Compartments and Polyclones in Insect Development Clones made in early development keep within certain fixed boundaries in the insect epithelium. F. H. C. Crick and P. A. Lawrence In this article our aim is to describe re- :ent work on the development of intact spithelia and in particular the important results and ideas of Professor Antonio Garcia-Bellido (1) and his group in Madrid which are not yet widely known. We try to explain as clearly as possible what these ideas are and what sort of experiments have been done to support them. Some of the more obvious questions arising from tbe results and how the new concepts may relate to other ideas such as "gradients" are listed. Development of DrosopLiln The development of an adult Drosophila is a complex process. The nucleus of the fertilized egg divides a number of times to form a compact mass of about 250 nuclei, near the center of the egg, without cell walls. These nuclei then migrate outward to the inner surface of the egg where for the first time cell membranes are formed. The cells divide several more times to form a single layer of cells, about 4000 in all, lin- Drs. Crick and Lawrence work in the Cell Biology Division of the Medical Research Council Laboratory of Molecular Biology, Cambridge, CB2 2QH. En- gland. 340 ing the inside of the egg. This is called the blastoderm. Behaving as a sheet of cells, the blastoderm undergoes complex folding movements generating a multilayered germ band, which soon becomes visibly segmented. The egg hatches after 24 hours and the animal then goes through three lar- val stages each separated by a molt. After these larval stages, lasting in all about 96 hours, the animal then pupates and meta- morphoses into the adult fly. This adult is formed mainly from special groups of cells in the larva which them- selves take little or no part in larval devel- opment or function. These are the histo- blasts and the imaginal discs. There are 19 of the latter (nine pairs of discs plus the single genital disc). We shall concentrate mainly on one pair of these, the so-called wing disc. The left wing disc, within the left side of the larva, produces the left wing of the insect and that part of the dorsal left side of the thorax next to the wing. The wing disc is seen in the first larval stage as a small patch of embryonic epider- ma1 cells (2). These cells remain diploid, while the surrounding larval epidermal cells become polyploid (3). There are prob- ably only about 15 to 30 cells forming the wing disc at this early stage (46). During the course of larval growth these disc cells divide in all about 10 or I I times (on aver- age) to give a total of some 50,000 cells (5). Shortly after puparium formation cell divi- sion of the disc stops. The disc has now a characteristic size and shape, being some- what like a flattened and heavily folded balloon (7). At metamorphosis a complicated set of cell movements occurs, and these result in the disc being turned inside out so that it can form the adult structure. The wing it- self, for example, is first formed as a bag. The bag is then collapsed to form the adult wing, which thus becomes a single sheet of epithelial cells folded and collapsed to form a double layer of epithelial cells. Basic Ideas of Clonal Analysis For the purposes of exposition we now temporarily leave the wing and describe a hypothetical sheet of "white" epithelial cells on the adult fly. We imagine that we have at our disposal a special technique that enables us to mark (say black), at ran- dom, a single cell in a developing disc. The mark is such that it does not interfere in any way with the normal development of the animal. Moreover, all the descendants of this marked cell retain the mark and can be recognized in the adult. The method of marking has the advantage that we can choose fairly precisely when, in devel- opment. we mark the cell; but it has the disadvantage that we cannot mark a par- ticular cell at that time, but only one cho- sen at random, and in early stages we usu- ally mark only one cell in any one individ- ual. If we assume that the significant fea- tures of the process are effectively the same in all individuals, we can piece together what is happening in development by com- bining experiments on many different indi- viduals. What do we find? Naturally, we see a set of black cells in the adult, but how many of them are there, and how are they ar- ranged? The first observation is what might be SCIENCE. VOL. 189 Fig. 1. A clone descending from a cell marked prior to the formation of the compartment border (XY). The clone is smooth at the edge of the structure (a) but rough elsewhere (6). Fig. 2. A clone descending from a cell marked after the formation of the compartment border. Fig. 3. The clones made by the two daughter cells of that cell generating the clone shown in Fig. 1. Fig. 4. The clone made by a cell marked prior to the formation of the compartment horder, both of whose daughter cells give rise to clones within one subcompartment. expected. In general, the earlier a cell is marked in development, the more black cells we find in the adult. A cell marked early leaves more descendants than a cell marked late. The next obvious question is: What frac- tion of the total cells are marked? By the total cells we mean the number of cells in that portion of the adult epithelium which has come from the set of cells under con- sideration in the larva (for example, the 50,000 epithelial cells that come from a single wing disc). The number of black cells produced by marking at a fixed time is not exactly con- stant, but the variation is such that we can usefully calculate its average value. If the average number of black cells in the adult is, say, a tenth of the total then making cer- tain reasonable assumptions there were, on average, about ten cells in the larval set at the time they were marked (8). As the time of marking gets later and later in devel- opment this fraction gets smaller and smaller, and the frequency of marked clones produced increases. From the arrangement of the black cells we can learn something about their move- ment during the interval between irradia- tion and observation. For instance, if there is a pepper and salt mixture, the cells must have been intermingling; while a coherent patch suggests that all the daughter cells have remained in contact during growth. The shape of the patch is also informative. For example, if it is long and thin this may result from the cell divisions being pre- dominantly oriented in one direction. In the case of the wing disc, it is found that the patches are usually both coherent and elongated so that the long axes of the patches are parallel to the long axis of the wing (4,5. 