Chapter 21:  Cellular Mechanisms of Development

Introduction

We begin by considering how the geometrical structure of the early vertebrate embryo is formed. The focus will be on the question of how cells move into the correct positions. Later sections will consider how cells adopt the correct differentiated characters. It is traditional to distinguish three phases in the development of a vertebrate - and indeed of many other types of animal. In the first phase the fertilized egg cleaves to form many smaller cells, and these become organized into an epithelium and perform a complex series of movements, called gastrulation and neurulation, that create the basic body plan, with a rudimentary gut cavity and a neural tube. In the second phase the rudiments of the various organs, such as limbs, eyes, heart, and so on, are formed - a process called organogenesis. In the third phase the tiny structures that have been generated in this way proceed to grow to their adult size. These phases are not sharply distinct but overlap considerably in time. To follow the course of events from the fertilized egg to the beginning of organogenesis, we take as our chief example the frog Xenopus laevis (Figure 21-1), whose early development has been particularly well studied. As in other amphibians, the entire process from fertilization onward takes place outside the mother, and the developing embryo is robust and easy to manipulate experimentally.

The Polarity of the Amphibian Embryo Depends on the Polarity of the Egg ²

The Xenopus egg is a large cell, just over a millimeter in diameter, enclosed in a transparent extracellular capsule, or jelly coat. Most of the cell's volume is occupied by yolk platelets, which are membrane-bounded aggregates chiefly of lipid and protein. The yolk is concentrated toward the lower end of the egg, called the vegetal pole; the upper end is called the animal pole. The animal and vegetal regions contain different selections of mRNA molecules as well as different quantities of yolk and other cell components, and they have different fates. Roughly speaking, the vegetal end of the egg is destined to form internal tissues (in particular, the gut), and the animal end, external ones (such as the skin). Fertilization initiates a complex series of movements that will eventually tuck vegetal regions into the interior to form the gut and in the process will establish the three principal axes of the body: anteroposterior, from head to tail; dorsoventral, from back to belly; and mediolateral, from the median plane outward to the left or to the right.

The animal-vegetal asymmetry of the unfertilized egg is sufficient to define only one of the eventual body axes - the anteroposterior - but fertilization triggers a distortion of the egg contents that creates an additional asymmetry defining a dorsoventral difference: the outer, actin-rich cortex of the egg cytoplasm abruptly rotates relative to the central core of the egg, so that the animal pole of the cortex is slightly shifted to the future ventral side (Figure 21-2). The direction of the rotation is determined by the point of sperm entry - perhaps through an effect of the centrosome that the sperm brings into the egg. Because pigment granules in the egg are displaced by the rotation, a band of slightly diminished pigmentation, called the gray crescent, becomes visible in some amphibian species opposite the sperm entry point. In the neighborhood of the gray crescent the cortex of the vegetal hemisphere has become juxtaposed with core cytoplasm of the animal hemisphere, creating a special region that is crucial in organizing the dorsoventral axis of the body, as we discuss later.

The sperm entry point corresponds, roughly speaking, to the future belly; the opposite side will form the back and dorsal structures, including the spinal cord. Treatments that block the rotation allow cleavage to occur normally but produce an embryo with a central gut and no dorsoventral asymmetry.

Cleavage Produces Many Cells from One ³

The cortical rotation is completed in about an hour after fertilization and sets the scene for cleavage, in which the single large egg cell subdivides by repeated mitosis into many smaller cells, or blastomeres, without any change in total mass (Figure 21-3). To survive, the embryo must quickly reach a stage where it can begin to feed, swim, and escape from predators, and these first cell divisions are extraordinarily rapid, with a cycle time of about 30 minutes. The very high rate of DNA replication and mitosis seems to preclude gene transcription (although protein synthesis occurs), and the cleaving embryo is almost entirely dependent on reserves of RNA, protein, membrane, and other materials that accumulated in the egg while it developed as an oocyte in the mother. The only crucial biosynthesis obviously required is that of DNA, and unusually rapid DNA replication is made possible by the use of an exceptionally large number of replication origins, closely spaced in the chromosomal DNA.

After about 12 cycles of cleavage (7 hours), the cell division rate slows down abruptly, and transcription of the embryo's genome begins. This change, known as the mid-blastula transition, seems to be triggered by attainment of a critical ratio of DNA to cytoplasm: the transition can be hastened or delayed by artificially increasing or decreasing the amount of DNA in the egg.

The Blastula Consists of an Epithelium Surrounding a Cavity ⁴

From the outset the cells of the embryo are not only bound together mechanically, they are also coupled by gap junctions through which ions and other small molecules can pass, conveying messages that may help to coordinate the behavior of the cells. Meanwhile, in the outermost regions of the embryo, tight junctions between the blastomeres create a seal, isolating the interior of the embryo from the external medium. At about the 16-cell stage, Na⁺ begins to be pumped across the cell membranes into the spaces between cells in the interior of the embryo, and water follows because of the resulting osmotic pressure gradient. As a result, the intercellular crevices deep inside the embryo enlarge to form a single cavity, the blastocoel, and the embryo is now termed a blastula (Figure 21-4). The cells that form the exterior of the blastula have become organized as an epithelial sheet, which will be crucial in coordinating their subsequent behavior.

Gastrulation Transforms a Hollow Ball of Cells into a Three-layered Structure with a Primitive Gut ^{5, 6}

Once the cells of the blastula have become arranged into an epithelial sheet, the stage is set for the coordinated movements of gastrulation. This dramatic process transforms the simple hollow ball of cells into a multilayered structure with a central gut tube and bilateral symmetry: by a complicated invagination, many of the cells on the outside of the embryo are moved inside it. Subsequent development depends on the interactions of the inner, outer, and middle layers of cells thus formed.

Gastrulation - the formation of a gut by tucking cells from the exterior of the early embryo into the interior - is a fundamental step in the development of practically every type of animal. The transparent embryo of the sea urchin provides one of the clearest and simplest illustrations of the process. Figure 21-5 shows the sequence of events, starting with a simple hollow blastula. Briefly, cells at the vegetal pole invaginate, forming a hollow tube that eventually makes contact with the epithelium near the opposite end of the embryo to form the mouth. Meanwhile, cells escape from the invaginating epithelium at certain sites and move into the body cavity to form embryonic connective tissue, or mesenchyme.

In the three-layered structure created by gastrulation, the innermost layer, the tube of the primitive gut, is the endoderm;the outermost layer, the epithelium that has remained external, is the ectoderm;and between the two, the looser layer of tissue composed of mesenchyme cells is the mesoderm.These are the three primary germ layers common to higher animals. The organization of the embryo into the three layers corresponds roughly to the organization of the adult - gut on the inside, epidermis on the outside, and connective tissue and muscle in between. Very crudely, these three layers of adult tissues may be said to derive from the endoderm, the ectoderm, and the mesoderm, respectively, although there are exceptions.

In Xenopus the geometry of gastrulation is more complex than in the sea urchin. But it is important to grasp the basic principles, for it is through the movements of gastrulation that the main axes of the vertebrate body are created. The details of the process are described in Figure 21-6. A central part is played by the tissue near the site of the gray crescent, to one side of the vegetal pole. Here, gastrulation starts with a short indentation that gradually extends to form the blastopore - a line of invagination that eventually curves around to encircle the vegetal pole. The site where the invagination starts defines the dorsal lip of the blastopore; this tissue plays a leading part in the ensuing complex series of movements and gives rise to the dorsal structures of the main body axis. As in the sea urchin, the end result of the whole process is a three-layered structure: an outermost sheet of ectoderm, an innermost tube of endoderm forming the rudiment of the gut, and between them a layer of mesoderm. Again, the mouth develops as a hole formed at an anterior site where endoderm and ectoderm come into direct contact without intervening mesoderm.

The transformation that is brought about by gastrulation can be summarized by plotting on the surface of the embryo at the beginning of gastrulation a fate map showing which regions are destined to give rise to which parts of the adult body; such a map is shown in Figure 21-6B.

Gastrulation Movements Are Organized Around the Dorsal Lip of the Blastopore ^{6, 7}

The dorsal lip of the blastopore plays a central role not just in a geometrical sense, but also as the source of a controlling influence. If the dorsal lip of the blastopore is excised from a normal embryo at the beginning of gastrulation and grafted into another embryo but in a different position, the host embryo initiates gastrulation both at the site of its own dorsal lip and at the site of the graft (Figure 21-7). The movements of gastrulation at the second site entail the formation of a second whole set of body structures, and a double embryo (Siamese twins) results.

By carrying out such grafts between species with differently pigmented cells, so that host tissue can be distinguished from implanted tissue, it has been shown that the grafted blastopore lip recruits host epithelium into its own system of invaginating endoderm and mesoderm. Evidently, the dorsal lip of the blastopore is the source of some signal (or signals) coordinating both the movements of gastrulation and, directly or indirectly, the pattern of specialization of the tissues in its neighborhood. Because of this crucial role in organizing the formation of the main body axis, the dorsal lip of the blastopore is known as the Organizer(or Spemann's Organizer, after its co-discoverer). It is the oldest and most famous example of an embryonic signaling center - a function we discuss later when we consider how cell diversification is controlled.

Active Changes of Cell Packing Provide a Driving Force for Gastrulation ^{1, 6, 8}

Gastrulation begins with changes in the shape of the cells at the site of the blastopore. In the amphibian these are called bottle cells: they have broad bodies and narrow necks that anchor them to the surface of the epithelium (Figure 21-8), and they may help to force the epithelium to curve and so to tuck inward, producing the initial indentation seen from outside. Once this first tuck has formed, cells can continue to pass into the interior as a sheet to form the gut and mesoderm. Just as in the sea urchin, the movement seems to be driven by a combination of mechanisms but mainly by active repacking of the cells, especially those in the dorsal part of the marginal zone neighboring the blastopore lip (see Figure 21-8). Here convergent extension occurs. Small square fragments of dorsal marginal-zone tissue isolated in culture will spontaneously narrow and elongate through a rearrangement of the cells, just as they would in the embryo in the process of converging toward the dorsal midline, turning inward around the blastopore lip, and elongating to form the main axis of the body. A current view of the cellular mechanism underlying convergent extension is illustrated in Figure 21-9.

The Three Germ Layers Formed by Gastrulation Have Different Fates ^{9, 10, 11, 12}

The endoderm forms a tube, the primordium of the digestive tract, from the mouth to the anus. It gives rise not only to the pharynx, esophagus, stomach, and intestines, but also to many associated glands. The salivary glands, the liver, the pancreas, the trachea, and the lungs, for example, all develop from extensions of the wall of the originally simple digestive tract and grow to become systems of branching tubes that open into the gut or pharynx. While the endoderm forms the epithelial components of these structures - the lining of the gut and the secretory cells of the pancreas, for example - the supporting muscular and fibrous elements arise from the mesoderm.

The differentiation of the mesoderm is guided by the Organizer at the dorsal lip of the blastopore, which is thought to be a source of signaling molecules that regulate choices between alternative mesodermal fates. Signals from the differently specialized groups of mesoderm cells in turn control the basic pattern of specializations of the endoderm and ectoderm and in particular initiate formation of the nervous system, as we shall see. The mesodermal layer is divided in the postgastrulation embryo into separate parts on the left and right of the body. Defining the central axis of the vertebrate body, and effecting this separation, is the very early specialization of the mesoderm known as the notochord. This is a slender rod of cells, about 80 µm in diameter, with ectoderm above it, endoderm below it, and mesoderm on either side (see Figure 21-12). It derives from the cells of the Organizer itself. As these pass around the dorsal lip of the blastopore and move into the interior of the embryo, they form a column of tissue that elongates dramatically by convergent extension. The cells of the notochord also become swollen with vacuoles, so that the rod elongates still further and stretches out the embryo. In the most primitive chordates, which have no vertebrae, the notochord persists as a primitive substitute for a vertebral column. In vertebrates it serves as a core around which other mesodermal cells gather to form the vertebrae. Thus the notochord is the precursor of the vertebral column, both in an evolutionary and in a developmental sense.

In general, the mesoderm gives rise to the muscles and to the connective tissues of the body - at first to the loose, space-filling, three-dimensional mesh of cells known as mesenchyme (see Figure 19-30) and ultimately to cartilage, bone, and fibrous tissue, including the dermis (the inner layer of the skin). In addition, the tubules of the urogenital system form from it and so does the vascular system, including the heart, the blood vessels, and the cells of the blood. These specialized mesodermal tissues derive from cells at different distances from the dorsal lip of the blastopore, with notochord having the most dorsal origin and blood cells the most ventral.

At the end of gastrulation the sheet of ectoderm covers the embryo and thus eventually forms the epidermis (the outer layer of the skin). It also gives rise to the entire nervous system. In a process known as neurulation, a broad central region of the ectoderm thickens, rolls up into a tube, and pinches off from the rest of the cell sheet (Figure 21-10). The tube thus created from the ectoderm is called the neural tube; it will form the brain and the spinal cord. The mechanics of neurulation depend, like gastrulation, on changes of cell packing and cell shape, and Figure 21-11 shows how the cytoskeleton can be organized to bring about cell shape changes that can make an epithelium roll up into a tube.

Neurulation is induced by an interaction with the underlying notochord and the mesoderm adjacent to it. If a piece of such dorsal mesoderm is taken from the area just beneath the future neural tube of one gastrulating amphibian embryo and implanted directly beneath the ectoderm of another gastrulating embryo in, say, the belly region, the ectoderm in that region will thicken and roll up to form a piece of misplaced neural tube.

Along the line where the neural tube pinches off from the future epidermis, a number of ectodermal cells break loose from the epithelium and migrate as individuals out through the mesoderm. These are the cells of the neural crest; they will form almost all of the peripheral nervous system (including most of the sensory and all of the sympathetic ganglia and the Schwann cells that make the myelin sheaths of peripheral nerves) as well as the pigment cells of the skin. In the head many of the neural crest cells will differentiate into cartilage, bone, and other connective tissues, which elsewhere in the body arise from the mesoderm. This is one of several instances that run counter to the general scheme in which the three germ layers give rise to cells in three corresponding concentric layers of the adult body.

The sense organs, by which light, sound, smell, and so forth impinge on the nervous system, also have ectodermal origins: some derive from the neural tube, some from the neural crest, and some from the exterior layer of ectoderm (see Figure 21-102). The retina, for example, originates as an outgrowth of the brain and so is derived from cells of the neural tube, while the olfactory cells of the nose differentiate directly from the ectodermal epithelium lining the nasal cavity.

The Mesoderm on Either Side of the Body Axis Breaks Up into Somites from Which Muscle Cells Derive ¹³

On either side of the newly formed neural tube lies a broad expanse of mesoderm (Figure 21-12). The thicker, more medial and dorsal part of this mesoderm gives rise to the muscular and skeletal tissues of the central body axis. It consists at first of a single continuous slab of tissue on each side of the body. To form the repetitive series of vertebrae and segmental muscles, this slab soon breaks up into separate blocks, or somites (Figure 21-13). The somites form one after another, starting in the head and ending at the tail (Figure 21-13). Segmentation is accompanied by changes in the connections between the mesoderm cells, but the mechanism that controls the regular spacing of the clefts that separate one somite from the next remains a mystery (although it is known that the physical process of somite formation is foreshadowed by a segmental pattern of expression of certain genes).

Each somite corresponds to one unit in the final sequence of articulated ele-ments. The bulk of the somite forms the skeletal muscles of the segment, while a subset of its cells go to form the corresponding vertebrae and other connective tissues such as dermis. The somites are also the source of almost all skeletal muscle cells elsewhere in the body: these derive from precursors that migrate away from the somites before differentiating overtly.

Changing Patterns of Cell Adhesion Molecules Regulate Morphogenetic Movements ¹⁴

The tissue movements in the embryo go hand in hand with changes in the chemical characters of the cells. By switching on production of a cytoskeletal protein, for example, a cell may alter its shape or the way it moves. By changing the set of adhesion molecules it displays on its surface, it may break old attachments and make new ones. Cells in one region may develop surface properties that make them cohere with one another and become segregated from a neighboring group of cells whose surface chemistry is different.

Classical experiments on early amphibian embryos showed that the effects of selective cell-cell adhesion can be so powerful that they bring about an approximate reconstruction of the normal structure even after the cells have been artificially dissociated into a random mixture (Figure 21-14). As discussed in Chapter 19, studies on chick and mouse embryos suggest that this behavior depends, at least in part, on a family of homologous Ca²⁺-dependent cell-cell adhesion glycoproteins - the cadherins. These molecules and other, Ca²⁺-independent cell-cell-adhesion molecules such as N-CAM are differentially expressed in the various tissues of the early embryo, and antibodies against them interfere with the normal selective adhesion between cells of a similar type.

Changes in the patterns of expression of the various cadherins correlate closely with the changing patterns of association among cells during gastrulation, neurulation, and somite formation (Figure 21-15); these transformations of the early embryo may be regulated and driven in part by the cadherin pattern. In particular, cadherins appear to have a major role in controlling the formation and dissolution of epithelial sheets and clusters of cells. They not only glue one cell to another, but also provide anchorage for intracellular actin filaments at the sites of cell-cell adhesion (discussed in Chapter 19): in this way they help to regulate the pattern of stresses and movements in the developing tissue according to the pattern of adhesions.

Besides sticking to one another, cells can stick to components of the extracellular matrix such as fibronectin and laminin. These adhesions are typically mediated by integrins, which, like cadherins, serve as transmembrane linkers between sites of attachment on the outside of the cell and actin filaments inside. Cell-matrix interactions of this sort are important for the movements of certain special classes of cells that lose adhesions to their neighbors and migrate as individuals through the embryo by crawling through the spaces between other cells. As a result of such invasions, to be discussed next, most tissues in the adult vertebrate body include admixtures of cells derived from widely separate parts of the early embryo.

Embryonic Tissues Are Invaded in a Strictly Controlled Fashion by Migratory Cells ^{11, 15, 16}

We have already mentioned two classes of migratory cells - those of the neural crest and those that leave the somites to give rise to skeletal muscle. Other important migrants are the precursors of the blood cells, of the germ cells, and of many groups of neurons within the central nervous system.

Cell migrations can be traced by marking the cells at the beginning of their journey, using either a nontoxic dye or, better, a heritable genetic label. Much of our knowledge has come from studies in which cells are grafted from quail embryos into chick embryos. Although the quail is similar in most respects to the chick, its cells can be distinguished in histological sections by a large, strongly staining mass of heterochromatin associated with the nucleolus. This nucleolar marker makes it possible to identify grafted cells that have migrated from the site where they were implanted. For example, if quail somite tissue is substituted for the somite tissue of a very young chick embryo before the limb buds appear, all the muscle cells in the limbs that subsequently develop have a quail origin (Figure 21-16). Evidently the future muscle cells migrate from the somites into the prospective wing region and remain there, inconspicuously mixed with the connective-tissue cells of the limb bud, until the time comes for them to differentiate.

In a similar way one can trace the dispersal of cells from the neural crest. These migrate along certain specific pathways through the embryo (Figure 21-17) and settle in precisely defined locations. As a migrant cell travels through the embryo, it repeatedly extends projections that probe its immediate surroundings (Figure 21-18), testing for subtle cues to which it is particularly sensitive by virtue of its specific assortment of cell-surface receptor proteins. Inside the cell these receptor proteins are connected to the cytoskeleton, which moves the cell along. Some extracellular matrix materials, such as fibronectin, provide adhesive sites that help the cell to advance; others, such as chondroitin sulfate proteoglycan, inhibit locomotion and repel immigration. The nonmigrant cells along the pathway may likewise have inviting or repellent surfaces, or may even extend filopodia that touch the migrant cell and affect its behavior. An incessant tug-of-war between opposing tentative attachments made by the migrant cell leads to a net movement in the most favored direction until the cell finds a site where it can form a lasting attachment. Other factors such as chemotaxis and interactions among the migratory cells may also play an important part.

Yet another means of controlling the distribution of migrant cells is through regulation of their survival and proliferation. Germ cells, blood cell precursors, and pigment cells derived from the neural crest all appear to be governed in this respect by the same basic control mechanism. This involves a transmembrane receptor, called the Kit protein, in the membrane of the migrant cells and a ligand, called the Steel factor, produced by the cells of the tissue through which the cells migrate and/or in which they come to settle. Individuals with mutations in the genes for either of these proteins are deficient in their pigmentation, their supply of blood cells, and their production of germ cells (Figure 21-19). The Steel factor appears to be required in a membrane-bound form in order to activate Kit correctly and enable all these cell types to survive and proliferate.

The Vertebrate Body Plan Is First Formed in Miniature and Then Maintained as the Embryo Grows ⁹

The embryo at the stage when the somites are forming and neural crest cells are setting off on their migrations is typically a few millimeters long and consists of about 10⁵ cells. While we have been speaking thus far mainly of Xenopus, the scale and general form are much the same for a fish, a salamander, a chick, or a human (see Figure 1-36). Later these species of embryo will grow to be very different in size and shape, but at this early stage they all share the basic vertebrate body plan. The central nervous system is represented by the neural tube, with an enlargement at one end for the brain; the gut and its derivatives, by a tube of endoderm; the segments of the trunk, by the somites; the other connective tissues, including the vascular system, by the more peripheral unsegmented mesoderm; and the epidermal layer of the skin, by the ectoderm. During subsequent development all of these components will enlarge, by a factor of as much as a hundred or more in length or a million or more in volume and cell number. But the same basic organization of the body will be preserved.

Summary

The eggs of most animals are large cells, containing stores of nutrients and other cell components specified by the maternal genome. In amphibians the first major movement after fertilization is a rotation of the cortex of the egg relative to its core. The asymmetry created by this rotation, together with the original asymmetry in the distribution of the contents of the egg before fertilization, defines the future antero-posterior and dorsoventral axes of the body. During the subsequent cleavage divisions the egg subdivides into many smaller cells, but no growth occurs.

