewed by Hengartner, 1997, 1998), suggesting that the apoptotic machinery identified in C. elegans is evolutionarily conserved and mediates at least a subset of apoptotic deaths in mammals. Unlike somatic cells, germ cells in C. elegans do not have a fixed lineage and populations of cells, rather than individual cells, are regulated, albeit through strict and well-characterized control mechanisms (reviewed by Schedl, 1997). The germline is one of the best characterized and most extensively studied cell types in C. elegans, as it is readily amenable to genetic and developmental analysis (Kimble and Ward, 1988; L’Hernault, 1997; Schedl, 1997).

The gonad of the C. elegans adult hermaphrodite consists of two symmetrical, U-shaped tubes, which are linked to a common uterus (Fig. 1A). The distal end of each gonadal tube is capped by a somatic distal tip cell (DTC). The LAG-2 growth factor, expressed by the DTC, maintains the stem cell potential of the adjacent germ cells and promotes their mitotic proliferation throughout the reproductive life of the animal. Germ cells positioned beyond the influence of the DTC signal enter meiosis and progress into the pachytene stage of meiosis I. Near the bend of the gonadal tube, cells exit pachytene in response to activation of the ras/MAPK pathway and begin to enlarge in size. In the proximal arm, germ cells progress to diakinesis of prophase I and increase in size to form oocytes as they migrate in a single row towards the uterus.

The mitotic and early meiotic germ cell nuclei are only partially enclosed by a plasma membrane and are thus part of a large syncytium (Figs 1A, 2A). Interestingly, nuclei within this syncytium are not synchronized, such that a given syncytium will contain at the same time nuclei at all the various stages of mitosis and early meiosis. Because syncytial germ cell nuclei appear to act as independent units, we will refer to them as ‘germ cells’ even though they share a common cytoplasm.

Previous work has established that germ cells have three fates available to them: they can either undergo mitosis, or they can enter meiosis and differentiate into either sperm or oocytes (Ellis and Kimble, 1994). Here we report our characterization of a fourth major germ cell fate: apoptotic cell death. We show that, in normal adult hermaphrodites, over half of all potential oocytes are eliminated during meiotic maturation. Germ cell death is mediated by the same core execution machinery, i.e. ced-3 and ced-4, as are all developmental deaths. However, activation of this machinery is regulated differently in the soma and the germline: loss-of-function mutations in egl-1, as well as a rare gain-of-function mutation in ced-9, both of which completely prevent somatic cell death, have little effect on germ cell death. Our results suggest that germ cell death in the adult hermaphrodite germline serves a homeostatic purpose by eliminating excess germ cell nuclei that acted as nurse cells.

Methods for culturing C. elegans have been described by Brenner (1974). All strains were grown at 20°C unless otherwise indicated. All mutants used in these studies were derived from the wild-type variety Bristol strain N2. Other wild-type species used were Caenorhabditis briggsae, Caenorhabditis vulgaris, Caenorhabditis remanei, Pelodera strongyloides, Panagrellus redivivus and Rhabditella axei. let-60(dx16) and lin-45(dx19) are recessive sterile mutations that were isolated after exposure to 310 nm UV light (E. L., unpublished). The following mutations are described by Riddle et al. (1997) or the references cited therein: LG I: mek-2(n2678), fog-1(q253ts), unc-15(e1214am), unc-13(e51), gld-1(oz10), gld-1(q485), gld-1(q93oz50), gld-1(q126), gld-1(q266), gld-1(q343), ces-1(n703gf), ced-1(e1735), glp-4(bn2), ces-2(n732); LG III: mpk-1(oz140), ced-4(n1162), lon-1(e185), ced-6(n1813), dpy-19(e1259), ced-7(n1892), mog-1(q223), unc-69(e587), ced-9(n1950gf), ced-9(n1653ts), ced-9(n1950n2161), ced-9(n1950n2077), ced-9(n2812), tra-1(e1575sd), tra-1(e1076) (Hodgkin and Brenner, 1977); LG IV: ced-2(e1752), ced-10(n1993), unc-5(e53), fem-1(e1965), mor-2(e1125), lin-45(dx19), let-60(n1046sd), let-60(s1124), let-60(dx16), unc-24(e138), fem-3(e1996sd), ced-5(n1812), him-8(e1489), dpy-20(e1282ts), ced-3(n717); LG V: her-1(n695sd,ts), him-5(e1490), egl-1(n986dm), egl-1(n1084n3082); LG X: ced-8(n1891), nuc-1(e1392am), sdc-1(n485); rearrangements and duplications: qC1(III), eDp6(III;f), hT2(I;III) and nDp4(I;V).

