Cell, Vol.

100, 2740, January 7, 2000, Copyright 2000 by Cell Press

Development: The Natural History of Genes

Review

Matthew P. Scott Departments of Developmental Biology and Genetics Howard Hughes Medical Institute Stanford University School of Medicine Stanford, California 94305-5329

Introduction Developmental biology is a powerful way to learn how genes and proteins operate in their natural habitats. In a growing animal, genes come to life. Just as animal structure becomes more understandable when ecology, physiology, and behavior are taken into account, genes and proteins are best understood in the context of developmental biologya natural history of genes. Developmental biology will help to exploit the current flood of descriptive knowledge from gene hunters. The gene flood is reminiscent of the first time plants and animals were extensively described and classified. Carolus Linnaeus (17071778), famed as Gods registrar, was a botanist and professor of medicine at the University of Uppsala (Goerke, 1973). Linnaeus confronted a chaotic natural world and organized it, inventing the system of binomial nomenclature (Linnaeus, 1735). Not all the relationships were clear. Observing organisms in terms of their ecology and physiology was helpful in sorting them out. Similar organizing power comes from learning how proteins work together in pathways and complexes, connecting proteins to particular tissues or changes in cell properties. Newfound genes can often be connected to known processes or pathways because mutations in the new genes give rise to familiar phenotypes. We have much to learn about the origins of families of proteins and how they became dedicated to their present day tasks. Linnaeus, also facing daunting complexity, began his classification with elegant simplicity: Minerals grow, plants grow and live, animals grow, live, and have feeling. The first edition of Systema Naturae in 1735 listed 549 species; by the eleventh edition, 5897 species were represented. Not all of them were in contigs. Linnaeus students and colleagues formed a worldwide web (www.org), supplying him with a constant flow of seeds, plants, eggs, animal skins, skeletons, descriptions, and lore. Animals were divided into six categories: quadrupeds, birds, amphibia (including reptiles), fishes, insects, and worms (including most invertebrates). Another category, the Paradoxa contained animals of uncertain type, including satyrs, phoenixes, and pelicans. Linnaeus promoted acceptance of his system by the clever political tactic of naming plants after eminent biologists. (Perhaps instead of naming categories of proteins cadherins or cyclins, we could have varmins and bishins, nussvolins and wieschins, and lewins.) Linnaeus was first to emphasize the relationship between humans and apes, causing an outcry. Much more startling relationships between humans and animals became evident upon investigating the mechanisms that build them. During the past 20 years developmental biology has been enriched by a profound recognition of

astonishing natural biological order, creating optimism about discovering new kinds of medicine. Natural biological order is familiar from the near universality of metabolic pathways and the genetic code. Developmental mechanisms underlying the morphology of diverse creatures are also remarkably universalmore than anyone had dreamed. There had been concern that molecules and regulatory mechanisms would vary tremendously through the animal kingdom, and that hundreds of separate signaling systems would be operative, but neither idea appears to be true. Darwin quoted von Baer: The feet of lizards and mammals, the wings and feet of birds, no less than the hands and feet of man, all arise from the same fundamental form (Darwin, 1871). In one sense von Baers form is the shape of proteins such as FGF and Hox and Hedgehog. The developmental mechanisms of each experimental organism used to seem impressively distinct; recently the emphasis has been on unifying themes and common genes. The field of development and evolution is reconciling these two conflicting view by finding out how a modest number of protein types can build a plethora of animal types. Linnaeus worked at a time when it was important (and still news) to affirm that every animal comes from an egg. We have come a long way. We consider here current knowledge of developmental biology, focusing on molecules that have maintained their connection to a process or pathway for half a billion years or so. Main findings include the following: (1) Different cell types transcribe different genes. Cytologically indistinguishable cells can have elaborate spatial and temporal gene expression patterns. (2) Although DNA rearrangements and deletions are used in specific instances for differential gene function, transcription is a more common type of gene regulation. (3) Transient bursts of transcription factor function, or of signals, lead to lasting changes in cell phenotype, often affecting generations of cells. Changes in chromatin structure, often reversible, perpetuate earlier regulatory decisions. (4) Localized RNA molecules create differences in egg cytoplasm and within cells for asymmetric cell divisions. Asymmetric cell division is crucial for both stem cells and differentiation. (5) Transcription factors regulate batteries of genes. Combinations of regulators govern most differentiation events, so the same regulator can be used repeatedly. The outcome depends on what other regulators are acting in concert or opposition. (6) Embryos are remarkably flexible in adapting to different cell numbers, environmental conditions, and altered gene function. Cells are sensitive to the character of their neighbors and tend to die or change if the neighboring cells are inappropriate. They are also sensitive to the number of neighbors, the community effect. (7) Many signals act over short distances, perhaps reflecting their origins in organisms with few cells. (8) Systems of checks and balances keep powerful signals in check. Intracellular and extracellular antagonists that limit the range or potency of the signals affect most or all signals. Some antagonists are inducible by their signals, thus providing feedback responses

Cell 28

in proportion to the initial intensity of the signal. Secreted antagonists could have evolved as a form of chemical warfare between organisms. (9) Cells are responsive to signals or to transcription factors at some times but not others, a property known as competence. Also, cells respond to a particular signal differently, depending on their history. History affects signal effectiveness, making certain genes responsive to signals by localizing molecules within cells or by polarizing, shaping, or moving the cell. (10) Spatial and temporal regulation of the cell cycle is crucial to forming organized tissues, allowing restorative events, and maintaining stem cells. (11) Cell death and aging are genetically governed processes, precisely scripted and regulated during normal development and alterable by disease or injury. (12) Most of the proteins that regulate development have been conserved through evolution. Often related genes have similar functions in vastly different animals. I have organized this review using the control of yeast matingtype to frame the discoveries of the past few decades. The ramifications of the MAT locus encompass phenomena common to many developing organisms: cell differentiation, differential gene transcription, DNA rearrangement, localized RNA molecules, coordinate regulation of gene batteries, inheritance of determined cell states, gene silencing, signals and sex, cell division, and cell asymmetry. Modern developmental biology has its roots in pioneer subjects of molecular biology. B. Polisky has proposed that the phage genome would be the most entertaining 50 kb of DNA to have along if one was to be marooned on a desert island with nothing to do but experiments. The lifecycle (Hendrix et al., 1983) is extraordinarily fascinating for a developmental biologist, with beautifully balanced regulation of operons, DNA-binding proteins, competing actions of different regulators, and the ordered assembly of a complex infectious particle. Similarly, yeast biology is a useful bridge to the processes of multicellular collaboration. Much developmental biology is included in accompanying reviews, including bacterial development, stem cells, neurogenesis, axon pathfinding, plant biology, and the cell cycle. A major review of induction, gradients, and organizers, among other topics, is provided in Fraser and Harland (2000 [this issue of Cell]). I will not emphasize those topics nor the revolution in understanding plant development (Hake and Meyerowitz, 1998) here. A celebration of pioneers such as Theodor Boveri, Hans Driesch, Oscar Hertwig, Wilhelm Roux, Hans Spemann, Walter Sutton, August Weismann, Edmund Wilson, and others is in Horder et al. (1985). Transcribing Different Genes The central mystery of development was simply stated by August Weismann in 1883: How is a single germ cell capable of reproducing the entire body with all its detail? Morgan connected this basic question with genetics in his Nobel lecture of 1935: If as is generally implied in genetic work (although not often explicitly stated), all of the genes are active all of the time, and if the characters of the individual are determined by the genes, then why are not all the cells of the body exactly alike? . . . Every cell comes to contain the same kind of genes. Why then is it that some cells become muscle

cells, some nerve cells, and others remain reproductive cells? (Sander, 1985). To these issues we must add the problem of pattern formation: how can a right hand reliably be made to mirror the left? That transcription is an important level of control was clear decades ago from studies of bacteria. The earliest views of differential transcription in an eukaryote came from watching polytene chromosomes from fly salivary glands change their puffing patterns with developmental time (Beermann, 1952). The puff patterns, showing visible changes at 194 loci, are indicators of differential activation of genes in time, in part under the control of steroid hormones (Ashburner, 1990). This was the first direct demonstration of differential gene activity in response to a steroid. More insights into multicellular development came when steady-state populations of RNA molecules were measured in different cell types using solution hybridization. In R0T curve experiments, hybridization rates were dependent on initial concentration (R0) and time (T). Elegant kinetic analyses were used to measure the complexity of the RNA and compare sequences from a variety of cells (Galau et al., 1974; Hough et al., 1975). This allowed the recognition of classes of RNAs that varied tremendously in their abundance, the detection of differences in RNA populations between cells of different types, and detection of the transcription of repetitive versus single-copy DNA. The favored human cell mix, the brain, took honors for the most complex RNA population. In the yeast Saccharomyces, transcription patterns distinguish three cell types. The two haploid matingtypes are named a and ; the third type of cell is a diploid heterozygote, a/ . The DNA sequence of the two haploid genomes is nearly the same. Certain genes are uniquely active in one or two of the three cell types, constituting one basis for cell differentiation. RNA transcription is one measure of the differences. Of 6164 genes analyzed, 32 genes were found to be differentially expressed according to mating-type (Galitski et al., 1999). Seventeen genes changed with ploidy, regardless of mating-type. Haploid cells have 32 genes transcribed at more than 2-fold higher levels than in haploid a cells. Haploid a cells have 50 genes transcribed at more than 2-fold higher levels than in haploid cells. The situation is similar in diploid cells. Many of the haploidspecific genes encode products that allow the cells to mate. Remarkable changes in phenotype can therefore involve a reasonably small fraction of the genes. As more experiments are done, we will learn how many genes do not change their transcription rate at all. In the yeast experiments, 2147/6164 gene transcription levels differed by less than 10% between a and . Sixteen hundred eighty-eight genes varied between haploid and diploid cells. During yeast sporulation, approximately 1000 of the approximately 6200 genes changed their levels of transcription (Chu et al., 1998). Many genes may be active only under certain nutritional, lifecycle, or temperature conditions. Indeed the genes that are active regardless of conditions may be small. Gene arrays applied to one particular cell culture change give a hint of what may be expected during development. When a fibroblast in culture is exposed to serum, more

