Proc. Nati. Acad. Sci. USA Vol. 89, pp. 3546-3550, April 1992 Evolution

Evidence for a programmed life span in a colonial protochordate (senescence/ascidian/aging/zooid/cell death)

BARUCH RINKEVICH*t, ROBERT J. LAUZONt§1¶, BYRON W. M. BROWNII, AND IRVING L. WEISSMAN** *Israel Oceanography and Limnological Research, Ltd., Tel Shikmona, P.O. Box 8030, Haifa 31080, Israel; *Hopkins Marine Station of Stanford University, Pacific Grove, CA 93950; and I1Department of Health Research and Policy, Division of Biostatistics, Stanford University School of Medicine, and **Department of Pathology and Developmental Biology, Beckman Research Center, Howard Hughes Medical Institute, Stanford University, Stanford, CA 94305

Contributed by Irving L. Weissman, January 2, 1992

containing one or more systems (5-15 zooids in each system). The experiments that form the basis of this communication originated when we observed that entire ramets derived from the same parent colony exhibited in-concert degenerative changes that led to their death within 1-2 weeks of one another, whether they were confined to one tank or kept in separate tanks. Within a tank containing ramets of several genotypes (genets), only particular parental colonies and their respective subcloned ramets underwent senescence and died within any one defined period of time. We have determined life spans of 41 independent genets and their respective ramets: 17 of 41 colonies displayed nonrandom mortality, and senescence was always accompanied by characteristic morphological changes in every dying Botryllus colony.

The variety of theories that have attempted to ABSTRACT defime the mechanisms of aging and life span can be broadly divided into two alternative but nonexclusive viewpoints. The fitrst stipulates that random changes of cellular and molecular structures lead to death following progressive "wear and tear." The second argues that life span is, at least in part, genetically programmed, and therefore aging may also result from time-dependent intrinsic processes. Here we demonstrate that ramets (clonal replicates) experimentally separated from colonies of the ascidian protochordate BobyUus schlosseri died months after their separation, almost simultaneously with their parent colony and sibling ramets. In addition, in experimentally joined chimeras between ramets of senescent and nonsenescent colonies, elements from different parent colonies displayed parent-colony-specific timing of mortality. Thus, the senescent phenotype was simultaneously expressed both in chimeras and in unfused ramets of the parent colony that was undergoing senescence, whereas control ramets from the other partner survived. These rmdings provide experimental evidence for a heritable basis underlying mortality in protochordates, unlinked to reproductive effort and other life history traits of this species.

EXPERIMENTAL PROCEDURES

The colonial ascidian Botryllus schlosseri (Tunicata, Ascidiacea) is a cosmopolitan filter-feeding metazoan inhabitant of shallow waters and harbors throughout the world (1). Following settlement, the free-swimming chordate tadpole metamorphoses to a founder individual, the oozooid. Colonies of genetically identical zooids subsequently develop by weekly cycles of asexual budding (blastogenesis), typically forming star-shaped modules called systems, which are embedded in a translucent, gelatinous matrix (tunic) (2). A common vascular network flows between systems comprised of blood vessels connecting individual zooids and terminating into ampullae at the periphery of the colony. Each blastogenic cycle culminates in a phase of programmed cell and module (zooid) death called takeover, in which all zooids in a single colony die and are replaced by a new generation ofzooids (3). Botryllid ascidians possess, as well, a unique histocompatibility system. When two genotypically different laboratory or field colonies come in contact, they either fuse with or reject each other (4). This fusibility/histocompatibility discrimination is controlled by a single gene locus or haplotype (Fu/HC; ref. 5) with multiple codominantly expressed alleles (6, 7). After the establishment of a common vascular system between a fusible pair reared in the laboratory, one member of the chimera often is resorbed by its partner (colony resorption) (8-10). During the course of experiments using animals born and reared in the laboratory (11, 12), colonies were separated (subcloned) into several ramets (clonal replicates), each