All ced-9 alleles except for n1653 and n1950 were maintained with the duplication qC1. fem-1(e1965) was maintained in trans to unc-5(e53) mor-2(e1125). fem-3(e1996) was maintained heterozygous with unc-24(e138) dpy-19(e1259). gld-1 alleles q266, oz10, and q93oz50 were maintained with nDp4(I;V). gld-1(q485)/unc-13(e51) gld-1(q126) maintained both q485 and q126. gld-1(q343) was maintained heterozygous with unc-15(e1214). mog-1(q223) was maintained in trans to dpy-19(e1259) unc-69(e587). tra-1(e1076) was maintained with eDp6(III;f).

We studied the appearance and number of cell corpses in the germline of nematodes by mounting animals in a drop of M9 salt solution containing 30 mM NaN₃ (Hodgkin, 1980) and observing the animals using Nomarski optics (Ellis et al., 1991a). Corpses are cellularized and more refractile than syncytial nuclei or oocytes and can readily be identified under high magnification. Average corpse numbers with the standard error of the mean (s.e.m.) were determined by the Statview II program (Abacus Concepts, Incorporated, Berkeley, California).

To study the kinetics of death and degradation of germ cells, adult wild-type nematodes (24 to 36 hours post L4/adult molt) were mounted in S Basal or 3 or 5 mM levamisole. Levamisole prevents animals from moving but does not affect the germline or the gonad (Sulston and Brenner, 1974). The germlines of individual animals were observed for 1 to 3 hours.

To study the number, timing and distribution of cell corpses in the germline, L4 stage larvae were transferred to new plates. Starting at the L4/adult molt, and every 12 hours thereafter, animals were anesthetized and observed using Nomarski optics (Ellis et al., 1991a). For each animal observed, only one arm was scored, as the other arm was usually concealed by the intestine.

To obtain an estimate of the relative numbers of corpses in different genetic backgrounds, 36-hour adult animals were stained with SYTO 12 (Molecular Probes, Eugene, OR), a vital dye that preferentially stains apoptotic germ cells. Animals were stained by incubating in a 33 μM aqueous solution of SYTO 12 for 4-5 hours at 23°C, then transferred to seeded plates to allow stained bacteria to be purged from the gut. After 30-60 minutes, animals were mounted on agarose pads and inspected using a Nikon Microphot-SA, equipped with standard epifluorescence filters and Nomarski optics. Only animals that stained brightly were scored. In wild-type animals, SYTO 12 has a granular cytoplasmic staining pattern in midto late-stage oocytes and in embryos. The identity of the stained granules was not determined. However, they might represent mitochondria and/or lysosomes. In wild-type animals, the posterior half gonad typically stained more brightly than the anterior half. SYTO 12 stains both corpses visible by Nomarski optics as well as late-stage germ cell corpses, which cannot be scored by Nomarski optics. Thus, the number of SYTO-positive germ cells is consistently higher than the number of corpses observed using Nomarski optics.

Double staining with SYTO 12 and Hoechst 33342 (Sigma) was done by incubating animals in an aqueous solution of 33 μM SYTO 12, 67 μM Hoechst 33342 for 6 hours at room temperature. In some cases, animals were anesthetized using 5 mM levamisole to facilitate photography. For image acquisition, a Sony AVC-D7 CCD camera was connected to a Scion LGIII frame grabber installed in a Quadra 700 computer and frames were taken using NIH Image software (as modified by Scion). Composite images were assembled and edited using Adobe PhotoShop 3.0 on a Power Macintosh 8100/80AV.

Adult hermaphrodites were fixed in 0.8% glutaraldehyde, 0.7% OsO₄, 0.1 M cacodylate buffer for 1 hour on ice. Subsequently, samples were cut and postfixed in 2% OsO₄, 0.1 M cacodylate buffer, mounted into an agar block, dehydrated in a series of alcohols and embedded in a mixture of Epon-Araldite. Thin sections (50 nm) were cut on an Ultracut E and pictures were taken with a JEOL 1200× electron microscope at 80 kV.