Review 29

than 500 genes of 8613 examined substantially change their transcription level (Iyer et al., 1999). The types of genes that become active suggest that exposure to serum signals the fibroblast to engage in wound repair, as it would in response to contact with serum in a wounded human. Spatial and Temporal Control of Gene Expression Another revelation came with the invention of in situ hybridization for detecting RNA fixed in tissues. Spatial differences in patterns of transcription, largely concealed when solution hybridization or RNA blots were used, suddenly became visible. Fields of apparently homogeneous cells took on great character as stripes and spots of RNA expression came to light. The Drosophila segmentation genes (Nusslein-Volhard and Wieschaus, 1980) have become a favorite example of patterned gene expression. At the stage when about 6000 cells on the embryos surface enclose the yolk, and most of the cells look the same, beautiful patterns of striped gene expression are detected for some of the gap, pair-rule, and segment-polarity genes. Most of the gap and pair-rule genes encode transcription factors that participate in a cascade of regulators that refines the stripes until they are just one cell wide. Gap proteins are produced in broad stripes under the control of localized regulators provided maternally. Gap proteins regulate pair-rule (and Hox) transcription, and in turn pair-rule proteins regulate striped segment-polarity gene transcription. As in yeast matingtype, combinations of enhancer-binding proteins govern transcription, in this case often combinations of different gap or pairrule proteins (Rivera-Pomar and Jackle, 1996). Because gap gene expression domains occur in broad peaks, graded toward anterior and posterior, and because they overlap, each location along the anteriorposterior (AP) axis has particular amounts and combinations of transcription factors. Those combinations govern where pair-rule genes become active, and also affect gap gene transcription in feedback loops. Some proteins repress, some activate, and some may do both. Ubiquitous cofactors assist in the regulation, as they do in yeast. The key to integrating the upstream regulators lies in enhancer sequences, where the arrangement of bound proteins determines the outcome. A pair-rule gene may have several enhancers, each one governing one or more stripes. The gene may be on in response to any of several different combinations of regulators. For example, typically such a successful combination of activating regulators occurs within each of seven stripes of cells along the AP axis. If an activating combination of proteins resides on any one of the enhancers, the gene will be on, while repressing combinations of proteins may be present on other enhancers. At some positions along the AP axis, no enhancer will host an activating protein combination. How a group of proteins bound to an integrating enhancer battle it out and then send a consensus signal to the promoter remains an active area of research (Mannervik et al., 1999). Some repressors act by sterically hindering binding of activators, while others actively interfere with bound activators, perhaps by competing for contacts with the basal transcription apparatus. One yearns for a microscopic picture of gene expression in action.

Temporal control of gene expression is implicit in the time it takes to produce a functional gap protein or pairrule protein; downstream genes will become active only after activating transcription factors are produced. The cell cycle is controlled by oscillating levels of functional cyclin/CDC complexes, controlled in turn by rates of phosphorylation and protein degradation (see review by Nurse, 2000 [this issue of Cell]). Any one clock, like the cell cycle oscillator, has the potential to affect and therefore schedule other events in development. The transcription of certain genes is linked to circadian rhythm, including c-hairy1, whose transcription cycles in time with the formation of somites (Pourquie, 1998). Other types of timers rely on the dilution or exhaustion of factors, as in the embryonic initiation of zygotic transcription in flies or frogs that occurs as the population of nuclei becomes sufficiently large. Patterns of cell division are controlled in time by heterochronic genes, as is clear from fascinating work on C. elegans. Mutations in heterochronic genes lead to premature or postponed sets of cell divisions. One of the heterochronic genes is related to circadian rhythm genes found first in flies (Jeon et al., 1999), suggesting a link between seemingly different types of timing mechanisms. DNA Rearrangement to Change Cell Type Yeast cells can change mating-type every generation, thus having the advantages of spore formation in starvation conditions and of recombination. The differences between the two types of haploid yeast cells are due to a locus called MAT that has unusual properties (Haber, 1998). Two alleles, a and , contain alternative versions of a 700 bp sequence. The locus is changed from one version to the other by a programmed DNA sequence change that transfers a sequence from either of two donor sites to the MAT locus, inserting different cassettes. The genes at the donor loci, HML and HMR, are kept silent by a chromatin protein repression system. A yeast cell buds to reproduce; a mother produces a daughter cell that is readily distinguishable morphologically. After such a cell division, the daughter cell will divide again without switching cell types, but the mother cell will divide to produce two cells, both of which have switched their mating-type. The resulting four cells will constitute two mating pairs, which will often form two diploid cells. Switching ability, an aspect of differentiation itself, is precisely controlled so that it occurs only in certain cell cycles, coordinated with cell lineage. Switching occurs only during the G1 phase of the cell cycle, which ensures that both cells formed at the next division will have the same genotype. Switching is initiated by production of the HO nuclease only in cells that will switch; the nuclease creates a DNA break to start the recombination event. While the MAT locus has special properties, echoes of those properties are heard in governance of immunoglobulin gene expression and switching (Cedar and Bergman, 1999; Oettinger, 1999). Again silent loci are rearranged to form an active gene, and the unused alleles are silent or deleted (Stavnezer, 1996). Another DNA change occurs in differentiation of egg follicle cells in Drosophila. The chorion genes must produce staggering amounts of eggshell protein in a short period of

Cell 30

time. They manage it by undergoing a cell typespecific overreplication of the genes (Calvi and Spradling, 1999). Trypanosomes have the challenge of tricking the immune system of the host. They have in their genome a large assortment of alternative versions of a surface protein, and they periodically rearrange their genes to express a different version (Cross, 1996). Gene rearrangement, in which part of the genome is excised, appears to be limited to special occasions when, for example, a large repertoire of related genes is created by combinatorial reassortment of gene pieces. Evidence for retention of a complete genome in differentiated cells is based in part on experiments in which somatic nuclei are transplanted into eggs or embryos and their ability to grow into animals is assessed. Some notable successesin growing complex animalshave shown that somatic nuclei can do a lot (Gurdon and Uehlinger, 1966). The difficulty of defining the differentiated character of any particular nucleus leaves the stability of the genome question incompletely resolved, but it is clear that adult cells exist that can be cloned into animals (Wakayama and Yanagimachi, 1999). Silent Genes Needham (1942, p. 101) noted that one of the most fundamental processes in development consists in the closing of doors, i.e. in determination, in the progressive restriction of the possible fates. In most cell types, only a fraction of genes is active at detectable levels. The mechanisms for keeping genes quiescent must be heritable, yet reversible in at least some cells such as the germline. Yeast MAT regulation has been a useful source of information about how genes are kept off, because the HML and HMR loci in fact contain complete genes that can become active under the mischievous control of a biologist. Normally the information at HML or HMR is expressed only after it is copied into the MAT locus. The HML and HMR genes are flanked by I and E sequences that silence the gene in cis. Proper repression requires histones, the DNA origin replication complex, acetylases and deacetylases, four Silent Information Repressor (SIR) proteins, and chromatin assembly complexes (Haber, 1998). By analogy to the visibly compact, relatively silent chromatin of other eukaryotes, the 3 kb of DNA that is silenced by this entourage of proteins is described as heterochromatin. This 3 kb is transcriptionally silent, resistant to endonuclease, and is associated with hypoacetylated histones. SIR proteins also affect mitotic chromatin structure and rDNA recombination, and are necessary for DNA ligation. These findings emphasize the links between chromatin structure and both transcription and DNA modification. In multicellular developmental contexts, gene silencing also depends on chromatin protein complexes. The complexes may activate or repress genes, and the regulation can be reversible. Genes of the Polycomb group (Pc-G) are required to repress Hox and other genes in Drosophila (Pirrotta, 1998). Related mammalian genes are implicated in Hox control, skeletal development, and hematopoiesis (Gould, 1997). Proteins encoded by Pc-G genes form complexes that maintain genes in repressed states. In the fly embryo, the initial repression of genes by transient spatial regulators is replaced by Pc-G maintenance systems. As more cells are produced, the Pc-G