Animals. Laboratory mariculture and mating procedures were carried out as described (8-10). The collection of oozoids was terminated 2 weeks after the release of the first hatch and resulted in the collection of 216 offspring. However, complete data are available on only 191 colonies, of which 61 were subcloned. In five of these colonies, only one ramet was subcloned. In 15 others, all ramets belonging to the genet died within 3 weeks of subcloning. The remaining 41 subcloned genets were then divided each into 3-11 ramets and distributed onto 218 glass slides. Of the original 216 colonies, 130 colonies were not subcloned. Of these, 77 died within less than 4 months and were excluded from analysis, since all subcloned colonies were long-lived (mostly subcloned only after the age of 4 months). Observations were carried out once weekly. A large Botryllus colony consists of several systems of zooids and can be easily subcloned repeatedly to several ramets by using the colony allorecognition assay (5). Each ramet was attached to a separate glass slide, and several hundred ramets of dozens of genets were maintained simultaneously. Dates of subcloning, colony number, number of ramets, and dates of death of each ramet were recorded. Colonies were subcloned on specific dates, and varying numbers of ramets were obtained according to colony size and availability. Lastly, vascular chimeras were generated by placing different but fusible genets side by side as described in the colony allorecognition assay (5, 8-10). HIsogy and Immunofuorescence Microscopy. Normal and senescent colonies of B. schlosseri were cut by a razor blade into systems with accompanying tunic and rinsed briefly in a mixture of 0.1 M sodium phosphate buffer and calcium-free artificial sea water at pH 7.0 (buffer A). The specimens were then fixed in periodate/lysine/paraformaldehyde (13) for 2.5 h at 4°C, rinsed in buffer A, dehydrated in ethanol, and embedded in JB-4 plastic (Polysciences) for 24

The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. §1734 solely to indicate this fact.

§Present address: Albany Medical College, Department of Pediatrics, Room ME 508, Albany, NY 12208. 1ITo whom reprint requests should be sent at the present address. 3546

Evolution: Rinkevich

et

al.

h. Sections 3 gm thick cut along the anteroposterior axis of Botryllus zooids were generated with a glass knife on a Reichert ultramicrotome. These sections were stained with a 0.05% toluidine blue/2.5% sodium bicarbonate solution for histological observations. B3F12.9 monoclonal antibody staining of senescent and normal colonies was done on representative sections from the dorsal and ventral planes of specimens. Plastic sections were incubated in a 0.1% trypsin/ 0.1% calcium chloride/20 mM Tris*HCl, pH 7.8 solution for 15 min at 370C to reexpose antigenic sites and rinsed briefly in 0.1 M sodium phosphate buffer at pH 7.5 (buff&i B). Sections were then incubated with the B3F12.9 antibody (IgM isotype, 1:1000 dilution of ascites fluid) diluted in buffer B, along with 5% normal goat serum and 0.1% gelatin for 90 min at 370C. After several washes in buffer B, sections were then labeled with a Texas Red-conjugated secondary goat anti-mouse IgM antibody (diluted 1:200 in the same buffer as B3F12.9) for 1 h at 370C. After three 5-min rinses in buffer B, the sections were dried and mounted with a solution of 90%o glycerol/10%o buffer B/1,4-diazabicyclo-octane (Sigma) at 100 mg/ml, and examined under an Olympus epifluorescence microscope.