Anecdotal observations of dead cells in the C. elegans germline have previously been made (Sulston, 1988; White, 1988; J. Kimble, personal communication). To corroborate these findings, we examined gonads from wild-type animals at various stages of development. We found that the germlines of adult hermaphrodites, but not those of larvae or adult males, consistently contained a small number of condensed structures when observed using Nomarski microscopy (Fig. 1B). These structures were restricted to a region of the germline occupied by syncytial germ cells in the pachytene stage of meiosis I (Fig. 1A). The number of these structures gradually increased throughout the reproductive life of the animals. Observations of individual cells indicated that these condensed structures arose from normal-looking germ cells. Over a time frame of about 20 minutes, the cytoplasm of these cells became increasingly refractive and finally melded with the nucleus to become a highly refractive uniform structure. These condensed structures were transient in nature and typically disappeared within an hour of their appearance (Fig. 1E-J). The radical morphological changes and subsequent elimination of these germ cells suggests that they are either degenerating or undergoing programmed cell death.

The condensed intermediates seen using Nomarski optics were highly reminiscent of the somatic programmed cell deaths that occur during C. elegans development (Sulston and Horvitz, 1977). To test the hypothesis that the germ cell deaths that we observed might be apoptotic in nature, we characterized them further using both morphological and genetic approaches.

We first tested whether dying germ cells show other features characteristic of apoptotic deaths, such as chromatin condensation and staining with vital dyes that are taken up by apoptotic cells (Kerr et al., 1972; Wyllie et al., 1980; Abrams et al., 1993; White et al., 1994). We adapted a Hoechst 33342 DNA-staining protocol for use on live animals and found that the nuclei of refractive cells (as determined by Nomarski microscopy) showed DNA condensation characteristic of apoptosis (Fig. 1B,C). Staining with the DNA intercalator dye 4′, 6-diamidino-2-phenyline (DAPI) gave similar results (data not shown). The vital dye acridine orange (AO) has been used successfully in Drosophila to specifically stain apoptotic cells in live animals (Abrams et al., 1993; White et al., 1994). We found that this dye, as well as the nucleic acid stain SYTO 12, specifically stained dying germ cells identified by Nomarski optics or the Hoechst dye (Fig. 1D; T. L. G., E. L., A. Samuelson, S. Milstein, and M. O. H., unpublished).

To obtain additional evidence confirming the apoptotic nature of dying germ cells, we analyzed cross sections of young adult hermaphrodite gonads by electron microscopy. While most sections contained only healthy, syncytial germ cells, we did find that a small fraction of germ cell nuclei exhibited morphological features characteristic of early stages of apoptotic cell death (Fig. 2B-D). Our ultrastructural studies also resolved the apparent dilemma of how individual cell nuclei within a syncytium can die while the rest of the syncytium survives. We found that cells fated to die rapidly cellularize away from the common syncytium, thereby physically isolating the doomed nucleus from its neighbors. The newly generated cell, which contains only limited amounts of cytoplasm (Fig. 2C,E), is subsequently recognized and engulfed by the gonadal sheath cells that surround the germline (Fig. 2C,D). The swift phagocytosis and degradation of apoptotic germ cells by sheath cells might explain why we failed to observe any dying germ cells at advanced stages of apoptosis in normal animals. Consistent with this hypothesis, if engulfment is inhibited (by genetic means, see below), persisting dead germ cells accumulate in the gonad and undergo further nuclear and cytoplasmic condensation (Fig. 2E,F).

14 genes have been identified that function in programmed cell death during C. elegans development (reviewed by Ellis et al., 1991b; Horvitz et al., 1994). We found that many of the developmental cell death genes also function in the germline, as animals mutant in these genes showed altered patterns of germ cell death.

For example, the genes ced-3 and ced-4 are required for programmed cell death in the C. elegans soma: strong loss-of-function (lf) mutations in either gene prevent all 131 cell deaths that normally occur during hermaphrodite development. We found that both genes are also required for germ cell death, as few if any germ cell corpses were seen in ced-3 or ced-4 mutant animals (Fig. 3A).