9). We must next ask: Even though a patch is irregularly shaped, is the shape the same in different individuals? The experimental results show that it is not so. Consider a set of experiments in which the mark was made at more or less the same time in the development of a number of different indi- I AUGUST 1975 viduals. Then it is found that the patches produced, when all drawn on the surface of a single idealized adult, do not neatly cover the entire epithelial surface, without either overlapping or leaving spaces (like a jigsaw puzzle). On the contrary, if two patches from separate individuals have ended up in roughly the same place, then it is always found that each partly overlaps the other and are usually of different size. This result shows that the cell lineage in Drosophila epithelium is not strictly determined in the same way in all individuals. After all these preliminaries we can now approach the important result. Let us as- sume that our hypothetical piece of epithe- lium is smooth in outline, as shown in Fig. I. Then perhaps it is not too surprising to learn that a black patch near the borders of this piece of epithelium has itself a rather smooth outline where it follows the bound- ary of the area but has a rough outline,else- where. We assume that at the earliest stage of marking (that is, when the disc is first formed) a black patch can be pro- duced anywhere within our area. In partic- ular it may have the size and shape shown in Fig. I. We now ask: Suppose the mark is made a little later, say, one cell genera- tion later, what will the patch be like? Naturally, it will, on an average, be half the size, and we expect it to have an ir- regular outline except where it touches the area border. But now in some cases a new and totally surprising restriction appears. When all the results from many different patches are combined, it is found that a rather smooth line (marked Xv) can be drawn, dividing our hypothetical area into two distinct parts, such that no black patch, made at this later time, will ever cross this line. Moreover, the outline of a patch touching this line is smooth where it runs along the line but rough elsewhere (Fig. 2). And this in spite of the fact that a patch marked one cell generation earlier can cross this special line. The surprising nature of this result can be seen by going back and considering the entirely irregular patch illustrated in Fig. I. We drew this particular patch (marked at the earlier stage) across the special boundary XY. We now ask: What would Fig. 1 look like if instead of just marking that particular cell we had been able to put a different mark on each of its two daugh- ter cells, produced one generation later? We should now find two adjacent patches, each with an irregular outline except where the patches touched. Along the line of con- tact their outlines would be smooth and fairly straight (Fig. 3). This result is true only if the double patch crosses the special line XY. Otherwise the contact outline of the two daughter patches would be irregu- lar (Fig. 4). Garcia-Bellido, Ripoll, and Morata (10) have called an area bounded by these special demarcation lines a compartmenf. The progeny of a cell marked at about the time of the drawing of boundary lines never fills a compartment completely, but often occupies an appreciable proportion of it. A compartment is thus made by the descendants of a small group of cells. We propose to call the cells in the compart- ment apolyclone. Just as a clone is a group of cells which are all, without exception, the descendants of a single cell, so a poly- clone is a group of cells that are descended from a certain (small) group of cells-the founder cells-which were present in the embryo at an earlier time. Moreover, in our terminology they are all the (surviving) descendants of that small group. This last point is vital since necessarily all the cells in a compartment are, for example, de- scendants of the fertilized egg. The dis- tinction is that some of the descendants of the egg make up other parts of the body; that is, they end up in other com- partments. The members of a polyclone, however, all fall within one compartment and account for all the cells in that compartment. This point can be made more sharply. Consider the small group of cells, the founder members of the polyclone, and then consider their immediate ancestors. Then (except in rare cases) this earlier 341 group will not form a polyclone for the compartment under consideration. That is, we will usually find that some of the de- scendants of these cells end up outside the compartment we are considering. The cells in the compartment are necessarily all de- scended from this smaller group, but they are not all the surviving descendants. Therefore, this earlier group are not the founder members of the polyclone for that compartment. The other side of the idea must also be mentioned: a compartment is never a clone, except perhaps accidentally in rare cases. That is, for most cases, the cells in a compartment cannot be traced back to any single cell, all of whose descendants fall within the compartment. This idea, which implies that for these properties cells are switched not singly but in groups, is impor- tant (I [). We thus see that the idea of a com- partment and the idea of a polyclone are, at the moment, intimately connected. As things stand at present we have no other reliable criterion for the sharply defined region we call a compartment except that a marked clone produced after a cer- tain time in development will never cross over the compartmental boundary and in- clude any part of any other