A cavity soon develops in the interior of the embryo, while the surrounding cells become organized into an epithelial sheet. Part of the epithelium then invaginates, transforming the embryo into a three-layered structure with an internal epithelial tube of endoderm, an external epithelial covering of ectoderm, and a middle layer of mesodermal cells that have broken loose from the original epithelial sheet. In this process of gastrulation the epithelial cells actively change their packing, and this is thought to provide a major driving force for the movements.

The endoderm will form the lining of the gut and its derivatives, the ectoderm will form the epidermis and the nervous system, and the mesoderm will form muscles, connective tissues, vascular system, and urogenital tract. The development of all these structures depends on interactions between the three germ layers and involves further cell movements. The dorsal mesoderm, for example, induces the overlying ectoderm to thicken, roll up, and pinch off to form the neural tube and neural crest. In the middle of the dorsal mesoderm a rod of specialized cells called the notochord elongates to form the central axis of the embryo. The long slabs of mesoderm on either side of the notochord become segmented into somites, from which the vertebrae and skeletal muscles will be derived. At several sites migrant cells, such as those of the neural crest, break loose from their original neighbors and migrate through the embryo to colonize new sites. Specific cell-adhesion molecules, such as cadherins and integrins, help to guide the migrations and control the selective cohesion of cells in epithelia.

Introduction

A fertilized egg may develop into a daisy or an oak tree, a sea urchin or a human being. The outcome is governed by the genome: the linear sequence of A, G, C, and T nucleotides in the DNA of the organism must direct the production of a variety of chemically different cell types arranged in a precise pattern in space. Developmental biology aims to explain how. The whole discussion of this problem, in this and subsequent sections, rests on one fundamentally important fact: the cells in the body inherit the same genome from the egg. No matter how different they may appear - in muscle, bone, or nerve, in root, stem, or leaf - they all contain the same set of genetic instructions.

One of the earliest and most powerful demonstrations of this principle came from experiments on nuclear transplantation using amphibian eggs (Figure 21-20). A typical amphibian egg is so large that, using a fine glass pipette, one can readily inject into it a nucleus taken from another cell. The nucleus of the egg itself is destroyed beforehand by ultraviolet irradiation. The egg is activated to begin development by the act of pricking with the fine pipette used to inject the transplanted nucleus. Thus one can test whether the nucleus from a differentiated somatic cell contains a complete genome equivalent to that of a normal fertilized egg and equally serviceable for development. The answer is yes: a complete swimming tadpole can be produced, for example, from an egg whose own nucleus has been replaced by a nucleus derived from a keratinocyte cell from an adult frog's skin or by a nucleus from a frog red blood cell. These experiments admittedly have limitations. They have been successful only with nuclei from a limited range of differentiated cell types and in only a few species. But there is now an overwhelming body of evidence pointing to the same conclusion. With just a few exceptions (see Figure 23-37), the genome remains intact during development. Genes can be switched on or off, and the cells of the body differ not because they contain different genes but because they express different genes. In Chapter 9 we examine the intracellular mechanisms for regulating gene expression. In this chapter we have to consider not only how the differences between cells originate, but how they are coordinated in space and time within a multicellular organism. The present section discusses how the first steps of cell diversification are coordinated in early embryos, taking frog and mouse as examples.

Initial Differences Among Xenopus Blastomeres Arise from the Spatial Segregation of Determinants in the Egg ^{2, 17, 19}

In most animal and plant species the egg itself is chemically asymmetrical, with certain components concentrated in specific regions of the cytoplasm or membrane. As a result, there are differences from the outset between the cells that form by cleavage because they receive different portions of the localized materials. The importance of such localized determinants in the egg varies from species to species. It is traditional to distinguish two theoretical extremes: in mosaic development the whole future pattern of the body is delineated by localized determinants in the egg, and subsequent cell-cell interactions count for nothing; in regulative development localized determinants in the egg count for nothing, and the body pattern is generated entirely by subsequent cell-cell interactions. In reality, most higher animals and plants lie between these extremes. None, so far as is known, is truly mosaic - regulative interactions always play an important part; mammalian eggs, as we shall see, appear to be entirely regulative. Xenopus represents a typical intermediate case.

The asymmetries of the Xenopus egg are manifest in several ways - in the eccentric location of the nucleus, in the distribution of yolk and pigment granules, in the cytoskeleton, and, perhaps most significantly, in the distribution of certain specific mRNAs. The egg asymmetries endow the early blastomeres with different characters according to whether they are animal or vegetal, dorsal or ventral. Treatments such as centrifugation or ultraviolet irradiation that displace the contents of the uncleaved egg or prevent the cortical rotation that usually follows fertilization lead to drastic disturbances of the embryonic body plan, and equally drastic disturbances result if the early blastomeres are artificially rearranged.

Inductive Interactions Generate New Types of Cells in a Progressively More Detailed Pattern ²⁰

The initial differences between the early blastomeres define only the crude beginnings of the pattern of the embryo. To generate the full range of cell types, the blastomeres must interact with one another. If the early Xenopus embryo is placed in a medium devoid of Ca²⁺ and Mg²⁺, the blastomeres lose their cohesiveness and can then be separated and allowed to develop on their own; some go on to develop features characteristic of ectoderm, while others develop features characteristic of endoderm, but none of them switches on expression of genes characteristic of mesoderm, such as the muscle-specific actin gene. But when cells from the animal pole of a blastula are placed next to vegetal cells, some of the animal pole cells are diverted from the ectodermal pathway of development into the mesodermal pathway (Figure 21-21). The switching of cells from one pathway into another by the influence of an adjacent group of cells is called induction. During normal development, inductive interactions may occur between cells that have been adjacent from the outset - as in mesoderm induction - or between cells that are brought together through morphogenetic movements such as gastrulation. By a series of successive inductions, it is possible to generate many different kinds of cells from interactions between a few kinds (Figure 21-22).

As we emphasized earlier, asymmetries in the Xenopus egg define not only the animal-vegetal axis, and thereby the partitioning of the embryo into ectoderm, mesoderm, and endoderm, but also the dorsoventral and anteroposterior axes of the body. For the organization of the dorsoventral axis, an inductive mechanism again seems to operate. Grafting experiments indicate that, while all vegetal blastomeres can induce mesoderm, they do not all do so in the same way: the dorsal vegetal blastomeres are unique in that they induce the cells above them to take on the special character of Spemann's Organizer. The Organizer in its turn, as we saw earlier, produces a signal that induces an array of specializations in the mesoderm next to it. Later still, the pattern created in the mesoderm will induce patterns of local specialization in the ectoderm and endoderm that it contacts.

Thus there seem to be at least three inductive signals at work in the earliest stages of Xenopus development: from ventral vegetal blastomeres, from dorsal vegetal blastomeres, and from the Organizer (Figure 21-23). What are these signals in chemical terms? Members of at least four families of secreted signaling proteins seem to be involved. Although their precise roles in normal development are still not clear, all are thought to be present in the early Xenopus embryo, and all have dramatic inductive effects when supplied artificially. For at least two of the four, artificial blockade of function produces embryos with major parts of the body missing (Figure 21-24).

Such observations do not explain, however, how the localization of the inductive signals that pass between blastomeres in the Xenopus embryo is governed by the pattern of asymmetries in the uncleaved egg. In the case of the Vg1 protein,which is a member of the TGF-β superfamily of secreted signaling factors, one can glimpse how this may come about. A store of maternal mRNA coding for the protein is localized in the vegetal part of the egg before fertilization. It is thought that the protein is produced in precursor form in the vegetal regions, and that it may be activated in the dorsal vegetal region and released from dorsal vegetal blastomeres to induce the Organizer (Figure 21-25).

A Simple Morphogen Gradient Can Organize a Complex Pattern of Cell Responses ²¹

There are many ways in which signals passing from one part of an embryo to another can control pattern formation (Figure 21-26). The Organizer exemplifies one strategy of particular interest: a small patch of tissue in a specific region acquires a specialized character and becomes the source of a signal that spreads into neighboring tissue and controls its behavior. The signal, for example, may take the form of a diffusible molecule secreted from the signaling center. Suppose that this substance is slowly degraded as it diffuses through the neighboring tissue. The steady-state concentration then will be high near the source and decrease gradually with increasing distance, so that a concentration gradient is established (Figure 21-27). Cells at different distances from the source will be exposed to different concentrations and may become different as a result. A substance such as this, whose concentration is read by cells to discover their position relative to a certain landmark or beacon, is termed a morphogen. Morphogen gradients are thought to be a common way of providing cells with positional information or controlling their pattern of differentiation, although there are still only a few cases where a morphogen has been identified chemically.

How do cells respond to a morphogen gradient? The concentration of a diffusible morphogen should be smoothly graded, but many of the important specializations in development are discrete: there is no graded series of mature kinds of cells intermediate between cartilage and muscle, or bone and nerve, for example. In theory, sharp distinctions can arise in a population of initially uniform cells through a threshold in their response to a smoothly graded signal. If there is a positive feedback in each responding cell that amplifies the effect of a small increment in the signal, cells exposed to only slightly different intensities of the signal can be launched on radically different courses of development according to whether their exposure is above or below a certain threshold level. If there are several thresholds of response to one signal, a single morphogen can control the pattern of several different cell choices. It has been shown, for example, that when cells from the animal pole of an early Xenopus embryo are exposed to the signaling molecule activin (see Figure 21-24), they will develop as epidermis if the activin concentration is low, as muscle if it is a little higher, and as notochord if it is a little higher still. The normal role of activin in the intact Xenopus embryo, however, is uncertain, and the nature of the signals emanating from Spemann's Organizer is still unclear.

Cells Can React Differently to a Signal According to the Time When They Receive It: The Role of an Intracellular Clock ²²

As development proceeds, embryonic cells generally change their character even if their environment is unchanged. If cells taken from the animal pole of a Xenopus blastula, for example, are kept in isolation in vitro, they will spontaneously differentiate into epidermis at roughly the normal time. In this sense, the cells behave as though governed by some sort of intracellular clock. Because cells are spontaneously changing their internal state, they may respond differently to an inductive signal according to the time when they receive it. If a fragment of animal pole epithelium is taken from an early gastrula and grafted over the eye rudiment of a later embryo, for example, it will be induced to differentiate (inappropriately) into a piece of tissue resembling neural tube; if it is allowed to age for a few hours in vitro before grafting into the same environment, it will be induced to differentiate (appropriately) into a lens; if it is cultured in vitro for a longer period still, it loses competence to respond to the inductive influence from the eye rudiment in either of these ways.

There is an important general lesson here: cellular diversity and spatial patterning can arise from a simple unchanging inductive signal acting on a succession of otherwise identical cells at different times (Figure 21-28). We have seen, for example, that the parts of the central body axis are formed sequentially during gastrulation, with anterior parts involuting around the blastopore lip first and posterior parts last. According to one theory, the difference in the age at which the cells pass the dorsal lip and are acted on by Spemann's Organizer could be the source of the differences of cell character between the anterior and posterior parts of the mesoderm and endoderm and therefore of the body as a whole.

Thus the general strategy of pattern formation can be summarized as follows: (1) patterns begin from simple asymmetries, (2) details are filled in sequentially through inductive cell-cell interactions, and the pattern of cell diversification that results depends both on (3) the positional signals between cells and on (4) intracellular programs that change a cell's response to these signals with time.

In different species these four basic elements may be combined in different ways. We now consider the special case of the early mammalian embryo, which has some remarkable regulative properties.

In Mammals the Protected Uterine Environment Permits an Unusual Style of Early Development ^{9, 23}

The mammalian embryo does many things differently from other animals. Developing in the protected environment of the uterus, it does not have the same need as the embryos of most other species to complete the early stages of development rapidly. Moreover, the development of a placenta quickly provides nutrition from the mother, so that the egg does not have to contain large stores of raw materials such as yolk. The egg of a mouse has a diameter of only about 80 µm and therefore a volume about 2000 times smaller than that of a typical amphibian egg. Its cleavage divisions occur no more quickly than the divisions of many ordinary somatic cells, and gene transcription has already begun by the two-cell stage. Furthermore, while the later stages of mammalian development are fundamentally similar to those of other vertebrates such as Xenopus, mammals begin by taking a large developmental detour to generate a complicated set of structures - notably the amniotic sac and the placenta - that enclose and protect the embryo proper and provide for the exchange of metabolites with the mother. These structures, like the rest of the body, derive from the fertilized egg but are called extraembryonic because they are discarded at birth and form no part of the adult.

The early stages of mouse development are summarized in Figure 21-29. The egg is surrounded initially by a transparent cell coat, the zona pellucida. Upon fertilization, the egg cleaves within this coat to form a mulberry-shaped cluster of cells called the morula. Sometime between the 8-cell and 16-cell stages, the surface of the morula becomes smoother and more nearly spherical as the cells change their cohesiveness and become compacted together (Figure 21-30), with tight junctions forming between the outer cells and sealing off the interior of the morula from the external medium. Soon after, the internal intercellular spaces enlarge to create a central fluid-filled cavity - the blastocoel. At this stage the morula is said to have become a blastocyst. The cells of the blastocyst form a spherical shell enclosing the blastocoel, with one pole distinguished by a thicker accumulation of cells. As shown in Figure 21-29, the entire outer cell layer is the trophectoderm; the cluster of cells inside the trophectoderm at the thicker pole is called the inner cell mass.

The whole of the embryo proper is derived from the inner cell mass. The trophectoderm is the precursor of the placenta and is the earliest component of the system of extraembryonic structures. Once the zona pellucida has been shed, the cells of the trophectoderm come into close contact with the wall of the uterus, in which the embryo becomes implanted. Meanwhile the inner cell mass grows and begins to differentiate. Part of it gives rise to some further extraembryonic structures, such as the yolk sac, while the rest of it goes on to form the embryo proper by processes of gastrulation, neurulation, and so on, that are largely homologous to those seen in other vertebrates, although extreme distortions of the geometry sometimes make the homology hard to discern.

All the Cells of the Very Early Mammalian Embryo Have the Same Developmental Potential ²⁴

Up to the eight-cell stage, each cell of the early mammalian embryo can form any part of the later embryo or adult. If the early embryo is split in two, a pair of identical twins can be produced - two complete normal individuals from a single cell. Similarly, if one of the cells in a two-cell mouse embryo is destroyed by pricking it with a needle and the resulting "half-embryo" is placed in the uterus of a foster mother to develop, in many cases a perfectly normal mouse will emerge.

Conversely, two eight-cell mouse embryos can be combined to form a single giant morula, which then develops into a mouse of normal size (Figure 21-31). Such creatures, formed from aggregates of genetically different groups of cells, are called chimeras. Chimeras can also be made by injecting cells from an early embryo of one genotype into a blastocyst of another genotype. The injected cells become incorporated into the inner cell mass of the host blastocyst, and a chimeric animal develops. It is even possible to make a chimera by injecting a single cell in this way; thus one can assay the developmental capabilities of the single cell. One of the major conclusions derived from these studies is that the cells of the very early mammalian embryo (up to the eight-cell stage) are initially identical and unrestricted in their capabilities: they are all totipotent. Localized determinants apparently have no part to play in the mammalian egg, and the pattern of cell diversification in the embryo is generated later, entirely through interactions of the cells with one another and with their environment.

Mammalian Embryonic Stem Cells Show How Environmental Cues Can Control the Pace as well as the Pathway of Development ²⁵

Mammalian early development is highly regulative. The fate of each cell is governed by interactions with its neighbors. The mouse experiments just described illustrate this well. The cells in a half-embryo or in a chimeric double embryo must adjust their behavior so as to generate an animal that is normal in both pattern and size. When the circumstances of development are more grossly abnormal, however, the embryonic cells can go wildly out of control. Some important lessons can be learned from these phenomena.

If a normal early mouse embryo is grafted into the kidney or testis of an adult, it rapidly becomes disorganized, and the normal controls on cell proliferation break down. The result is a bizarre growth known as a teratoma, which consists of a disorganized mass of cells containing many varieties of differentiated tissue - skin, bone, glandular epithelium, and so on - mixed with undifferentiated stem cells that continue to divide and generate yet more of these differentiated tissues. Teratomas with similar properties can also arise spontaneously from germ cells in the gonads as the result of various developmental accidents.

It is possible to derive transplantable cancers from teratomas. Such teratocarcinomas will grow without limit until they kill their host. They can be maintained indefinitely by grafting samples of the tumor cells serially from one host to another, and they always include some undifferentiated stem cells, together with a variety of differentiated cell types to which the stem cells give rise. The teratocarcinoma stem cells can also be maintained in culture as permanent cell lines.

One might think that teratocarcinoma stem cells originate, as in other cancers, through mutations in genes responsible for the normal controls of cell behavior (discussed in Chapter 24). The following observations, however, suggest that this is not the case. Stem cells with very similar properties can be derived by placing a normal inner cell mass in culture and dispersing the cells as soon as they proliferate. Once dispersed, some of the cells, if kept in suitable culture conditions, will continue dividing indefinitely without altering their character. The resulting embryonic stem (ES) cell lines are similar to teratocarcinoma-derived cell lines, but they can be generated at such high frequency from normal embryos that it is unlikely that they arise by mutation. Instead, it appears that separating the cells from their normal neighbors and placing them in the appropriate culture medium has arrested the normal program of change of cell character with time and so enabled the cells to carry on dividing indefinitely without differentiating. The presence in the medium of a protein growth factor known as leukemia inhibitory factor (LIF) seems to be critical for this suspension of developmental progress. With a slightly more complex cocktail of growth factors, embryonic germ cells can be induced to behave in the same way in culture.

The state in which the ES, teratocarcinoma, or germ-cell-derived stem cells are arrested seems to be equivalent to that of normal inner-cell-mass cells. This can be shown by taking the cells from their culture dish and injecting them into the blastocoel cavity of a normal blastocyst (Figure 21-32). The injected cells become incorporated in the inner cell mass of the blastocyst and can contribute to the formation of an apparently normal chimeric mouse. Descendants of the injected stem cells can be found in practically any of the tissues of this mouse, where they differentiate in a well-behaved manner appropriate to their location and can even form viable germ cells. This capability of ES cells forms the basis for a widely used technique that allows mice to be generated with a genetically engineered mutation in any chosen gene whose DNA has been cloned. To produce such "gene-knockout" mice, mutant ES cells are made by selecting for a DNA insertion that replaces the chosen gene by an artificially altered version; the mutant ES cells are then used to produce chimeric mice that carry the mutation in their germ cells (see p. 329).

The extraordinarily adaptable behavior of ES cells shows that environmental cues not only guide choices between different pathways of differentiation, but in certain cases, they can also stop or start the developmental clock - the processes that drive a cell to progress from an embryonic to an adult state.

Summary

In the course of embryonic development many types of cells are generated from the fertilized egg. The genomes of the differentiated cells remain the same; it is the pattern of gene expression that changes. Some of the differences between cells in the early embryo generally originate from the unequal distribution of cytoplasmic determinants localized in the egg before cleavage, but most of them arise later from local differences in the environments of the cells in the embryo. In Xenopus, for example, the animal and vegetal cells of the early embryo inherit different cytoplasmic determinants from the egg, and an influence from the vegetal cells then induces some of the animal cells to develop as mesoderm instead of ectoderm. This mesoderm induction seems to be mediated by families of growth factor proteins that also help regulate growth and differentiation in the mature organism.

Mammalian eggs are exceptional in that they are essentially symmetrical. Thus all the cells in an early mammalian embryo are initially alike and become different only through their interactions with one another. Through cell-cell interactions cells from two different early mouse embryos can adjust their fates and collaborate to form a single chimeric mouse. Early mouse embryo cells removed from the normal influences of their neighbors can proliferate inappropriately to give rise to teratocarcinomas, from which embryonic stem cells can be obtained. But when implanted into a normal early embryo, such cells revert to normal behavior, and their progeny differentiate according to their environment and can contribute to the formation of a healthy chimeric animal.

Introduction

Cells must not only become different, they must also remain different after the original cues responsible for cell diversification have disappeared. Despite the continual turnover and resynthesis of almost all cell components, most cell types in the adult body have at least some distinctive features that are stably and heritably maintained even when the environment is changed. Thus, when a pigment cell divides, its daughters remain pigment cells; when a keratinocyte from the skin divides, its daughters remain keratinocytes; even though a fibroblast may be convertible into some other sort of connective-tissue cell such as a cartilage cell, it never changes into a neuron or a liver cell; and so on. Such durable differences between cell types are ultimately due to the different influences that the cells have been subjected to in the embryo, but the differences are maintained because the cells somehow remember the effects of those past influences and pass them on to their descendants. As we discuss in this section, cell memory - and the types of information, especially positional information, that cells retain as a consequence - are central elements of the patterning mechanisms that make a complex multicellular organism possible.

Cells Often Become Determined for a Future Specialized Role Long Before They Differentiate Overtly ^{15, 26}

Cell memory is most obvious in the persistence and stability of the differentiated states of cells in the adult body (discussed in Chapter 22). But the final character of a cell has usually been decided by a complex sequence of cues delivered to its progenitors during development and is often fixed long before differentiation becomes manifest. Through a series of decisions taken before, during, and just after gastrulation, for example, certain cells in the somites of a vertebrate become specialized at a very early stage as precursors of skeletal muscle cells; they then migrate from the somites into various other regions including those where the limbs will form (see Figure 21-16). These muscle cell precursors lack the large quantities of specialized contractile proteins found in mature muscle cells; indeed, they look superficially just like the other cells of the limb rudiment. But after several days they begin manufacturing large quantities of specialized muscle proteins, whereas the other limb cells with which they are mingled differentiate into various types of connective-tissue cells. Thus the developmental choice between muscle and connective tissue has been made by each cell long before it is expressed in overt differentiation, and it is meanwhile recorded in each cell as a molecular change that has no obvious effect on the cell's outward appearance.