The ced-9 gene protects C. elegans cells from programmed cell death: loss of ced-9 function results in the death of many cells that normally survive (Hengartner et al., 1992). In ced-9(lf) animals, we observed increased levels of germ cell death, indicating that ced-9 also has a protective function in the germline (Fig. 3B and data not shown).

Efficient phagocytosis of apoptotic cells in C. elegans requires the function of at least six genes: ced-1, ced-2, ced-5, ced-6, ced-7 and ced-10 (Ellis et al., 1991a; Horvitz et al., 1994). We found that animals mutant for any one of these genes accumulated many more cell corpses in the germline than wildtype animals (Fig. 3C). In older animals, many unengulfed cell corpses swelled and eventually lysed, suggesting that these lingering corpses were undergoing secondary necrosis, as has been observed for mammalian corpses that escape ingestion (Wyllie et al., 1980). In addition, cellular debris accumulated in front of the spermatheca and in the uterus of these mutants. We surmise that this debris was the result of necrotic lysis or possibly the collapse of corpses forced through the spermatheca by the progression of oocytes. A seventh gene, ced-8, is involved in the timely phagocytosis of embryonic, but not larval, corpses (Ellis et al., 1991a; Horvitz et al., 1994). A time course study of germ cell death in ced-8 mutant animals revealed no significant difference from that in the wild type (Fig. 3C), suggesting that ced-8 is not required for the rapid engulfment of germ cell corpses.

The nuc-1 (NUClease abnormal) gene is required for the degradation of DNA within engulfed corpses (Hedgecock et al., 1983). We found that, in nuc-1 mutants, sheath cells contained DAPI-positive vacuoles with undegraded DNA. The absence of DAPI-positive vacuoles in the germline of nuc-1 mutants suggests that dying germ cells are not reabsorbed by the germline syncytium but rather are engulfed by somatic sheath cells. DAPI-positive vacuoles mildly increased in size and number as animals aged, consistent with our observation that germ cell deaths occur continuously throughout the reproductive life of the animal (Fig. 3A and data not shown). DNA degradation also failed to occur when engulfment was blocked (e.g., in ced-5 mutants), indicating that, similar to somatic apoptosis (Hedgecock et al., 1983), germ cell apoptosis requires phagocytosis for nuc-1-mediated DNA degradation (data not shown).

Mutations in the genes ces-1 and ces-2 (CEll death Specification abnormal) affect a specific subset of somatic programmed cell deaths (Ellis and Horvitz, 1991). These genes could be required to specify the identity of the affected cells, or could participate in a cell-type-specific pathway that controls activation of the apoptotic machinery (reviewed by Driscoll, 1992; Hengartner and Horvitz, 1994b). We found no difference in germline corpse numbers between any of these mutants and wild type (Fig. 3E), suggesting that ces-1 and ces-2 play no role in germ cell death.

Despite the many similarities in the genetic regulation of somatic and germ cell deaths, we observed a few striking differences. We found that the ced-9(n1950) gain-of-function mutation, which completely prevents cell death during hermaphrodite development (Hengartner et al., 1992), has little if any effect on germ cell death (Fig. 3D). To confirm this result, we repeated our time-course analysis in a ced-1(e1735) background. The ced-1(e1735) mutation allows corpses to persist and thus provides a more sensitive assay for germ cell death, as it amplifies the difference in germ cell corpse number between wild-type and ced-3 mutant animals. Again we found that ced-9(n1950) is essentially indistinguishable from the wild-type allele in the ced-1(e1735) background.

The gene egl-1 is a negative regulator of ced-9 and is required for all somatic deaths (Conradt and Horvitz, 1998). Surprisingly, egl-1(lf) mutants are indistinguishable from the wild type with respect to germ cell apoptosis (Fig. 3D; Conradt and Horvitz, 1998), suggesting that egl-1 is not involved in this process. In summary, our data indicate that the molecular mechanisms that regulate the core apoptotic complex differ significantly between the soma and the germline.