compartment; whereas clones formed earlier may well do so. Reciprocally, we cannot say that a group of cells form a polyclone unless we first define the compartment to which the polyclone refers. We must now consider the second major fact about certain compartments, namely that as time goes on they become subdi- vided. Let us call a certain compartment camp I; at a later time it will be subdivided into two compartments which we may call camp IA and camp 1B. These two sub- compartments are not necessarily equal ei- ther in area or in number of cells but to- gether they add up exactly to camp 1. By definition all these compartments are polyclones. That is, the ancestors of all the cells contained in each compartment can be traced back to a founder group, early in the embryo, all of whose descendants end up in a compartment being considered. It is an experimental fact that one marked clone of cells, started from a single cell at a certain early stage, may stay entirely within camp I and yet go across the border between camp 1A and camp IB. A marked clone made at a slightly later stage, how- ever, will never cross this boundary. This implies that in any particular case the cells that are the founder members of the poly- clone for camp I form three classes: those whose descendants will fall (i) wholly with- in camp IA, (ii) wholly within camp lB, and (iii) partly into camp 1A and partly into camp IB. 342 It is this third class which explains why early clones can cross a subcompartment boundary whereas later ones cannot. How- ever, at a slightly later stage in embryogen- esis some further developmental step must take place, since at that time the descend- ants of the founder members of camp I will fall strictly into the first two classes listed above. No cell will then be found with the properties of class iii. Every cell in this enlarged group will be either a founder member for camp IA or a founder member for camp IB. In short, whereas before only one polyclone existed, that polyclone can now be considered as the sum of two distinct polyclones. The work of Garcia-Bellido and his col- leagues shows that this process of forming subcompartments within larger com- partments can happen several times in suc- cession. The data suggest, but do not prove, that the division takes place each time into just two parts. The Methods We shall now illustrate the methods used in clonal analysis by describing in outline the techniques employed by Gar- cia-Bellido er 01. (IO) in their detailed stud- ies of the wing disc of Drosophila melano- gosfer. The wing disc is strictly called the dorsal mesothoracic disc. There are two of them in each larva, one for each side of the adult animal. Each disc produces the entire epithelium for a wing and that part of one side of the thorax near the wing. The dor- sal part of the thorax is called the notum and the lateral part the pleura. The method used to mark a clone is mitotic recombination produced by x-rays delivered at a chosen time in development, usually during the larval stages. The ge- netic makeup of the animal is designed so that certain mitotic recombinants will be phenotypically different. For example, if the animal is heterozygous for the reces- sive gene yellow (y/ +) then mitotic recom- bination may produce two daughter cells. One of these (+/+), will be phenotypically wild-type and therefore indistinguishable from unaltered cells, but the other will be homozygous for yellow b/y). All the de- scendants of this cell will also be (y/y). If such a descendant in the adult is colored at all then it will be yellow rather than the normal darker color. An ideal genetic marker would be easily scored in all types of cell, have complete expression, and be cell-autonomous. That is. the phenotype would depend only on the genetic makeup of the cell in question and not at all on that of neighboring cells. Un- fortunately few such markers are known. Markers often used are: multiple wing hairs fmwh) which produces groups of two to five hairs (trichomes) on the wing in- stead of one per cell as in the wild type; and forked v) and singed (sn), which produce deformed bristles and hairs. To assist rec- ognition, the mutant allele with the most extreme phenotype among those available is usually used; and to minimize mistakes more than one marker is often employed. Double marking also allows the degree of expression and cell autonomy to be checked. The markers used so far in this work do not allow a marked cell to be recognized when it is first produced in the imaginal disc, or even after a few divisions. The cell phenotypes employed can only be scored by the observer at the adult stage when the cells have digerentiated. Moreover, only cells that form (or can be induced to form) hairs or bristles can be scored at all easily, so that if these are lacking or sparse in some particular area it is often difficult to find the exact edges of a marked clone in such regions. Fortunately most of the wing disc derivatives, being covered with hairs, are relatively easy to score. If the growing disc in the larva is irra- diated at the early stages of development, there will be few target cells and most indi- viduals examined will not show any mutant patches. This cannot be overcome by in- creasing the x-ray dose (which is usually loo0 roentgens) as too big a dose will inter- fere with development. One simply has to examine a fairly large number of flies. If the x-rays are given later in development, more mutant clones are produced (since there are more target cells); but the aver- age size of each clone will be smaller since a cell altered at a later stage produces few descendants. This small clone size means that it is more difficult to recognize com- partment boundaries since most of the clones will be in the middle of a com- partment rather than near its edge and even those at the boundary, being small, will not display the boundary so graph- ically. This is somewhat offset by the subdivisions making the compartments smaller as time goes on but in spite of this it becomes progressively more difficult to recognize compartment boundaries. It would in any case involve much more work if enough patches are to be