A cell that has made a developmental choice in the above sense is said to be determined. Since the concept is a basic part of the language of developmental biology, it is useful to have a formal definition: a cell is determined if it has undergone a self-perpetuating change of internal character that distinguishes it and its progeny from other cells in the embryo and commits them to a specialized course of development. The term differentiation is generally reserved for overt cell specialization, that is, for a specialization of cell character that is grossly apparent. Usually, a cell becomes determined before it differentiates, although in some cases the two processes occur simultaneously. Indeed, it is possible for differentiation to occur without determination, if the overt specialization of cell character is reversible.

The Time of Cell Determination Can Be Discovered by Transplantation Experiments ²⁷

To prove that a cell or group of cells is determined, one must show that it has a distinctive character that is maintained even when its circumstances are altered by experimental manipulation. The standard technique is to transplant the cells to a test environment (Figure 21-33).

A simple example of such an experiment comes from studies on amphibian embryos. As noted earlier, one can plot a fate map for a blastula or an early gastrula, showing which of its parts will normally develop into what. The cells in one region, for example, are fated to become epidermis if development proceeds normally, while those in another region are fated to form brain. To establish when these two groups of cells become determined to follow their particular modes of differentiation, a block of cells is cut from the prospective epidermal region and put in the position of prospective brain, and vice versa. If the cells are transplanted at the early gastrula stage, they show no memory of their origins and differentiate in the fashion appropriate to their new locations. If, however, the same experiment is done at a somewhat later stage, in the late gastrula, the prospective brain cells transplanted to an epidermal site will differentiate as misplaced neural tissue, and the prospective epidermal cells transplanted to a brain site will differentiate there as misplaced epidermis. This shows that both groups of cells have become determined sometime between the early and late gastrula stages.

Cell Determination and Differentiation Reflect the Expression of Regulatory Genes ²⁸

The phenomenon of determination raises three molecular questions: what molecule or molecules define a cell's state of determination; what is the memory mechanism that maintains that state; and how is determination coupled to differentiation? In general, the character of a cell is governed by the combination of gene regulatory proteins that it contains. These control its pattern of gene expression. In the well-studied case of muscle, as discussed in Chapter 9, a critical part is played by the MyoD family of closely related myogenic proteins (MyoD, Myf5, MRF4, and myogenin). In suitable circumstances these can activate the expression of muscle-specific genes such as muscle actin and muscle myosin, and introduction of a MyoD family member into fibroblasts and various other cell types can convert them into muscle precursor cells. In normal development genes coding for proteins of the MyoD family begin to be switched on very early in the muscle precursor cells as they leave the somites, suggesting that the presence of these proteins defines the cells' state of determination. And if the myogenin gene is deleted by targeted genetic recombination, for example, muscle cells fail to develop.

The set of genes subject to activation by MyoD family members includes at least some of the genes of that family themselves. For this reason, expression of one member of the family generally leads to expression of others as well. In addition, at least some of these regulatory proteins act back directly on their own gene, so as to maintain expression of the gene once it has been turned on. The positive feedbackresulting from mutual activation and self-activation provides a possible mechanism for cell memory, as discussed in Chapter 9.

This still leaves a problem. The muscle precursor cells do not start to manufacture large quantities of muscle-specific proteins until days, weeks, or even years after leaving the somites. How can they remain undifferentiated for so long after they have become determined? The mechanism is thought to depend on other proteins that interact with MyoD family members and regulate their action. As discussed in Chapter 9, MyoD and its relatives belong to the helix-loop-helix superfamily, whose members dimerize with one another in order to bind to DNA and activate gene expression. The efficacy in gene activation depends on the choice of partner for dimerization. By regulating the availability of appropriate dimerization partners for a protein of the MyoD family, the cell can apparently switch from a determined state, where the protein is able to maintain production of MyoD family members only, to a differentiated state, where the protein activates the full panoply of muscle-specific genes (Figure 21-34).

The State of Determination May Be Governed by the Cytoplasm or Be Intrinsic to the Chromosomes ²⁹

Cell memory, as manifested in the phenomenon of determination, presents one of the most challenging problems in molecular biology. In Chapter 9 we discuss some of the molecular mechanisms by which certain patterns of gene expression can become self-sustaining. In the context of cell determination three broad categories of cell memory can be distinguished, which may be called cytoplasmic, autocrine, and nuclear memory, respectively. The mechanism that has just been outlined for the myogenic proteins is an example of cytoplasmic memory. Here, components encoded by the set of active genes are present in the cytoplasm and act back on the genome, directly or indirectly, to maintain the selective expression of that specific set of genes. An implication of this mechanism is that if a nucleus is taken from one type of differentiated cell and injected into the cytoplasm of another type, the pattern of gene expression should alter to match the character of the host cytoplasm. The nuclear transplantation experiments on amphibian eggs that we discussed earlier provide an example of this sort of behavior.

The autocrine memory mechanism is a variant of the cytoplasmic. It depends again on the synthesis of products that stimulate their own production, but with the special feature that these products are secreted into the extracellular medium and act back on the cell's exterior to keep the cell in the state where it produces them. This mechanism has an important side effect: since neighboring cells share the same extracellular environment, they will tend to behave cooperatively, adopting the same state because they are exposed to the substances that they themselves produce, and an individual cell transplanted into a new environment will tend to switch its character to match that of the cells that surround it on all sides. Thus a group of cells may behave as determined, even though an individual cell in isolation does not. Such "community effects" in cell determination seem to be common and have been especially well documented in the early Xenopus embryo.

In contrast with cytoplasmic and autocrine memory, nuclear memory depends on self-sustaining changes that are intrinsic to the chromosomes - changes that define the selection of genes to be expressed and yet leave the DNA sequence unaltered. X-chromosome inactivation (see p. 446) and genomic imprinting (see p. 451) are well-established examples. Nuclear memory is based on inherited modifications in the chromatin or the DNA; unlike cytoplasmic memory, it allows two identical genes to coexist in different states in a single cell, one being expressed and the other not, even though both are exposed to the same intracellular environment.

Our ignorance is still profound concerning cell memory, and it is not yet possible in most cases even to classify the memory mechanism as cytoplasmic or nuclear.

Cells in Developing Tissues Remember Their Positional Values ³⁰

In an animal embryo positional signals and interactions operate over small distances, on the order of a millimeter or less, and through cell memory these influences leave their mark on cell character. As the body grows, further influences act locally in each of its parts, creating new distinctions within each class of cells and embroidering progressively finer levels of detail on the original basic body plan.

Thus before cells become committed to a particular mode of differentiation, they usually become regionally specified: they acquire distinct biochemical address labels, or positional values, that reflect their location in the body. The positional value of a cell will guide its behavior in subsequent steps of pattern formation - the way it responds to later positional signals, the ways in which it interacts with its neighbors, and the range of modes of differentiation ultimately open to it and its progeny. The cues that control the choice of positional value are said to provide the cell with positional information.

The existence and nature of remembered positional values is dramatically demonstrated by grafting experiments that have been carried out between the developing leg and wing of the chick embryo. The leg and the wing of the adult both consist of muscle, bone, skin, and so on - almost exactly the same range of differentiated tissues. The difference between the two limbs lies not in the types of tissues, but in the way in which those tissues are arranged in space. So how does the difference come about? At first sight it might seem simplest to explain the difference in terms of the presence of a different spatial distribution of signals in the developing forelimb and hindlimb, which directly tells cells which differentiated state to adopt. A simple grafting experiment shows that this view is profoundly wrong.

In the chick embryo the leg and the wing originate at about the same time in the form of small tongue-shaped buds projecting from the flank (Figure 21-35). The cells in the two pairs of limb buds appear similar and uniformly undifferentiated at first (see Figure 19-30). A small block of undifferentiated tissue at the base of the leg bud, from the region that would normally give rise to part of the thigh, can be cut out and grafted into the tip of the wing bud. Remarkably, the graft forms not the appropriate part of the wing tip, nor a misplaced piece of thigh tissue, but a toe (Figure 21-36). This experiment shows that the early leg-bud cells are already determined as leg but are not yet irrevocably committed to form a particular part of the leg: they can still respond to cues in the wing bud so that they form structures appropriate to the tip of the limb rather than the base. The signaling system that controls the differences between the parts of the limb is apparently the same for leg and wing. The difference between the two limbs results from a difference in the internal states of their cells at the outset of limb development. Even though the cells look the same and are destined to give rise to the same range of differentiated cell types, they are nonequivalent, with different positional values. In this way the final specification of how a limb cell should behave is built up combinatorially: first it is supplied with information as to whether it is to be leg or wing; then signals within the growing limb bud specify more fine-grained components of positional value, reflecting the precise position within the limb.

One of the most remarkable revelations of modern molecular genetics has been that almost all animals seem to use the same highly conserved molecular machinery to record positional values along the head-to-tail axis of the body, and some of these same gene products also operate to specify positional values in the limbs of vertebrates. We shall postpone the discussion of these master regulators of the body pattern until we have introduced the fruit fly, Drosophila, where the machinery was first discovered and characterized.

The Pattern of Positional Values Controls Cell Proliferation and Is Regulated by Intercalation ³¹

A crucial aspect of pattern formation is the regulation of cell proliferation, through which the parts of the pattern attain their appropriate sizes. In many cases growth and the pattern of positional values both depend in a closely coupled way on continuing cell-cell interactions. A simple rule has been deduced from studies of the regeneration that occurs in various organisms when fragments of tissue with different positional values are juxtaposed and allowed time to grow and adjust. The principles appear to be general, but they are perhaps most clearly illustrated by studies on the leg of the cockroach (Figure 21-37).

Cockroaches belong to the class of insects in which there is no radical metamorphosis from larva to adult but a gradual progression through a series of juvenile forms separated by molts, in which the old coat of cuticle is shed and a larger one is laid down. The juvenile cockroach has well-differentiated limbs, but the differentiated cells - unlike those in human limbs - are still able to respond to the cues that governed the development of the limb pattern, and they can regenerate that pattern if it is disturbed. Thus the workings of the pattern-formation system can be tested by operations done long after the period of embryonic development.

If two cockroach legs are amputated through one of their middle segments - through the tibia, say - but at different levels, the distal fragment of the one can be grafted onto the proximal stump of the other in such a way that the composite leg heals with the middle part of the tibia missing. Yet the leg that emerges after the animal has molted appears normal: the missing middle part of the pattern has regenerated (Figure 21-38A). More surprising is the result of a variant of this operation. The tibia of one cockroach leg is cut through near the proximal end and that of another leg near the distal end. The large detached portion of the first leg is then stuck onto the large remaining stump of the second leg to give an excessively long leg with a middle part present in duplicate (Figure 21-38B). The animal is left to molt. The leg that results, far from being more nearly normal, is now even longer because a third middle part of a tibia has developed between the two already present. As shown in Figure 21-38B, the bristles on this freshly formed region point in the direction opposite to that of the bristles on the rest of the tibia.

Many different operations of this type can be performed. All of them point to the existence in the insect epidermis of a system of positional values that makes the cells at different positions along the limb axis nonequivalent, and that is intimately coupled to the control of cell proliferation. It is convenient to describe the positional value by a number that goes from a maximum at one end of the limb segment to a minimum at the other. In the operations described above, cells with widely different positional values are brought together. As a result, new cells are formed by proliferation of the cells in the neighborhood of the junction. These new cells acquire positional values interpolated between those of the two sets of cells that were brought into confrontation (Figure 21-38). This behavior is summed up in the rule of intercalation: discontinuities of positional value provoke local cell proliferation, and the newly formed cells take on intermediate positional values so as to restore continuity in the pattern. Cell proliferation ceases only when cells with all the missing positional values have been intercalated in the initial gap and have become spread out to the normal spatial separation from one another. This process as a whole is called intercalary regeneration.

The rule of intercalation, with the corollary that cell proliferation continues until a certain spacing of positional values has been attained, is a powerful organizing principle in those systems to which it applies. Beginning with a pattern specified approximately and in miniature - for example, by a morphogen gradient - it can bring about the construction of a complete accurate pattern of positional values and regulate the growth of each part of the pattern to a standard size: all that is necessary is that the initial pattern should be qualitatively - that is, topologically - correct. The same rule appears to govern many processes of organogenesis and regeneration not only in insects but also in crustaceans and amphibians. Even in creatures such as mammals, where lost structures generally do not regenerate in the adult, the rule of intercalation may help to regulate growth and pattern formation during embryonic development. Unfortunately, the molecular mechanisms that underlie this crucial form of growth control are unknown.

Summary

Embryonic cells must not only become different, they must also remain different even after the influence that initiated cell diversification has disappeared. This requires cell memory, which enables cells to become determined for a particular specialized role long before they differentiate overtly. The mechanisms of cell memory may be cytoplasmic, involving molecules in the cytoplasm that act back on the nucleus to maintain their own synthesis, autocrine, involving secreted molecules that act back on the cell, or nuclear, involving processes of chromatin or DNA modification. In some cases the state of determination has been related to the expression of specific regulatory genes, such as the myogenic genes for muscle cells.

The different kinds of cells in an embryo are produced in a regular spatial pattern. The formation of this pattern usually begins with asymmetries in the egg and continues by means of cell-cell interactions in the embryo. The spatial signals that coordinate pattern formation supply cells with positional information, and a cell's remembered record of this information is called its positional value. Cells in the early forelimb and hindlimb rudiments of a vertebrate embryo, for example, acquire different positional values, making forelimb and hindlimb cells nonequivalent in their intrinsic character, long before the detailed pattern of cell differentiation has been determined.

In many animals the pattern of positional values is closely coupled to the control of cell proliferation according to a simple rule of intercalation. According to the rule, discontinuities of positional value provoke local cell proliferation, and the newly formed cells take on intermediate positional values that restore continuity in the pattern. This mechanism is likely to operate in normal embryonic development to correct inaccuracies in the initial specification of positional information.

Introduction

For cells, as for computers, memory makes complex programs of behavior possible, and many cells together, each one stepping through its complex developmental program, can generate a very complex adult body. Some of the steps that a cell takes in the course of development are autonomous, while others are affected by signals from other cells. Thus the cells of the embryo can be likened to an array of little computers, or automata, operating in parallel and exchanging information with one another. The rules that determine cell behavior are encoded in the cell's genes. Each cell contains the same genome and therefore behaves according to the same rules, but it can exist in a variety of states; the rules direct development along various alternative paths according to a combination of the past information the cell has remembered and the present environmental signals it receives. Computer modeling shows that even a very simple set of rules for the individual automata (cells) in such a system can lead to the production of astonishingly complex patterns; one cannot deduce the rules simply by observing the normal development of the pattern. The challenge, therefore, is to decipher the underlying cellular rules of development by experimentation and to find out how they are specified by the genes.

In this enterprise the nematode worm Caenorhabditis elegans offers some exceptional advantages, and it has become one of the foremost model systems in developmental genetics. We use it here to illustrate some general principles. A detailed discussion of the developmental genetics of pattern formation, however, is reserved for the next section, on Drosophila, where more years of research and a much larger army of research workers have provided a fuller picture.

Caenorhabditis elegans Is Anatomically and Genetically Simple ³³

As an adult, C. elegans is about 1 mm long and consists of only about 1000 somatic cells and 1000-2000 germ cells (exactly 959 somatic cell nuclei plus about 2000 germ cells are counted in one sex; exactly 1031 somatic cell nuclei plus about 1000 germ cells in the other) (Figure 21-39). The anatomy has been reconstructed, cell by cell, by electron microscopy of serial sections. The body plan of this simple worm is fundamentally the same as that of most higher animals in that it has a roughly bilaterally symmetrical, elongate body composed of the same basic tissues (nerve, muscle, gut, skin) organized in the same basic way (mouth and brain at the anterior end, anus at the posterior). The outer body wall is composed of two layers: the protective hypodermis, or "skin," and the underlying muscular layer. A simple tube of endodermal cells forms the intestine. A second tube, located between the intestine and the body wall, constitutes the gonad; its wall is composed of somatic cells, with the germ cells inside it. C. elegans has two sexes - a hermaphrodite and a male. The hermaphrodite can be viewed most simply as a female that produces a limited number of sperm: she can reproduce either by self-fertilization, using her own sperm, or by cross-fertilization after transfer of male sperm by mating. Self-fertilization allows a single heterozygous worm to produce homozygous progeny, a special feature that helps to make C. elegans an exceptionally convenient organism for genetic studies.

The relative simplicity of C. elegans anatomy is reflected in a similar simplicity of its genome. The animal has six homologous pairs of chromosomes, estimated to carry a total of 3000 "essential" genes (that is, genes in which mutations are lethal or have an easily observable effect on the phenotype) and four or five times that number of nonessential genes. The haploid genome consists of approximately 10⁸ nucleotide pairs of DNA, which is about 20 times more than E. coli, about the same as Drosophila, and 30 times less than humans. Currently, more than 900 essential genes have been identified by mutation. These include genes that influence visible features such as the shape or behavior of the worm, genes that code for known proteins such as myosin, and genes that control the course of development. Nearly the entire genome has been mapped as a large set of overlapping DNA segments, represented by a library of ordered genomic clones (see p. 314), and a systematic effort has begun to determine the complete DNA sequence of the organism.

Nematode Development Is Almost Perfectly Invariant ³⁴

C. elegans begins life as a single cell, the fertilized egg, which gives rise, through repeated cell divisions, to 558 cells that form a small worm inside the egg shell. After hatching, further divisions result in the growth and sexual maturation of the worm as it passes through four successive larval stages separated by molts. After the final molt to the adult stage, the hermaphrodite worm begins to produce its own eggs. The entire developmental sequence, from egg to egg, takes only about three days.

Because C. elegans is small and transparent, its individual cells can be followed as they divide, migrate, differentiate, and die in the living embryo, and their pedigree can be traced from egg to adult organism. By this simple technique of direct observation, the behavior and lineage of all of the cells from the single-cell egg to the adult animal have been described. This has made possible a detailed lineage analysis that would be very difficult in larger animals, where individual cells at early stages usually must be specially marked if they and their progeny are to be identified later. Moreover, in larger animals the details of cell lineage show many random variations, even between genetically identical individuals. In the nematode, by contrast, the somatic structures develop by an invariant, predictable cell lineage, and each of the many cell divisions is precisely timed. This means that a given precursor cell follows the same pattern of cell divisions in every individual, and with very few exceptions the fate of each descendant cell can be predicted from its position in the lineage tree (Figure 21-40).

The full description of cell lineage in C. elegans leads to an immediate answer to a fundamental question. The nematode, like most animals, is formed from a relatively large number of cells that can be classified into a much smaller number of differentiated cell types. Given the importance of cell ancestry, one might be tempted to guess that all the cells of a given type are descendants of a single "founder cell" committed exclusively to that developmental pathway. Lineage analysis shows, however, that this is not generally true, either for nematodes or for other animals. Thus in C. elegans (with a few exceptions such as the intestinal cells and the germ-line cells) each class of differentiated cells - hypodermal, neuronal, muscular, gonadal - is derived from several founder cells originating in separate branches of the lineage tree (see Figure 21-40). Thus cells of similar character need not be close relatives. Conversely (but rarely), cells of very different character may be closely related by lineage; for example, some of the neurons in C. elegans are sisters of muscle cells.

The problem, then, is to understand the rules that operate in each branch of the lineage tree to generate a specific array of cell types, each in appropriate numbers.

Developmental Control Genes Define the Rules of Cell Behavior That Generate the Body Plan ³⁵

To explain how the genome specifies the developmental rules, one has to be able to identify the genes that control the cells' developmental choices. Mutations in such genes will disturb development, but they are not the only mutations that do so. Some mutations, for example, will cut short all cell lineages and cause premature death of the embryo simply because they disrupt "housekeeping" genes that every cell needs in order to survive and proliferate. Other mutations will affect genes for proteins that particular types of differentiated cells require in order to carry out their specialized function; the body plan will then be essentially normal, but certain cell types, though still identifiable, will malfunction. Mutations in genes that are involved specifically in controlling developmental choices, by contrast, will disturb the body plan: they typically give rise to cells of the normal differentiated types arranged in an abnormal pattern or in abnormal numbers as a result of specific alterations in the lineage tree. Developmental control genes identified in this way can be classified according to the parts of the lineage tree that are affected and, hence, if we know the rules of cell behavior that generate that part of the lineage tree, according to the rules of cell behavior for which they are responsible.

To illustrate the principles of genetic analysis of a developmental mechanism, we discuss one example of a cell-cell interaction in the nematode - the induction of the vulva.

Induction of the Vulva Depends on a Large Set of Developmental Control Genes ³⁶

The vulva - the egg-laying orifice in a hermaphrodite - is a ventral opening in the hypodermis (skin) formed by 22 cells that arise by specific lineages from three precursor cells in the hypodermis. A single nondividing cell in the gonad, called the anchor cell, attaches, or "anchors," the developing vulva to the overlying gonad (the uterus) to create a passageway through which the eggs can pass to the outside world. Microsurgical experiments show that the anchor cell is responsible for inducing the three nearest hypodermal cells to form a vulva (Figure 21-41). If the anchor cell is destroyed by focusing a laser beam on it, these cells, instead of forming a vulva, give rise to ordinary hypodermal cells. And if the anchor cell is shifted relative to the hypodermal cells, there is a corresponding shift in the site at which the vulva develops: flanking the three cells that normally give rise to the vulva lie three others that are also capable of doing so if exposed to the anchor-cell signal. Thus the anchor cell induces vulval differentiation in C. elegans just as the vegetal blastomeres induce mesodermal differentiation in the early Xenopus embryo. Only the anchor cell is necessary for this induction: if all the gonadal cells except the anchor cell are destroyed, the vulva still develops normally.