What triggers germ cell death? In wild-type nematodes, germ cell deaths occur exclusively in adult hermaphrodites; we observed no germ cell deaths in either adult males or in larvae of either sex (Fig. 3A and data not shown). The timing and location of germ cell deaths suggest that they are associated with oogenesis, which is also restricted to adult hermaphrodites. To test this hypothesis, we examined the effect on germ cell death of mutations that affect sexual identity or germ cell differentiation (Table 1).

In C. elegans, the fundamental difference between males and hermaphrodites is the ratio of sex chromosomes to autosomes (the X:A ratio): males carry one X chromosome, hermaphrodites two (Meyer, 1997). To assess whether the X:A ratio determines the extent of germ cell apoptosis, we analyzed the germline of masculinized XX animals, which only produce sperm instead of oocytes (sdc-1(lf), tra-1(lf)). These animals exhibited no germline corpses. In contrast, feminized XO animals (fem-1(gf), tra-1(gf)), which produce oocytes, had germ cell death. Together, these results suggest that physiological germ cell apoptosis is determined not by sex chromosome dosage but rather by the sex of the animal (Table 1).

Because the soma controls many germline events (Clifford et al., 1994), we next asked whether germ cell apoptosis is determined by the sex of the soma or the sex of the germline. We found that germ cell death still occured in XO animals with a male soma but a feminized germline (fog-1(lf) and gld-1(q126)) (Table 1). Conversely, mutations (gld-1(oz10) and mog-1(lf)), which masculinize the germline of XX hermaphrodites such that only sperm are produced, resulted in the absence of germ cell corpses. We also observed dying germ cells in animals in which the germline was incompletely masculinized and still generated both sperm and oocytes (e.g., in XX intersex or pseudomale animals; sdc-1(lf), her-1(gf), and tra-1(lf)). These results indicate that germ cell death requires a female germline and is independent of the sex of the soma. Finally, we asked whether entry into meiosis and differentiation into oocytes are required for germ cells to die. Mutations have been identified that affect the ability of germ cells to proliferate and differentiate (Ellis and Kimble, 1994). For example, in gld-1(q485) hermaphrodites, meiotic differentiation is blocked, resulting in a tumorous germline filled with mitotically proliferating germ cells (Francis et al., 1995). We found that these animals did not contain germ cell corpses in the tumorous mitotic region (Table 1). However, some germ cells entered meiosis in older gld-1(q485) hermaphrodites; such cells were capable of undergoing cell death. Similarly, we found no germ cell corpses in glp-4 mutants, in which germ cells were blocked in mitotic prophase and failed to proliferate and differentiate (Table 1). Together, these results suggest that mitotic germ cells are protected from apoptosis. Since we observed germ cell deaths if, and only if, oogenesis was also occurring, both in the wild type and in the various mutant backgrounds tested, we infer that germ cell death is either an integral part or a direct by-product of the oogenic program.

In wild-type animals, germ cell death occurs near the region where cells exit the pachytene stage (Fig. 2A and data not shown). To test whether apoptotic germ cell death might be functionally linked to meiotic progression, we examined whether mutations that affect differentiation or meiotic cell cycle progression influence germ cell death. Previous studies have shown that the ras/MAPK signaling cascade is required for exit of C. elegans germ cells from pachytene arrest (Church et al., 1995). Known members of this signaling pathway include LET-60 Ras, LIN-45 Raf, MEK-2 MAPK/ERK kinase and MPK-1 MAP kinase (MAPK) (Kayne and Sternberg, 1995). Strong loss-of-function alleles in any of these four genes block exit from pachytene during oogenesis, resulting in sterility (Church et al., 1995, and E. L., T. Schedl, R. Francis, M.-H. Lee, and K. Kornfeld, unpublished data). We found that these mutations also block programmed germ cell death (Table 2). Thus, activation of the ras/MAPK pathway, which is required for germ cells to exit pachytene, is also a prerequisite for germ cell death.

The failure of pachytene-arrested cells to undergo programmed cell death could be explained either by an absence of the death machinery in these cells or by the presence of a death-suppressing activity. To distinguish between these two possibilities, we constructed double mutants carrying strong loss-of-function mutations in a ras/MAPK pathway gene and in the cell death suppressing gene ced-9. If the cell death machinery were already present in pachytene-arrested cells, then the absence of ced-9 might result in its activation. Consistent with this hypothesis, germ cell death reappeared in ced-9; lin-45 and mek-2; ced-9 double mutants (Table 2). These results suggest that pachytene-arrested cells have the ability to die, but are normally protected from apoptosis through a ced-9-dependent mechanism.

