Somite number and vertebrate evolution

The vertebrate body plan is made up of repeating and non-repeating patterns of organ primordia (Cooke, 1975). Repeating patterns include somites, feathers, pharyngeal arches, rhombomeres and, less obviously, the digits. At least two mechanisms are involved in generating repeating patterns: a segmentation mechanism and, superimposed on this, a positional field for making the segments or units different (Ingham and Martinez Arias, 1992). Raff has pointed out that it is relatively easy for evolution to vary the total number of segments in some repeating series (Raff, 1996). A striking example of this is the variation in the number of body segments in vertebrates (Fig. 1).

Vertebrae develop from the sclerotomes of the somites, which themselves segregate from mesoderm laid down during gastrulation and tail development (Verbout, 1985; Keynes and Stern, 1988; Christ and Wilting, 1992). Our aim in this article is to assess how evolutionary changes in somite number relate to modern concepts of evolution and development such as the developmental hourglass, the phylotypic stage and the zootype. Developmental mechanisms controlling somite number are discussed. We argue that uncoupling or dissociation between repeating patterns and positional fields, as exemplified by somite development, is an important feature of development and evolution in a number of systems.

Only vertebrates and other chordates have a tailbud. After gastrulation, the vertebrate tailbud can continue to make somites (Gont et al., 1993; Tucker and Slack, 1995). Vertebrate segmentation is therefore, in principle, an open-ended system. This may help explain how total somite number can vary widely between vertebrate species. Somitogenesis in vertebrates is analogous to the process of segment formation in many annelids and arthropods, including ‘short germ’ insects, where segments are added from a growth zone that lies in front of the anal segment or telson (Anderson, 1973; Nagy and Carroll, 1994).

In some of these lineages, the number of segments is indefinite and varies between individuals. In many other groups of invertebrates, the final number of segments in the adult body is invariant. Some precise, but as yet unknown, mechanism must terminate segment formation. For example, in leeches, an indefinite number of segment precursor cells is budded from the teloblasts, but only 32 segments differentiate; supernumerary cells die or fuse with the yolk syncytium (Weisblat et al., 1988).

As may be the case in the backbone of vertebrates, segment differentiation in centipedes is more consistent with position along the body than with segment number (Minelli, 1992). Segmentation is best understood in Drosophila, but this model system provides little insight into the general problem of segment numbers. Drosophila and other ‘long-germ’ insects have no posterior growth zone. Segments are generated almost simultaneously by the subdivision of an existing field of cells, using a cascade of patterning interactions that are triggered by maternal gradients (Pankratz and Jackle, 1993). Growth and segmentation are uncoupled. This mechanism – which is presumably highly derived – has more in common with the segregation of rhombomeres in the vertebrate embryo than with somitogenesis from a blastema.

In cartilaginous fishes such as sharks, there may be several hundred vertebrae (Springer and Garrick, 1964). In teleosts, a typical vertebral number is 48, although the range is very wide (Baer, 1964). Long-bodied teleosts such as eels may have as many as 200 vertebrae. At the other extreme the platyfish has around 6 trunk and 20 tail segments (Tavolga, 1949). Like teleosts, amphibians also show wide variations. Adult frogs have 6-9 presacral vertebrae, the lowest count of any terrestrial vertebrate (Young, 1962). By contrast the caecilian amphibians have worm-like bodies and 95-285 trunk vertebrae (Duellman and Trueb, 1994). A cautious estimate, from incomplete fossil data, is that when some of these long-bodied vertebrates evolve, they show an average increase of one vertebra every million years (R. Carroll, personal communication).

The highest vertebral numbers are found in some reptiles. Snakes may have hundreds of vertebrae – as many as 565 in the extinct Archaeophis. In modern birds, vertebral number is more stable, a typical range being 37-53 (Starck, 1996). In mammals, the number of precaudal vertebrae is also relatively stable, but the number of vertebrae in the tail varies from 3 to 47 (Baer, 1964; Anderson and Jones, 1984). Interestingly, vertebral number is much more plastic in some dolphins (genus Phocoenoides) than it is in terrestrial mammals, and may approach 100 (Anderson and Jones, 1984).

In this section, we discuss developmental mechanisms that might be targets for evolutionary changes in somite number.