To identify genes involved in a given step of development, one searches for mutations that disrupt the process by screening the progeny of a large population of animals that have been exposed to mutagens. In this way many mutants are found that have a "vulvaless" phenotype, where none of the hypodermal cells behave as though they have received the anchor-cell signal. Another large group of mutants have a converse "multivulva" phenotype, in which all six hypodermal cells capable of responding to the anchor-cell signal behave as though they have actually received it, so that the worm forms several vulvalike structures instead of one. Individual mutations giving a similar phenotype are then tested in pairs to see whether they affect the same or different genes, as explained in Panel 21-1, pages 1072-1073. Once a set of relevant genes has been identified in this way, still more components of the system usually can be discovered by searching for mutations in other genes that will suppress the ill effects of mutations in an already identified gene. Such extragenic suppressor mutations can be rare, and it is only in genetically favorable organisms such as C. elegans that one can easily find them; but when found, they often identify genes whose protein products interact directly with those of the already identified gene (because the alteration in the shape of the one protein molecule, for example, can be compensated for by a complementary alteration in the shape of its partner). More than 30 distinct identified genes have been implicated in the control of vulval development.

Genetic and Microsurgical Tests Reveal the Logic of Developmental Control; Gene Cloning and Sequencing Help to Reveal Its Biochemistry ³⁷

We focus here on just five of the vulval control genes, called lin-3, let-23, sem-5, let-60,and lin-45. Impairment of the function of any one of them by mutation has the same consequence - a vulvaless phenotype. Conversely, a genetic change causing excess of any of the gene products or excessive or unregulated activity - in other words, a gain of function - can have an opposite, multivulva effect. Each of the five genes therefore is needed for induction of the vulva, in a way that suggests they might all be links in a single chain of cause and effect; all of them, that is, might belong to a single genetic pathway. We saw in Chapter 15 how the signaling pathway that controls specialization of a particular cell type in the Drosophila eye has been defined by genetic analysis. A similar kind of analysis has been used to determine the order in which the vulval control genes act, as explained in Figure 21-42. The five genes do indeed appear to lie in a single genetic pathway, with lin-3 the most upstream, then let-23, sem-5, let-60, and lastly lin-45. Thus, for example, a gain-of-function mutation in lin-3 has no effect on the phenotype in an animal that also carries a loss-of-function mutation of let-23; the double mutant is vulvaless because the upstream component can do nothing when the downstream component upon which it should operate is missing.

The next problem is to relate the gene actions to specific cells in the embryo. Is lin-3, for example, needed in the anchor cells that produce the inductive signal or in the hypodermal cells that respond to it? For lin-3a simple answer has come from molecular genetics: the gene has been cloned and has been shown to be expressed in the anchor cell and nowhere else in the neighborhood (Figure 21-43). The other four genes, by contrast, appear to function in the hypodermal cells, and a gain-of-function mutation in one of them can cause a multivulva phenotype even when the anchor cell has been destroyed.

To complete the picture, we have to relate the genetically defined pathway to protein molecules and biochemistry. The five genes lin-3, let-23, sem-5, let-60, and lin-45 have all been cloned and sequenced, and in each case the sequence indicates the probable function: lin-3 codes for a protein similar to a secreted signaling molecule well known in vertebrates - epidermal growth factor (EGF); let-23 codes for a receptor tyrosine kinase homologous to the members of the vertebrate EGF receptor family; sem-5, as we saw in Chapter 15, codes for a protein containing the SH2 and SH3 domains, found in many proteins that directly bind to such receptors and mediate their effects on other intracellular components; and let-60 and lin-45 are respectively homologous to the vertebrate ras and raf genes, whose products relay signals intracellularly from such receptors into the cell interior, as discussed in Chapter 15. Presumably, therefore, the Lin-3 protein is the signal molecule secreted by the anchor cell, the Let-23 protein is the transmembrane receptor in the hypodermal cells to which it binds, and the Sem-5, Let-60, and Lin-45 proteins are links in the intracellular signaling chain through which binding of the ligand to the receptor exerts its ultimate effects on gene expression and cell determination (Figure 21-44). In fact, the genetic analysis of this system in the developing nematode worm provides one of the clearest accounts we have of the organization of a signaling pathway that appears to have been conserved throughout most of the animal kingdom. A very similar pathway, as we saw in Chapter 15, emerges from analysis of the sevenless mutant in Drosophila (see p. 764).

Heterochronic Mutations Identify Genes That Specify Changes in the Rules of Cell Behavior as Time Goes By ³⁸

As computer programmers know, small changes in a program can have drastic effects on the output produced when a program is executed. Likewise, a mutation in a control gene that alters a single rule of cell behavior can result in a grossly abnormal cell lineage tree. This is well illustrated by heterochronic mutations in C. elegans, which cause certain sets of cells to behave in a way that would be appropriate for normal cells at a different stage in development. A daughter cell may behave like its parent or grandparent, for example, and the offspring of the daughter may behave again in the same way, and so on, with the result that a portion of the lineage pattern is reiterated indefinitely.

Figure 21-45 shows lineage diagrams for a set of mutations in a gene called lin-14, illustrating this phenomenon: instead of progressing through the normal series of cell divisions characteristic of the first, second, third, and fourth larval stages and then halting, many of the cells in gain-of-function lin-14 mutants repeatedly go through the patterns of cell divisions characteristic of the first larval stage, continuing through as many as five or six molt cycles and persisting in the manufacture of an immature type of cuticle. Loss-of-function mutations in the lin-14 gene have the reverse effect, causing cells to adopt mature states precociously, skipping intermediate stages, so that the animal reaches its final state prematurely and with an abnormally small number of cells.

The lin-14 gene has been cloned, and the protein it encodes has been found to be concentrated in cell nuclei. In a normal individual the protein is present in most of the somatic cells of the late embryo and early first larval stage, but its concentration then declines to near zero by the second larval stage. Those lin-14 mutants that enter an adult state precociously are found to have a reduced level of the Lin-14 protein, whereas those mutants that carry on with repeated first larval stage cycles are found to express the Lin-14 protein for an abnormally long time (because of a mutation in a regulatory portion of the gene). Thus the effect of the Lin-14 protein is to keep the cells in an immature state, and normal maturation depends on its disappearance. This gene product is presumably only one of many whose changing concentrations in cells specify changes in the rules of cell behavior as development proceeds.

The Tempo of Development Is Not Controlled by the Cell-Division Cycle ³⁹

The example we have just discussed brings us to a fundamental general problem in development. The genome has to define a set of rules for cell division as well as for cell specialization, and the two processes have to be coordinated. How is the division cycle regulated in development, and how is it coordinated with cell specialization?

One suggestion is that changes of internal state might be locked to passage through the division cycle: the cell would click to the next state as it went through mitosis, so to speak. This seems a tempting idea, especially when one pictures development in terms of lineage diagrams, but the evidence is largely against it. Cells in developing embryos frequently go on to differentiate in an almost normal way even when cell division is artificially prevented. Necessarily, there are some abnormalities, if only because a single undivided cell cannot differentiate normally in two ways at once. But in most cases that have been studied, it seems clear that cell divisions are not the ticks of a clock that sets the tempo of development. Rather, the cell changes its chemical state with time regardless of cell division, and this changing state controls both the decision to divide and the decision as to when and how to specialize.

Cells Die Tidily as a Part of the Program of Development ⁴⁰

A C. elegans hermaphrodite generates 1030 somatic cell nuclei in the course of its development, but 131 of the cells die. These programmed cell deaths occur in an absolutely predictable pattern, and they create no mess. Whereas cells that die from damage or poisoning typically swell and burst, spilling their contents over their neighbors, these normal cell deaths occur by a process known as apoptosis, in which the cell nucleus becomes condensed, the cell itself shrivels, and the shrunken corpse is rapidly engulfed and digested by neighboring cells (Figure 21-46). Programmed cell death is a regular feature of normal animal development and is probably the fate of a substantial fraction of the cells produced in most animals.

Because apoptosis occurs quickly and leaves no trace, the deaths easily go unnoticed. Yet cell death may be as important as cell division in generating an individual with the right cell types in the right numbers and places. In vertebrates, for example, it regulates the numbers of neurons (as we discuss later), eliminates undesirable types of lymphocytes (discussed in Chapter 23), disposes of cells that have finished their job (as when a tadpole loses its tail at metamorphosis), and helps to sculpt the shapes of developing organs (creating the gaps between digits by doing away with the cells that lie between the digit rudiments in the limb bud, for example).

Normal cell deaths are thought to be suicides in which the cell activates a death program and kills itself. The best evidence that animal cells have an intrinsic death program comes from genetic studies in C. elegans, where two genes, called ced-3 and ced-4 (ced stands for "cell death abnormal"), have been identified that are required for the 131 normal cell deaths to occur. If either gene is inactivated by mutation, the cells that are normally fated to die survive instead, differentiating as recognizable cell types such as neurons. Conversely, overexpression or misplaced expression of ced-3 and ced-4 (as a result of loss-of-function mutations that inactivate another gene, ced-9, which normally represses the death program) causes many cells to die that would normally survive.

The amino acid sequences of these three Ced proteins are known. The Ced-4 protein is novel and is thought to act upstream of Ced-3, which is a protease. Ced-9 is 23% identical in amino acid sequence to a mammalian protein called Bcl-2 (the product of the proto-oncogene bcl-2), which acts like Ced-9 to suppress programmed cell death in many types of mammalian cells. Remarkably, when the human bcl-2 gene is transferred to C. elegans, it acts to inhibit normal cell death in the worm and is even able to rescue ced-9 mutants that otherwise die early in development. These important findings indicate that both the mechanism of programmed cell death and its regulation have been highly conserved in evolution from worms to humans, confirming that the ability to commit suicide in this way is a fundamental property of animal cells.

Summary

Two things make the nematode Caenorhabditis elegans an attractive organism for investigating the genetic basis of development: first, genetic analysis is easy because the generation time is short and the genome small; second, the normal course of development is extraordinarily reproducible and has been chronicled in detail, so that a cell at any given position in the body has the same lineage in every individual, and this lineage is fully known. As in other organisms, development depends on an interplay of cell-cell interactions and cell-autonomous processes. Cell-destruction experiments show, for example, that the development of the vulva depends on an inductive signal, and the genes required for this induction can be identified through mutations that disrupt vulval development. Molecular genetic analysis reveals the individual functions of these genes and shows that several of them code for components of a signaling pathway that operates in vertebrates too. Lineage analysis of mutants leads to the discovery of many other important classes of genes, including genes whose products serve to specify changes in the rules of cell behavior with time during development and genes that are responsible for programmed cell death - an invariable feature of development in all animals.

Introduction

The structure of an organism is controlled by its genes: classical genetics is based on this proposition. Yet for almost a century, and even long after the role of DNA in inheritance had become clear, the mechanisms of the genetic control of body structure remained an intractable mystery. In recent years this chasm in our understanding has begun to be filled. In the previous section we used the nematode worm to illustrate some of the general principles of how developmental control genes orchestrate the events of development. But it is the fly Drosophila melanogaster (Figure 21-47), more than any other organism, that has really transformed our understanding of how genes govern the patterning of the body. Decades of genetic study, culminating in massive systematic searches, have yielded a large catalogue of developmental control genes in the fly whose specific function is to define the spatial pattern of cell types and body parts. It has become possible not only to identify the key genes, but also to watch them at work: by in situ hybridization using DNA or RNA probes, one can observe directly how the internal states of the cells in the embryo are defined by the sets of regulatory genes that they express. By analyzing mutants, transgenic animals, and animals that are a patchwork of mutant and nonmutant cells, one can go on to discover how each gene operates as part of a system to specify the organization of the body. Moreover, the fly has provided a crucial key to our own development; for the genes controlling the pattern of the body in Drosophila turn out to have close counterparts in higher animals, including ourselves.

Our account of Drosophila developmental genetics is divided into two sections. The first deals with events in the early embryo and describes how the basic body plan is created, with a head rudiment at one end, a posterior rudiment at the other, and in between them an ordered series of segments - the basic modular units from which all insects are constructed. The second section deals with later events and discusses the genetic apparatus that endows cells with positional values that make the cells of one segment different from those of the next; these processes ensure that, for example, the head will develop antennae and the thorax legs - and not, as happens in some mutants we shall encounter, the other way around.

The Insect Body Is Constructed by Modulation of a Fundamental Pattern of Repeating Units ^{41, 42}

The timetable of Drosophila development, from egg to adult, is summarized in Figure 21-48. The period of embryonic development begins at fertilization and takes about a day, at the end of which the embryo hatches out of the egg shell to become a larva. The larva then passes through three stages, or instars, separated by molts in which it sheds its old coat of cuticle and lays down a larger one. At the end of the third instar it pupates. Inside the pupa a radical remodeling of the body takes place, and eventually, about nine days after fertilization, an adult fly, or imago, emerges.

The fly consists of a head, three thoracic segments (numbered T1 to T3), and eight or nine abdominal segments (numbered A1 to A9). Each segment, although different from the others, is built according to a similar plan. Segment T1, for example, carries a pair of legs, T2 carries a pair of legs plus a pair of wings, and T3 carries a pair of legs plus a pair of halteres - small knob-shaped balancers important in flight, evolved from the second pair of wings that more primitive insects possess. The quasi-repetitive segmentation develops in the early embryo during the first few hours after fertilization, but it is more obvious in the larva, where the segments look more similar than in the adult. In the embryo it can be seen that the rudiments of the head, or at least the future adult mouth parts, are likewise segmental (Figure 21-49). At the two ends of the animal, however, there are highly specialized terminal structures that are not segmentally derived.

It is partly a matter of convention where one draws the boundary between one segmental unit and the next. In discussing patterns of gene expression, we shall see that it is convenient to speak in terms of a total of 14 parasegments (numbered P1 to P14) that are half a segment out of register with traditionally defined segments (Figure 21-50).

Drosophila Begins Its Development as a Syncytium ^{41, 43}

The egg of Drosophila is about 400 µm long and about 160 µm in diameter, with a clearly defined polarity. Like the eggs of other insects, it begins its development in an unusual way: a series of nuclear divisions without cell division creates a syncytium. The early nuclear divisions are synchronous and extremely rapid, occurring about every 8 minutes. The first nine divisions generate a cloud of nuclei, most of which migrate from the middle of the egg toward the surface, where they form a monolayer called the syncytial blastoderm. After another four rounds of nuclear division, plasma membranes grow inward from the egg surface to enclose each nucleus, thereby converting the syncytial blastoderm into a cellular blastoderm consisting of about 6000 separate cells (Figure 21-51). A small subset of nuclei populating the extreme posterior end of the egg are segregated into cells a few cycles earlier; these pole cells are the primordial germ cells that will give rise to eggs or sperm.

As in a cleaving amphibian egg, the very rapid cycles of DNA replication seem to hinder transcription, so that up to the cellular blastoderm stage development depends largely - although not exclusively - on stocks of maternal mRNA and protein that accumulated in the egg before fertilization. After cellularization, cell division continues in a more conventional way, asynchronously and at a slower rate, and the rate of transcription increases dramatically.

The cellular blastoderm corresponds to the hollow blastula of an amphibian or a sea urchin, even though its interior is filled with yolk rather than being a fluid-filled cavity. Gastrulation follows as soon as cellularization is complete, and although the geometry of this process is very different in the insect, the general outcome is similar. Through coordinated cell movements, endodermal cells are invaginated into the interior to form the gut extending along the axis of the embryo. Mesoderm surrounds the gut rudiment and occupies the space between it and an enveloping layer of ectoderm on the exterior.

By marking and following the cells through their complex gastrulation movements, one can draw a fate map for the monolayer of cells on the surface of the blastoderm (Figure 21-52). The fate map is especially simple for a cross-section through the middle of the embryo, with prospective mesoderm ventrally and ectoderm on each side above it. As in a vertebrate, the cords of nerve cells that run the length of the body derive from part of the ectoderm: a subset of the cells in this neurogenic ectoderm will detach from their neighbors, escape from the epithelial sheet, and move into the interior of the embryo as neuronal precursors. For mesoderm, ectoderm, and nerve cord, the position of the cells along the anteroposterior axis is roughly preserved during gastrulation because their movements are in the transverse plane. The gut, however, is formed by invagination of two groups of cells from the opposite extremities of the embryo; these two invaginations meet in the middle to form eventually a continuous gut tube.

As gastrulation nears completion, a series of indentations and bulges appear in the surface of the embryo, marking the subdivision of the body into parasegments along its anteroposterior axis (see Figure 21-49). More subtle tests show that the main features of this segmental pattern are already established at the cellular blastoderm stage, before gastrulation begins.

Two Orthogonal Systems Define the Ground Plan of the Embryo ⁴⁴

Two coordinates are needed to define each position in the blastoderm, and, correspondingly, one can distinguish two sets of egg-polarity genes that act independently at the outset of development to specify the two main axes of the embryo - the dorsoventral and the anteroposterior. These genes define the spatial coordinates of the embryo by setting up morphogen gradients in the egg.

The egg-polarity genes were found by exhaustive searches for mutants in which the polarity of the embryo is disrupted. In this way 12 dorsoventral egg-polarity genes were discovered. All but one of these have the same loss-of-function mutant phenotype, in which the embryo is dorsalized - that is, all its cells take on a dorsal character, so that the normal ventral structures fail to form. The remaining gene has the opposite loss-of-function phenotype - the embryo is ventralized. We shall see that all these genes are components of a single system that sets up a dorsoventral morphogen gradient in the early embryo.

The anteroposterior set of genes, by contrast, can be subdivided according to their mutant phenotypes into three subsystems, responsible for specifying different parts of the anteroposterior axis (Figure 21-53). The anterior group (4 genes) governs the anterior part of the axis. The posterior group (11 genes) governs the posterior part of the axis. Lastly, the terminal group (6 known genes) governs the two extreme ends of the embryo, comprising the specialized nonsegmental terminal structures and in particular the pair of regions - one anterior, one posterior - from which the gut is derived. Like the dorsoventral system, each of these three subsystems sets up a morphogen gradient - one in the anterior half of the embryo, one in the posterior half (although this is somewhat controversial), and one operating symmetrically at both of the extreme ends of the embryo. Loss-of-function mutations that inactivate a particular subsystem cause a loss of the corresponding anterior, posterior, or terminal structures.

The four primary spatial signals - anterior, posterior, terminal, and ventral - organize the subsequent patterning of the embryo by governing the expression of other sets of genes, which serve to interpret, refine, and record the positional information that the primary signals supply.

The Patterning of the Embryo Begins with Influences from the Cells Surrounding the Egg ^{44, 45}

The egg-polarity genes are transcribed from the maternal genome during oogenesis, and their products act very soon after fertilization or in some cases even before. Thus the phenotype of the embryo is determined by the alleles present in the mother (and in her oocytes) rather than by the combination of maternal and paternal genes possessed by the embryo itself. Genes acting in this way are called maternal-effect genes. They are discovered by looking for the appropriate mutant phenotypes in the embryos produced from eggs laid by mothers who themselves appear normal but who carry a genetic mutation that makes their eggs abnormal (Figure 21-54). Most often, the maternal-effect mutation is recessive, and the mothers who make the defective eggs are homozygous for the mutant gene.

Once a gene has been identified, its site of action can be investigated by creating and analyzing genetic mosaics - flies containing marked clonal patches of cells in which the gene of interest is missing or mutated. (We explain later how this astonishing trick of genetic microsurgery is performed.) In the case of the egg-polarity genes it can be shown in this way that, while most are required in the oocyte lineage itself, a few crucial ones are required instead in the follicle cells that surround the oocyte in the ovary. The genes required in the follicle cells supply cues that act on the outside of the egg to localize the sources of the dorsoventral and terminal morphogen gradients that will develop inside it (Figure 21-55). In addition, localized products supplied to the growing oocyte by the giant nurse cells connected to it at one end serve to define the anteroposterior polarity of the egg.

To see how the patterns are set up inside the egg, we focus first on the dorsoventral system.

The Dorsoventral Axis Is Specified Inside the Embryo by a Gene Regulatory Protein with a Graded Intranuclear Concentration ^{44, 45, 46}

The role of the follicle cells in establishing the dorsoventral gradient in the Drosophila egg is to provide a localized signaling molecule that binds to a receptor on the outside of the egg and thereby controls the distribution of a gene regulatory protein inside the egg. The system can be analyzed genetically in much the same way as described earlier for the system mediating vulval induction in the nematode worm. Seven of the genes in the dorsoventral system are concerned with producing the localized extracellular signal; one, called Toll, encodes the transmembrane receptor for the signaling molecule, and the products of the remaining three act inside the embryo, downstream from Toll. The final maternal-effect gene in the signaling pathway codes for a gene regulatory protein and is called dorsal. The extracellular signaling molecule produced by the follicle cells is generated in active form only at the ventral surface of the egg and forms a gradient that is reflected in a graded activation of the Toll protein and ultimately in a graded concentration of the Dorsal protein in the nuclei of the embryo.

The Dorsal protein belongs to the same family as the NF-kB gene regulatory protein of vertebrates (see Figure 15-32) and is thought to act in a similar way. In the newly laid egg both the dorsal mRNA (detected by in situ hybridization) and the protein it encodes (detected with antibodies) are distributed uniformly in the cytoplasm. After the nuclei have migrated to the surface of the embryo to form the blastoderm, however, a remarkable redistribution of the Dorsal protein occurs: dorsally the protein remains in the cytoplasm, but ventrally it is concentrated in the nuclei, and between these two extremes there is a smooth gradient of nuclear localization (Figure 21-56). The partitioning of Dorsal protein between nucleus and cytoplasm appears to be governed, in part at least, by the product of a gene called cactus. The Cactus protein is homologous to the I-kB protein that inhibits NF-kB in vertebrate cells by preventing it from migrating into the nucleus (see Figure 15-32). By analogy, the Cactus protein is thought to bind to the Dorsal protein, trapping it in the cytoplasm; the signal transmitted by the Toll protein is thought to lead to the phosphorylation of the Dorsal protein, causing it to dissociate from the Cactus protein so that it can enter nuclei.