In adult hermaphrodites, germ cell differentiation and progression through the gonad are coordinated in such a way that the position of a germ cell within the gonad is a reliable indicator of its developmental status. Cell death appears to be required to maintain this steady-state structure: in strong ced-3 and ced-4 loss-of-function mutants, the absence of germ cell death results in a gradual increase in the number of syncytial germ cells (Fig. 4). As the syncytium expands to accommodate the extra germ cell nuclei, the area devoted to fully grown oocytes decreases over time (Fig. 4C). In old ced-3 or ced-4 animals, this syncytial hyperplasia is so extensive that occasionally only one or two full-sized oocytes are present within a gonad, as opposed to greater than a dozen oocytes in wild-type animals (Fig. 4A,B).

To determine whether germ cell death also occurs in other nematodes, we looked for corpses in the germlines of other nematode species. We found that C. briggsae, C. vulgaris and C. remanei, three close relatives of C. elegans, as well as Pelodera strongyloides, a more distant relative (Fitch and Thomas, 1997), had programmed cell deaths in their germlines (data not shown). Another distant nematode relative, Rhabditella axei, is distinctive among the species that we observed because its germline is composed of germ nuclei within large compartments. We failed to observe any germ cell death in this species. Panagrellus redivivus, another free-living nematode, has germ cell death in both males and females (Sternberg and Horvitz, 1981). These observations indicate germline programmed cell death is evolutionarily conserved among nematode species.

Apoptosis is a common feature of metazoan germline development (Tilly, 1996). Despite its prevalence, little is known about the molecular regulation of programmed cell death in the germline. Here we show that cell death is a major outcome for germ cells in the adult C. elegans hermaphrodite and provide the first detailed description of the morphology, genetics and biology of germ cell death in this organism. We establish that germ cell death is extensive, apoptotic in nature and strictly dependent on oogenesis. We also provide genetic evidence that regulation of the apoptotic machinery occurs via mechanisms distinct from those used in the soma.

Cell death is a very common fate in the C. elegans female germline. A wild-type hermaphrodite generates about 2000 germ cells over its lifetime and produces about 300 sperm and a slightly larger number of mature oocytes (Schedl, 1997). We have observed over 70 germline corpses in gonad arms of old animals in which cell deaths are not efficiently removed because of a mutation that blocks cell-corpse engulfment (Fig. 3C), and over 100 in some animals with two such mutations (data not shown), indicating that at least 100 cells must die in each gonad arm. However, the total number of germ cell deaths is likely to be much higher, for two reasons. First, the mutations that affect engulfment are incompletely penetrant and only prevent a fraction of all corpses from being engulfed (Hedgecock et al., 1983; Ellis et al., 1991a). Second, germ cell corpses are also eliminated by passage into the uterus or by secondary necrosis (data not shown). Thus, we estimate that programmed cell death is the fate of about half of female germ cells.

This estimate is supported by our analysis of the kinetics of germ cell apoptosis. It takes approximately 1 hour for a cell to die and disappear, and about 2-3 corpses can be observed at any given time (Figs 1E-J, 3A). Integration of the number of corpses observed over time in wild-type adult hermaphrodites suggests that approximately 150 germ cells die within the first 3-4 days of adulthood in each gonad arm (Fig. 3A).

Several lines of evidence indicate that the massive amounts of germ cell deaths observed in adult wild-type hermaphrodites are apoptotic in nature. First, the dying germ cells resemble the programmed cell deaths that occur during C. elegans development and exhibit many of the morphological and ultrastructural features characteristic of apoptosis, including cytoplasmic and nuclear condensation, chromatin aggregation, and rapid recognition and phagocytosis by surrounding cells (Figs 1, 2). Second, dying germ cells can be labeled with fluorescent dyes that specifically stain apoptotic cells (Fig. 1D and data not shown). Third, the genetic pathway that mediates apoptosis during development also functions in the germline: ced-3 and ced-4 are required for germ cells to die (Fig. 3A) and ced-9 is required to protect germ cells from apoptotic death. Mutants homozygous for a weak loss-of-function mutation of ced-9, n1653, have many more cell deaths in the germline than wild-type animals (Fig. 3B). In strong (putative null) ced-9 mutants, almost all potential oocytes die by programmed cell death (T. L. G. and M. O. H., unpublished), which probably accounts for the low brood size observed of ced-9(lf) mutants (Hengartner et al., 1992).