chromatin complex constitutes a form of cellular memory. The still mysterious workings of the Pc-G genes have immense relevance to plans for modifying gene activities to promote healing, because stimulating regeneration processes will very likely require derepression of genes normally active only during embryonic development. Chromatin protein complexes that oppose repression, such as the Trithorax group (Trx-G) of proteins first found in Drosophila, may be useful in manipulating genes for reactivation and healing. One of the members of the Trx-G is brahma, which encodes an ATPase subunit of a large complex necessary for proper homeotic gene transcription. The Brahma complex is related to the Swi/Snf complex as well as to other activating chromatin complexes found in yeast. The HO locus, encoding the endonuclease required for matingtype switching, is among the genes activated by Swi/ Snf. A plethora of large chromatin protein complexes battles for control of genes; how they interact with sequence-specific DNA-binding proteins is a research area of great importance (Mannervik et al., 1999). An additional hurdle for gene activation, not present in Drosophila or C. elegans, is the clamping off of gene expression resulting from methylation of DNA, an interesting process that is also linked to cellular memory. The imprinting of genes, so that they are selectively accessible for activation depending on whether they are transmitted from the mother or father, is among the most dramatic kinds of regulation by methylation (Reik and Walter, 1998). Localized Intracellular mRNA in Differentiation The regulation of the HO nuclease during yeast cell type switching has been traced to a localized mRNA that codes for a regulator of HO. HO transcription is regulated by, among other things, a protein called Swi5p. Swi5p is found in mother cell nuclei, due to the regulation of SWI5 transcription by the Ash1p transcription factor, which may directly repress SWI5 in daughter cells. In this way the restriction of HO function to precise cells in the lineage has been reduced to understanding how Ash1p is restricted to daughter cells. ASH1 mRNA is localized to one region of cytoplasm in mother cells prior to cell division (Bobola et al., 1996; Sil and Herskowitz, 1996). The localization is controlled by actin and the myosin-like Myo4p. Presumably ASH1 RNA localization is one aspect of the many structural differences between mother cell and bud. Localized mRNAs are absolutely critical for the development of many organisms, most notably in the maternal dowry provided to oocytes (Bashirullah et al., 1998). Dramatic examples include bicoid RNA localized in head primordia to polarize the AP axis of the Drosophila egg, and nanos RNA, which is a posterior determinant. The encoded proteins are often transcription factors, RNAbinding proteins, or both. Some RNA molecules are moved to specific locations within cells, including those coding for many cytoskeletal proteins. The relevant control elements, usually in the 3 untranslated regions, work in multiple species, suggesting ancient origins for the elements. Asymmetric cell divisions of the sort that occur in yeast also occur frequently in neurogenesis, and again localized mRNAs or proteins such as prospero

Review 31

are prominent in controlling the different dowries of the two daughter cells (Hawkins and Garriga, 1998). Regulation of Batteries of Genes by Transcription Factors DNA rearrangement, though found in diverse organisms, is still viewed as the exception to having identical genomes in every cell. Transcription appears to be a much more prevalent level of regulation than gene rearrangement. The transcription factors encoded by MAT are well understood, but new information is providing a more complete picture of what they do (Johnson, 1998). MAT is active in cells and codes for two proteins, MAT 1p and MAT 2p. MAT 1p is a transcription factor that activates the transcription of -specific genes. MAT 2p, a repressor protein that contains a homeodomain, represses a-specific genes. In a cells, a-specific genes are activated by STE12p transcription factor, which is made by a and cells. The combined actions of Ste12p and Mcm1p, a protein that is made in all three cell types, turns on a-specific genes. a-specific genes are not active in cells despite the presence of Ste12p and Mcm1p, due to the MAT 2p repressor. Yeast MAT products work together with other proteins. For example, repression by MAT 2p in a cell requires Mcm1p and a repressor composed of Tup1p and Ssn6p, all proteins that are produced in a, , and a/ cells. The other possible resident at the MAT locus, MATa, is active in a and a/ cells and codes for MATa1p. In diploids MATa1p cooperates with MAT 2p and the Ssn6/Tup1 repressor to repress haploid-specific genes and MAT 1. The complex binds to a target sequence that is distinct from the sequence recognized by MAT 2p alone. Thus, Mcm1p is particularly interesting, since it acts as an activator in a cells and as a repressor in the other two cell types. When neither MAT allele is active, the default state is a, since the relevant target genes are constitutive without repressor around. In this situation -specific genes are silent due to the absence of MAT 1p. From the MAT studies, some generalities emerge (Johnson, 1998): transcription factors are fairly ineffective alone but powerful in complexes, the DNA acts as a nucleation site for assemblies of protein complexes, and proteins (like scientists) can have positive or negative effects depending on their collaborators. With the advent of recombinant DNA, specific genes could be assessed in differentiated cell types. The myogenic transcription factors are a notable case of coordinate gene regulation. In muscle cells transcripts for all the structural proteins appear with similar kinetics upon stimulation by activating genes such as myogenin and MyoD in susceptible cell types. We now recognize two main families of proteins that stimulate muscle development: the bHLH family that includes myogenin and MyoD, and the myocyte-enhancing factor (MEF) family (Yun and Wold, 1996). Members of the two families cooperate, in both vertebrates and invertebrates (Baylies et al., 1998), to promote muscle differentiation by activating muscle-specific genes. The regulation of muscle differentiation has parallels with regulation in the growth of the peripheral nervous system. As in yeast mating

type switching, promoters and enhancers of genes activated by myogenic proteins are assembly sites for protein complexes (Firulli and Olson, 1997). Once the right genes are activated, the cells can start to have fun, mating or differentiating. Wild Genes and Mating Habits The romance between haploid yeast cells of opposite mating-type begins with the differential gene expression described above. Yeast cells of opposite mating-type must find each other to combine their haploid genomes to form diploids. The process occurs through a system of signals. The a cells secrete a protein signal called a factor that binds to the Ste2p receptor on cells. When the signal is received, the receptor, a transmembrane protein, triggers a signal transduction process that culminates in changes in cell shape and formation of surface structures suitable for mating. Concomitantly, factor protein secreted by cells is received by the Ste3p receptor on a cells, and triggers the appropriate differentiation events for mating in those cells. The two signals and their two receptors are transcribed in the proper haploid cell types under the control of MAT alleles. In cells, MAT 1p, in combination with the non cell type specific transcription factor Mcm1p, activates the -specific genes factor and STE2, while MAT 2p (working with Mcm1p, Tup1p, and Ssn6p) represses the a-specific genes a factor and Ste3p. In a cells, the a-specific genes are constitutive and the -specific genes are silent because MAT 1 is repressed by MATa1p. Mating behavior in animals involves evolutionary flourishes ranging from the huge antlers of Irish elk to the elaborate nests of bowerbirds. Genes controlling sex differentiation also exhibit rapid evolution and a remarkable range of mechanisms. Intensive studies of sexual differentiation in worms and flies have provided exceptionally well-understood pathways (Cline and Meyer, 1996; Marin and Baker, 1998) and major progress has been made with mammals (Swain and Lovell-Badge, 1999). In flies and worms, the X chromosome to autosome ratio is measured with known counting elements, setting in motion a cascade of RNA splicing and transcription regulatory events that has been traced all the way to the production of differentiation products such as yolk protein. Although most of the molecules involved in controlling sexual differentiation and dosage compensation differ between worms, flies, and mammals, some of the genetic logic is similar. Critical elements for sex determination, in particular Sry, have been identified on the mouse and human Y chromosome, and now will come the filling in of the pathway between this most upstream regulatory element and sexual differentiation. Elaborate dosage compensation systems are necessary to provide the developing animal with the proper ratio of X chromosome products and autosome-derived products (Lyon, 1999). The mechanisms are quite different in mammals, flies, and worms, with one X chromosome inactivated in mammalian females, transcription heightened on one X chromosome in male flies, and lowered transcription from one X chromosome in worm hermaphrodites. Different chromatin regulators have been adapted, different ones by different organisms, in order to regulate sex-specific levels of gene expression.