Once inside a nucleus the Dorsal protein turns on or off the expression of different sets of genes depending on its concentration. In this way the gradient of nuclear localization of the protein creates a dorsoventral series of territories - distinctive bands of cells that run the length of the embryo. Most ventrally, where the concentration of Dorsal protein is highest, it switches on, for example, expression of a gene called twist, specific for mesoderm. Most dorsally, where the concentration of Dorsal protein is lowest, a gene called decapentaplegic (dpp) is permitted to switch on, specifying dorsal structures. And in an intermediate region, where the concentration of Dorsal protein is high enough to repress dpp but too low to activate twist, the cells are specified to become neurogenic ectoderm (see Figure 21-56).

Products of the genes directly regulated by the Dorsal protein generate in turn more local signals that define finer subdivisions of the dorsoventral axis. In particular, dpp codes for a secreted protein of the TGF-β superfamily that is thought to form a local morphogen gradient in the dorsal part of the embryo. The action of this protein is reminiscent of the action of activin, also a TGF-β family member, in early Xenopus development. From experiments with injected dpp mRNA, it seems that the highest concentrations of Dpp protein cause development of the most dorsal tissue of all - extraembryonic membrane - intermediate concentrations cause development of dorsal ectoderm, and very low concentrations allow development of neurogenic ectoderm.

Like the dorsoventral system, the terminal system depends on a transmembrane receptor that detects localized signals provided by follicle cells to generate gradients of gene regulatory proteins inside the embryo (Figure 21-57). These gradients serve to specify gut endoderm, as well as some specialized terminal structures, and so can be viewed, with the dorsoventral system, as part of the apparatus for defining the three basic germ layers of the insect. The dorsoventral and terminal systems in the fly, therefore, employing secreted molecules that act as inductive signals, are comparable with the inductive mechanisms for specifying germ layers in the early Xenopus embryo.

The anterior and posterior systems of egg-polarity genes, by contrast, set up gradients that depend instead on localized accumulations of specific mRNAs inside the egg (see Figure 21-57). These gradients govern the differences between head and rear and specify the series of body segments along the head-to-rear axis, as we shall see in detail for the anterior system. First, we pause briefly to discuss a special role of the posterior system: the specification of germ cells.

The Posterior System Specifies Germ Cells as well as Posterior Body Segments ⁴⁷

In practically all animals that have been studied, the primordial germ cells - the precursors of the next generation of gametes - are singled out at a very early stage of development from the somatic cells - those that will form all the other tissues of the body (see Figure 21-51). In many species the egg contains localized cytoplasmic components - visible as polar granules in C. elegans and Drosophila, or as germ plasm in Xenopus - that are segregated into the primordial germ cells during egg cleavage and are suspected to include or to be associated with the determinants of germ-cell character. These components are generally concentrated at the posterior or vegetal end of the egg, and the cells that inherit them migrate from that site to colonize the gonads.

In Drosophila maternal-effect genes required for the formation of germ cells can be identified through the discovery of mutants that produce offspring in which germ cells are lacking. These abnormal offspring are found to lack posterior body segments also, indicating that the genes belong to the posterior system of egg-polarity genes. The products of many of these genes turn out to be localized at the posterior pole - among them, presumably, the determinants of germ-cell character. The morphogen gradient that organizes the posterior body segments depends on the machinery that creates and localizes the germ cell determinants. A key gene in this system is called oskar. Normally, oskar mRNA and protein are localized at the posterior pole of the egg. In their absence no germ cells develop there, and if oskar mRNA is artificially misdirected to the anterior end of the egg, germ cells will form there instead. The localized oskar mRNA, moreover, can be shown to control the localization of other components - products of other posterior-group genes involved in the development of posterior body segments as well as germ cells. By their localization at the posterior pole of the egg, these products can become specifically incorporated in the cells that form there, determining their fate as germ cells.

mRNA Localized at the Anterior Pole Codes for a Gene Regulatory Protein That Forms an Anterior Morphogen Gradient ^{44, 48}

If a Drosophila egg is carefully punctured at its anterior end, allowing a small amount of the most anterior cytoplasm to leak out, the embryo fails to develop head segments. And if cytoplasm from the posterior end of another egg is injected into the site from which the anterior cytoplasm has leaked, a second set of abdominal segments will develop, with reversed polarity, in the anterior half of the recipient egg (Figure 21-58). This experiment shows that the segmental patterning of the anteroposterior axis is controlled by substances localized at the ends of the egg. These substances have been identified by the genetic approach, starting with a search for mutations that mimic the effects of losing anterior or posterior cytoplasm. Most notably, mothers that are homozygous for a mutation in the egg-polarity gene bicoid produce embryos that lack head and thoracic structures and have abdominal structures extended over an abnormally large fraction of the body length. Such a mutant embryo can be rescued from abnormal development, however, if cytoplasm from the anterior end of a normal egg is injected into its anterior end. Thus the normal bicoid gene is required to make some product at the anterior end of the egg that can act as the source of a long-range influence controlling the pattern of development of the anterior parts.

In situ hybridization studies show that bicoid mRNA is originally synthesized in the ovary by the nurse cells connected with the oocyte (see Figure 21-55). As the bicoid mRNA passes through the cytoplasmic bridges into the oocyte, it becomes anchored by part of its 3' untranslated tail to a component of the cytoplasm - presumably a part of the cytoskeleton - at the oocyte's anterior end. Translation of this mRNA begins only when the egg is laid, giving rise to a concentration gradient of Bicoid protein with its high point at the anterior end of the embryo. The concentration gradient can be altered genetically by constructing mutants that contain multiple copies of the normal bicoid gene: as the gene dosage increases in the mother, so does the protein concentration increase in the egg. The segments of the resultant embryo are correspondingly shifted toward the posterior pole, as though their locations were determined by positional information derived from the local concentration of the Bicoid protein (Figure 21-59). This protein therefore fits exactly the definition of a morphogen. Like Dorsal, the Bicoid protein binds to DNA and functions by regulating the expression of other genes.

Three Classes of Segmentation Genes Subdivide the Embryo ⁴⁹

Graded global cues are thus provided inside the egg by the products of the egg-polarity genes. For the anterior system the cues derive from the bicoid mRNA that is localized at the anterior end of the egg before fertilization, and they take the form of an anteroposterior gradient of the Bicoid gene regulatory protein. The gradient guides the creation of a series of discrete body segments. This process depends on a collection of segmentation genes, about 25 of which have been characterized. Mutations in any one of these genes will alter the number of segments or their basic internal organization without affecting the global polarity of the egg. The segmentation genes act at later stages than the egg-polarity genes, when the embryo is transcribing its own genome instead of relying on stored maternal mRNA. Because the embryonic gene transcripts, rather than maternal transcripts, determine the phenotype, these genes are classed as zygotic-effect genes rather than maternal-effect genes.

The segmentation genes fall into three groups according to their mutant phenotypes and the stages at which they act (Figure 21-60). First come a set of at least six gap genes, whose products mark out the coarsest subdivisions of the embryo. Mutations in a gap gene eliminate one or more groups of adjacent segments, and mutations in different gap genes cause different but partially overlapping defects. In the mutant Krüppel, for example, the larva lacks eight segments, from T1 to A5 inclusive.

The next segmentation genes to act are a set of eight pair-rule genes. Mutations in these cause a series of deletions affecting alternate segments, leaving the embryo with only half as many segments as usual. While all the pair-rule mutants display this two-segment periodicity, they differ in the precise positioning of the deletions relative to the segmental or parasegmental borders. The pair-rule mutant even-skipped, for example, which is discussed in Chapter 9, lacks the whole of each even-numbered parasegment, while the pair-rule mutant fushi tarazu (ftz) lacks the whole of each odd-numbered parasegment, and the pair-rule mutant hairy lacks a series of regions that are of similar width but out of register with the parasegmental units.

Finally, there are at least 10 segment-polarity genes. Mutations in these genes produce larvae with a normal number of segments but with a part of each segment deleted and replaced by a mirror-image duplicate of all or part of the rest of the segment. In gooseberry mutants, for example, the posterior half of each segment (that is, the anterior half of each parasegment) is replaced by an approximate mirror image of the adjacent anterior half-segment (see Figure 21-60).

We see later that, in parallel with the segmentation process, a further set of genes, the homeotic selector genes, serve to define and preserve the differences between one parasegment and the next.

The phenotypes of the various segmentation mutants suggest that the segmentation genes form a coordinated system that subdivides the embryo progressively into smaller and smaller domains along the anteroposterior axis distinguished by different patterns of gene expression. Again, molecular genetics provides the tools to investigate how this system works.

The Localized Expression of Segmentation Genes Is Regulated by a Hierarchy of Positional Signals ^{50, 51}

Most of the segmentation genes have been cloned, and cDNA sequencing reveals that about three-quarters of them, including all of the gap genes, code for gene regulatory proteins. Their actions on one another and on other genes can therefore be observed by comparing gene expression in normal and mutant embryos. Using appropriate probes to detect the gene transcripts, one can, in effect, take snapshots as genes switch on or off in changing patterns. By analyzing in this way mutants that lack a particular segmentation gene, one can begin to deduce the logic of the gene control system.

We have already seen how in situ hybridization in normal embryos has helped to show that the bicoid gene transcripts are the source of a positional signal: the transcripts are localized at one end of the egg, even though the effects of a mutation in the gene are spread over a large part of the embryo. In a similar way it can be shown that the gap genes in their turn generate (directly or indirectly) positional signals that help to control the pattern of development in neighborhoods extending beyond their own expression domain. Mutants that are defective in the gap gene Krüppelor hunchback, for example, show abnormalities within the region where the gene transcripts are detected in a normal embryo and also for several segments beyond (Figure 21-61). As with the Bicoid protein, it is thought that the gene regulatory proteins encoded by gap genes such as Krüppel and hunchback spread out as diffusible morphogens from the sites where the genes are transcribed.

The next finer level of spatial patterning is marked out by the pair-rule genes. Some of these, too, may code for proteins that spread by diffusion to exert effects on cells neighboring the site of gene transcription; others, by contrast, appear to affect the development only of those regions in which they are transcribed. Transcripts of the normal ftz gene, for example, occur in seven circumferential "zebra stripes" at the blastoderm stage (Figure 21-62), each of the stripes being roughly four cells wide, matching in width and location the rudiments of the even-numbered parasegments that would be missing in a ftz mutant.

Taken together, these observations imply that the products of the egg-polarity genes provide global positional signals that cause particular gap genes to be expressed in particular regions, and the products of the gap genes then provide a second tier of positional signals that act more locally to regulate finer details of patterning by influencing the expression of yet other genes, including the pair-rule genes. In this way the global gradients produced by the egg-polarity genes organize the creation of a fine-grained pattern through a process of sequential subdivision using a hierarchy of sequential positional controls. This is a reliable strategy: because the global positional signals do not have to specify fine details, the individual nuclei that respond to them do not have to react with extreme precision to small differences of signal concentration (Figure 21-63).

The Product of One Segmentation Gene Controls the Expression of Another to Create a Detailed Pattern ^{41, 50, 51, 52}

The hierarchy of control relationships between the successive tiers of segmentation genes can be demonstrated by observing the expression pattern of one such gene when another is inactivated by mutation. In a mutant embryo that lacks the normal Krüppel product, for example, the usual ftz stripes fail to develop in just that region of the blastoderm corresponding to the defect in the Krüppel mutant. Thus the Krüppel product, directly or indirectly, regulates ftz gene expression. In a ftz mutant, by contrast, the distribution of the normal Krüppel product is not disturbed, indicating that the ftz product does not regulate Krüppel gene expression.

There are also interactions between genes in the same tier of the regulatory hierarchy. The gap genes Krüppel and hunchback, for example, are expressed in adjacent regions of the blastoderm, with a sharp boundary between the hunchback territory anteriorly and the Krüppel territory posteriorly (see Figure 21-61). A repression of Krüppel gene expression by the Hunchback gene regulatory protein helps to establish this boundary, ensuring that the expression domains of the two genes are properly correlated. Interactions of this sort also guide the regular periodic pattern of expression of the pair-rule genes, setting up an exactly reproducible arrangement of mutual exclusions and overlaps that repeats itself reliably in every double-segment unit in the blastoderm of every normal embryo (Figures 21-64 and 21-65). In this way different bands of cells around the blastoderm are distinguished by different combinations of pair-rule gene expression, down to the finest possible level of detail - the width of a single cell, which corresponds to about a quarter of the width of a prospective segment or parasegment.

This whole elaborate patterning process depends on the long stretches of DNA sequence that control the expression of each of the segmentation genes. These regulatory regions bind multiple copies of the gene regulatory proteins produced by a subset of other segmentation genes, and the gene is turned on or off according to the combination of proteins bound. In Chapter 9 (see p. 426) we focus on one particular segmentation gene and discuss how the decision whether to transcribe the gene is made on the basis of all these inputs.

Egg-Polarity, Gap, and Pair-Rule Genes Create a Transient Pattern That Is Remembered by Other Genes ^{41, 50, 52}

Within the first few hours after fertilization, the gap genes and the pair-rule genes are activated one after another. Their mRNA products appear first in patterns that only approximate the final picture; then, within a short time - through a series of interactive adjustments - the fuzzy initial distribution of gene products resolves itself into a regular, crisply defined system of stripes (see Figure 21-65). But this system itself is unstable and transient. As the embryo proceeds through gastrulation and beyond, the regular segmental pattern of gap and pair-rule gene products disintegrates. Their actions, however, have stamped a permanent set of labels (positional values) on the cells of the blastoderm. These positional labels are recorded in the persistent activation of certain of the segment-polarity genes and of the homeotic selector genes, which serve to maintain the segmental organization of the larva and adult.

Segment-Polarity Genes Label the Basic Subdivisions of Every Parasegment ⁵³

Segment-polarity genes are expressed in a pattern that repeats itself from one parasegment to the next. The gene engrailed provides a good example (Figure 21-66). Its RNA transcripts are seen in the cellular blastoderm in a series of 14 bands, each approximately one cell wide, corresponding to the anteriormost portions of the future parasegments. These bands appear in a fixed relationship to the bands of expression of the pair-rule genes (see Figure 21-64). Again, the pattern is governed in a combinatorial fashion by the products of the previous set of genes in the hierarchy and is refined and elaborated by interactions among the segment-polarity genes themselves. Through expression of different segment-polarity genes in different bands of cells, each future parasegment is already subdivided at the cellular blastoderm stage into at least three distinct regions. The chemical distinctions will persist, maintained by continued transcription of at least some of the segment-polarity genes, after the pair-rule gene products have largely disappeared (see Figure 21-66). Some of the segment-polarity genes thus expressed - including, in particular, one called wingless - encode secreted proteins that act also during subsequent development as spatial signals within the parasegment to regulate the details of its internal patterning and growth.

Besides regulating the segment-polarity genes, the products of pair-rule genes collaborate with the products of gap genes (and perhaps egg-polarity genes) to cause the precisely localized activation of a further set of spatial labels - the homeotic selector genes, which permanently distinguish one parasegment from another. In the next section we examine these selector genes in detail and consider their role in cell memory.

Summary

Like other insects, Drosophila is constructed from a series of repeating modular units called segments, with specialized nonsegmental structures at each end of the body. Each major subdivision of each segment is distinguished by the expression of a particular selection of control genes that defines its "address." The pattern originates with asymmetry in the egg: positional information is supplied by four gradients set up by the products of four groups of maternal-effect genes called egg-polarity genes. The four groups of genes control four distinctions fundamental to the body plan of animals: dorsal versus ventral, endoderm versus mesoderm and ectoderm, germ cells versus somatic cells, and head versus rear. The egg-polarity genes operate by setting up graded distributions of gene regulatory proteins in the egg and early embryo, but the gradients are set up differently for the different egg axes.

The dorsoventral polarity is defined by a localized signal from the follicle cells that surround the egg. The signal molecule binds to transmembrane receptors in the ventral surface of the egg, leading ultimately to a graded intranuclear concentration of the gene regulatory protein Dorsal along the dorsoventral axis of the early embryo. The Dorsal protein regulates expression of other genes, including dpp, whose product acts in turn as a morphogen to specify finer subdivisions of the dorsoventral axis, like the early inductive signals that operate in Xenopus.

In the case of the anterior group of egg-polarity genes, the gradient arises from a localized deposit of mRNA, the product of the bicoid gene, at the anterior end of the egg. Because the insect egg develops initially as a syncytium, the Bicoid protein translated from this mRNA is able to diffuse in the cytosol along the length of the embryo, guiding the global organization of its anterior half. The Bicoid concentration gradient initiates the orderly expression of gap genes, pair-rule genes, segment-polarity genes, and homeotic selector genes. These, through a hierarchy of interactions, become expressed in some regions of the embryo and not others, progressively subdividing the body into a regular series of segmental and subsegmental units.

Introduction

The first glimpses of the system of genes for pattern formation came over 70 years ago, with the discovery of the first of a set of mutations in Drosophilathat cause bizarre disturbances of the organization of the adult fly. In the mutation Antennapedia, for example, legs sprout from the head in place of antennae (Figure 21-67), while in the mutation bithorax, portions of an extra pair of wings appear where normally there should be the much smaller appendages called halteres. These mutations transform parts of the body into structures appropriate to other positions and are called homeotic. A whole set of homeotic selector genes determines the anteroposterior character of the segments of the fly. In this section we follow Drosophila development through to the final steps in the formation of the adult fly to see how the homeotic selector genes do their job. At the end of the section we see that the same genes have a central role in patterning the body parts of other animals, including ourselves.

The Homeotic Selector Genes of the Bithorax Complex and the Antennapedia Complex Specify the Differences Among Parasegments ^{54, 55}

The homeotic selector genes of interest to us here all lie in one or the other of two tight gene clusters known as the bithorax complex and the Antennapedia complex. Each complex contains several genes with analogous functions: those in the bithorax complex control the differences among the abdominal and thoracic segments of the body, while those in the Antennapedia complex control the differences among thoracic and head segments. In some other insects the corresponding groups of genes all lie in a single complex, called the HOM complex;the Antennapedia and bithorax complexes are thus thought to be the two halves of a single HOM complex that has become split in the course of the fly's evolution. Each homeotic selector gene has a characteristic domain of action, defined as the region of the body that is transformed as a result of mutation in that gene. Typically, this domain has sharp boundaries that are roughly half a segment out of register with the conventional segment boundaries, indicating that the domain is a parasegment or a block of parasegments (see Figure 21-50).

Many of the mutations of homeotic selector genes have a recessive lethal phenotype and allow the embryo to survive only to around the time of hatching. Observations of embryos or very early larvae therefore give the clearest and in some respects most complete picture of the role of the homeotic selector genes.

Larvae that are deficient in all the genes of the bithorax complex have a particularly simple structure: the head and anterior thorax are normal as far as the P4 parasegment, but all of the remaining 10 parasegments are converted to the character of P4. Partial deletions of the bithorax complex cause transformations that are less extensive (Figure 21-68). These observations, and analogous findings for the Antennapedia complex, illustrate the essential role of the homeotic selector genes in defining the differences among the parasegments: when the genes are missing, the distinctions between one parasegment and another are not made.

Homeotic Selector Genes Encode a System of Molecular Address Labels ^{50, 56}

Like the segmentation genes, the homeotic selector genes are first activated in the blastoderm. Since all of the DNA in the Antennapedia and bithorax complexes has been cloned, nucleic acid probes are available to map the spatial pattern of transcription of each of the homeotic selector genes by in situ hybridization. The conclusions from these studies are striking: to a first approximation each homeotic selector gene is normally expressed in just those regions that develop abnormally, as though misplaced, when the gene is mutated or absent.

The products of the selector genes can thus be viewed as molecular address labels possessed by the cells of each parasegment. If the address labels are changed, the parasegment behaves as though it were located somewhere else. Because the segmentation genes help to control the activation of the homeotic selector genes, the pattern of homeotic selector gene expression is in exact register with the parasegmental boundaries defined by the pair-rule and segment-polarity gene products. In this way the combination of a particular homeotic selector gene product (or set of such products) with a particular set of segmentation gene products reliably defines a unique address carried only by the cells in one subdivision of one segment.

Although the pattern of expression of the homeotic selector genes undergoes complex adjustments as development proceeds, these genes continue to play a crucial part throughout the subsequent development of the fly. They somehow equip cells with a memory of their positional value.

The Control Regions of the Homeotic Selector Genes Act as Memory Chips for Positional Information ^{54, 57, 58, 59}

The products of the homeotic selector genes, as discussed in Chapter 9, are gene regulatory proteins, all homologous to one another and all containing a highly conserved homeobox sequence, which codes for a DNA-binding homeodomain (60 amino acids long) in the corresponding proteins. Although many other genes also contain a homeobox, the particular type of homeobox sequence found in the homeotic selector genes is characteristic.

There are eight homeotic selector genes in the Antennapedia and bithorax complexes (which, for convenience, we shall refer to collectively as the HOM complex). Their coding sequences are interspersed amid a much larger quantity - a total of about 650,000 nucleotide pairs - of regulatory DNA. This DNA includes binding sites for the products of egg-polarity and segmentation genes - genes such as bicoid, hunchback, and even-skipped. The regulatory DNA in the HOM complex acts as an interpreter of the multiple items of positional information supplied to it by all these factors, and, in response to them, it makes a decision to transcribe or not to transcribe a particular set of homeotic selector genes. There are, however, some deep mysteries about how the HOM control system is organized and how it operates.