scored to make an apparent boundary statistically signifi- cant. However, Garcia-Bellido er al. devised a method of overcoming this difficulty. There exist a series of dominant Minute loci (12) which are lethal when homo- zygous. When heterozygous, the insects grow slowly and the bristles are small. They needed a mutant which (after mitotic recombination) would make the marked clone grow faster than the unaltered cells SCIENCE, VOL. 189 , 05mm , , 05mm , Fig. 5 (left). Drawings of Drosophila wing to show the position of the aotero-posterior compartment border. Fig. 6 (right). Outline drawing of Dm- sophila wing to show the area covered by a typical M+/M+ clone in a M/M+ background. and thus produce a much bigger patch. Morata and Ripoll (13) showed that homozygous wild-type cells (M+ /M +) produced by mitotic recombination di- vided more rapidly than the slow-growing heterozygous Minure (M/&f+) back- ground, which was the effect they needed. in addition, for reasons which are obscure, the frequency of mitotic recombination for (M/M+) larvae after irradiation is appar- ently increased (14). This is especially use- ful in the early stages of development when the normal rate is inconveniently low. A second somewhat unexpected result was that in spite of the (M+/M f) clones being much larger than normal, the overall size and shape of the wing was not altered (13). This implies that there are special mecha- nisms to regulate size and shape which can cope with differential cell division rates- an important result in its own right. These mechanisms can also regulate for the loss of cells both due to x-rays and the forma- tion of M/M cells. The Results Having given an indication of the meth- ods used in this type of clonal analysis we must now mention some of the earlier re- sults. Becker (15) was the first to use x-rays to produce clones at particular stages of development, in his study of the Drosoph- ila eye. Later Garcia-Bellido (9) noted that clones produced after the 1st instar larva never crossed from dorsal to ventral on the wing; and Bryant and Schneider- man (8) that they were confined to single leg segments when larvae older than early third-stage larva were irradiated. Bryant (4) made the important observation that clones in the wing disc could cross from dorsal to ventral if produced early enough, but not when produced late. Similarly, in OncopelU.s (16-18) up until the late blastoderm stage, clones may extend to two or more abdominal segments, but after that stage clones are strictly confined to a I AUGUST 1975 single segment. These observations all show that within three different discs of Drosophila and in the Oncopeltus abdo- men the "anlagen are represented by sepa- rate populations of proliferating cells" (4). The most detailed results so far have been obtained by Garcia-Bellido and his colleagues studying the development of the wing disc. As might have been expected the earliest clones (irradiation of first-stage larvae) are contained exclusively within the fairly large area of the adult cuticle pro- duced by the entire disc. This shows that effective separation of the wing disc from the other discs producing the adult epithe- lium must have occurred before the first larval stage. However, even at these early times a compartment boundary is appar- ent within the disc. This was first clearly demonstrated by the Madrid school using the Minute technique. The boundary, which separates anterior regions from posterior regions, runs along the middle of the wing between the third and fourth vein. The actual demarcation line is near the fourth vein but is distinct from it (Fig. 5). The line runs along both surfaces of the wing and continues on the body where it divides the notum into two distinct areas. Even a very large clone (Fig. 6) will ob- serve this demarcation line although at this stage it may well cross the wing mar- gins, thus appearing on both dorsal and ventral surfaces and extending onto the notum. The edges of the clone are somewhat ir- regular except where they run along the de- marcation line. It is not very likely that this line marks the frontier where two ini- tially remote and separate groups of cells have moved together, since both anterior and posterior regions are within the same nascent imaginal disc and thus proba- bly fairly close together (6). Since about twice as many clones appear in the an- terior compartment as in the posterior one, it is surmised that at this early time there are about twice as many anterior as posterior cells. That is, the antero- posterior division is not exactly into two equal parts but more like a 2 : 1 ratio (IO). Some time later, during larval devel- opment (the exact time is not quite clear), each of these two compartments is found to be divided into four parts, giving eight compartments in all. The demarcation lines divide dorsal from ventral areas and wing from thorax. The final size of these compartment areas varies somewhat (from IO' cells to 10' cells or less). The evidence that late clones really observe these demar- cation lines is very strong. They are ob- served by very large clones, which in some cases make up as much as 90 percent of a compartment. Such clones may border a demarcation line for as many as a thou- sand cells. Nor is the effect solely due to the fact that clones are often elongated in a direction roughly parallel to a demarcation line. The main axis of these clones meets the demarcation line at various angles, sometimes even perpendicularly. Nor on any simple model can the demarcation lines be lines of fusion of quite separate groups of cells if only for the fact that marked clones made at a slightly earlier stage will go straight across these lines. As development proceeds the recogni- tion of new subcompartments again be- comes somewhat more difficult because the effects of differential growth (due to M+/M+ cells in a M/M+ background) have less time to produce larger clones. Garcia-Bellido, Ripoll, and Morata sug- gest that there may be two further demar- cation lines formed about the same time. On the adult fly these separate two areas on the body, one dividing the notum into two parts and the other the pleura. These compartments were all discovered by the use of MJM+ flies, but similar experi- ments on non-Minute flies (which have, of course, smaller clones), show that the de- marcation lines are also observed in this more normal situation. The Minute flies thus serve to make the subcompartments more easy to observe: the phenomenon it- self is not peculiar to them alone. 