Cell 32

Sex determination is often noted as an example of rapid evolution, and indeed the organization of heterotypic chromosomes and the regulators of sexual differentiation appear to change quickly (Marin and Baker, 1998). However, some regulators are conserved, and more parallels will probably come to light. The fly doublesex gene encodes a transcription factor that controls sex-specific gene products. Alternative splicing of dsx RNA causes production of one protein form in males and a different protein form in females. In worms the mab-3 gene, which controls sex determination, encodes a Dsx-related protein (Raymond et al., 1998). Animal Communication Systems The yeast signaling proteins have characteristics similar to signaling proteins involved in development of multicellular animals. Indeed, gonadotrophin releasing hormone (GnRH) is similar in structure to yeast factor (Loumaye et al., 1982). A fascinating area of developmental biology is the exploration of the origins of higher eukaryotic signals as revealed by tracing related molecules in a wide variety of organisms. The first evidence for chemical factors that could alter development was the production of galls in plants under the influence of chemicals from insect salivary gland extracts. In 18941895, C. Herbst set out to identify external stimuli that alter development, including calcium, light, gravity, salts, and temperature (Needham, 1942). Hints of chemical influence also came from exogastrulation caused by lithium treatment of amphibian embryos. Herbst found that one stimulus might have quite different responses if it acted on different tissues. Hormone signals drive both sexual differentiation and the equally astonishing process of metamorphosis, a process central to some of the most remarkable natural history of animals and genes. In Darwins words, Many insects, and especially certain crustaceans, show us what wonderful changes of structure can be effected during development. Such changes, however, reach their acme in the so-called alternate generations of some of the lower animals. It is, for instance, an astonishing fact that a delicate branching coralline, studded with polypi and attached to a submarine rock, should produce, first by budding and then by transverse division, a host of huge floating jelly-fishes; and that these should produce eggs, from which are hatched swimming animalcules, which attach themselves to rocks and become developed into branching corallines; and so on in an endless cycle. Metamorphic transformations are especially significant among human disease parasites. Advances in the developmental biology of these organisms will be significant in disease control as well as interesting in their own right. The most remarkable finding from decades of work on developmental signaling is that few new types of signals have been found. The routine approach to the development of any tissue is now to check for the involvement of FGF, TGF , Hedgehog, Wnt signals, a few others, and their transduction machinery. Not all these signals are involved in every process, but often several of them are relevant. The natural history of the Drosophila wingless (wg) gene, a Wnt family member, shows how signals are used

repeatedly. Early in Drosophila embryogenesis, wg is activated by pair-rule transcription factors in transverse stripes that are located near the center of each segment primordium. Maintenance of wg transcription requires a Hedgehog signal originating from stripes of cells just posterior to each wg stripe. wg function is necessary for formation of segmental divisions, for keeping cells alive, and for formation of the proper segmental pattern of denticles (bristles) on the surface of the cuticle. Wg protein carries a signal to the underlying visceral mesoderm cells to subdivide the mesoderm into repeating units. Wg is used again in the formation of appendages. The small clusters of cells set aside during embryogenesis to form legs and wings are determined in their fates, and spatially localized, by the intersection of transverse stripes of Wg with longitudinal stripes of the TGF molecule Dpp. Homeotic genes, which govern segment-specific appendage pattern, modulate the effect of the Wg/ Dpp combination. In another phase of its function, wg is required after early segmentation events for the heart precursors to develop. The name wingless derives from the original partial loss-of-function mutation that abolishes wing development. Wg actually has multiple roles in wing development, first subdividing the thoracic primordium into body wall and wing, then distinguishing dorsal and ventral wing surfaces, then stimulating production of distinctive bristles along the wing margin. Wg regulates cell division in the wing imaginal disc. Without wg function, half the flys brain does not develop. Wg is needed to pattern sensory organs on the body surface, eyes, and appendages. Proper dorsalventral (DV) patterning in the eye requires wg. The Wingless saga is typical of many important regulatory molecules; they are used repeatedly during development but with different outcomes. The specific actions of a regulator depend on what other regulators are acting in concert, and on the competence of cells to respond. Wg is one of the few proteins for which good evidence has been obtained for concentration-dependent effects on cell fate determination. Morphogens are defined as substances that direct different cell fates depending on their abundance, and although candidate morphogens have been endlessly proposed and discussed, it is in fact challenging to show that under normal developmental circumstances, different concentrations of a protein direct different cell fates (Neumann and Cohen, 1997). The difficulty is partly in making quantitative measurements in vivo, and partly in distinguishing the effects of multiple factors. Nonetheless, several regulators, including activin and hedgehog in vertebrates as well as bicoid, dorsal, decapentaplegic, and wingless in flies, seem to have morphogen properties under some circumstances. It is worth noting that most signal transduction systems are similar in animals separated by vast evolutionary distances. Similar does not mean identical. Distinctive components and distinctive regulatory relationships have been found. Increasingly, linear pathway diagrams are being replaced by networks of arrows representing a virtual ecosystem of interacting proteins. Often the social groupings have considerable complexity. We are in great need of better ways to monitor and represent the flow of information, changes in protein modifications and activities, and changes in subcellular localization

Review 33

that happen rapidly and concurrently as signals are processed. Competence At one moment a cell will respond to a signal, at the next it will not. At one time production of a transcription factor will send a cell into a differentiation pathway, at another time it will not. The difference is called competence, a word first used in the context of development to describe timing-dependent susceptibilities of tissues to inductive influences (Waddington, 1932). Needham (1942, p. 119) put competence into molecular terms, noting that an inductor may be a substrate, and that during the unavoidable reaction which will follow its entry into the competent cell, the physical conformation of the enzyme protein may itself be changed. An animal will respond only to signals it recognizes, and gene responses have similar restrictions: (1) Sensory abilities: Signals will be sensed only if receptors are available. Yeast a cells do not respond to the secreted a factor because they do not have the right receptor. (2) The need for specific social interactions: If a transcription factor requires a partner to work, the time when the partner is present determines competence. Most gene activation events during development require multiple transcription factors, as in yeast differentiation, so competence depends on the presence of all the necessary factors. Nuclear receptor proteins often act in combination, as do segmentation proteins. (3) Interfering signals: Some genes are inactivated, for example, by methylation, during differentiation. In some cases only part of a gene, such as a particular enhancer, need be blocked. A compact inaccessible chromatin state will make a cell incompetent for anything requiring one of the repressed genes. (4) Tribal conflict: Modifying agents, such as proteases or phosphatases that destroy or modify signals or other proteins, can make a cell incompetent to respond. (5) Environmental impacts: A cell may sense a regulator but be unable to carry out a differentiation program due to nutritional or other environmental deficits. A variation of this concept may account for temperature-dependent regulation of sex in reptiles. Limb development provides nice examples of the integration of multiple signals, all of them acting on competent cells but not others (Johnson and Tabin, 1997). In limbs and other tissues, far more progress has been made in identifying signals and regulators than in defining competence. For example, the application of FGF molecules to the flank of a chick embryo can induce formation of an extra leg (Cohn et al., 1995), but we do not understand why those cells can do it while cells in other places or at other times in development cannot. When we fully understand the answer, limb regeneration might become possible in humans. Evolutionary Conservation of Gene Function In addition to the similarity of yeast factor to GnRH, many yeast proteins are related to well-known vertebrate regulators, such as ras. In 1857, two years before The Origin of Species was published, Thomas Huxley wrote, The most important of all the generalizations of natural history, and, indeed, one of the most brilliant additions which the progress of modern science has

made to human knowledge, is the law that all animals and plants, however infinitely various their form may seem to be, are, in reality, constructed upon a very few plans; in other words, that the parts of animals and plants are associated and arranged according to certain fixed laws. This comment was about morphology, and the rational basis for taxonomy, but applies equally well to the gene systems that control morphology. Huxley would have enjoyed the conceptual simplification that comes from knowing the similar crystal structures of a homeodomain, repressor, and yeast MAT 2p. The remarkable similarity of the genetic regulation of development in distant organisms has heralded a new conception of evolution. It was a big surprise when evolutionary conservation of the Krebs cycle, the genetic code, and classes of structural proteins was extended to regulation of development. The diversity of organisms had fooled everyone into thinking that the evolution of completely different regulatory processes, or at least completely different uses of the same genes, was likely to be responsible for evolutionary change. Embryos of different vertebrate species are more similar to each other than are the corresponding adults, a useful hint of relatedness. Darwin complained in 1860: Embryology is to me by far the strongest single class of facts in favour of change of forms and not one, I think, of my reviewers has alluded to this (Oppenheimer, 1967, p. 222). Retention of working regulatory systems for animal development would seem to be much more likely than invention of dramatically new systems. We can tally up some of the outstanding cases of evolutionary conservation of gene function. First came the Hox genes, clustered genes that are differentially activated along the head to tail axis of the embryo (Lewis, 1978; McGinnis and Krumlauf, 1992). The combination of Hox genes active in a certain part of the animal governs pattern formation there. For example, the shape of vertebrae is altered in mice carrying Hox gene mutations, and Hox genes govern the type of appendages formed on insects. Hox genes are necessary for regional neuronal differentiation in the Drosophila brain and in the mammalian hindbrain (Reichert and Simeone, 1999). Repeating units formed by the action of segmentation genes (Figure 1, top) become different (Figure 1, bottom) due to Hox gene action. Hox gene action is not limited to distinguishing repeating body units. Complex overlapping patterns of Hox gene expression contribute to limb patterning and Hox genes are active in blood cells. Hox genes encode homeodomain transcription factors similar in structure to the yeast MAT 2 protein. Tinman homeodomain protein is another case of striking evolutionary conservation. Tinman is required for heart and visceral muscle development in flies; the related gene Nkx 2.5 is necessary for proper heart development in mice and humans (Evans, 1999). Heart development has a central astounding aspect: the need for circulation early in the embryogenesis of many species means that the heart must begin functioning when it is tiny and continue functioning as it grows. The phenomenon is akin to keeping a rowboats outboard motor running while changing it into an engine for a cruise ship. Another example is the Pax6 homeodomain protein that is required for eye development in humans, mice, and flies. In flies Pax6 works with the sine oculis and eyes

Cell 34

Figure 2. Example of Conserved Genes Affecting DorsalVentral Cell Fates in Insect and Mammalian Nervous Systems In situ hybridizations from serial sections are superimposed at the left and diagrammed at the right. Corresponding genes are shown in the same color. The order of expression of the genes is preserved from mice (left) to flies (right). Modified from Weiss et al. (1998).