RESULTS in Senescent Colonies ofB. schlosAlterations Morphological seri. The death of individual botryllid zooids commonly occurs abruptly at the takeover stage of the blastogenic cycle. In this process, only blastozooids from the previous generation die; morphologically, the common event appears to be a predictable topographic and temporal apoptosis of visceral tissues (R.J.L., C. W. Patton, and I.L.W., unpublished results). In contrast, the senescence of entire botryllid colonies or their subclones documented here involved the death of all zooids and buds. This ramet-wide event was preceded by a characteristic pattern of morphologic changes, which included (i) constriction of blood vessels and ampullae, resulting in a sluggish blood flow between zooid systems and within zooids, and (ii) zooid shrinkage followed by a dense accumulation of pigment cells and other blood cells in both zooids and ampullae (Fig. 1 A and B). Zooids gradually collapsed, and systems became disorganized; they lost their star-shaped configuration and vascular connections with neighbors (Fig. 1B). When senescence occurred during the takeover phase of the blastogenic cycle, parental zooids never completely resorbed; in contrast, in nonsenescent colonies at takeover, parental zooids regressed simultaneously as the new buds grew (data not shown). The tunic of senescent colonies was soon exploited by fouling organisms and gradually disintegrated. Histological sections revealed a generalized compaction of visceral organs spanning the dorso-ventral plane of the zooid and congestion of the peribranchial cavity and blood lacunae by blood cells and degenerating tissues (Fig. 1 C and D). Transmission electron microscopy demonstrated that cell death within visceral organs occurred principally by necrosis: the cells swelled and cell membranes subsequently ruptured (R.J.L. and I.L.W., unpublished results). In zooids of nonsenescent colonies, blood cells were confined to the blood space separating the epidermis from the perivisceral epithelium and to the blood lacunae of the viscera (Fig. 1C). During senescence, the perivisceral epithelium gradually degenerated, allowing the influx of blood cells and dying tissues into the peribranchial cavity (Fig. 1D). This was further demonstrated by immunostaining of tissue sections from senescent and nonsenescent colonies with a monoclonal antibody (B3F12.9) that has been shown to localize to the extracellular matrix of the perivisceral epithehum and on the surface of blood cells (R.J.L., K. J. Ishizuka, and I.L.W., unpublished results). Zooids of nonsenescent colonies labeled with B3F12.9 displayed a uniform immu-

Proc. Natl. Acad. Sci. USA 89 (1992)

3547

nostaining pattern of the perivisceral epithelium along their anteroposterior axis (Fig. 1E). In contrast, animals undergoing senescent regression gradually lost this characteristic staining pattern along the entire anteroposterior axis (Fig. 1F). These results suggested that extracellular matrix breakdown was a hailniark of senescence. The Heritability of Senescence Among Ramets. To test whether senescence occurred randomly or not, we prepared defined crosses of B. schlosseri and compared life spans among ramets subcloned from different colonies (genets) and among subcloned versus intact colonies. Forty-one colonies were divided into 3-11 ramets apiece. The life spans of ramets from the 41 different genets are shown in Fig. 2 and Table 1. Another 53 colonies were not subcloned and served as controls. The mean life span of laboratory-maintained control colonies was -9 months (273 + 117 days), whereas the original systems of the 41 subcloned colonies had a mean life span of 10 months (308 + 104 days). The longevity of these two groups of colonies (P > 0.15; single classification analysis of variance) as well as variances of the two groups (P > 0.25; F distribution, two-tailed test) did not differ significantly, indicating that the subcloning procedure did not affect survivorship and that both groups of colonies were equally variable in their life span. This allowed us to use the original (and still-maintained) parents of the 41 subcloned colonies as controls to determine whether mortality within a set of ramets derived from a single genet was significantly clustered. We calculated the Mann-Whitney rank sum test to compare absolute deviations of life spans within each sample of ramets from a single genet to the absolute deviation from the mean of the 41 individual genets. If the life span of all ramets of a single genet is nonrandom, as would be the case if it were under genetic control, then the variation in life spans among these ramets should be significantly smaller than the variation of life span among ramets of different genets. Genets that demonstrated random mortality (P > 0.05; Mann-Whitney) are shown in Table 1 and Fig. 2A, whereas the remainder, exhibiting nonrandom mortality (P < 0.05), are presented in Fig. 2B. These results may also be expressed graphically in terms of the standard deviation as a function of the mean life span of ramets. The data shown in Fig. 3 demonstrate that in 17 of the 41 subcloned colonies (41.5%, P < 0.05; Fig. 2B) ramets displayed nonrandom mortality; the "within-genet" variation in life span was significantly smaller than genet-togenet variation. Our results fell into a bimodal distribution, with some overlap between nonrandom and random mortality. Interestingly, even some of the statistically random colonies showed clustered mortality (e.g., nos. 7, 32, and 112 in Fig. 2A). Most strikingly, the low mode of variation was concentrated in those genets with either low or high mean ramet life spans (Fig. 3). Furthermore, in several colonies (nos. 53, 63, 104, 115, and 183 in Fig. 2B), all ramets, including the original unfragmented parent colony, underwent senescence simultaneously and died within a span of 7-10 days; this happened up to 5 months after subcloning. In two other cases (colonies 18 and 72 in Table 1), the early death of a single ramet resulted in a higher coefficient of variation. In the last group of 21 subcloned genets (no. 155, Fig. 2A and Table 1; P > 0.05), the life span of ramets from separate genets was clearly variable. The Onset of Senescence in Vascular Chimeras. The expression of nonrandom mortality was further tested under controlled laboratory mariculture in experimental vascular chimeras comprising senescent and nonsenescent genets. Could a parabiotic union potentially extend the life span of the aging colony or, in contrast, induce mortality in the nonsenescent partner? Mortality studies were performed involving eight different original colonies (genets), wherein the life spans of ramets alone and in chimeric unions were recorded. The