343 Further Problems Having now described the results on compartments in outline we must ask how widely the idea is applicable and what are its limitations. One limitation is that the evidence obtained so far relates only to epi- dermal structures. This is mainly because in insects they are so easy to observe and so rich in detail. Internal structures, for ex- ample, the exact arrangement of the mus- cles, cannot be studied satisfactorily with- out the use of more difficult experimental methods. However, the properties of internal tis- sues may be partly imposed by the pattern of the enclosing epithelium (19) and they may well also be compartmented. With regard to compartments in imagi- nal discs, there are a series of outstanding questions that need answering. Are all sub- divisions binary? We have seen how the first division of the wing disc, after the very early antero-posterior divisions, appears to yield four parts rather than two. It is natu- ral to ask if this is really two separate bi- nary steps in quick succession, and this question focuses attention on the exact timing of the subdivisions. Even for an ob- viously binary step one can ask whether the decision is an abrupt one or is spread over a period. Does it necessarily require cell division? Are compartment boundaries always smooth? The edge between the dor- sal and ventral surface of the wing is very well defined, and clones that border it are smooth to the nearest cell (4, 5). but is this true for all boundaries? The problem of how a compartment boundary is formed and how it gets so straight appears to be a dithcult one. Fac- tors that may have to be considered are strictly oriented mitoses near the boundary (17), straightening effects due to differ- ential cell affinities, and possibly cell death for cells which get themselves into the wrong places, so that the compartment edges are trimmed. It is claimed (20) that extensive cell death is unlikely because oth- erwise clone size near the boundary would be smaller, which is apparently not the case. Nor is it completely clear where the process of the subdivision of compart- ments stops. Even the technique for spot- ting compartment boundaries, using rel- atively fast-growing marked clones, has its limitations as, at later times, even these clonal patches are rather small. How can we be sure that these are not further sub- compartments? Even the definition of a compartment becomes difficult at this point. Although formally, for example, the descendants of a single bristle mother cell [for example, the trichogen, the tor- mogen, the sense cell, and the neu- rilemma cell making up a bristle in On- cope/rus (21)] which are most certainly a clone and which stay together, could perhaps be regarded as a compartment, we feel that this is stretching the term too far. It would seem sensible to restrict the term "compartment" for the moment to fairly large groups of cells and to those groups which form a polyclone rather than a clone. Fig. 7. A metathoracic appendage from a Drosophila carrying an extreme allele for bifhorox. The posterior haltere develops normally @) while the anterior haltere is transformed into an apparently normal and complete anterior wing compartment. 344 Other Possible Characteristics of Compartments We have seen that, at the moment, a compartment is defined by its boundaries and these alone, since clones, made after a certain time in development, never cross them. Are there other properties that allow us to identify a compartment? One such property may be the area af- fected by a homeotic mutant. There are mutants that shift an imaginal disc, or part of an imaginal disc, into another devel- opmental pathway. For example, oris- tapedia (ssa) transforms part of the an- tenna into leg segments (22). It is rather rare for a mutant to turn one whole disc into another whole disc. Possi- bly such a drastic change would be lethal and thus escape observation. It is more common for a part of one disc to be turned into part of another one. Even in these cases the transformation is not always complete, because of partial and variable expressivity. We can, however, ask the gen- eral question: In such cases do the (maxi- mum) boundaries of the transformation coincide with a compartment boundary found by the clonal method? Morata and Garcia-Bellido (23) have shown by clonal analysis that the haltere disc (the metathoracic disc) has within it an antero-posterior boundary; but locating it precisely is difficult because of the ab- sence of suitable landmarks on the haltere. It has been known for many years (24) that various mutants in the bithorax system turn various parts of haltere into wing (or vice versa) with different degrees of ex- pressivity. A number of mutants appear to respect the antero-posterior boundary of wing with some precision and probably also of the haltere although here the pre- cision is more difficult to judge. For ex- ample, an extreme allele of bithorux (bx') turns the anterior part of the haltere into anterior wing while leaving the posterior part of the haltere (which is much smaller) unaltered. The boundary of this trans- formed half-wing is very close or identical to the antero-posterior boundary found by clonal methods in the wild-type wing (Fig. 7). Another mutant in this complex locus (postbifhorax) also delineates this bound- ary because its effect is restricted to the posterior part of the haltere. The gene engruiled also delineates the boundary, and in an especially interesting way. In flies mutant for engruiled the pos- terior part of the wing is transformed and resembles a mirror image of the anterior part (25). The Minute technique has re- cently been used to show that the realm of action of the engrailed gene precisely coin- cides with the posterior compartment, SCIENCE. VOL. Is9 there being