Figure 1. Segmentation and Homeosis; the Same Two Houses at Different Times After segmentation produces repeating structural elements (top), action of Hox genes in specific segments gives them individual properties (bottom). Photographs by Robert S. Brantley (Brand, 1995).

absent genes, and at least some of their homologs are required for fish and mouse eye development (Gehring and Ikeo, 1999). Dorsal-ventral patterning in the early vertebrate and fly embryo requires localized expression and action of signaling proteins of the TGF class, such as Dpp in flies and BMP2 and 4 in vertebrates. The proteins direct polarization of the embryo and are restricted in their actions by antagonist proteins of the Sog/Chordin class working from the opposite pole. In flies Dpp acts in dorsal cells and Sog in ventral cells, whereas in vertebrates BMPs are ventral and Chordin dorsal. This and other evidence has suggested that the axes may have been flipped during evolution (Holley et al., 1995). Further support for this view comes from conserved expression patterns of three homeobox genes in early neural development (Chan and Jan, 1999). In flies the order of gene expression is ventral-vnd-ind-msh-dorsal, and in the vertebrate neural tube, the order for the genes most closely related in sequence is ventral-Nkx2.2(vnd)Gsh1,2(ind)-Msx1(msh)-dorsal (Figure 2). The apparent discrepancythe axis does not look flippedis resolved because the invagination of the neural plate to

form the neural tube does the flipping. Other types of regulation in neural development, such as the determination of neuronal subtypes by bHLH proteins, also seem similar in Drosophila and mammals (Chan and Jan, 1999). What exactly does the dedication of particular transcription factors to particular tissues or patterning events mean? Let us take Tinman as an example. Presumably, more than half a billion years ago, Tinman became irreversibly associated with one or more target genes involved in building some sort of pump, or perhaps even earlier to define a particular type of mesoderm. An ancestral Nk-class homeobox gene would activate some generally useful target genes in many cell types. If Tinman arose as a duplicate of another NK gene, it might have been free to change either its spatial regulation or its binding specificity for DNA or cofactor proteins. Since modern day Tinman can bind in vitro to sequences that other NK proteins bind, either the binding differences are subtle or they are irrelevant. So, let us suppose that the original distinguishing change in the tinman gene was due to the acquisition of a regulatory element (responsive to Hox genes or TGF inducers?) to make tinman active in only a subset of the mesoderm cells. This might turn up transcription of a Tinman target gene in that subset of cells, resulting in changed morphology, electrical conduction, or whatever. Once any useful change along these lines occurred, Tinmans association with the particular subset of cells would be selectable. Now other genes could fall under Tinmans influence in those cells simply by acquisition of the short DNA enhancer sequences necessary for Tinman regulation. Point mutations might do it, as would transposition of a piece of DNA containing Tinman target enhancer sequences. There would be strong selection for animals where Tinman had acquired targets that enhanced heart or visceral mesoderm function. Whatever the original course of events, the dedication of a transcription factor to a particular organogenesis event implies conservation of at least some target genes, which we could call the primeval target genes. They may be but a small subset of the current array of targets, and distinguishing them from subsequently acquired targets may unveil the process of evolutionary recruitment of genes for a task.

Review 35

Some signals also appear to have conserved their dedication to a particular developmental process, though this is the exception rather than the rule. The rule is that the same signal is used for many events that are not obviously linked in different organisms. The use of a particular signal, FGF, in controlling branching morphogenesis appears to have been conserved (Metzger and Krasnow, 1999). FGF signals are critical for branching of the trachea in flies and for branching of the lung buds in vertebrates. In both cases respiratory requirements interact with genetically programmed events to ensure proper access by all cells to oxygen. FGF actions must be driven by local low oxygen tension. FGF molecules are used for many purposes, but branching morphogenesis is one that is recognizably common to animals separated by hundreds of millions of years of evolution. By late in his life, Linnaeus fully appreciated the value of examining different stages of development as a guide to classification, since relationships may become evident with one criterion but not another. Similarly, a gene discovered for one of its roles may in fact have originally been dedicated to another, more evolutionarily conserved, purpose. Linnaeus was stern in his view of sloppy taxonomists who did not pay attention to multiple criteria: Paying too little regard to Nature, they disunited natural genera, on account of the most trifling distinctions. This made their continuance in the science of very short duration; our business here is not to suppose but to examine what nature will allow of, and what she will not. Knowledge of this sort, built on opinion only, will not stand. We are therefore to look into the science with great accuracy; and the larva of the insect, its manner of changing, and other things of moment, are to be known, before we presume to form a new genus, as men of experience will readily admit. Daily experience in botany teaches us that none are more apt to form new genera, than those least qualified for it. A good example of the importance of paying sufficient regard to the natural history of genes comes from the Toll transmembrane protein. Toll was identified as a key component of the regulatory system that polarizes the Drosophila oocyte along the DV axis. Proteins related to Toll have now been identified in many organisms including mammals. DV patterning is not the evolutionarily conserved function. By examining adult as well as larval gene functions, observing the genes at multiple stages of their life cycle, Toll-related proteins were found to share a role in immunity to bacterial and fungal infection. That role is common to humans, flies, and tobacco (Meister et al., 1997). Toll is accompanied in its conserved role by the rest of its ecosystem. For example, NF B proteins are transcription factors whose entry into the nucleus is regulated by extracellular signals, coordinating the inflammation response to infection. Dorsal, a protein related to NF B, transduces the Toll signal and is also linked to the flys immune response (Lemaitre et al., 1996). The genome projects will complete documentation of a pattern that is now familiar: many genes present as single copies in worms or flies are present as families in vertebrates (Miklos and Rubin, 1996). It is difficult to assess whether proliferation of genes into families was essential for evolving the parts of vertebrates that are

more complex than parts of insects. Some of the increased complexity is illusory; simple organisms often are not. What is the most important change: gene regulation, with members of families expressed in different patterns, or protein diversification within a family? Few tests have been done to rigorously assess the equivalence of different members of a gene family. The coding region of the mouse engrailed1 gene has been replaced by that of engrailed2, with few detrimental effects (Hanks et al., 1995). On the other hand, fetal and adult hemoglobin genes are expressed at different times for a clear reason: the proteins provide appropriate and distinct oxygen affinities. In cases where the proteins really are equivalent, one can imagine a different history in which a single coding region gradually acquired a plethora of cis regulatory sequences, and this seems to have happened in numerous cases. One can therefore view gene families as essential for protein diversification but not necessarily for regulatory diversification. Families could be essential for regulatory diversification if incompatible enhancers and silencers must control the same proteincoding sequence. It seems possible that the existence of gene families is to some extent a smokescreen and that most of the important changes during evolution happened with gene families present but do not require that there be gene families. Much of the current emphasis on similarities between vertebrate and invertebrate regulators of development is driven by amazement; much less similarity was expected. There are tremendous differences as well, and exploring them and their origins will unveil a more complete picture of how all Linnaeus animals came to exist. Quite a few rapidly evolving genes are known. For example, the bicoid protein is required as a concentrationdependent regulator, a morphogen, of anteriorposterior polarity in fly embryos, but this function may have arisen only during the evolution of Diptera. We need to understand rapid evolution as well as conservation. Origins of Developmental Regulators A billion years ago or so, differentiation was accomplished using components that worked so well that the same system is still used. It is striking that homeodomain proteins, like MAT 2p, are used to control cell differentiation in yeast just as they are in humans. The origins of many other regulators is far less clear. Hedgehog proteins are like other classes of signaling proteins in that no sign of them can be found in bacteria or yeast, and it is not clear where they first arose. However, yeast cells have a gene related to Niemann-Pick C1 (NPC1), a human disease syndrome gene essential for proper transport of cholesterol and other lipids in mammalian cells. The amino acid sequence of the Hedgehog receptor, Patched, is closely related to NPC1. During Hedgehog synthesis cholesterol is covalently joined to the signaling portion of the Hedgehog protein. Also, certain cholesterol analogs are powerful inhibitors of the reception of the Hedgehog signal. These and other data point at the possibility that some parts of the Hedgehog signaling pathway are derived from proteins needed for intracellular lipid trafficking (Beachy et al., 1997). The recruitment of Toll to DV patterning may have been a