3548

Evolution: Rinkevich et al.

Proc. Natl. Acad. Sci. USA 89 (1992)

FIG. 1. The morphology of senescence in B. schlosseri. Nonsenescent (A, C, and E) and senescent (B, D, and F) ramets from the same genet were processed for histological sectioning and also for immunohistochemical staining with a monoclonal antibody specific to the perivisceral epithelium of botryllid ascidians (B3F12.9). The nonsenescent ramet died 12 days later. (A and B) Depiction of the colonial morphology of normal and senescent zooids (and buds), respectively, in the active feeding phase of blastogenesis. Note the constriction and congestion of blood vessels in the senescent colony. Zooid morphology and configuration of star-shaped systems are also greatly altered in the latter. (C and D) Depiction of the histological architecture of normal and senescent zooids, respectively, stained with toluidine blue. Note the generalized visceral necrosis (arrowheads) and blood cell congestion in the lacunae and peribranchial cavity of senescent zooios. The perivisceral epithelium is also undergoing degenerative changes: it swells and loses its squamous morphology (compare C and D). (E and F) Depiction of immunohistochemical profiles of normal and senescent zooids, respectively, as determined by immunofluorescence microscopy. The B3F12.9 monoclonal antibody localizes to the surface of all blood cells and to the extracellular matrix of the perivisceral epithelium (R.J.L., K. J. Ishizuka, and I.L.W., unpublished results). During the growth phase of the asexual cycle, normal zooids display a uniform pattern of staining of the perivisceral epithelium along their anteroposterior axis (arrows in E). On the other hand, senescent zooids exhibit a gradual loss of this staining pattern (arrows in F). bl, Blood lacuna; bK, blood vessel; b, bud; end, endostyle; epi, epidermis; h, heart; int, intestine; pv, perivisceral epithelium; sto, stomach; t, tunic; z, zooid. (A and B, bar = 1 mm; C-F, bar = 100 ,um.)

results are shown in Table 2. Nonrandom mortality was expressed simultaneously in all ramets belonging to the specific genet first undergoing senescence, whether those ramets were growing alone or were combined witi other ramets in chimeras. Both partners of the chimera died at this time, whereas isolated control ramets of the nonsenescent genet survived. In this small sample, the induced mortality correlated with the outcome of colony resorption (5) in two of three sets of experimental chimeras: the ramets of genet 1 determined chimera mortality, although earlier studies on the same pairs of ramets from these two genets had established that colony resorption consistently was of genet 2 by genet 1. In other words, genet 1 dominated genet 2 in these chimeras.