no effect on the anterior: If large engmifed (en/en) clones are made in a wide-type wing (en/+) they may fill the anterior compartment right up to the an- tero-posterior boundary but never cross it. They are completely without effect on the pattern. However, all engrailed clones in the posterior part express the phenotype (26) and, as discussed later, may cross the antero-posterior boundary. Another possible correlation is between gradient discontinuities and compartment boundaries, These discontinuities can be of at least two kinds. The first has a discon- tinuity in the value of the gradient but not its slope, as shown in Fig. 8. The other has no discontinuity in the value but a change of slope, in particular a change of sign of the slope to give the mirror-image situ- ation shown in Fig. 9. The first of these is found between the segments of the insect cuticle in Rhodnius and Oncopeltus. Lawrence (16, 18) has shown in Oncopeltus that marked clones do not cross the intersegmental boundary, so here at least we have one clear case where a clonal boundary coincides at least approximately with a gradient disconti- nuity (27). Another possible case is sug- gested by the mutant engruiled mentioned above. Since this produces a rough mirror image across the antero-posterior com- partment boundary of the wing, one might be tempted to think that the underlying gradient (or "prepattern") might have the mirror image form shown in Fig. 8 both in the mutant and the wild-type. Otherwise the experimental evidence for this possible correlation is either scanty or absent. There are several other properties which we can speculate about. Experiments de- signed to show how mixtures of cells from imaginal discs appear to sort out show clearly that cells from different discs will segregate, suggesting rather strongly that they have different surface properties (28). Moreover, such segregation also occurs be- tween marked cells from different parts of the same disc. For example, cells from the anterior part of the wing disc will segregate from those of the posterior part (29). This obviously suggests the generalization that each compartment has characteristic cell surface properties, different from every other compartment, which allow cells from any two compartments to segregate. Thus the normal development and maintenance of the antero-posterior boundary in the wing might depend on the confrontation of cells of a different type, that is "anterior" with "posterior" cells. If so, one might ex- pect that boundary to be malformed or nonexistent in engrailed flies where the posterior cells are partially transformed into those of the anterior type. Clonal I AUGUST 1975 A /v (8) (9) Fig. 8. The probable gradient situation in Drosophila wing; the slope, but not the altitude. changes near the antero-posterior compartment border. Fig. 9. The probable gradient situ- ation in two adjacent abdominal segments of Hemiplera. The step probably coincides with the intersegmental compartment boundary. analysis of engrailed flies has recently shown that clones do frequently cross the line where the border normally is (26). This never happens in flies wild-type for the engmiled locus, a result that strongly supports the idea that the role of the en+ allele is both to control the development of the posterior pattern and to instruct the cells so that they do not intermingle with cells of the neighboring anterior we. An additional possibility is that there is a gradient of cell surface properties within each compartment. This is certainly sug- gested by the observation (30) that in the epidermis of Oncopelrus a graft takes bet- ter if it is from the same level in the seg- mental gradient, even if from a different segment, than if moved to a different posi- tion in the gradient in the same segment. These speculations go far beyond the ex- perimental data now available, but they do suggest that direct methods of character- izing cell surface properties, preferably in situ, would be very valuable. If such a method could be developed it would have the enormous advantage that it might work for the cells of the developing imagi- nal disc so that one could spot com- partments and their boundaries at the mo- ment, or soon after, they are formed. It is also possible that. even though all the epithelial cells of a disc appear very similar, the compartments within them could differ by a particular enzyme or set of enzymes. For this reason there is a case for testing all the imaginal discs, both in their mature and their developing states, by as many histochemical tests as are available. A beginning has already been made in this approach by Janning (32) us- ing a test for aldehyde oxidase. Another histological feature that may correlate to some extent with com- partments is the distribution of nerve axons. Hasenfuss (32). studying the epi- dermis of Galleria and Rhyacophila. no- ticed that the nerve axons of the sensillae in the abdominal epidermis were collected into groups each of which went to one seg- mental ganglion only. He suggested that this was because each group came only from a single epidermal segment, This, however, was true only of the axons since the dendrites were observed to extend over considerable distances and thus could not be confined within one segment. A similar phenomenon has also been observed by Lawrence (33) in the abdomen of Onco- peltus. In this case. the intersegmental boundaries are clearly delineated by color and cell shape. No axons have been ob- served to cross these boundaries, although they do cross the midline. (It is known that the midline is formed in the embryo by the fusion of two separate groups of cells.) One is thus led to the speculation that the fields outlined by well-defined groups of nerve axons may perhaps coincide in certain cases with compartments or sub- compartments. This might be because compartmentalization may often occur be- fore the separation of the neuroblasts from the presumptive epidermis, so that any cell surface differences or other labels asso- ciated with a compartment may be shared by both the epidermal cells and the neu- rons. The hypothetical properties so far dis- cussed would be