Cell 36

relatively recent theft (and duplication) of an immunological function. Other genes have probably been coopted from basic metabolic processes. One of the most interesting questions is whether there is any logic to which types of regulators are used in which processes. Why is a zinc finger protein used here and a homeodomain protein there? Is the present scenario a historical accident, or does it reflect features of the proteins that make them especially suitable for certain tasks? The origins of the machinery that regulates development remain mostly unknown. Death As young biologists who assisted the explosive advances in developmental biology have become middleaged biologists, the developmental biology community has become fascinated with death. This at first took the mild form of an interest in cell death (Horvitz, 1999), which was comfortably separate from real death. Cell death is an extremely important mechanism during development, for example in shaping digits or allowing only functional neurons or the right lymphocytes to survive, and in disease, for example by regulating (or failing to regulate) inappropriate growth. Now the trend is a greater interest in avoiding death altogether; the field of aging research is flourishing as never before. It is fascinating that both cell death and aging appear to be programmed genetically just as the construction of the organism is programmed (Kenyon, 1996; Defossez et al., 1998; Guarente et al., 1998; Lin et al., 1998). Individual cells age too. The earliest known event associated with the aging of yeast cells is genetic instability in the cluster of ribosomal DNA genes (Defossez et al., 1998). Nontransformed eukaryotic cells in culture also have a lifetime timer running, quite possibly linked to telomerase function and chromosome shortening. Reactivation of telomerase allows cells to continue dividing in culture beyond the time they would otherwise enter crisis and die. In accordance with this view, most human tumor cells contain active telomerase, whereas most somatic cells do not (Hodes, 1999). The rapid advances in aging research demand frequent review; I will review themagain in 2050 and 2100. Developmental Biology, in Sickness and in Health Genes discovered for their roles in development are also critical in human disease. Linnaeus was interested in nosology, the classification of diseases. Continuing discoveries of the roles of developmental genes in disease will allow more rigorous and unambiguous diagnosis. Excitement has come from finding genes that guide cell differentiation and pattern formation and that are linked to human disease. For example, mutations in -catenin contribute to human colon and other cancers (Morin et al., 1997). Mutations in human Sonic hedgehog result in holoprosencephaly; mutations in the hedgehog receptor Patched (Figure 3) lead to polydactyly and spina bifida in Gorlins syndrome and to basal cell carcinoma, medulloblastoma, and rhabdomyosarcoma (Goodrich and Scott, 1998). Mutations in the tinman-related human gene Nkx2.5 cause heart disease (Schott et al., 1998). Cell cycle regulators such as APC and Rb are critical in human cancer and are now viewed in the context of their

Figure 3. High Precision in Spatial Gene Regulation and Signaling Spatial regulation of a Hedgehog target gene. Mouse embryo with patched1 expression shown. Hedgehog signals induce expression of patched1 and other target genes in precisely controlled spatial and temporal patterns, as is shown here using a lacZ gene inserted into the patched1 locus. The regulatory relationships in Hedgehog signaling are largely conserved in all animals where they have been tested. From Ljiljana Milenkovic.

normal developmental roles, APC in the Wnt signaling pathway and Rb in eye development. The presenilin proteases that are crucial in familial Alzheimers disease have been linked to roles in processing Notch signaling proteins that are active in many developmental events (Levitan and Greenwald, 1998). An insulin receptor-like gene, daf-2 of C. elegans, was discovered due to its role in inhibiting dauer formation (a developmental form of the worm adapted for diapause) but has also been implicated in affecting longevity (Kimura et al., 1997; Tissenbaum and Ruvkun, 1998). In worms insulin signaling may be required to steer metabolism away from fat formation and is also involved in embryonic development. Dauer formation also requires a worm homolog of the PTEN gene, which is a tumor suppressor gene in humans (Rouault et al., 1999). Genes required to determine neuronal cell type in nematodes are closely related to genes involved in human polycystic kidney disease; the worms genetic pathway including these genes provides ideas about what other proteins are relevant (Emmons and Somlo, 1999). A gene related to the Drosophila epithelial polarity gene crumbs is the cause of a form of retinitis that affects 1.5 million people (den Hollander et al., 1999). In each of these cases, the excitement comes from connecting a protein involved in human disease to a whole pathway of interacting components. The elucidation of the ras pathway, so important in human cancer and development, has benefited dramatically from its connections to developmental events in flies and worms (Kayne and Sternberg, 1995; Wassarman et al., 1995). Developmental biology has great potential for impact on the clinic. Already, treatments involving skin grafts and growth hormones have helped many people. Isolation of stem cells for a variety of tissues and better

Review 37

manipulations of stem cells based on newfound knowledge of growth and differentiation controls will surely make regeneration more useful for healing and tissue replacement (see reviews by Fuchs and Segre, 2000 and Weissman, 2000 [both in this issue of Cell]). We see progress in identifying signals for producing neurons that are lost in Parkinsons disease, in inducing growth of hair, in following pancreas differentiation to learn about diabetes, in stimulating growth of lung buds in culture, and in growing arteries in culture. The biggest obstacle is likely to be finding out how to create cells with the ability to respond rather than to find the best cocktail of signals. The developmental biology of parasites also holds enormous promise for new approaches to dread diseases (Teixiera, 1998). Trypanosomes, leishmania, and malaria parasites together account for hundreds of millions of infected people. The lifecycles of parasites are fascinating for their evasions of the immune response and for the adaptability of many parasites to multiple host organisms, as well as for unusual molecular features such as the polycistronic transcription and intronless genome of trypanosomes. What Can We Build? One measure of progress in developmental biology will be whether we can direct the growth of a useful tissue. Here is a test of whether we really know how things work: take cells growing in culture and manipulate them to make either of two alternative organs by manipulating signals and gene activities. Organ development means not just activating differentiation markers but making a fully functional, well-shaped tissue that works. Make the epithelium fold where a fold would be useful; cause an organizing center to appear in a certain spot; make some cells delaminate and form another layer; make differentiated cell types appear; make some cells migrate to a useful new position; direct the formation of a left-right symmetry and then asymmetry; and make a colorful pattern! If we could direct the same cell population into either of two completely different paths, we must have learned something. How can such knowledge be useful? The shaping of tissues during healing and regeneration is mostly done surgically now, and major limitations leave people in therapy or with reduced opportunities for decades. Facilitating a patients natural healing processes to promote more complete healing, more rapid healing, or better organized healing would help tremendous numbers of people. Altering properties of cells to make them more resistant to infection or more aggressive in fighting infection will also become possible. Stimulating natural growth processes to repair nerve damage, to grow new teeth, to replace blood without a marrow transplant, or even to regenerate limbs would be of clear benefit. What are the risks? One is cancer. By stimulating growth processes necessary for repair at a time in life when (perhaps) normal restraints are no longer operating, cancer might be a higher risk. A second concern is misuse of the knowledge. The science fiction literature is replete with examples of body modifications chosen to be repugnant. Many people find blood donation or cosmetic repair after injury acceptable, whereas the

limits to child design or selection, which many people find offensive, will almost certainly become a considerable social problem. The present controversies over modification of tomato ripening properties, bovine growth hormone produced in bacteria, and pesticides genetically introduced into plants are forerunners of debates about the proper limits of intervention in medicine. The current tremendous inequities in access to medical care may or may not be improved by new types of developmental medicine. What New Tools Are Needed? Our present picture of gene activation during development is largely dependent on static views of embryos. Major limitations include the following: (1) Detection of markers like fluorescent proteins is slow because the time needed for the protein to fold delays detection for significant periods of development. (2) At present we cannot watch more than four proteins, or gene expression patterns, at a time. In the seething cytoplasm, thousands of changes occur, while we watch a few events at a time. We need new ways to see, and new representations of what we see. (3) It is difficult to quantitate the concentration of any of the relevant proteins as a function of time, let alone their activity states. (4) Modifications of proteins, such as phosphorylation or glycosylation, are hard to monitor or manipulate. (5) The chromatin is preset to allow a gene to respond or not, and this cannot readily be viewed, particularly not at the single cell level. With in situ hybridization to RNA in tissues, and detection of proteins with antibody stains, we can discriminate among previously indistinguishable cells. This has been fantastically revealing about cell determination, as it gives us one view of what the cell is thinking, long before morphological changes occur. This has led to a few markers being used to indicate ventral-ness or liverish-ness. While this may be safe, cells might activate a few indicator molecules without turning on the full set characteristic of a tissue type or stage of development. The more complete views of gene activation that have arrived with microarrays will help to clarify whether this is a real problem. Until recently gene expression during development has been viewed by reconstructing events from different individual animals. Progress here will come from noninvasive imaging techniques such as MRI applied to animals (Jacobs et al., 1999) and watching gene expression using proteins tagged with fluorescent markers, such as green fluorescent protein (GFP) (Chalfie et al., 1994). GFP has already been extremely useful in watching the movements of proteins and cells in living embryos. Fluorescence resonance energy transfer (FRET) is a technique that detects interactions between proteins based on the proximity of an emitting fluorescent molecule and a sensing fluorescent molecule that emits at a new wavelength if sufficiently stimulated by the first emitter (Periasamy and Day, 1999). GFP together with FRET, and refinements and variations to come, will allow the associations of proteins to be monitored in living embryos. Our view of development, and what matters in development, might very well be differently skewed if we