One remarkable feature is that the sequence in which the genes are ordered along the chromosome in both the Antennapedia and the bithorax complexes corresponds almost exactly to the order in which they are expressed along the axis of the body (Figure 21-69). It is as though the genes are activated serially by some process that spreads farther and farther along the chromosome in proportion to some intracellular indicator of distance along the body axis. It is not clear whether this ordering is merely an accident of evolution or truly reflects involvement of some activation mechanism that propagates along the chromosome, although we shall see later that it is a feature of the HOM complex that has been highly conserved in the course of evolution.

There is a further puzzle. The HOM complex serves to make each parasegment different from the next, but the number of homeotic selector genes is smaller than the number of parasegments. The bithorax complex, for example, contains just three genes, but it is responsible for the differences between 10 parasegments (see Figure 21-69). Moreover, there are many mutations, mapping to different sites in the complex, that alter the anteroposterior character of only a single parasegment or even of a part of a parasegment. Most of these mutations lie in noncoding control regions and are also ordered along the chromosome in a sequence that matches in detail the anatomical ordering of the regions they affect. This suggests that the differences between body regions are defined not simply by the presence of different homeotic selector gene products but, more subtly, by persistent differences of some sort in the states of the control regions associated with those genes. A control region, in this view, is to be pictured not as a simple on-off switch but as something more like a computer microchip: it receives inputs (in the form of gene regulatory factors and other molecules that bind to it), it produces an output (in the form of a directive to transcribe or not to transcribe the homeotic selector gene), and it can store a memory trace (a record of positional information) that affects the way the output is computed from the inputs. The positional value of a cell thus will not necessarily be reflected in a certain fixed level of expression of the homeotic selector gene but rather in a particular way of regulating that gene in response to changing conditions.

All this remains speculative as long as we have no answer to a third and most fundamental question about the HOM complex: what mechanism maintains the memory trace? As discussed earlier (see p. 1062), one possibility is that the mechanism involves positive feedback, where the product of a gene, once it is made, stimulates its own transcription. At least some of the homeotic selector genes seem to have this property. The gene Deformed (in the Antennapedia complex), for example, has multiple binding sites for the Deformed protein in its upstream control region, and in some cells these are sufficient for it to keep itself activated once activity has been triggered. Such self-stimulatory effects, however, are not sufficient by themselves to maintain the memory trace in most cells. A whole additional set of genes, called the Polycomb group, have been found to be required to keep silent those homeotic selector genes that should not be expressed: if any of the Polycomb-group genes are inactivated by mutations, the homeotic selector genes are initially switched on in a normal pattern but then become activated indiscriminately all over the embryo (Figure 21-70A). The Polycomb protein is bound to the chromatin of the genes it controls (Figure 21-70B). Moreover, related genes appear to be involved elsewhere in the control of chromatin structure, suggesting that the memory of positional value may be carried by some persistent local modification of the chromatin in the HOM gene complex.

The Adult Fly Develops from a Set of Imaginal Discs That Carry Remembered Positional Information ⁶⁰

The basic pattern of expression of the homeotic selector genes is established in the Drosophila embryo and determines the structure not only of the larva, but also, much later, that of the adult fly. To appreciate fully the role of these genes as carriers of a positional memory, it is necessary to have some idea of the curious way in which the adult, or imago, finally develops.

The adult fly is formed largely from groups of cells, called imaginal cells, that are set aside, apparently undifferentiated, in each segment of the larva. The imaginal cells for most of the adult body originate from the embryonic epidermis - the epithelium that covers the body. They remain connected with the epidermis of the larva, and they will form mainly the epidermal structures of the adult fly. The imaginal cells for the head, thorax, and genitalia are organized into imaginal discs; other clusters of imaginal cells will form the abdomen. There are also groups of imaginal cells in the viscera of the larva to give rise to the internal organs of the fly. Detailed studies have focused chiefly on the imaginal discs. There are 19 of these, arranged as 9 pairs on either side of the larva plus 1 disc in the midline (Figure 21-71). The discs are pouches of epithelium, shaped like crumpled and flattened balloons, that evaginate (turn inside out), extend, and differentiate at metamorphosis. The eyes and antennae develop from one pair of discs, the wings and part of the thorax from another, the first pair of legs from another, and so on.

The cells of one imaginal disc look like those of another, and when they differentiate, they will give rise to generally similar sets of specialized cell types. But grafting experiments show that they are in fact already regionally determined and nonequivalent. If one imaginal disc is transplanted into the position of another in the larva and the larva is then left to go through metamorphosis, the grafted disc is found to differentiate autonomously into the structure appropriate to its origin, regardless of its new site. This implies that the imaginal disc cells are governed by a memory of their original position. By an ingenious grafting procedure that lets the imaginal disc cells proliferate for an extended period before differentiating, it can be shown that this cell memory is stably heritable (with rare lapses) through an indefinitely large number of cell generations (Figure 21-72).

The homeotic selector genes are essential components of the memory mechanism. If they are eliminated from imaginal disc cells at any stage in the long period leading up to differentiation at metamorphosis, the cells will differentiate into incorrect structures, as though they belonged to a different segment of the body. This can be demonstrated by the very powerful technique of x-ray-induced mitotic recombination - in effect, a form of genetic surgery on individual cells by means of which mutant clones of cells of a specified genotype can be generated at a chosen time in development, as we now explain.

Homeotic Selector Genes Are Essential for the Memory of Positional Information in Imaginal Disc Cells ⁶¹

A short pulse of x-irradiation, as a side effect of the damage it does to DNA, can provoke crossing over between homologous chromosomes in a dividing cellan event that would normally occur only at meiosis. As explained in Figure 21-73, if the cell is heterozygous for a gene in the crossed-over chromosomal region, the process can result in a pair of daughter cells that are homozygous, the one receiving two copies of the maternal allele of the gene, the other receiving two copies of the paternal allele. The occurrence of the cross-over can be detected if the animal is chosen to be also heterozygous for a mutation in a marker gene - a pigmentation gene, for example - that lies near the gene of interest and so undergoes crossing over in company with it. In this way marked homozygous mutant clones of cells can be created to order (Figure 21-74).

The major effects of mutations in homeotic selector genes are generally recessive: only the homozygous mutant organism shows the homeotic transformation. By exploiting mitotic recombination, one can create a clonal patch of marked homozygous homeotic mutant cells in an imaginal disc and examine their behavior in a heterozygous, phenotypically normal background. The finding is that the marked cells, and only the marked cells, show the homeotic transformation (provided that they lie in the normal domain of action of the homeotic selector gene), and this applies whether the recombination event was provoked early in development or late. A 2-day larva heterozygous for a mutation that destroys the function of the Ultrabithorax (Ubx) gene (in the bithorax complex), for example, can be x-irradiated to produce isolated clones of homozygous cells in its imaginal discs that contain no functional Ubx gene. These clones, if they lie in the haltere disc, will give rise to patches of wing-type tissue in the haltere. These and other observations indicate that each cell's memory of positional information depends on the continued activity of the normal homeotic selector gene. This memory, furthermore, is expressed in a cell-autonomous fashion - each cell maintains its state independently, depending on its own history and genome, regardless of its neighbors.

The Homeotic Selector Genes and Segment-Polarity Genes Define Compartments of the Body ^{53, 62}

The remembered distinctions specified by the homeotic selector genes are discrete: there is an abrupt difference of gene expression between cells in adjacent parasegments. The same is true for at least some of the segment-polarity genes, such as engrailed (see Figure 21-66), whose differential expression corresponds to an abrupt difference between cells in the posterior part of a parasegment and cells in its anterior part. Thus, through the differential expression of these two classes of genes, the body is subdivided into a series of discrete regions comprising cells in different states of determination. At the frontier between one such region and the next, the cells appear to be prevented from mixing, as though selective cohesion between cells with the same molecular address label keeps them segregated from cells with a different label (Figure 21-75). Thus, for example, when a clone of genetically marked but otherwise normal cells is created in the wing by mitotic recombination, the clone is observed to be confined strictly to one side or the other of a precisely specified boundary at the frontier between the two parasegments from which the wing is constructed. A subdivision of the body defined in this way - in the wing or any other organ - is called a compartment (Figure 21-75).

By definition, a compartment boundary is a frontier where two populations of cells in different states of determination are prohibited from mixing. Because the state of determination is not normally reversible, each compartment has to be a self-sufficient unit. It cannot recruit cells from the adjacent compartment or transfer surplus cells into it. It can and does, however, regulate its internal organization and its size in obedience to the rule of intercalation, discussed earlier, by adjustments that do not violate this constraint. Thus, in the regulation of pattern and growth, each compartment seems to behave as a more or less independent module during normal development (although during regeneration after a drastic disturbance cells sometimes do switch their character and their compartmental allegiance).

Some of the morphogenetic signals operating in the imaginal disc to control these processes have been identified. They appear to include products of the dpp and wingless genes, which, as we saw, are both active in patterning the early embryo also (see Figure 21-56). But we do not yet know in molecular genetic terms how these signaling systems are organized or how they collaborate with the homeotic selector genes to give each compartment its characteristic internal pattern and make it stop growing when it has reached its proper size.

Thus, in following the genetic pathways of pattern formation to later and later stages and finer and finer levels of detail, we come to a point where the chain of cause and effect becomes obscure. At the very last stage in the process, however, as cells prepare for terminal differentiation, the trail can be picked up again, and we can trace the genetic mechanisms that control some of the most minute details of patterning of the fly's body surface as displayed in its array of sensory bristles.

Localized Expression of Specific Gene Regulatory Proteins Foreshadows the Production of Sensory Bristles ⁶³

Flies have many bristles on their body - some big, some small. The big ones are landmark structures on the surface of the fly: they are relatively few and far between and occupy exactly predictable positions. The small ones are more closely spaced and occur in fields covering precisely defined regions of the body surface. The bristles are miniature sense organs - components of the peripheral nervous system. Some respond to chemical stimuli, others to mechanical stimuli, but they are all constructed in a similar way. The structure is seen at its simplest and most stereotyped in the mechanosensory bristles. Each of these, whether big or small, consists of exactly four cells: a shaft cell, a socket cell, a glial sheath cell, and a neuron (Figure 21-76). Movement of the shaft of the bristle excites the neuron, which sends a signal to the central nervous system.

The cells of the bristle of the adult fly derive from the imaginal disc epithelium, and all four of them are granddaughters of a single sensory mother cell that becomes distinct from the neighboring prospective epidermal cells during the last larval instar (Figure 21-77). To account for the pattern of bristle differentiation, we have to explain first how the genesis of sensory mother cells is controlled and then how the four granddaughters of each such cell become different from one another.

Two genes, called achaete and scute, are crucial in initiating the formation of bristles in the imaginal disc epithelium. These genes have similar and overlapping functions and code for closely related gene regulatory proteins of the helix-loop-helix class (discussed in Chapter 9). They belong to a group of closely linked homologous genes, all located in the achaete-scute complex. In situ hybridization shows that achaete and scute are expressed in the imaginal disc in precisely the regions where bristles will form. Mutations that eliminate the expression of these genes at some of their usual sites block development of bristles at just those sites, and mutations that cause expression in additional, abnormal sites cause bristles to develop there. But expression of achaete and scute is transient, and only a minority of the cells initially expressing the genes go on to become sensory mother cells; the others become ordinary epidermis. The state that is specified by expression of achaete and scute is called proneural. The proneural cells are primed to take the neurosensory pathway of differentiation, but which of them will actually do so depends on competitive interactions among them.

Lateral Inhibition Regulates the Fine-grained Pattern of Differentiated Cell Types ^{63, 64}

Proneural cells, expressing achaete or scute or both genes together, occur in groups in the imaginal disc epithelium - a small, isolated cluster of fewer than 30 cells for a big bristle, a broad, continuous patch of hundreds or thousands of cells for a field of small bristles. In the former case just one member of the cluster becomes a sensory mother cell; in the latter case many cells scattered throughout the proneural region do so. The sensory mother cells are almost always separated from one another by a certain minimum number of epidermal cells. Experiments with genetic mosaics show that a cell that becomes committed to the sensory-mother-cell pathway of differentiation sends a signal to its neighbors not to do the same thing: it exerts a lateral inhibition (Figure 21-78). If a cell that would normally become a sensory mother is genetically disabled from doing so, a neighboring proneural cell, freed from lateral inhibition, will become a sensory mother instead.

The genes responsible for lateral inhibition were first identified as such through studies of mutant embryos. In the embryo, both the achaete-scute complex and the genes for lateral inhibition govern development of the central and peripheral nervous system in just the same way that they later govern development of the sense organs of the peripheral nervous system in imaginal discs. In both situations mutations abolishing lateral inhibition have a simple and striking effect: neural cells are produced in vast excess at the expense of epidermal cells (Figure 21-79). Genes are generally named according to their mutant phenotype; hence, the genes responsible for lateral inhibition are called, confusingly, neurogenic genes. They form a genetic system with at least seven members.

The best-known neurogenic gene is called Notch. It codes for a transmembrane protein that is thought to serve as the receptor for the lateral-inhibition signal. Experiments with genetic mosaics show that cells lacking Notch are blind to the signal and follow a neural pathway of differentiation. Another related transmembrane protein, encoded by the neurogenic gene Delta, appears to be a ligand that binds to Notch and activates it; lateral inhibition, it seems, is transmitted via direct cell-to-cell contact. Downstream from Notch the products of other neurogenic genes act intracellularly to interpret the signal and suppress neural differentiation.

The same lateral inhibition mechanism dependent on Notch can be shown to operate twice in the formation of bristles - first, to force the neighbors of sensory mother cells to follow a different pathway and become epidermal and, second, to make the four granddaughters of the sensory mother cell follow different pathways of differentiation so as to form the four components of the bristle. At both stages the default pathway is the neural pathway, and lateral inhibition mediates a competitive interaction that forces adjacent cells to differentiate in contrasting ways.

The same set of neurogenic genes in Drosophila not only mediates lateral inhibition repeatedly during development of the nervous system but also is required for the detailed patterning of many other tissues of the fly. Indeed, lateral inhibition is a key strategy in the control of multicellular patterns of differentiation throughout the animal world and almost certainly in plants also; the types of spacing patterns that it can generate are ubiquitous, from the stomata on a leaf to the photoreceptors in the eye. As homologues of the neurogenic genes are found in vertebrates, it may be that the same conserved molecular mechanisms operate in at least some of these cases. In the final part of this section we consider how far Drosophila does actually provide a universal model for the molecular genetics of pattern formation.

The Developmental Control Genes of Drosophila Have Homologues in Vertebrates ⁶⁵

The theory of evolution tells us that all animals are our cousins. It is easy enough to see the family resemblances between a human being and a mouse, or even a fish, and to chart the homologies between the parts of their bodies and the parts of our own. But when we compare ourselves with flies or worms, from which we are separated by about 600 million years, the correspondences are far from clear. True, one can recognize some familiar cell types - neurons, striated muscle cells, and spermatozoa, for example. With a little less confidence, one can see similarities in the body plan, with its central gut tube and its head at one end. But how deep do these similarities go? The fossil record gives us no clear answer, but molecular genetics has begun to supply one.

Comparisons of gene sequences show that an astonishingly large proportion of the genes in an animal such as a fly have unmistakable homologues in vertebrates, and vice versa. Such homologies have been recognized for a majority of the developmental control genes we have mentioned in this chapter. But are these control genes used in the same combinations and for homologous purposes, so that the genetic system governing development is conserved? When we compare a human being with a fly, there seem at first sight to be fundamental differences, in development as well as final structure. Vertebrate eggs do not, for example, develop through a syncytial stage as insects do, and their initial multicellular patterning therefore cannot be controlled by morphogen gradients such as that of Bicoid in Drosophila, set up by intracellular diffusion of a protein through a cytoplasm that is shared by many nuclei. And yet, when we turn to slightly later stages, we encounter a remarkable pattern of anatomical correspondences. These could never have been discerned without the help of molecular genetics, which reveals in very different animals similar positional markers expressed in body parts that we might not otherwise judge to have anything in common. The HOM gene complex has been central to this new appreciation of our relation to flies and worms.

Mammals Have Four Homologous HOM Complexes ^{59, 66}

Because the homeodomain of the homeotic selector genes has been highly conserved in evolution, it has been relatively easy to discover homologues of the Drosophila genes in other classes of animals. They have been found in almost every sort of creature - in Hydra, in nematodes and earthworms, in beetles and mollusks and sea urchins, in fish, frogs, birds, and mammals. Remarkably, in those cases that have been investigated adequately, these genes seem to be grouped in complexes similar to the insect HOM complex. In the mouse there are four such complexes - called the HoxA, HoxB, HoxC, and HoxD complexes - each on a different chromosome. Individual genes in each complex can be recognized by their homeobox sequences, as counterparts of specific members of the Drosophila set. It appears that each of the four mammalian Hox complexes is, roughly speaking, the equivalent of a complete insect HOM complex (that is, an Antennapedia complex plus a bithorax complex) (Figure 21-80). The ordering of the genes within each Hox complex is essentially the same as in the insect HOM complex, suggesting that all four vertebrate complexes originated by duplications of a single primordial complex and have preserved its basic organization. Most tellingly, when the expression patterns of the Hox genes are examined in the vertebrate embryo by in situ hybridization, it turns out that the members of each complex are expressed in a head-to-tail series along the axis of the body, just as they are in Drosophila (Figure 21-81). The pattern is most clearly seen in the neural tube. With minor exceptions this anatomical ordering matches the chromosomal ordering of the genes in each complex, and corresponding genes in the four different Hox complexes have almost identical anteroposterior domains of expression.

The gene expression domains define a detailed system of correspondences between insect body regions and vertebrate body regions. As shown in Figure 21-82, the parasegments of the fly correspond to a similarly labeled series of segments in the anterior part of the vertebrate embryo. These are most clearly demarcated in the hindbrain, where they are called rhombomeres. In the tissues lateral to the hindbrain the segmentation is seen in the series of branchial arches, prominent in all vertebrate embryos - the precursors of the system of gills in fish and of the jaws and structures of the neck in mammals; each pair of rhombomeres in the hindbrain corresponds to one branchial arch (see Figure 21-82). In the hindbrain, as in Drosophila, the boundaries of the expression domains of the Hox genes are aligned with the boundaries of the anatomical segments. And as in Drosophila compartments, the cells of one rhombomere do not mix with those of the next rhombomere.

It is not yet clear, however, how similar in detail the mechanisms that set up the hindbrain and branchial arch segmentation of a vertebrate are to those that generate the parasegments of an insect. Although, for example, vertebrates have homologues of the engrailed and wingless genes, these are not expressed in a repetitive segmental fashion in the hindbrain.

Hox Genes Specify Positional Values in Vertebrates as in Insects ⁶⁷

Despite uncertainties over mechanisms of segmentation, there can be little doubt that our head-to-tail axis is homologous to that of an insect and that essentially the same sets of genes mark out the anteroposterior positional values of our cells. The Hox genes appear to have not only similar expression patterns to the insect HOM genes but also similar controlling functions. Because the vertebrate has four Hox gene complexes acting more or less in parallel along its body axis, in place of the insect's single HOM complex, it is not enough to eliminate or misexpress a single Hox gene to produce a full-blown homeotic transformation of one region into the character of another. Nevertheless, genetically engineered mice with alterations in single Hox genes do show localized abnormalities that can be interpreted as incomplete homeotic transformations.

This illustrates one of the fundamental difficulties in analyzing the genetics of developmental control systems in vertebrates. The vertebrate genome is very big, and it owes its size, in large measure, to gene duplications in the course of evolution. Thus it contains multiple variant copies of genes that are represented singly in a fly or a nematode: the four Hox complexes corresponding to the single HOM complex are typical in this respect. The multiple versions of a gene have overlapping and partially interchangeable functions, and this partial redundancy makes it very difficult to identify the basic role of any single gene, just as it is hard, by removing or inserting a single screw, to demonstrate the function of the multiple screws that hold a door on its hinges. Herein lies the cardinal importance of the insights that simpler model organisms such as Drosophila and Caenorhabditis elegans have to offer.

This is not to say that an individual gene in a vertebrate set is superfluous; for as evolution proceeds, the duplicated genes diverge and begin to take on new and more specialized functions that distinguish them from one another. Old components can be adapted to organize the development of new types of structures in addition to the old. The limbs of higher vertebrates provide a beautiful example.

Subsets of Hox Genes Are Expressed in Order Along Two Orthogonal Axes in the Vertebrate Limb Bud ⁶⁸

Earlier in this chapter we used the developing limb buds of the chick embryo to show that cells in different regions are distinguished from one another by a property that we called their positional value. This remembered characteristic of the cells controls whether they will form the structures appropriate to leg or wing, upper arm or forearm, thumb or little finger. Molecular genetics has revealed what "positional value" means in molecular terms in the limb bud.

We have seen that along the main body axis, both in flies and in vertebrates, positional values are defined by the state of expression of HOM/Hox genes. In situ hybridization shows that the same is true in the limb buds of a mouse or chick embryo - but with a twist. Instead of finding the corresponding genes of all four Hox complexes expressed in similar, overlapping patterns, as in the hindbrain, one finds a subset of members of the HoxD complex expressed in a series of domains ordered along one limb axis (very roughly, the anteroposterior) and a subset of members of the HoxA complex expressed in series along a different axis (more or less proximodistal) (Figure 21-83). To test whether these genes actually control limb patterning, a retrovirus has been used as an expression vector in the chick embryo to introduce a particular Hox gene into the limb bud cells and force expression of the gene in an inappropriate site. When cells in the region from which the first toe will develop are thus caused to express the Hox gene characteristic of the second toe (Hoxd-11), their behavior is transformed, and at the site of the first toe a duplicate of the second toe develops. Evidently, when vertebrates evolved limbs, they co-opted the different sets of Hox genes in different ways to control limb patterning as well as the patterning of the main body axis.