However, ramets from genet 3 resorbed genet 4, and ramets of genet 5 resorbed genet 6; in both instances, the chimera's mortality was determined by its subordinate genet. The only two chimeras in which one partner survived the death of the other (Table 2) were those in which the partners disconnected about 1 month prior to death, thereby precluding colony resorption. Although these experiments do not rule out the possibility that chimeric death resulted from colony resorption, this appears unlikely. The phenotypes of independent senescent subcloned colonies and that of chimeras (data not shown) were identical to other colonies that underwent senescence and died at that time (Fig. 1). Furthermore, the salient features of senescence and mortality were easily

Evolution: Rinkevich et al.

Proc. Nail. Acad. Sci. USA 89 (1992)

A 268 ± 101 $

112

7

155

32 332± 50 $

350±119 $

L 406±62 $ I

I--

--E B 13

402 _±53

_±1 * 7878226e6 523 ± 3 * * L-

'

104

210±0 - t 39 393+38*-

556 ±17 106 4

28

40

l

*;492±+0 * ± 125 209 6 * * 191 + 12 * 131 Lt

219 ± 19 * *

*1t

L-

51 187± 1 . 53 498±0 . .

162

487 ± 18 ...

**

487 ± 18

429± 27 77

--I

183

0 2 4 6 8 10 12 14 16 18 AGE (months)

353+ (0

E

.

1--

L

0 2 4 6 8 10 12 14 16 18 AGE (months)

FIG. 2. Longevity of ramets subcloned from four representative colonies ofB. schlosseri undergoing random mortality (A) and the 17 genets expressing nonrandom mortality (B). The colony numbers are marked on the left margins of the lines. The original part of each colony is marked by a thick horizontal line; a ramet subcloned from the original part, by a thin line; and the secondary ramet subcloned from a ramet, by a thinner line. When more than one ramet was subcloned on a given day, their longevities are marked by parallel horizontal lines originating from one perpendicular line. The righthand end of each horizontal line represents the age at death of each ramet derived from a specific genet. Numbers above each colony represent the average age ± SD of all ramets belonging to a specific genet. Levels of significance between the absolute deviations in mortalities of ramets from each genet, compared to the sample mean of the 41 parent genets (see text for further details), are *, P < 0.05; **, P < 0.01; ***, P < 0.001; $, not significant (P > 0.05).

distinguished from those of colony resorption: after resorption, one of the partners survived while the zooids of the other partner were eliminated, and the surviving partner retained the tunic's functional blood vessels and ampullae of the resorbed genet (8). Senescent chimeras, in contrast, underwent overall collapse as one partner died.

3549

Table 1. Life span of genets undergoing random mortality Life span of Life span of ramets, days Colony original part, Colony original no. days Range Mean ± SD (n) 7 210 90-345 268 ± 101 (10) 18 325 273-465 413 ± 93 (4) 23 225 192-440 318 ± 92 (5) 31 216-495 1% 393 ± 110 (6) 32 271 260-372 332 ± 50 (5) 35 205 263-433 329 ± 65 (9) 41 203 269-465 356 ± 83 (5) 46 227 186-403 261 ± 97 (4) 59 420 361-480 420 ± 60 (3) 68 233 266-475 369 ± 111 (4) 72 501 205-501 438 ± 104 (7) 95 380 183-302 255 ± 63 (3) 98 288 104-469 359 ± 138 (6) 103 302 323-469 396 ± 103 (2) 109 327 175-297 236 ± 86 (2) 112 455 210-465 350 ± 119 (7) 128 252 180-439 310 ± 183 (2) 144 198 200-429 278 ± 131 (2) 155 453 290-453 406 ± 62 (7) 159 220 212-441 278 ± 110 (4) 171 345 302-445 374 ± 101 (2) 176 339 411 ± 101 (2) 339-482 182 259 213-503 366 ± 102 (7) 186 389 321 ± 153 (2) 213-429 The absolute deviation in life span between ramets from different genets is not significantly different from the mean life span of the 41 original parent colonies (P > 0.05; Mann-Whitney rank sum test on absolute deviations from the sample mean).

tunately, the onset of senescence precludes further fertility of the genet. Therefore, a prospective mating study will be required to test whether the trait is transmitted from genet to genet in Mendelian fashion, as well as within a genet by asexual proliferation of ramets.