Cell 38

could observe activation of kinase cascades in space and time rather than gene transcription and protein accumulation. The ability to detect only activated forms of a kinase with specific antibodies (e.g., Gabay et al., 1997) is valuable, but does not allow imaging fast enough to keep up with the rapid dynamics of living cell metabolism. There is a dearth of small molecules in our understanding of development. We have steroids in hormonal coordination of metamorphosis and growth, cAMP in many events, nitric oxide in vascularization, and a few others, yet it seems likely that small molecules are as ubiquitous and crucial in development as they are in regulation of bacterial metabolism. How could we understand Trp repressor without knowing about tryptophan? Where Did We Come From? How can developmental biology help us look into the past and understand human origins? Our own place in the animal kingdom was noted in a letter written by Linnaeus to a friend in 1747: You disapprove my having located Man among the Anthropomorphi. But man knows himself . . . I demand of you, and of the whole world, that you show me a generic characterone that is according to generally accepted principles of classificationby which to distinguish between Man and Ape. I myself most assuredly know of none. I wish somebody would indicate one to me. But, if I had called man an ape, or vice versa, I should have fallen under the ban of all the ecclesiastics. It may be that as a naturalist I should have done so (Greene, 1909). Now we know ourselves somewhat better and realize that to know ourselves better still we must know all life better. At age 25 Linnaeus took an adventurous 3000 mile journey through Lapland that had an influence on his science reminiscent of the effect of the Beagle voyage on young Darwin. Linnaeus observed the natural world intensely despite storms, whitewater boat accidents, and near starvation. Along the way he encountered frog development while helping a woman who believed she had three frogs in her stomach (she heard them croaking) as a result of drinking water containing frog spawn. Even without such medical interludes, we can learn as Linnaeus did, by searching for natural surprises. Our few model organisms provide an incomplete view. We must first preserve what remains of the natural world around us and then, like Linnaeus and Darwin, walk through and watch developments and development with open eyes and minds. Developmental biology is a celebration of natural beauty combined with the elegance of molecular control systems. The more we know, the more astounding the renewal of life becomes. Now the genome projects have stalked a host of completely untamed genes, never sighted before. As with animals, wild genes are even more fun than domesticated ones. To appreciate and explore the new genes in their full natural splendor, fresh from the jungle, is the delightful task facing developmental biologists.
Acknowledgments With a subject of this scope it is impossible to credit all the people who have made major contributions to developmental biology. I

thank Roel Nusse and the anonymous reviewers for helpful comments on the manuscript, Drs. Tim Galitski and Gerry Fink for helping with microarray analysis, and Drs. Ljiljana Milenkovic and Joseph Weiss for figures. I am particularly grateful to Allan Spradling and Mary Lou Pardue for inspiration, guidance, and friendship, and I thank my students and colleagues for all the excitement they have brought. The Howard Hughes Medical Institute and grants from the NIH, DARPA, and the Parseghian Foundation support research in my laboratory. References Ashburner, M. (1990). Puffs, genes, and hormones revisited. Cell 61, 13. Bashirullah, A., Cooperstock, R.L., and Lipshitz, H.D. (1998). RNA localization in development. Annu. Rev. Biochem. 67, 335394. Baylies, M.K., Bate, M., and Ruiz Gomez, M. (1998). Myogenesis: a view from Drosophila. Cell 93, 921927. Beachy, P.A., Cooper, M.K., Young, K.E., von Kessler, D.P., Park, W.J., Hall, T.M., Leahy, D.J., and Porter, J.A. (1997). Multiple roles of cholesterol in hedgehog protein biogenesis and signaling. Cold Spring Harb. Symp. Quant. Biol. 62, 191204. Beermann, W. (1952). Chromomeren Konstanz und Spezifische Modifikation der Chromosomenstruktur in der Entwicklung und Organ Differenzierung von Chironomus tentans. Chromosoma 5, 139198. Bobola, N., Jansen, R.P., Shin, T.H., and Nasmyth, K. (1996). Asymmetric accumulation of Ash1p in postanaphase nuclei depends on a myosin and restricts yeast matingtype switching to mother cells. Cell 84, 699709. Brand, S. (1995). How Buildings Learn: What Happens After Theyre Built (New York: Penguin Books). Calvi, B.R., and Spradling, A.C. (1999). Chorion gene amplification in Drosophila: a model for metazoan origins of DNA replication and S-phase control. Methods 18, 407417. Cedar, H., and Bergman, Y. (1999). Developmental regulation of immune system gene rearrangement. Curr. Opin. Immunol. 11, 6469. Chalfie, M., Tu, Y., Euskirchen, G., Ward, W.W., and Prasher, D.C. (1994). Green fluorescent protein as a marker for gene expression. Science 263, 802805. Chan, Y.-M., and Jan, Y.N. (1999). Conservation of neurogenic genes and mechanisms. Curr. Opin. Neurobiol. 9, 582588. Chu, S., DeRisi, J., Eisen, M., Mulholland, J., Botstein, D., Brown, P.O., and Herskowitz, I. (1998). The transcriptional program of sporulation in budding yeast. Science 282, 699705. Erratum: Science 282 (5393), 1998. Cline, T.W., and Meyer, B.J. (1996). Vive la difference: males vs females in flies vs worms. Annu. Rev. Genet. 30, 637702. Cohn, M.J., Izpisua-Belmonte, J.C., Abud, H., Heath, J.K., and Tickle, C. (1995). Fibroblast growth factors induce additional limb development from the flank of chick embryos. Cell 80, 739746. Cross, G.A. (1996). Antigenic variation in trypanosomes: secrets surface slowly. Bioessays 18, 283291. Darwin, C. (1871). The Descent of Man and Selection in Relation to Sex (New York: Hurst and Co.). Defossez, P.A., Park, P.U., and Guarente, L. (1998). Vicious circles: a mechanism for yeast aging. Curr. Opin. Microbiol. 1, 707711. den Hollander, A.I., ten Brink, J.B., de Kok, Y.J., van Soest, S., van den Born, L.I., van Driel, M.A., van de Pol, D.J., Payne, A.M., Bhattacharya, S.S., Kellner, U., et al. (1999). Mutations in a human homologue of Drosophila crumbs cause retinitis pigmentosa (RP12). Nat. Genet. 23, 217221. Emmons, S.W., and Somlo, S. (1999). Signal transduction. Mating, channels and kidney cysts. Nature 401, 339340. Evans, S.M. (1999). Vertebrate tinman homologues and cardiac differentiation. Semin. Cell. Dev. Biol. 10, 7383. Firulli, A.B., and Olson, E.N. (1997). Modular regulation of muscle gene transcription: a mechanism for muscle cell diversity. Trends Genet. 13, 364369.

Review 39

Fraser, S.E., and Harland, R.M. (2000). The molecular metamorphosis of experimental embryology. Cell 100, this issue, 4155. Fuchs, E., and Segre, J.A. (2000). Stem cells: a new lease on life. Cell 100, this issue, 143155. Gabay, L., Seger, R., and Shilo, B.Z. (1997). In situ activation pattern of Drosophila EGF receptor pathway during development. Science 277, 11031106. Galau, G.A., Britten, R.J., and Davidson E.H. (1974). A measurement of the sequence complexity of polysomal messenger RNA in sea urchin embryos. Cell 2, 920. Galitski, T., Saldanha, A.J., Styles, C.A., Lander, E.S., and Fink, G.R. (1999). Ploidy regulation of gene expression [see comments]. Science 285, 251254. Gehring, W.J., and Ikeo, K. (1999). Pax 6: mastering eye morphogenesis and eye evolution. Trends Genet. 15, 371377. Goerke, H. (1973). Linnaeus (Charles Scribners Sons). Goodrich, L.V., and Scott, M.P. (1998). Hedgehog and patched in neural development and disease. Neuron 21, 12431257. Gould, A. (1997). Functions of mammalian Polycomb group and trithorax group related genes. Curr. Opin. Genet. Dev. 7, 488494. Greene, E.L. (1909). Linnaeus as an evolutionist. Proc. Washington Acad. Sciences 11, 1726. Guarente, L., Ruvkun, G., and Amasino R. (1998). Aging, life span, and senescence. Proc. Natl. Acad. Sci. USA 95, 1103411036. Gurdon, J.B., and Uehlinger, V. (1966). Fertile intestine nuclei. Nature 210, 12401241. Haber, J.E. (1998). Mating-type gene switching in Saccharomyces cerevisiae. Annu. Rev. Genet. 32, 561599. Hake, S., and Meyerowitz, E.M. (1998). Growth and development: growing up green. Curr. Opin. Plant Biol. 1, 911. Hanks, M., Wurst, W., Anson-Cartwright, L., Auerbach, A.B., and Joyner, A.L. (1995). Rescue of the En-1 mutant phenotype by replacement of En-1 with En-2. Science 269, 679682. Hawkins, N., and Garriga, G. (1998). Asymmetric cell division: from A to Z. Genes Dev. 12, 36253638. Hendrix, R.W., Roberts, J.W., Stahl, F.W., and Weisberg R.A. (1983). Lambda II (Cold Spring Harbor, NY: Cold Spring Harbor Laboratory Press). Hodes, R.J. (1999). Telomere length, aging, and somatic cell turnover. J. Exp. Med. 190, 153156. Holley, S.A., Jackson, P.D., Sasai, Y., Lu, B., De Robertis, E.M., Hoffmann, F.M., and Ferguson, E.L. (1995). A conserved system for dorsalventral patterning in insects and vertebrates involving sog and chordin. Nature 376, 249253. Horder, T.J., Witkowski, J.A., and Wylie C.C. (1985). A History of Embryology (Cambridge: Cambridge University Press). Horvitz, H.R. (1999). Genetic control of programmed cell death in the nematode Caenorhabditis elegans. Cancer Res. 59, 1701s-1706s. Hough, B.R., Smith, M.J., Britten, R.J., and Davidson E.H. (1975). Sequence complexity of heterogeneous nuclear RNA in sea urchin embryos. Cell 5, 291299. Iyer, V.R., Eisen, M.B., Ross, D.T., Schuler, G., Moore, T., Lee, J.C.F., Trent, J.M., Staudt, L.M., Hudson, J., Jr., et al. (1999). The transcriptional program in the response of human fibroblasts to serum. Science 283, 8387. Jacobs, R.E., Ahrens, E.T., Meade, T.J., and Fraser, S.E. (1999). Looking deeper into vertebrate development. Trends Cell. Biol. 9, 7376. Jeon, M., Gardner, H.F., Miller, E.A., Deshler, J., and Rougvie, A.E. (1999). Similarity of the C. elegans developmental timing protein LIN-42 to circadian rhythm proteins. Science 286, 11411146. Johnson, A.D. (1998). Molecular mechanisms of cell-type determination in yeast. Curr. Opin. Genet. Dev. 5, 552558. Johnson, R.L., and Tabin, C.J. (1997). Molecular models for vertebrate limb development. Cell 90, 979990. Kayne, P.S., and Sternberg, P.W. (1995). Ras pathways in Caenorhabditis elegans. Curr. Opin. Genet. Dev. 5, 3843. Kenyon, C. (1996). Ponce delegans: genetic quest for the fountain of youth. Cell 84, 501504.