In insects, body segmentation and segment identity are controlled by different genes (Ingham and Martinez Arias, 1992; Lawrence, 1992). The same appears to be true for the vertebral column. Although vertebrate segmentation genes have proved elusive, several promising candidate genes have recently been identified. Unfortunately, we are a long way from understanding the relation of these genes to the control of somite number. In the mouse, the Eph-related receptor tyrosine kinase Sek-1 is expressed in two stripes in the unsegmented mesoderm immediately caudal to the last-formed somite (Nieto et al., 1992). The first narrow stripe may correspond to the anterior half of the next somite to form, and a broader stripe caudal to that may represent the next but one somite. Expression patterns which appear before segments have formed, and which correspond with the future position of segments, are known as isomorphic prepatterns (Wolpert and Stein, 1984). While these patterns are suggestive of segmentation genes, functional studies are needed for a definitive answer.

Signalling by homologues of Drosophila Delta, and its receptor Notch, is important for normal somite development in vertebrates. Targeted disruption of Notch-1 leads to defective somitogenesis (Conlon et al., 1995) as does disruption of RBP-Jκ, a transcription factor in the Notch signalling pathway (de la Pompa et al., 1997). The correct expression of Notch-1, and the mouse Delta homologue Dll1, is dependent on the transmembrane protein presenilin 1 (Wong et al., 1997). In Xenopus, X-Delta-2 is expressed in a segmental pattern in the paraxial mesoderm that precedes overt somite formation, suggesting a role as a segmentation gene. However, targeted disruption shows that, in mice, the Delta homologue Dll1 is probably involved in the maintenance of somite borders rather than in the establishment of the primary segmental pattern (Hrabe de Angelis et al., 1997).

her1, the zebrafish homologue of the Drosophila pair-rule gene hairy, is expressed in alternating somites (Müller et al., 1996). Homologues of the Drosophila segment-polarity gene engrailed may be involved in the formation of enterocoelous somites, such as the first 8 segments of amphioxus (Holland et al., 1997) and the mandibular head cavity of the lamprey, Lampetra japonica (Holland et al., 1993; De Robertis, 1997). This and other evidence suggests a common evolutionary origin of segmentation in vertebrates and at least some invertebrate taxa (De Robertis, 1997).

Other putative segmentation genes have been identified in the vertebrate hindbrain. The zinc-finger genes kreisler/valentino and Krox-20 are important for normal segmental patterning of rhombomeres (reviewed by Lumsden and Krumlauf, 1996). However, in the Krox-20 null mutant mouse, although the normal pattern of rhombomeres is disrupted, the underlying segmentation of the hindbrain is maintained (Schneider-Maunoury et al., 1997). Sek genes are also likely to have a role in hindbrain segmentation (Irving et al., 1996). Disruption of Sek-1 signalling by injection of a tk⁻ truncated RNA disrupts hindbrain segmentation in zebrafish and Xenopus embryos (Xu et al., 1995).

Tam (1981) proposed a relationship between the rate of axial growth, the rate at which somites are laid down and the size of somites. Thus tail somites in the mouse are small and numerous, and the rate of axial elongation when they are laid down is relatively slow. By contrast, the lumbar and sacral somites are larger and fewer, and are segregated at a time of rapid axial elongation. Thus, in the mouse, there may be a negative correlation between the rate of axial elongation, and somite number. Other species appear to show a positive correlation.

Caecilian embryos undergo rapid elongation during neurula stages. As the body lengthens, large numbers of somites are segregated (Sammouri et al., 1990; Duellman and Trueb, 1994). The hagfish (a chordate) also has a highly elongated primary axis and forms a high somite number: 75 by the tailbud stage (Fig. 2). Recent studies on reptiles support the idea of a positive correlation between axial elongation and somite number. Raynaud (1994) examined somitogenesis in the ‘slow worm’ Anguis fragilis and the green lizard Lacerta viridis. The slow worm is a limbless lizard with a long body, rather like a snake. The lizard, by contrast, has well-developed limbs and does not have a highly elongated body. When the embryos were compared, it was found that the slow-worm embryo elongated, and laid down somites, more rapidly than the lizard.

It is not clear whether an increase in body length is primary or secondary to an increase in somite number. In other words, does somite number regulate according to the length of the strip of presomitic mesoderm? Experiments indicate that this may not be the case (reviewed in Tam, 1981). For example, if Xenopus embryos are made smaller by ablating tissue at early stages, the same number of somites is formed, but the somites are smaller (Cooke, 1975). These conclusions were reached by comparing operated embryos with stage-matched controls. It will be interesting to determine whether these findings are applicable to other amphibians, such as the direct-developing frogs, which show different patterns of development to Xenopus (Elinson, 1987).