A central problem now for vertebrate embryology is to find out how the Hox genes themselves are regulated. Several studies show that retinoic acid can control Hox gene expression both in the limb bud and along the main body axis, but how this control is exerted and what part it plays in normal development are as yet open questions.

The HOM/Hox genes provide at present the most spectacular example of conserved developmental control machinery. But the flood of genetic homologies discovered through gene sequencing in the past few years gives every reason to expect that many further developmental parallels between vertebrates and invertebrates, no less profound, will soon become apparent.

Classical and molecular genetic studies of small, tractable organisms such as flies and worms give us a key to unlock the mysteries of development in the animal world as a whole. But can we take the generalization a step further still, to the world of plants, or does plant development rest on an entirely different set of principles and mechanisms? This is the question that we tackle in the next section.

Summary

Homeotic selector genes specify the differences between body segments along the head-to-rear axis: they provide the cells with a record of their positional value. Mutations in homeotic selector genes can convert one body segment to the character of another, and deletion of the genes en masse results in a larva whose body segments are all alike. Similar transformations are seen in the external structures of the adult fly, which are derived from the imaginal discs of the larva. Transplantation experiments show that the cells in the discs retain a long-term memory of their positional value, and this memory depends on the continued presence of the homeotic selector genes.

The homeotic selector genes all code for DNA-binding proteins containing a characteristic highly conserved homeobox sequence. They are grouped in two clusters in the genome, thought to be the separated parts of a single ancestral gene cluster called the HOM complex. The chromosomal ordering of the genes in each part of the complex matches the spatial ordering of their expression domains in the body. The molecular mechanism of the memory phenomenon is unknown, but it is thought to depend on self-perpetuating changes in the state of the control regions in the HOM complex.

The expression patterns of the HOM genes and segment-polarity genes jointly subdivide the body into compartments whose cells do not mix. Subsequent processes generate a fine-grained pattern of cell differentiation inside each compartment. Lateral inhibition, mediated by the so-called neurogenic genes, plays a key part in this final stage of cell diversification, causing cells that are in contact with one another to differentiate in different ways and so helping to organize the creation of minutely specialized sets of cells forming structures such as sensory bristles.

A large proportion of the developmental control genes identified in flies and worms have homologues in other types of animals, including vertebrates. In some cases the corresponding genes have been shown to have corresponding developmental functions, implying that fundamental mechanisms of animal development have been conserved even where the outward appearance of the body has evolved out of all recognition. Practically all animals appear to have HOM gene complexes organized in a similar way to those of insects: in mammals there are four such complexes, called Hox complexes, and their products are thought to specify positional values that control the anteroposterior pattern of parts in the region of the hindbrain and trunk. The Hox complexes have also acquired new functions as specifiers of positional information in the more recently evolved parts of the vertebrate body, in particular in the limbs.

Introduction

Plants and animals are separated by about a billion years of evolutionary history. They have evolved their multicellular organization independently but using the same initial tool kit - the set of genes inherited from their common unicellular eucaryotic ancestor. Most of the contrasts in their developmental strategies spring from two basic peculiarities of plants. First, they get their energy from sunlight, not by ingesting other organisms. This dictates a different body plan. Second, their cells are encased in semirigid cell walls that are cemented together, preventing them from moving as animal cells do. This dictates a different set of mechanisms for shaping the body and different developmental mechanisms to cope with a changeable environment.

Animal development is largely buffered against environmental changes, and the embryo generates the same genetically determined body structure unaffected by external conditions. The development of most plants, by contrast, is dramatically influenced by the environment: because they cannot match themselves to their environment by moving to another place, plants adapt instead by altering the course of their development. Their strategy is opportunistic. A given type of organ - a leaf, a flower, or a root, say - can be produced from the fertilized egg by many different paths according to environmental cues. A begonia leaf pegged to the ground may sprout a root; the root may throw up a shoot; the shoot, given sunlight, may grow leaves and flowers.

The mature plant is typically made of many copies of a small set of standardized modules, as described in Figure 21-84. The positions and times at which those modules are generated are strongly influenced by the environment, causing the overall structure of the plant to vary. The choices between alternative modules and their organization into a whole plant depend on external cues and long-range hormonal signals that play a much smaller part in the control of animal development.

But although the global structure of a plant - its pattern of roots or branches, its numbers of leaves or flowers - can be highly variable, its detailed organization on a small scale is not. A leaf, a flower, or indeed an early plant embryo, is as precisely structured as any organ of an animal. The internal organization of a plant module raises essentially the same problems in the genetic control of pattern formation as does animal development, and they are solved in analogous ways. In this section we focus on the cellular mechanisms of development in flowering plants. We examine both the contrasts and the similarities with animals.

Embryonic Development Starts by Establishing a Root-Shoot Axis and Then Halts Inside the Seed ⁷⁰

Flowering plants, despite their staggering variety, are of relatively recent origin. The earliest known fossil examples are 125 million years old, as against 350 million years for vertebrate animals. This helps to explain why certain features of their form and development are remarkably constant. Their basic strategy of sexual reproduction is briefly summarized in Panel 21-2, page 1109. The fertilized egg, or zygote, of a higher plant begins by dividing asymmetrically to establish the polarity of the future embryo. One product of this division is a small cell with dense cytoplasm, which will become the embryo proper. The other is a large vacuolated cell that divides further and forms a structure called the suspensor, which in some ways is comparable to the umbilical cord in mammals. The suspensor attaches the embryo to the adjacent nutritive tissue and provides a pathway for the transport of nutrients.

During the next step in development the diploid embryo cell proliferates to form a ball of cells that quickly acquires a polarized structure. This comprises two key groups of proliferating cells - one at the suspensor end of the embryo that will generate a rootand one at the opposite pole that will generate a shoot (Figure 21-85). The main root-shoot axis established in this way is analogous to the head-to-tail axis of an animal. At the same time it begins to be possible to distinguish the future epidermal cells, forming the outermost layer of the embryo, the future ground tissue cells, occupying most of the interior, and the future vascular tissue cells, forming the central core. These three sets of cells can be compared to the three germ layers of an animal embryo. Slightly later in development, the rudiment of the shoot begins to produce the embryonic seed leaves, or cotyledons - one in the case of monocots and two in the case of dicots. Soon after this stage, development usually halts and the embryo becomes packaged in a seed, specialized for dispersal and for survival in harsh conditions. The embryo in a seed is stabilized by dehydration, and it can remain dormant for a very long time - even hundreds of years. When rehydrated, the seeds germinate and embryonic development resumes.

The Repetitive Modules of a Plant Are Generated Sequentially by Meristems ⁷¹

Roughly speaking, the embryo of an insect or a vertebrate animal is a rudimentary miniature scale model of the later organism, and the details of body structure are filled in progressively as it enlarges. The plant embryo grows into an adult in a quite different way: the parts of the adult plant are created sequentially by groups of cells that proliferate to lay down additional structures at the plant's periphery. These all-important groups of cells are called apical meristems (see Figure 21-84). Each meristem consists of a self-renewing population of stem cells. As these divide, they leave behind a trail of progeny that emerge from the meristem region, enlarge, and finally differentiate. Although the shoot and root apical meristems generate all the basic varieties of cells that are needed to build leaves, roots, and stems, many cells outside the apical meristems also retain a capacity for further proliferation. In this way trees and other perennial plants, for example, are able to increase the girth of their stems and roots as the years go by.

The rudiments of the apical meristems of root and shoot are already determined in the embryo. As soon as the seed coat ruptures during germination, a dramatic enlargement of nonmeristematic cells occurs, driving the emergence first of a root, to establish an immediate foothold in the soil, and then of a shoot (Figure 21-86). This is followed by rapid and continual cell divisions in the apical meristems: in the apical meristem of a maize root, for example, cells divide every 12 hours, producing 5 x 10⁵ cells per day. The rapidly growing roots and shoots probe the environment - the roots increasing the plant's capacity for taking up water and minerals from the soil, the shoots increasing its capacity for photosynthesis (see Panel 21-2, p. 1109).

The Shaping of Each New Structure Depends on Oriented Cell Division and Expansion ⁷²

Plant cells, imprisoned within their cell walls, cannot crawl about and cannot be shuffled as the plant grows, but they can divide, and they can swell, stretch, and bend. The morphogenesis of a developing plant therefore depends on orderly cell divisions followed by strictly oriented cell expansion. Most cells produced in the root-tip meristem, for example, go through three distinct phases of development - division, growth (elongation), and differentiation. These three steps, which overlap in both space and time, give rise to the characteristic architecture of a root tip. Although the process of cell differentiation often begins while a cell is still enlarging, it is comparatively easy to distinguish in a root tip a zone of cell division, a zone of oriented cell elongation (which accounts for the growth in length of the root), and a zone of cell differentiation (Figure 21-87).

Because it affects the direction of cell elongation, the exact plane in which cells divide is crucial to plant morphogenesis, and changes in the plane of division are often associated with morphogenetic events such as the production of a leaf or petal primordium (Figure 21-88). The special intracellular mechanisms controlling the plane of cell division in plants are discussed in Chapter 18.

In the phase of controlled expansion that generally follows cell division, the daughter cells may often increase in volume by a factor of 50 or more. This expansion is driven by an osmotically based turgor pressure that presses outward on the plant cell wall, and its direction is determined by the orientation of the cellulose fibrils in the cell wall, which constrain expansion along one axis (see Figure 19-68). The orientation of the cellulose in turn is apparently controlled by the orientation of arrays of microtubules just inside the plasma membrane, which are thought to guide cellulose deposition (discussed in Chapter 19). These orientations can be rapidly changed by plant growth regulators, such as ethylene and gibberellic acid (Figure 21-89), but the molecular mechanisms underlying these dramatic cytoskeletal rearrangements are still unknown.

Each Plant Module Grows from a Microscopic Set of Primordia in a Meristem ⁷³

The apical meristems are self-perpetuating: they carry on with their functions indefinitely, as long as the plant survives, and they are responsible for its continuous growth and development. But apical meristems also give rise to a second type of outgrowth, whose development is strictly limited and culminates in the formation of a structure such as a leaf or a flower, with a determinate size and shape and a short lifespan. Thus, as a vegetative shoot elongates, its apical meristem lays down behind itself an orderly sequence of nodes, where leaves have grown out, and internodes (segments of stem). In this way the continuous activity of the meristem produces an ever increasing number of similar modules, each consisting of a stem, a leaf, and a bud (see Figure 21-84). The modules are connected to one another by supportive and transport tissue, and successive modules are precisely located relative to each other, giving rise to a repetitively patterned structure. This iterative mode of development is characteristic of plants and is seen in many other structures besides the stem-leaf system (Figure 21-90).

Although the final module is large, its organization, like that of an animal embryo, is mapped out at first on a microscopic scale. At the apex of the shoot, within a space of a millimeter or less, one finds a small, low central dome surrounded by a set of distinctive swellings in various stages of enlargement (Figure 21-91). The central dome is the apical meristem itself; each of the surrounding swellings is the primordium of a leaf. This small region, therefore, contains the already distinct rudiments of several entire modules. Through a well-defined program of cell proliferation and cell enlargement, each leaf primordium and its adjacent cells will grow to form a leaf, a node, and an internode. Meanwhile, the apical meristem itself will give rise to new leaf primordia, so as to generate more and more modules in a potentially unending succession. The serial organization of the modules of the plant is thus controlled by events at the shoot apex. Local signals within this tiny region determine the pattern of primordia - the position of one leaf rudiment relative to the next, the spacing between them, and their location relative to the apical meristem itself.

Almost nothing is known of the mechanisms that mediate these central patterning processes in the plant kingdom. All the strategies that we discussed for animal pattern formation, such as those based on local morphogens, timing mechanisms, and lateral inhibition, are possibilities here too. Detailed studies of the fate and lineage of cells in the shoot apex and the root apex are beginning to provide some of the essential background information, however, and some of the key developmental control genes are beginning to be identified. The gene regulatory protein encoded by the gene Knotted,for example, is expressed in the central part of the meristem, and overexpression in tobacco causes leaf cells to behave as meristem, generating new organs from the leaf itself.

Long-range Hormonal Signals Coordinate Developmental Events in Separate Parts of the Plant ⁷⁴

If a stem is to branch, new meristems must be created, and it is through control of this process that the environment exerts an important part of its influence over the form of a plant. At each node, in the acute angle, or axil, between the leaf branch and the stem, a bud is formed. This contains a nest of cells, derived from the apical meristem, that have kept a meristematic character (and express Knotted). They have the capacity to become the apical meristem of a new branch, but they also have the alternative option of remaining quiescent. The plant's pattern of branching is regulated through this choice, which the environment helps to dictate. Separate parts of the plant experience different environments and react to them individually by changes in their mode of development. The plant, however, must continue to function as a whole. This demands that developmental choices and events in one part of the plant should affect developmental choices elsewhere. There must be long-range signals to bring about such coordination.

As gardeners know, for example, by pinching off the tip of a branch one can stimulate side growth: removal of the apical meristem relieves the quiescent axillary meristems of an inhibition and allows them to form new twigs. In this case the long-range signal from the apical meristem, or at least a key component of the system of signals, has been identified. It is an auxin, a member of one of five known classes of plant growth regulators (sometimes called plant hormones), all of which have powerful influences on plant development. The four other known classes are the gibberellins, the cytokinins, abscisic acid, and the gas ethylene. As shown in Figure 21-92, all are small molecules that readily penetrate cell walls. They are all synthesized by most plant cells and can either act locally or be transported to influence target cells at a distance. Auxin, for example, is transported from cell to cell at a rate of about 1 cm per hour from the tip of a shoot toward its base. Each growth regulator has multiple effects, and these are modulated by the other growth regulators as well as by environmental cues and nutritional status. Thus auxin alone can promote root formation, but in conjunction with gibberellin it can promote stem elongation, with cytokinin it can suppress lateral shoot outgrowth, and with ethylene it can stimulate lateral root growth. The receptors that recognize these growth regulators are only now being characterized, and their mechanisms of action remain unknown.

Arabidopsis Serves as a Model Organism for Plant Molecular Genetics ⁷⁵

By screening systematically for mutations affecting the pattern of the plant embryo, it has been possible to begin to identify the genes that govern plant development and to start to work out how they function. This approach requires a plant that is, like Drosophila or Caenorhabditis elegans, small, quick to reproduce, and convenient for genetics. The role of "model plant" has fallen on a small weed, the common wall cress Arabidopsis thaliana (Figure 21-93), which can be grown indoors in test tubes in large numbers and produces thousands of offspring per plant after 8 to 10 weeks. Arabidopsis also has the advantage for molecular analysis of having one of the smallest plant genomes known (7 x 10⁷ nucleotide pairs), comparable to yeast (2 x 10⁷ nucleotide pairs), C. elegans (10⁸ nucleotide pairs), and Drosophila (10⁸ nucleotide pairs). Cell culture and genetic transformation methods have been established, large numbers of interesting mutants have been isolated, and an ordered collection of genomic DNA clones is now available. Arabidopsis has, in common with C. elegans, one significant advantage over Drosophila for genetics: like many flowering plants, it can reproduce as a hermaphrodite because a single flower produces both eggs and the pollen that can fertilize them. Therefore, when a flower that is heterozygous for a recessive lethal mutation is self-fertilized, one-fourth of its seeds will display the homozygous embryonic phenotype (Figure 21-94).

By using mutagens to create tens of thousands of mutant plants and inspecting their progeny in this way, a total of about 50 distinct genes governing embryonic pattern formation in Arabidopsis have thus far been identified. As in Drosophila, the patterning genes can be grouped according to their homozygous mutant phenotypes (Figure 21-95). Some are required for formation of the seedling root, some for the seedling stem, and some for the seedling apex with its cotyledons. Another class is required for formation of the three major tissue types - epidermis, ground tissue, and vascular tissue - and yet another class for the organized changes of cell shape that give the embryo and seedling their elongated form. Given this catalogue of key genes, it should soon be possible to take the next step and discover how they function. But from the range of mutant phenotypes, it already seems likely that the initial patterning of the plant embryo will be largely explicable within the same conceptual framework that we have presented for animals. As we now see, the same can be said for the later developmental processes by which a flower is made.

Homeotic Selector Genes Specify the Parts of a Flower ⁷⁶

Meristems face other developmental choices besides that between quiescence and growth, and these also are frequently regulated by the environment. The most important is the decision to form a flower (Figure 21-96).

The switch from meristematic growth to flower formation is typically triggered by light. By poorly understood mechanisms based on light absorption by specific proteins known as phytochromes, the cells in the meristem are able to alter their pattern of gene expression in response to a change in day length and thereby undergo the change of state that initiates flower development. By this switch in its state the apical meristem abandons its chances of continuing vegetative growth and gambles its future on the production of gametes. Its cells embark on a strictly finite program of growth and differentiation: by a modification of the ordinary mechanisms for generating leaves, a series of whorls of specialized appendages are formed in a precise order - typically sepals first, then petals, then stamens carrying anthers containing pollen, and lastly carpels containing eggs (see Panel 21-2, p. 1109). By the end of this process the meristem has disappeared, but among its progeny it has created germ cells.

The series of modified leaves forming a flower can be compared to the series of body segments forming a fly. In plants, as in flies, one can find homeotic mutations that convert one part of the pattern to the character of another. The mutant phenotypes can be grouped into three classes (Figure 21-97) in which different but overlapping sets of organs are altered. The first class, exemplified by the apetala2 mutant of Arabidopsis, has its two outermost whorls transformed: the sepals are converted into carpels and the petals into stamens. The second class, exemplified by apetala3, has its two middle whorls transformed: the petals are converted into sepals and the stamens into carpels. The third class, exemplified by agamous, has its two innermost whorls transformed, with a more drastic consequence: the stamens are converted into petals, the carpels are missing, and in their place the central cells of the flower behave as a floral meristem, which begins the developmental performance all over again, generating another abnormal set of sepals and petals nested inside the first and, potentially, another nested inside that, and so on, indefinitely. These phenotypes identify three classes of homeotic selector genes, which, like the homeotic selector genes of Drosophila, all code for gene regulatory proteins. These define the differences of cell state that give the different parts of a normal flower their different characters. In situ hybridization confirms that the genes are expressed in the patterns expected on this interpretation (Figure 21-98). In a triple mutant where all three genetic functions are absent, one obtains in place of a flower an indefinite succession of tightly nested leaves. Leaves therefore represent a "ground state" in which none of these homeotic selector genes are expressed, while the other types of organ result from expressing the genes in different combinations.

Similar studies have been carried out in the snapdragon Antirrhinum majus, and a similar set of phenotypes and genes have been identified. Gene sequencing reveals that, despite the large evolutionary distance between Antirrhinum and Arabidopsis, the corresponding homeotic phenotypes arise from mutations in homologous genes: plants, no less than animals, have conserved their homeotic selector gene systems. Again, the set of these genes appears to have arisen through gene duplication: several of them, required in different organs of the flower, have clearly homologous sequences. These are not of the homeobox class but are related to another family of gene regulatory proteins (the so-called MADS family) found in yeast and in vertebrates.

Investigation of the molecular genetics of plant development has only just begun. So far, almost nothing is known, for example, about the genetic systems responsible for local cell-cell communication and positional signaling in plant pattern formation. Yet it is clear already that plants and animals, despite their differences, have independently found very similar solutions to many of the fundamental problems of multicellular development.

Summary

The development of a flowering plant, like that of an animal, begins with division of a fertilized egg to form an embryo with a polarized organization: the apical part of the embryo will form the shoot, the basal part, the root, and the middle part, the stem. At first, cell division occurs throughout the body of the embryo. As the embryo grows, however, addition of new cells becomes restricted to small regions known as meristems. Apical meristems, at shoot tips and root tips, will persist throughout the life of the plant, enabling it to grow by sequentially adding new body parts at its periphery. Typically, the shoot generates a repetitive series of modules, each consisting of a segment of stem, a leaf, and an axillary bud. An axillary bud is a potential new meristem, capable of giving rise to a side branch, and the environment can control the development of the plant by regulating bud activation. Environmental cues can also cause the apical meristem to switch from a leaf-forming to a flower-forming mode. Long-range signaling mediated by plant hormones coordinates such developmental events occurring in separate parts of the plant.

The internal organization of each plant module, however, is controlled through strictly local pattern formation mechanisms analogous to those that govern animal development. These operate in the neighborhood of the apical meristem, where the relative positions of the rudiments of leaves and other organs are initially mapped out on a microscopic scale. The pattern of modified leaves - sepals, petals, stamens, and carpels - in a flower is set up similarly. The genetic basis of pattern formation in plants can be analyzed in the same way as in animals. The small weed Arabidopsis thaliana is widely used as a "model plant" for such studies. Genes governing the organization of the embryo, analogous to the egg-polarity and segmentation genes of Drosophila, can be identified. And the sequence of parts in a flower is controlled by homeotic selector genes closely analogous (although not homologous) to those of animals.

Introduction

Nerve cells, or neurons, are among the most ancient of all specialized animal cell types, as important to jellyfish and sea anemones as they are to worms, flies, and people. Their structure is like that of no other class of cells, and the development of the nervous system poses problems that have no parallel in other tissues. A neuron is extraordinary above all for its enormously extended shape, with a long axon and dendrites connecting it through synapses to other cells (Figure 21-99). The central challenge of neural development is to explain how the axons and dendrites grow out, find their right partners, and synapse with them selectively to create a functional network (Figure 21-100).