(n

0 %..180-

0

z

0

0

!z1200

OD

0 0

000 0 0

rv60-

0 0

0cno 0

2

2. 0 *0

0

0

z c/)

*

210

*

m

490

560

0

0 0

0-

280

350

420

MEAN LIFESPAN (DAYS)

DISCUSSION Bancroft (4) originally reported that field colonies of B. schlosseri exhibited a series of regressive changes prior to death, including decreased budding potential and shrinkage of individual zooids. Here, we confirm this observation and in addition show that several morphological characteristics are reproducibly associated with senescent laboratory-reared ramets and vascular chimeras. Most importantly, our findings led us to conclude that in a significant fraction of genets a heritable program readout in all asexually derived ramets may determine the onset of senescence and death in this animal. These experiments do not prove that the heritable trait is transmitted sexually via chromosomal genes. Unfor-

FIG. 3. Standard deviation (in days) of ramets from the 41 genets expressed as a function of the mean life span (in days) of ramets. Genets undergoing nonrandom (17 of 41 genets; P < 0.05) (e) and random (P > 0.05) (o) mortality were defined as such by using the Mann-Whitney rank sum test to compare absolute deviations within groups of ramets from single genets to the absolute deviations from the sample mean (41 individual genets). Note that random and nonrandom senescent genets are bimodally distributed, although there is some overlap. Most significantly, however, genets having standard deviation values of zero are endowed with both low and high life spans. Because of the large sample size, ramets from several colonies were placed in different tanks and maintained under various conditions (field vs. laboratory, running sea water tanks vs. standing sea water tanks). The open circles with the number 2 inside indicate two ramets with the same values.

3550

Evolution: Rinkevich et al.

Proc. Natl. Acad. Sci. USA 89 (1992)

Table 2. The onset of senescence in Botryllus chimeras occurs at the time of the earliest senescent partner Time to mortality, weeks Genet 1 vs. genet 2 Genet 3 vs. genet 4 Genet 5 vs. genet 6 Genet 7 vs. genet 8 Chimera/ ramet number Chimeras Gir G2r Chimeras G3r G4r Chimeras GSr G6r Chimeras G8r G7r 1 4.5 4.5 >16.5 17.5* 5 >26 17.5 5 >13 3 >11 3 2 4.5 4.5 >16.5 5 17.5* >26 5 17.5 >13 3 >11 3 3 5 4.5 4.5 >16.5 17.5 >26 17.5 5 >13 4 4.5 4.5 >16.5 17.5 5 >13 5 5 4.5 >16.5 Age of genet 40 16 36 16 10 36 12 28 The time to mortality represents the time from the day the assay was initiated (postfusion in the case of chimeras). The age of each genet (in weeks) when the experiments were initiated is indicated at the bottom of the table. Within the four experimental groups of chimeras, genet 1 resorbed genet 2, genet 3 resorbed genet 4, and genet S resorbed genet 6. In genet 7 and 8 chimeras, the outcome of resorption could not be determined. "G" and "r" are abbreviations for genet and ramet, respectively (e.g., Glr refers to ramets of genet 1). In all four groups, the nonsenescent controls were either used for other experiments not reported in this manuscript or were discarded. *These experimental chimeras disconnected 82 and 93 days postfusion, respectively. The genet 4 partners disconnected and died at 17.5 weeks postfusion, whereas genet 3 partners survived and were followed for an additional period of 8 weeks.