Experimental data support a model in which somite number is controlled in part by species-specific cyclical properties of somite precursor cells (Cooke, 1981). Using clonal analysis, Stern and colleagues have shown that precursors in the mesoderm of the chicken embryo give rise to progeny in every 5-7th somite (Selleck and Stern, 1991; Stern et al., 1992). Heat shock of chick embryos produces defects in every 6-7th somite, a periodicity that corresponds quite well with the predicted cell cycle time of somite precursors (Primmett et al., 1988, 1989; Stern et al., 1988). Heat shock in anuran and zebrafish embryos also gives rise to defects in the somite series, but these are single rather than repeated (reviewed by Keynes and Stern, 1988; Kimmel et al., 1988).

It would be interesting to determine whether somite precursors show different cell-cycle times in species with different somite numbers. For example, do snake somite precursors have a very short cell cycle? Tam (1981) has drawn attention to the possible parallels between cell cycle time in somitogenesis and skeletogenesis in the limb. According to the progress zone model, there is a relationship between the number of skeletal elements laid down along the proximodistal axis of the limb and the number of cell cycles completed by mesoderm cells in the progress zone (Summerbell et al., 1975). Whether or not the tailbud behaves in this way remains to be determined.

While differences in segment number among species are due principally to evolutionary changes in somitogenesis, other factors can be involved. The low vertebral number in adult frogs is due to fusion of vertebrae, and the loss of the tail at metamorphosis (Duellman and Trueb, 1994). Although anurans such as Xenopus have only 6-9 presacral vertebrae as adults, more than 40 somites are segregated in the Xenopus embryo (Nieuwkoop and Faber, 1967). Tucker and Slack (1995) have shown that the trunk-tail boundary slides anteriorly along the primary axis during Xenopus development. We suggest that this may contribute to the low vertebral number in anurans; somites that were initially truncal could become caudal and therefore be resorbed during metamorphosis.

In the chick, segments are lost when the tailbud regresses following a wave of apoptosis (Mills and Bellairs, 1989). An interesting finding in this context is that Wnt-3a is required for normal development of the tail region in mice. Homozygous null mutant Wnt-3a^−/− embryos show truncation of the body at the axial level of the hindlimbs (Greco et al., 1996). These experimental data show how mutations could result in a major evolutionary change such as loss of caudal vertebrae. Other genes are important for axial elongation and tail development, notably the T-box genes including Brachyury (Chapman et al., 1996). Differential growth of the tailbud in the embryo can contribute to a high vertebral number in the adult. In the thresher shark Alopias vulpinus (Fig. 1), the tail contains nearly 300 vertebrae, more than the trunk itself (Springer and Garrick, 1964).

Snakes are a mystery because there is relatively little published data on their development. It has been suggested that the high somite number in snakes is due to the modification of phylotypic stages (Raff, 1996; Gerhart and Kirschner, 1997). In support of this view, a tailbud stage snake (Natrix maura) embryo has been seen with over 100 somites (A. Raynaud, unpublished). By contrast, the snake Vipera aspis is reported to have only 36 somites at the tailbud stage (Hubert and Dufaure, 1968), a similar number to birds or mammals. However, it is not clear whether the tailbud stage used in that study is comparable to that in other amniotes.

If snakes do have a normal somite number at the tailbud stage, compared to other amniotes, then the cloaca must slide posteriorly along the primary axis during development. Otherwise, adult snakes would have a short trunk and a long tail – which is not what is observed. In summary, the potential developmental targets for evolutionary changes in vertebral number include segmentation genes, tailbud growth and apoptosis, axial growth rate, somite size, vertebral fusion and cloacal displacement.

Since somite number is a key variable in vertebrate evolution, it is important to determine which developmental stages are targets for this change. A widely accepted model of the relationship between developmental stage and phenotypic divergence is the ‘developmental hourglass’ or ‘phylotypic egg timer’ (Elinson, 1987; Wolpert, 1991; Duboule, 1994; Collins,

1995, Raff, 1996; Gerhart and Kirschner, 1997).