Most of the components of a typical nervous system - the various classes of neurons, sensory cells, and muscles - originate in widely separate locations in the embryo and are initially unconnected. Thus, in the first phase of neural development (Figure 21-101), the different parts develop according to their own local programs, following principles of cell diversification common to other tissues of the body, as already discussed. The next phase involves a type of morphogenesis unique to the nervous system: a provisional but orderly set of connections is set up between the separate parts of the system through the outgrowth of axons and dendrites along specific routes, so that the parts can begin to interact. In the third and final phase, which continues into adult life, the connections are adjusted and refined through interactions among the far-flung components in a way that depends on the electrical signals that pass between them.

Stocks of Neurons Are Generated at the Outset of Neural Development and Are Not Subsequently Replenished ⁷⁸

The nervous system develops from the ectoderm in all animals. In vertebrates, on which we concentrate here, it derives chiefly, as we saw earlier in this chapter, from two sets of cells - those of the neural tube (an invagination of the ectoderm) and those of the neural crest (a population of cells that break loose from the neural ectoderm and migrate to other regions of the embryo). The neural tube forms the central nervous system (the spinal cord and brain, including the retina of the eye), while the neural crest gives rise to most of the neurons and supporting cells of the peripheral nervous system (Figure 21-102).

The neural tube, with which we shall be mainly concerned, consists initially of a single-layered epithelium (Figure 21-103). This will generate both the neurons and the associated supporting, or glial, cells of the central nervous system. In the process it becomes transformed into a thicker and more complex structure with many layers of cells of various types.

Because differentiated neurons do not divide, each one can be assigned a "birthday," defined as the time of the final mitosis that generated it from a dividing neuronal precursor cell. In both higher vertebrates and invertebrates, the birthdays of the neurons of a given type generally all occur within a strictly limited period of development, after which no further neurons of that type are produced. Each region of the developing neural tube has its own program of cell divisions, and neurons with different birthdays are generally destined for different functions. Since neural stem cells usually do not persist once the production of nerve cells is complete, nerve cell numbers thereafter can only be regulated downward, through cell death, as we shall see.

The Time and Place of a Neuron's Birth Determine the Connections It Will Form ⁷⁹

Before sending out its axon and dendrites, the immature neuron or its precursor commonly migrates from its birthplace and settles in some other location. In the central nervous system glial cells often provide a pathway for the migration. The neural tube of a vertebrate embryo, for example, contains a scaffolding of radial glial cells. Each of these cells extends from the inner to the outer surface of the tube, a distance that may be as much as 2 cm in the cerebral cortex of the developing brain of a primate. Prospective neurons go through their final cell division close to the lumen of the neural tube and then travel outward by crawling along the radial glial cells (Figure 21-104).

Successive cohorts of migrant cells, born at different times, settle in different positions. In the cerebral cortex, for example, the neurons become arranged in layers according to their birthdays as a result of a migration in which the cells that are born later migrate outward past those born earlier. By transplanting cells between young and old embryos, it can be shown that these different choices of destination are already specified before the cells set off on their migration; they reflect differences in the intrinsic characters of the cells produced at different times - differences that will also dictate the synaptic connections that the cells later form. Thus, in the cerebral cortex the early-born cells (in inner layers) will send their axons to regions outside the cortex, while the late-born cells (in outer layers) will send their axons to regions within the cortex. This relationship between birthday and axonal connections is maintained even in a mutant mouse in which the migrations are abnormal and the final positions of the early- and late-born cells are inverted, confirming that the connections reflect the intrinsic character, rather than the final location, of the neurons (Figure 21-105).

No less important than the time of birth of a neuron is the place of its birth. Cells in different regions of the neural tube have different positional values that govern the connections they will form. These position-dependent differences are evident in the pattern of expression of the Hox genes, as we have already seen, and of a large number of other genes that code for gene regulatory proteins and other regulatory molecules. The mechanisms that create the molecular differences between prospective neurons are poorly understood, but they seem, where known, to be similar in principle to the mechanisms of pattern formation discussed earlier. The question we have to confront now, however, is a different one: how do the newborn nerve cells, equipped with their specific markers, proceed to set up an orderly pattern of connections?

Each Axon or Dendrite Extends by Means of a Growth Cone at Its Tip ^{77, 80}

As a rule the axon and the dendrites begin to grow out from the nerve cell body soon after the cell body has reached its final location. The sequence of events was originally observed in intact embryonic tissue by the method of Golgi staining (Figure 21-106). This technique, and other methods developed subsequently, reveal an irregular, spiky enlargement at the tip of each developing nerve cell process. This structure, which is called the growth cone, appears to be crawling through the surrounding tissue. It comprises both the engine that produces the movement and the steering apparatus that directs the tip of each process along the proper path.

Much of what we know about the properties of growth cones has come from studies in tissue or cell culture. One can watch as a neuron begins to put out its processes, all at first alike, until one of the growth cones puts on a sudden turn of speed, identifying its process as the axon, with its own axon-specific set of proteins (Figure 21-107). The contrast between axon and dendrite established at this stage will cause the two types of process to grow out for different distances, to follow different paths, and to play different parts in synapse formation.

For an isolated neuron in culture the distinction between axon and dendrite is not always easy to see, and it is convenient to refer to both types of process as neurites. The growth cone at the end of a typical rapidly growing neurite moves forward at a speed of about 1 mm per day. It consists of a broad, flat expansion, like the palm of a hand, with many long microspikes or filopodia extending from it like fingers (Figure 21-108). These are continually active: some are retracting back into the growth cone while others are elongating, waving about, and touching down and adhering to the substratum. The "webs" or "veils" between the filopodia form lamellipodia with a typical ruffling membrane. All these features, as well as the configuration of the cytoskeleton internally, suggest that the growth cone is crawling forward in much the same way as the leading edge of a cell such as a neutrophil or fibroblast, as discussed in Chapter 16.

With its filopodia and lamellipodia the growth cone explores the regions that lie ahead and on either side. When such a protrusion contacts an unfavorable surface, it withdraws; when it contacts a more favorable surface, it persists longer, steering the growth cone as a whole to move in that direction. In this way the growth cone can be guided by subtle variations in the surface properties of the substrata over which it moves.

The Growth Cone Pilots the Developing Neurite Along a Precisely Defined Path in Vivo^{81, 82, 83, 84}

In living animals growth cones generally travel toward their targets along predictable routes, exploiting a multitude of different cues to find their way. Most often, they take routes that have been pioneered by other neurites, which they follow by contact guidance. As a result, nerve fibers in a mature animal are usually found grouped together in tight parallel bundles (called fascicles or fiber tracts). Such crawling of growth cones along axons is thought to be mediated by homophilic cell-cell-adhesion molecules - membrane glycoproteins that help a cell displaying them to stick to any other cell that displays them also. As discussed in Chapter 19, two of the most important classes of such molecules are those that belong to the immunoglobulin superfamily, such as N-CAM, and those of the Ca²⁺-dependent cadherin family, such as N-cadherin. Members of both families are generally present on the surfaces of growth cones, of axons, and of various other cell types that growth cones crawl over, including glial cells in the central nervous system and muscle cells in the periphery of the body. Growth cones also migrate over components of the extracellular matrix, especially laminin, which they bind to by means of cell-surface matrix receptors of the integrin family (discussed in Chapter 19).

In some cases one can demonstrate the importance of a given cell-cell or cell-matrix adhesion molecule by blocking its function with an antibody and observing a disturbance of axon outgrowth. But usually a growth cone employs several adhesion systems to migrate, and antibodies against any single one of them have little effect; only when multiple antibodies are applied, so as to block all of them together, is the growth cone severely hindered in its navigation. In principle, different combinations of adhesion molecules allow for great variety in the surface properties of growth cones and for subtle and complex pathway selection according to the combinations of molecules on the surfaces of cells along the way.

It is still uncertain how far different combinations of adhesion proteins such as N-CAM, N-cadherin, and integrins in the growth cone membrane are sufficient to explain why some growth cones take one route while others take another or how a set of axons, on reaching their target region, are able to form synapses there in an orderly array. Adhesion molecules are certainly not the only influences at work. The contacts a growth cone makes with cell surfaces and matrix can give rise to intracellular signals that can, for example, actively inhibit forward movement. Substances that diffuse through the extracellular medium can also give rise to gradients that provide guidance. In the developing spinal cord, for example, there is a group of neurons whose axons travel ventrally, toward the floor plate of the neural tube, to cross by that route to the other side of the tube. When these neurons are placed in culture a short distance from an explanted fragment of floor plate, their axons will again orient their outgrowth toward it, implying that the specialized cells in the floor plate secrete molecules that have a chemotactic guiding effect.

Target Tissues Release Neurotrophic Factors That Control Nerve Cell Growth and Survival ^{82, 85}

Most types of neurons in the vertebrate central and peripheral nervous system are produced in excess; up to 50% or more of them then die soon after they reach their target, even though they appear perfectly normal and healthy up to the time of their death. About half of all the motor neurons that send axons to skeletal muscle, for example, die within a few days after making contact with their target muscle cells. This large-scale death of neurons is thought to reflect the outcome of a competition. Each type of target cell releases a limited amount of a specific neurotrophic factor that the neurons innervating that target require to survive: the neurons apparently compete to take up the factor, and those that do not get enough die by programmed cell death. This seemingly wasteful process provides a simple and elegant means of adjusting the number of neurons of each type to the number of target cells that they innervate.

The first neurotrophic factor to be identified, and still the best characterized, is known simply as nerve growth factor, or NGF. It was discovered by accident in the course of experiments in which foreign tissues and tumors were transplanted into chick embryos. Transplants of one particular tumor became exceptionally densely innervated and caused a striking enlargement of certain groups of peripheral neurons in the vicinity of the graft. Just two classes of neurons were affected: sensory neurons and sympathetic neurons (a subclass of peripheral neurons that control contractions of smooth muscle and secretion from exocrine glands). The cause of this phenomenon was traced to a specific protein, NGF, and it was shown that if anti-NGF antibodies are administered to mice while the nervous system is still developing, most sympathetic neurons and some sensory neurons die. Sympathetic neurons and some sensory neurons also die in culture in the absence of NGF; if NGF is present, they survive and send out neurites (Figure 21-109). Some classes of neurons in the central nervous system are dependent on NGF in a similar way.

NGF is produced by the tissues that are innervated by NGF-dependent neurons. Experimental manipulations confirm that the larger the quantity of target tissue, the larger the number of surviving neurons, and this effect can be shown to be mediated by NGF because it can be mimicked by direct manipulation of NGF concentrations. Later in life, after the phase of cell death is over, NGF has a continuing role in regulating the density of innervation by controlling the extent of local sprouting of axon branches. This mechanism is important in restoring innervation in tissues such as skin and smooth muscle after an injury. NGF acts in the intact animal just as it does in a culture dish (see Figure 21-109), both to sustain cell survival and as a local stimulus for growth cone activity, thus adjusting the supply of innervation according to the requirements of the target.

NGF is only one of a family of homologous neurotrophic factors (called neurotrophins) that are responsible for this type of regulation in different parts of the vertebrate nervous system. They bind to a complementary family of transmembrane receptor proteins (named after a proto-oncogene called trk that codes for one of them), which belong to the tyrosine-kinase class of receptors discussed in Chapter 15. It is hoped that the neurotrophic factors will prove useful in the treatment of neurological diseases, such as Alzheimer's disease and motor neuron disease (Lou Gehrig's disease), in which neurons degenerate and die inappropriately.

We now return to the problem of the spatial patterning of nerve connections.

The Positional Values of Neurons Guide the Formation of Orderly Neural Maps: The Doctrine of Neuronal Specificity ⁸⁶

The inputs from sense organs are generally mapped or projectedin an orderly way onto the sensory regions in the central nervous system, and the outputs from the motor regions of the central nervous system are mapped in an orderly way onto the muscles. Thus, similar nerve cells in different regions of the vertebrate retina send their axons to synapse with neurons in correspondingly different regions of the optic tectum in the midbrain (Figure 21-110), and similar motor neurons at different locations in the spinal cord send their axons to different muscles.

In principle, the growth cones could be simply channeled to different destinations as a direct consequence of their different starting positions, like drivers on a multilane highway where it is forbidden to change lanes. This possibility was tested in the visual system by a famous experiment in the 1940s. If the optic nerve of a frog is cut, it will regenerate. The retinal axons grow back to the optic tectum, restoring normal vision. If, in addition, the eye is rotated in its socket at the time of cutting of the nerve, so as to put originally ventral retinal cells in the position of dorsal retinal cells, vision is still restored, but with an awkward flaw: the animal behaves as though it sees the world upside down. This is because the misplaced retinal cells make the connections appropriate to their original, not their actual, positions (Figure 21-111). The cells are evidently endowed with positional values, carrying a record of their original position, so that cells on opposite sides of the retina are intrinsically different. As in the cortex of the reeler mouse (see Figure 21-105), it is the intrinsic character, rather than the position, that decides the choice of target site. Such nonequivalence among neurons is referred to as neuronal specificity.

Axons from Opposite Sides of the Retina Respond Differently to a Gradient of Repulsive Molecules in the Tectum ⁸⁷

On reaching the tectum, the retinal axons must choose, according to their individual character, which region of tectum to innervate. Axons from the nasal retina (the side closest to the nose), for example, project to the posterior tectum, and axons from the temporal retina (the side farthest from the nose) project to the anterior tectum. This choice is governed by differences in the intrinsic characters of the cells in different parts of the tectum. Thus the neuronal map depends on a correspondence between two systems of positional markers, one in the retina and the other in the tectum.

Experiments in vitro with tissues from the chick embryo give some insight into the nature of the tectal markers and the way in which the retinal axons respond to them. Fragments of retina are placed in culture and allowed to send out axons over a substratum that is carpeted with membrane vesicles prepared from tectal cells (Figure 21-112). The carpet is laid out in stripes, with bands of anterior tectal membrane alternating with bands of posterior tectal membrane. Axons from nasal retina, depending on details of the preparation, either show no preference and grow indiscriminately in all of the bands or show a preference, appropriately, for posterior tectal membrane. Axons from temporal retina consistently grow only along the bands of anterior tectal membrane, in accordance with their normal destiny. Surprisingly, this is not because the anterior tectal membrane is particularly adhesive or attractive to them but because the posterior tectal membrane is particularly repellent: filopodia that touch it withdraw and collapse. In fact, the growth cones of the temporal axons (but not those of nasal axons) will collapse and retract if a suspension of posterior tectal membranes is dripped onto them. No such collapse occurs in response to anterior tectal membrane.

The peculiar effects of the posterior tectal membrane on the temporal retinal cells have been traced to a specific inhibitory glycoprotein that is distributed in a gradient from posterior to anterior in the tectum. In other parts of the nervous system other surface molecules can be shown to have analogous functions as growth cone repellents. These crude systems of markers are adequate to define the anteroposterior orientation of the map in the frog optic tectum. Other mechanisms of an entirely different sort, however, are required to make the map precise.

Diffuse Patterns of Synaptic Connections Are Sharpened by Activity-dependent Synapse Elimination ^{88, 89}

In a normal animal the retinotectal map is initially fuzzy and imprecise. Studies in frogs and fish show that each retinal axon at first branches widely in the tectum and makes a profusion of synapses, distributed over a large area of tectum that overlaps with the territories innervated by other axons. These territories are subsequently trimmed back by elimination of synapses and retraction of axon branches. This refinement of the map through synapse elimination is governed by two competition rules that jointly create spatial order: (1) axons from separate regions of retina, which tend to be excited at different times, compete to dominate the available tectal territory, but (2) axons from neighboring sites in the retina, which tend to be excited at the same time, innervate neighboring territories in the tectum because they collaborate to retain synapses on shared tectal cells (Figure 21-113). The mechanism underlying both these rules depends on electrical activity and signaling at the synapses that are formed. If all action potentials are blocked by a toxin that binds to voltage-gated Na⁺ channels, synapse elimination is inhibited and the map remains fuzzy.

This phenomenon of activity-dependent synapse elimination is encountered in almost every part of the developing vertebrate nervous system. Synapses are first formed in abundance and distributed over a broad target field; then the system of connections is pruned back by competitive processes that depend on electrical activity and synaptic signaling. The elimination of synapses in this way is distinct from the elimination of surplus neurons by cell death, and it occurs after the period of normal neuronal death is over.

The cellular mechanisms of synapse elimination are beginning to be clarified by experiments on the innervation of skeletal muscle in vertebrate embryos, where typically each muscle cell at first receives synapses from several neurons but in the end is left innervated by only one. Co-cultures of motor neurons with muscle cells can be used to analyze the mechanism in vitro. One can identify a muscle cell that is innervated by a single neuron and then directly excite the muscle cell repeatedly with puffs of acetylcholine delivered through a micropipette close to its surface. The synapse made on the muscle cell by the neuron is found to be permanently weakened by this treatment unless the neuron itself is stimulated electrically so that it fires in synchrony with the acetylcholine puffs delivered to the muscle cell, in which case the synapse remains strong (Figure 21-114). Weakening, or repression, of the synapse reflects a change on its presynaptic side, which causes the axon terminal to release less neurotransmitter when the neuron fires. It can be shown that this synaptic repression depends on the entry of Ca²⁺ into the muscle cell through the cation channels associated with the acetylcholine receptors. Somehow, a sudden rise in intracellular Ca²⁺ causes the postsynaptic cell to send a rebuff to any axon terminals synapsing on its surface in that neighborhood, but the axon terminals are immune to this rebuff if they themselves have just been active.

These and many other findings suggest a simple interpretation of the competition rules for synapse elimination in the retinotectal system. Axons from different parts of the retina fire at different times and so compete. Each time one of them fires, the synapse(s) made by the other on a shared tectal target cell are weakened, until one of the axons is left in sole command of that cell. Axons from neighboring retinal cells, on the other hand, tend to fire in synchrony with one another: they therefore do not compete but instead maintain synapses on shared tectal cells, creating a precisely ordered map in which neighboring cells of the retina project to neighboring sites in the tectum (see Figure 21-113).

Experience Molds the Pattern of Synaptic Connections in the Brain ^{89, 90}

The same "firing rule" relating synapse maintenance to neural activity helps to organize our developing brains in the light of experience. In the brain of a mammal axons relaying inputs from the two eyes are brought together in the visual region of the cerebral cortex, where they form two overlapping maps of the external visual field, one as perceived through the right eye, the other as perceived through the left. The organization and development of the cortical projections from the two eyes have been studied in great detail, both by anatomical tracing and by physiological tests in which single cortical cells are monitored to find out what kinds of visual stimulus will excite them. These studies reveal an extraordinary sensitivity to experience early in life: if, during a certain critical period, one eye is kept covered so as to deprive it of visual stimulation, while the other eye is allowed normal stimulation, the deprived eye loses its synaptic connections to the cortex and becomes almost entirely, and irreversibly, blind. In accordance with the firing rule, a competition has occurred in which synapses in the visual cortex made by inactive axons are eliminated while synapses made by active axons are consolidated. In this way cortical territory is allocated to axons that carry information and is not wasted on those that are silent.

But the firing rule also operates in more subtle ways to establish the nerve connections that enable us to see. For example, the ability to see depth - stereo vision - depends on the presence in the visual cortex of cells that receive inputs from both eyes at once, conveying information about the same part of the visual field as seen from two slightly different angles. These binocularly driven cells allow us to compare the view through the right eye with that through the left so as to derive information about the relative distances of objects from us. If, however, the two eyes are prevented during the critical period from ever seeing the same scene at the same time - for example, by covering first one eye and then the other on alternate days or simply as a consequence of a childhood squint - almost no binocularly driven cells are retained in the cortex, and the capacity for stereo perception is irretrievably lost. Evidently, in accordance with the firing rule, the inputs from each eye to a binocularly driven neuron are maintained only if the two inputs are frequently triggered to fire in synchrony, as occurs when the two eyes look together at the same scene.

We saw in Chapter 15 that synaptic changes underlying memory in many parts of the brain hinge on the behavior of a particular type of receptor for the neurotransmitter glutamate - the NMDA receptor. Ca²⁺flooding into the postsynaptic cell through the channels opened by this receptor triggers lasting changes in the strengths of the synapses on that cell, just as Ca²⁺ entering a muscle cell via acetylcholine-receptor channels during development affects the synapses made on it by motor neurons. The changes that are induced by the NMDA-dependent mechanism in the adult brain obey rules closely akin to the developmental firing rule. In fact, the refinement and remodeling of synaptic connections that we have just described in the developing visual systems of mammals and amphibians can be blocked by an inhibitor of the NMDA receptor. Both memory and the developmental adjustments, therefore, may depend on essentially the same machinery. The molecular basis of this device through which experience molds our brains is one of the central challenges that the nervous system presents to cell biology.

Summary

The development of the nervous system proceeds in three phases: first, nerve cells are generated through cell division; then, having ceased dividing, they send out axons and dendrites to form profuse synapses with other, remote cells so that communication can begin; last, the system of synaptic connections is refined and remodeled according to the pattern of electrical activity in the neural network.

Axons and dendrites grow out by means of growth cones at their tips, following specific pathways delineated by cells and extracellular matrix along the way. The guidance depends on many different classes of adhesion molecules and intercellular signals as well as on factors that inhibit and repel growth cones. Growth cones from different, nonequivalent neurons respond differently to these cues, and in this way neural maps are set up - orderly projections of one array of neurons onto another. After the growth cones have reached their targets, two major sorts of adjustment occur. First, many of the innervating neurons die as a result of a competition for survival factors such as NGF (nerve growth factor) secreted by the target tissue. This cell death adjusts the quantity of innervation according to the size of the target. Second, individual synapses are pruned away in some places, reinforced in others, so as to create a more precisely ordered pattern of connections. This process depends on electrical activity: synapses that are frequently active are reinforced, and different neurons contacting the same target cell tend to maintain their synapses on the shared target only if they are both frequently active at the same time. In this way the structure of the brain can be adjusted to reflect the connections between events in the external world. The underlying molecular mechanism may be similar to that responsible for the formation of memories in adult life.