Interestingly, although ced-9 is vital for germ cell survival, the gain-of-function mutation ced-9(n1950), which completely suppresses somatic cell death, does not prevent germ cell death (Fig. 3D). The n1950 mutation (G169E) alters an invariant glycine residue located in the conserved BH1 domain, which is found in most CED-9/Bcl-2 family members (Hengartner and Horvitz, 1994b). The requirement of this domain for interaction between pro-survival Bcl-2 family members and the BH3 domain of pro-apoptotic Bcl-2 family members (Reed, 1994; Yin et al., 1994) suggests that n1950 might affect the interaction of CED-9 with pro-apoptotic C. elegans proteins. An excellent candidate for such an interactor is the BH3 domain protein EGL-1, which physically interacts with CED-9 and is essential for all developmental cell deaths (Conradt and Horvitz, 1998). Since ced-9(gf) and, more importantly, egl-1 mutants do not affect germ cell apoptosis (Fig. 3D, E), we propose that distinct CED-9 partners might be used to control germline and somatic cell death (Fig. 5).

Germ cell death is observed only during a specific stage of oogenesis, shortly before germ cells exit the pachytene stage of prophase I. Mutations that inactivate the ras/MAPK signaling cascade prevent cells from exiting pachytene arrest (Church et al., 1995) and also prevent germ cell death (Table 2). The ras/MAPK pathway determines the fate of several somatic cell lineages. For example, activation of this pathway by the EGF receptor homolog LET-23 is necessary and sufficient for the six vulval precursor cells to differentiate and form vulval tissue (reviewed by Kayne and Sternberg, 1995). In contrast, the ras/MAPK pathway appears to play a permissive (rather than instructive) role in the specification of germ cell death, as activation of the pathway is necessary, but not sufficient, for germ cells to die, as germ cell death levels are not increased in mutations that hyperactivate the pathway (Table 2).

The ras/MAPK pathway might directly regulate the cell death machinery, presumably via phosphorylation. An interaction between raf and Bcl-2 has been reported in mammalian cells (Wang et al., 1996), and phosphorylation modulates the activity of both Bcl-2 and Bad (Datta et al., 1995; Chang et al., 1997; del Peso et al., 1997; reviewed by Chao and Korsmeyer, 1998). Alternatively, ras/MAPK pathway-dependent signaling might indirectly affect germ cell apoptosis, e.g., by promoting the progression of pachytenestage cells, which are resistant to apoptosis (Table 2), to a later differentiation stage that is more sensitive to pro-apoptotic signals (Fig. 5). Presumably, further differentiation restores death resistance to the mature oocytes (Fig. 5). Alternation between death-resistant and death-sensitive stages during differentiation is a common tactic used to selectively eliminate subpopulations of cells and has been reported to occur during germ cell maturation in chicken ovaries, as well as during B and T cell development (MacLennan, 1994, Robey and Fowlkes, 1994).

Does the extensive apoptosis that occurs during oogenesis serve any purpose? One possibility is that an oogenesis-specific checkpoint uses apoptosis to eliminate unfit cells, e.g., germ cells with damaged DNA. This hypothesis would imply that over half of the female germ cells are so defective that they have to be removed. However, a number of observations suggest that the cells that die under normal growth conditions are healthy. First, all germ cell nuclei appear grossly similar when stained for DNA or observed using electron microscopy (except for cells already undergoing programmed cell death; Figs 1, 2). Second, ced-3 and ced-4 mutants (in which no germ cell deaths occur) do not show any increase in the number of defective oocytes or in embryonic lethality (Ellis and Horvitz, 1986; Hengartner et al., 1992; and data not shown). Thus, germ cells that are rescued from death in these mutants seem to be healthy and functionally equivalent to the cells that normally survive.