Kimura, K.D., Tissenbaum, H.A., Liu, Y., and Ruvkun G. (1997). daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans. Science 277, 942946. Lemaitre, B., Nicolas, E., Michaut, L., Reichhart, J.M., and Hoffmann, J.A. (1996). The dorsoventral regulatory gene cassette spatzle/Toll/ cactus controls the potent antifungal response in Drosophila adults. Cell 86, 973983. Levitan, D., and Greenwald, I. (1998). Effects of SEL-12 presenilin on LIN-12 localization and function in Caenorhabditis elegans. Development 125, 35993606. Lewis, E.B. (1978). A gene complex controlling segmentation in Drosophila. Nature 276, 565570. Lin, Y.J., Seroude, L., and Benzer, S. (1998). Extended life-span and stress resistance in the Drosophila mutant methuselah [see comments]. Science 282, 943946. Linnaeus, C. (1735). Systema Naturae. Reprinted in 1964 by Nieuwkoop and B. De Graaf, Holland. Loumaye, E., Thorner, J., and Catt, K.J. (1982). Yeast mating pheromone activates mammalian gonadotrophs: evolutionary conservation of a reproductive hormone? Science 218, 13231325. Lyon, M.F. (1999). X-chromosome inactivation. Curr. Biol. 9, R235 R237. Mannervik, M., Nibu, Y., Zhang, H., and Levine M. (1999). Transcriptional coregulators in development. Science 284, 606609. Marin, I., and Baker, B.S. (1998). The evolutionary dynamics of sex determination. Science 281, 19901994. McGinnis, W., and Krumlauf, R. (1992). Homeobox genes and axial patterning. Cell 68, 283302. Meister, M., Lemaitre, B., and Hoffmann, J.A. (1997). Antimicrobial peptide defense in Drosophila. Bioessays 19, 10191026. Metzger, R.J., and Krasnow, M.A. (1999). Genetic control of branching morphogenesis. Science 284, 16351639. Miklos, G.L., and Rubin, G.M. (1996). The role of the genome project in determining gene function: insights from model organisms. Cell 86, 521529. Morin, P.J., Sparks, A.B., Korinek, V., Barker, N., Clevers, H., Vogelstein, B., and Kinzler, K.W. (1997). Activation of beta-cateninTcf signaling in colon cancer by mutations in beta-catenin or APC. Science 275, 17871790. Needham, J. (1942). Biochemistry and Morphogenesis (Cambridge: Cambridge University Press). Neumann, C., and Cohen, S. (1997). Morphogens and pattern formation. Bioessays 19, 721729. Nurse, P. (2000). The long twentieth century of the cell cycle and beyond. Cell 100, this issue, 7178. Nusslein-Volhard, C., and Wieschaus, E. (1980). Mutations affecting segment number and polarity in Drosophila. Nature 287, 795801. Oettinger, M.A. (1999). V(D)J recombination: on the cutting edge. Curr. Opin. Cell. Biol. 11, 325329. Oppenheimer, J.M. (1967). Essays in the History of Embryology and Biology (Cambridge, MA: MIT Press). Periasamy, A., and Day, R.N. (1999). Visualizing protein interactions in living cells using digitized GFP imaging and FRET microscopy. Methods Cell Biol. 58, 293314. Pirrotta, V. (1998). Polycombing the genome: PcG, trxG, and chromatin silencing. Cell 93, 333336. Pourquie, O. (1998). Clocks regulating developmental processes. Curr. Opin. Neurobiol. 8, 665670. Raymond, C.S., Shamu, C.E., Shen, M.M., Seifert, K.J., Hirsch, B., Hodgkin, J., and Zarkower, D. (1998). Evidence for evolutionary conservation of sex-determining genes. Nature 391, 691695. Reichert, H., and Simeone, A. (1999). Conserved usage of gap and homeotic genes in patterning the CNS. Curr. Opin. Neurobiol. 9, 589595. Reik, W., and Walter, J. (1998). Imprinting mechanisms in mammals. Curr. Opin. Genet. Dev. 8, 154164. Rivera-Pomar, R., and Jackle, H. (1996). From gradients to stripes in Drosophila embryogenesis: filling in the gaps. Trends Genet. 12, 478483.

Cell 40

Rouault, J.P., Kuwabara, P.E., Sinilnikova, O.M., Duret, L., ThierryMieg, D., and Billaud, M. (1999). Regulation of dauer larva development in Caenorhabditis elegans by daf-18, a homologue of the tumour suppressor PTEN. Curr. Biol. 9, 329332. Sander, K. (1985). The role of genes in ontogenesis-evolving concepts from 18831983 as perceived by an insect embryologist. In A History of Embryology, T.J. Horder, J.A. Witkowski, and C.C. Wylie, eds. (Cambridge: Cambridge University Press), pp. 363395. Schott, J.J., Benson, D.W., Basson, C.T., Pease, W., Silberbach, G.M., Moak, J.P., Maron, B.J., Seidman, C.E., and Seidman, J.G. (1998). Congenital heart disease caused by mutations in the transcription factor NKX25. Science 281, 108111. Sil, A., and Herskowitz, I. (1996). Identification of asymmetrically localized determinant, Ash1p, required for lineage-specific transcription of the yeast HO gene. Cell 84, 711722. Stavnezer, J. (1996). Immunoglobulin class switching. Curr. Opin. Immunol. 8, 199205. Swain, A., and Lovell-Badge, R. (1999). Mammalian sex determination: a molecular drama. Genes Dev. 13, 755767. Teixiera, S.M.R. (1998). Control of gene expression in trypanosomatidae. J. Med. Biol. Res. (Brazil) 31, 15031516. Tissenbaum, H.A., and Ruvkun, G. (1998). An insulin-like signaling pathway affects both longevity and reproduction in Caenorhabditis elegans. Genetics 148, 703717. Waddington, C.R. (1932). III. Experiments on the development of chick and duck embryos, cultivated in vitro. Philos. Transactions Royal Soc. B. 221, 179230. Wakayama, T., and Yanagimachi, R. (1999). Cloning the laboratory mouse. Semin. Cell. Dev. Biol. 10, 253258. Wassarman, D.A., Therrien, M., and Rubin, G.M. (1995). The Ras signaling pathway in Drosophila. Curr. Opin. Genet. Dev. 5, 4450. Weissman, I.L. (2000). Stem cells: units of development, units of regeneration, and units of evolution. Cell 100, this issue, 157168. Weiss, J.B., von Ohlen, T., Mellerick, D., Dressler, G., Doe, C.Q., and Scott, M.P. (1998). Dorsal-ventral patterning in the Drosophila central nervous system: the intermediate neuroblast defective homeobox gene specifies intermediate column identity. Genes Dev. 12, 35913602. Yun, K., and Wold, B. (1996). Skeletal muscle determination and differentiation: story of a core regulatory network and its context. Curr. Opin. Cell. Biol. 8, 877889.