possessed by all or most of the cells within a given compartment or subcompartment. They could be described as area properties. Another rather differ- ent property would be one which charac- terized boundaries between compartments, that is. an edge property. For example, the cells on one side of the intersegmental boundaries in Oncopelrus are markedly elongated in the direction of the boundary (17). Do all compartment boundaries have this property? For the antero-posterior wing boundary it seems that the adult cells have no unusual appearance; but never- theless a detailed scrutiny of several such boundaries might be worthwhile. Another obvious hypothesis is that whereas there may be free diffusion of certain chemicals within compartments it may be greatly re- stricted across compartment boundaries. This suggests that compartments might not be electrically coupled to each other, but a direct test across the intersegmental boundary in Rhodniw (34) showed cou- pling to be normal. Moreover, a careful cy- tological study by electron microscopy has shown no observable difference in the vari- ous types of cell junctions (gap junctions, septate desmosomes, attachment desmo- somes) for the corresponding intersegmen- tal boundary in Oncope/fus (35). One is thus not exactly encouraged to look for these same differences at compartment boundaries in structures from imaginal discs. Nevertheless, it would be surprising if there were not some important cytolog- ical difference at compartment bounda- ries. 345 Possible Mechanisms for Compartment Formation We must now consider the nature of the step which partitions the cells that are the ancestors of one compartment in such a way that some of them become the founder cells of one subcompartment while the oth- ers become the founders of the other sub- compartment. As we have seen, this step is often a partition into two parts (rather than three, four, or more), and it is possible that this is always the case. For the mo- ment we will only consider the case of bi- nary partition. At present, little can be said about any underlying biochemical mechanism, but we can usefully discuss the problem at the cellular level. Unfortunately, we have rather few facts to go on. In view of the existence of size and shape regulation (as shown by the experiments in which a rela- tively fast-growing clone within a com- partment does not alter its dimensions), it is not obviously a requirement that the partition need be always exactly the same, since any variation, if it is not too big, can probably be corrected by subsequent growth. We consider three possible types of mechanism. 1) The partition of daughrers. All the cells divide once, one daughter of each di- vision being allocated to one sub- compartment and one to the other. 2) Random allocation. The cells are al- located at random, with a fixed probability which we shall assume to be about one- half. Because of the number of cells in- volved, the chance of all the cells being ac- cidentally allocated to one subcompart- ment is so small that it can be ignored (for example, for 20 cells this chance is 1 in 2rp or about 2 in 106). Even if all cells but one are allocated to one subcompartment, the single cell allocated to the other could, conceivably, compensate for this numeri- cal handicap by an increased rate of multi- plication. 3) Geographical partition. The patch of epithelial cells is divided. the dividing line separating the founder cells of one sub- compartment from those of the other. The difficulty with the first two mecha- nisms is that, in order to get the cells of each subcompartment together in one patch, a certain amount of relative cell movement would have to take place. Since the partitioning into subcompartments takes place several times in succession, one would not expect marked clones to stay in one piece, as they usually do. Thus these two mechanisms seem unlikely, except per- haps for the first of the several partitioning steps. The mechanisms can be saved to some extent by an additional hypothesis; that 346 any cell which is surrounded by cells of the other type commits suicide. It is difficult to make this model precise, but it would ap- pear to lead to a fair amount of cell death. Moreover, the cells which migrated would still have to move to the correct place in the epithelium relative to other surround- ing tissues. The third proposed mechanism-geo- graphical partition-seems to us to be by far the most likely one, especially as it does not need to be extremely precise. Consider, say. a patch of 20 cells. Let each cell divide once to give 40 cells. Each of these cells will be surrounded in the epithelium by several other cells (the average number is usually a little above five), one of which will be its sister cell. Now draw an arbi- trary (but moderately straight) line par- titioning the patch into two parts. This line will separate some cells which are sisters. The problem is to estimate the fraction (averaged over many cases) of the original 20 cells which will have daughters sepa- rated by the line. It is only these particular cells that can produce a clone of descend- ants which will go across the boundary between the subcompartments. Several approximate estimates have been made by Ripley (36) using various simplifications. The fraction defined above can be written as equal to C/N"2 when N is thenumber of cells at the time the line is drawn (40 in the example above) and C is a parameter which is approximately con- stant. The values of C found were not far from 0.55. Thus for N = 25 the fraction is about 11 percent. This calculation shows rather clearly that on this simple mecha- nism the existence of clones which cross the subcompartment boundary will not be a rare event if they are marked one generation before the compartment is divided. A more detailed mathematical study of this problem would be worthwhile since it is important to compare the detailed ex- perimental data (what fraction of clones crosses a border, what fraction runs along- side one, and the like, as a function of ex- act time of irradiation) with what would be expected on the various theoretical models. Conclusion We have seen that the work