A second possibility consistent with our results is that apoptosis is used to cull germ cells that are created in excess to synthesize cytoplasmic components required by mature oocytes. Unlike mammals and Drosophila, C. elegans has no morphologically distinct nurse cells. Rather, this function is performed by early meiotic germ cells, which show a high level of transcriptional activity (Starck, 1977). Transcripts generated by these cells are exported to the common cytoplasmic core and subsequently incorporated into the maturing oocytes (Gibert et al., 1984). As is the case in other species, C. elegans presumably needs many nurse cells to provide for each oocyte. Because C. elegans nurse cells are also gametic precursors, there are many more candidate oocyte nuclei than can be accomodated by the available cytoplasm. We propose that C. elegans uses programmed cell death to solve this problem: once their function as nurse cells is fulfilled, excess nuclei might be eliminated by a stochastic selection mechanism that matches the number of oocyte precursors that are allowed to survive and differentiate to the amount of common cytoplasm. The nature of this putative selection process is at present unclear but should be amenable to genetic analysis.

The idea that germ cell death is used to eliminate meiotic cells that act as nurse cells can account for several apparently unrelated observations. First, it explains why germ cell death is only observed in adult hermaphrodites. Larvae have no germ cell death because oogenesis starts only very late in larval development. Males have no germ cell death because sperm contain little cytoplasm and therefore all germ cells can be allowed to form gametes. Second, the timing of germ cell deaths within the oogenic pathway – shortly before germ cells enlarge in size to form mature oocytes – would maximize the time that germ cells can function as nurse cells, while still allowing for the elimination of excess cells before they accumulate too much cytoplasm. Indeed, the suggestion that nuclei are expendable but cytoplasm is valuable might explain why dying germ cells pump almost all of their cytoplasm into the common cytoplasmic core before pinching off the syncytium (Fig. 2C,D).

While, in C. elegans, germ cell deaths are normally restricted to pachytene stage pre-oocytes, it might be possible under special circumstances or in other nematodes for other germ cell types to die. For example, in the nematode Panagrellus redivivus, germ cell deaths occur in both the male and the female germlines (Sternberg and Horvitz, 1981). The biological rationale for these deaths is likely to be distinct, suggesting that germ cell death in nematodes can be regulated at multiple levels.

How does the germline syncytium prevent the apoptotic machinery from spreading from a condemned cell to other nuclei? Dying nuclei rapidly cellularize away from the common syncytium. This process may sequester pro-apoptotic factors, such as active caspases, away from the cells that should survive and become oocytes. However, cellularization and presentation of signals for engulfment must temporally follow the initial activation of ced-3 and ced-4, as these events do not occur in animals mutant for either gene. The nature of the cellularization machinery is unknown; an attractive hypothesis is that its activation is triggered by cleavage of specific substrates by the CED-3 protease.

Genetic analysis of programmed cell death in C. elegans has been used with great success to identify the molecular basis of apoptosis. These studies revealed that the central machinery that controls all programmed cell deaths has been conserved through evolution (Yang et al., 1995; Zou et al., 1997; Hengartner, 1997, 1998). We have shown here that the soma and germline use a common apoptotic execution machinery. However, these two types of tissues must use different regulatory mechanisms to control activation of this machinery, as neither the BH3-domain protein EGL-1 nor the gain-offunction mutation in ced-9 affects germ cell death. We propose the existence of a distinct genetic pathway that specifically controls the decision between differentiation and death in the germline. Further analysis of C. elegans germ cell death may thus identify additional regulatory mechanisms that control apoptosis in mammals.

We thank Andy Samuelson and Stuart Milstein for optimizing the acridine orange staining procedure. We appreciate the critical reading of this manuscript by our colleagues. We are grateful to Tim Schedl for providing several gld-1 mutant strains and stimulating discussions, and David Fitch for discussions about nematode evolution. Some strains were obtained from the Caenorhabditis Genetics Center, which is funded by the NIH National Center for Research Resources (NCRR). This work was supported by Public Health Service Grant GM525420 and generous support from the Donaldson Charitable Trust to M. O. H.; E. L. is supported by NIH grant GM49785; H. R. H. is a Investigator of the Howard Hughes Medical Institute; M. O. H. is a Rita Allen Scholar. M. O. H. dedicates his contribution to this paper to Walter Hengartner, on the occasion of his retirement.