An intriguing result presented here is that, under controlled laboratory conditions, the senescent partner in a chimera between senescent and nonsenescent genets induced mortality in the nonsenescent partner (Table 2). These findings suggest that organismal senescence in Botryllus is a dominant trait, independent of reproductive effort. Colonies of B. schlosseri populations from Woods Hole (Massachusetts) may be either semelparous (death occurs immediately after sex) or iteroparous (colonies reproduce several times) (14). Laboratory and field experiments as well as mating data indicate that there is a genetic component to this life history polymorphism, the traits of which could be transmitted to offspring (14). The Monterey (California) colonies followed here (controls and clonal replicates) were strictly iteroparous: mortality occurred independently of reproductive output. Botryllus colonies grow by blastogenesis, which includes a recurrent process of programmed cell death called takeover. This developmentally regulated event involves the simultaneous and selective regression of all "parent" zooids every 6 days at 18°C (2, 3), in turn leaving their asexually budded progeny to mature. Recent findings indicate that the principal mode of cell death during takeover occurs via apoptosis (R.J.L., C. W. Patton, and I.L.W., unpublished results). The morphology and kinetics of cell death within senescent ramets and chimeras reported here were distinct from those occurring in either takeover or colony resorption (R.J.L., unpublished observations). These findings suggest that zooidal life span and colonial life span may be under separate genetic control. Several observations indicate that colonies in field populations of B. schlosseri from Monterey Bay (California) (15), the Venetian lagoon (Italy) (16), and the Mediterranean coast of Israel (B.R., unpublished observations) may also exhibit finite life spans. Colonies whose founding larvae settle during the summer live about 3 months and then develop regressive features identical to those described here (15, 16). Under laboratory mariculture, the ease with which clonal replicates of Botryllus are obtained may render this colonial ascidian a promising model system in which molecular components regulating mortality and senescence may be charac-

terized. The observation that the majority of genets (24 of 41) did exhibit a somewhat random array of deaths suggests that the heritability of senescence could be complex. But this very complexity of control of senescence should be elucidated in genetically inbred lines of B. schlosseri, as these become available. B.R. and R.J.L. contributed equally to this manuscript. We thank K. Ishizuka and K. Palmeri for their tedious task of animal maintenance; Chris Patton for photographic assistance; and S. Sorger, N. Chadwick, K. Palmeri, and K. Ishizuka for critical reading of the manuscript. B.R. was a Lucille P. Markey fellow of the Life Sciences Research Foundation. R.J.L. was the recipient of a Medical Research Council of Canada fellowship. This study was supported by U.S. Public Health Service Grants GM 25902 and CA 42551 to I.L.W., by a grant from the Israel Academy of Sciences and Humanities to B.R., and by a Career Development Grant from the Israel Cancer Research-U.S. Foundation to B.R.

1. Kott, P. (1985) Mem. Queensl. Mus. 23, 1-440. 2. Sabbadin, A. (1978) in Marine Organisms: Genetics, Ecology and Evolution, eds. Battaglia, B. & Beardmore, J. A. (Plenum, New York), pp. 195-208. 3. Mukai, H. & Watanabe, H. (1976) J. Morphol. 148, 337-362. 4. Bancroft, F. W. (1903) Proc. Cal. Acad. Sci. (3rd Ser.) 3, 137-186. 5. Weissman, I. L., Saito, Y. & Rinkevich, B. (1990) Immunol. Rev. 113, 227-241. 6. Scofield, V. L., Schlumpberger, J. M., West, L. A. & Weissman, I. L. (1982) Nature (London) 240, 499-502. 7. Taneda, Y., Saito, Y. & Watanabe, H. (1985) Zool. Sci. 2, 433-442. 8. Rinkevich, B. & Weissman, I. L. (1987) J. Zool. 213, 717-733. 9. Rinkevich, B. & Weissman, I. L. (1987) Biol. Bull. 173, 474488. 10. Rinkevich, B. & Weissman, I. L. (1987) Symbiosis 4, 117-134. 11. Sabbadin, A. (1969) Publ. Staz. Zool. Napoli. 37, Suppl. 62-72. 12. Boyd, H. C., Brown, S. K., Harp, J. A. & Weissman, I. L. (1986) Biol. Bull. 170, 91-109. 13. McClean, I. W. & Nakane, P. K. (1974) J. Histochem. Cytochem. 22, 1077-1083. 14. Grosberg, R. K. (1988) Evolution 42, 900-920. 15. Carwile, A. H. (1989) Ph.D. thesis (Univ. of California, Los

Angeles). 16. Brunetti, R. (1974) Boll. Zool. 41, 225-251.