of Garcia- Bellido and his colleagues has clearly brought out the formation of compart- ments and successive subcompartments in the epithelium produced by the wing disc of Drosophila and that there is evi- dence that a similar process occurs in the production of other regions of the insect epithelium. We have also seen that the phenomena, although clearly demonstrat- ed in outline. need further detailed study, especially quantitative study. The mecha- nism that produces these subcompart- ments is obscure although a plausible model can be suggested for the general nature of the process, It is therefore pertinent to ask what is the novelty of these ideas, viewed from the general perspective of development stud- ies. To do this we must ask what the ex- periments show does not happen. We are not talking about the determina- tion of cell type in the usual sense-for ex- ample, a muscle cell as opposed to a fibro- blast-but about cell position. In this sys- tem the determination and differentiation of cell types-for example, bristles as op- posed to epithelial cells-probably comes later and may well also be dependent on the compartment to which the cells belong. What we are concerned with is geographi- cal position in the organism and, more- over, not about exact geographical posi- tion but whether a cell is somewhere within one well-defined region or another one. What has been demonstrated in this sys- tem is that once a major developmental step of this type has been taken by a cell it is not reversed in the progeny of that cell, at least in normal development. If reversal was possible, a cell which had been deter- mined for the dorsal side of the wing and which found itself on the ventral side could be reprogrammed to be a ventral cell. What clonal analysis has shown is that this never happens. Either such a cell cannot get to the wrong side of the wing, or, if it does so, it must either move back to the right side or be killed. The exact mecha- nism is obscure. Whatever it is, it is clearly of interest, even though the basic concept, the irreversibility of major developmental steps, is not in itself especially novel. But it would be both novel and exciting if it turns out that the compartments and subcompartments are used by the orga- nism as units for the control of shape and size; if gradient systems meet at com- partment boundaries, if cell surface prop- erties changed abruptly there, if size regu- lation occurred partly independently within each of these domains, and so forth. It may be that the normal development of each imaginal disc can usefully be divided into a precise succession of major steps each of which produces a set of new com- partments. If so, by studying compartment formation, one could both enumerate these steps and determine their times of action. On this picture each compartment would be specified by a unique combination of a small number of controlling genes [selec- tor genes (I)] that are active in it. (The steps that follow-for example, the deter- mination of a bristle in a particular posi- tion within a compartment-may be of a SCIENCE, VOL. 189 somewhat different and more complex character.) For the first time there is the real prospect of understanding the logic behind gene deployment in pattern forma- tion. As we have seen, the speculative ideas about compartments in this section are not supported by hard evidence. The best we have so far is a series of hints. But it is ex- actly this possibility, that compartments may have a wider significance, which makes the study of them at the present time so important and so interesting. Reference and Nom 1. A. Garcia-Bellido, in Cell Pwerding (Associated Scientific Publishers, London, 1975). p. 161. 2. C. Auerbach, Trans. R. Sot. Edinb. S-S, 787 (1936). 3. M. J. Pcarson,J. CellSci. 16, II3 (1974). 4. P. J. Bryant, DRY. Biol. 22,389 (1970). 5. iqb$aa&lhdo and J. R. Menam. ibid. 24, 61 6. P. Rjpoll. Wilhelm Roux' Arch. Enrwicklungs- mech. Org. 169.20 (1972). 7. C. A. Poodry and H. A. Schneiderman. ibid. 168, I~1971, 8. 9. IO. II. 12. 13. 14. IS. 16. 17. 18. 19. +,`j: `j&ant and H. A. Schneiderman, Dev. &cd. 20.263 (1969). A. Garcia-Bellida, Genetics 60. IS I (1968). ; P. Ripoll, G. Morata. Nor. New Bid. 245.251 (1973); Dev. BioL, in press. W. Gehring, Dev. Biol. 16,438 (1967). D. L. Lindslcy and E. H. Grcll. Generfc Variarions of Drosophila mdanogarter (Carnegie lnstitulion ofWashington, Washin G. Morata and P. Riool f ton, D.C.. 1968). . Dev. Biol. 42.21 I (1975). A. Fcrrus, Generics, in ;a. H. J. Becker. Z. In ukr. AbHammwgs-Verer- B bungs/. 88,333 (1957). P. Lawrence.Symp. Sot. Exp. Biol. 25,379(1971). ___ Nor. Newb'iol. 242.31 (1973). -J. Embryo/, Exp. Morphol. 30,681 (1973). T. S. Sahota and W. E. Bcckel. Con. J. Zool. 45. 407(1967j. 20. P. Santamaria and A. Garcia-Bcllido, in prepara- tion. 21. P. A. Lawrence, 1. Cc//Sri. 1,475 (1966). 22. W. Gehring and R. Nothiger, in Dave/u men&l Syslcm: Imccrs, S. Counce and C. H. ddding- ton, Eds. (Academic Press, New York, 1973), vol. 2,p.Zll. 23. G. Morur and A. Garcia-Bellido, in 24. E. B. Lewis, Am. Zoo/. 3. 33 (196 P rcpdrstion. ); in Role o/ 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. Chromosomes m Development, M. Locke, Ed. (Academic Press. New York, 1964). p. 2!l; in He.r- irqefmm Mcndel, P. A. Brink. Ed. (U~IV. OfWw consin Pr.w Mndimn 196-n n 17 A. Garb 87 (1972). G. Mar&a and P. A. Lawrence. Narure ILond.1 25536 18 4 (1975). P. A. L awrcncc, F. H.C. Crick, M. Munro. J. Cell Sci. 11.815 ( 1972). R. N0thige r. Wilhelm Roux' Arch. Enrwicklungs- mech. Org. 155,269 (19.54). A. Garcia-Bellido. D le,v:,~i@ ;l;;78 (1966). P. A. Lawrence, unpubt~sned W. Janning. Drosophila In/:! ierv 50. I51 119731. 1. Hasenfuss, Verb. Drsch. ZooI. Gez'66,il (l9i3); Zoo/. Jahrb. Anal. 90. I (1973); ibid., p. 175. P. A. Lawrence. in Cc/I Patlcming (Associated Scientific Publishers, London, 1975) .3. A. E. Warner and P. A. Lawrence, Ep mwe ILond.) 245,47 "nT-' P. A. L .`,`Gr%e and S. M. Green, 1. Cell Biol. 63. --\ 373 (190,. 9. Ripley, personal communication. We thank Professor A. Garcia-Bellido and Dr. G. Morata and our collcagues in the MRC Labora- tory for many helpful cornmenu and criticisms.