The hourglass model predicts that adult differences between species arise through divergence from a conserved (phylotypic) stage of embryonic development. For many anatomical characters this is clearly true, since embryos of different species often look far more alike than their respective adults. The phylotypic stage is thought to be the tailbud stage (Slack et al., 1993). There can be difficulties with identifying the tailbud stage in different species (Richardson et al., 1997). The tailbud is a compact, ectoderm-covered swelling containing mesoderm, the caudal end of the notochord and neural tube, and part of the primitive gut. When the tailbud is a distinct projection, but is not yet segmented, the tailbud stage has been reached.

‘Phylotypic’ stage indicates that embryos all show the body plan characteristic of their phylum, or comparable higher taxon. However, recent studies suggest that the vertebrate tailbud stage may not be as highly conserved as originally thought (Richardson, 1995; Richardson, et al., 1997). Variations in developmental timing (heterochrony) mean that the various parts of the body plan develop at different times in different species (Richardson, 1995). For example, zebrafish embryos have no branchial arches at the tailbud stage, and the heart primordia have not fused into a tube (Richardson et al., 1997). Thus if all one knew about the zebrafish body plan came from the tailbud stage, this species would not even be placed in the same phylum as vertebrates.

It is important to determine precisely which features of development are highly conserved and which can vary. Somitogenesis is a useful character to examine in this context because it can be easily quantified. According to the hourglass model, one would expect somite number to be similar in all species of vertebrates at a conserved stage. Adult differences in segment number should then become apparent through divergence at later stages. But this is not the case in the species examined here. The short-bodied platyfish Xiphophorus maculatus develops a total of 26 body segments (Tavolga, 1949). By contrast the caecilian amphibian Typhlonectes compressicaudus has a very long body and 98-100 segments (Sammouri et al., 1990). The development of these differences does not obey the hourglass model.

In the platyfish, somitogenesis is retarded from the earliest stages of development. At late neurula stages (Fig. 3A), no somites have developed at all and only 12-13 somites have formed by the tailbud stage (Fig. 3B). By contrast the worm-like caecilian (Amphibia: Gymnophiona) develops from an embryo which has an elongated primary axis and accelerated somitogenesis. This extreme elongation of the body is seen at the late neurula stage (Fig. 3C). Even at this early stage, 17 somites have segregated. By the tailbud stage, the primary axis of the caecilian is so elongated that it reaches right round the large egg. Over 50 somites have segregated by the tailbud stage (Fig. 3D), many more than in the platyfish at the same stage.

Thus somitogenesis appears to show an earlier onset and increased rate in the caecilian. With respect to somite number and body length, divergence between these two species begins before the tailbud stage and continues throughout it. This is in contrast to the predictions of the developmental hourglass (Fig. 4). These two species do not share a common, conserved embryonic stage that has the same, or even similar, number of somites.

Raynaud (1994) reached the same conclusion. At the tailbud stage the green lizard has 27 somites, while the slow worm has 60 and is almost twice as long. These emerging differences in body form are apparent from the earliest stages of somite segregation and cannot therefore be attributed to divergence from a shared, conserved stage. In summary, changes in somite number during evolution can, in some species, be attributed to patterns of development showing divergence beginning before the putative phylotypic stage and continuing to diverge through this and subsequent stages. It is important to realise that one does not have to use the tailbud stage as the yard stick for showing differences in somitogenesis between species. Richardson (1995) showed that a variety of developmental characters, plotted against somite number in different species, showed dissociation from somitogenesis.

We now consider how it is possible for somite number to vary so widely between species at the tailbud stage – the putative phylotypic stage – when this stage is thought to be constrained by conserved patterns of developmental gene expression.

A key advance in evolutionary developmental biology has been the identification of the zootype – a conserved pattern of homeobox gene expression seen in a wide range of animal taxa (Slack et al., 1993; Manak and Scott, 1994). These genes may encode positional value along the primary axis. The existence of the zootype is consistent with Wolpert’s concept of a universal positional field (Wolpert, 1989). Zootypic genes are thought to be expressed most strongly at phylotypic (tailbud) stages. However, recent work has shown that phylotypic stages are more variable than is widely assumed (Richardson, 1995; Richardson et al., 1997). Furthermore, the variations in somite number at the phylotypic stage described above are not consistent with evolutionary constraint in this feature. Changes in phylotypic stages are perhaps to be expected; the zootype has itself evolved – by gene duplication and loss.

Raff (1996) has identified two opposing positions in evolutionary developmental biology, one favouring developmental conservatism and the other stressing the evolutionary lability of embryonic stages. We suggest that, at least with respect to certain features such as somitogenesis, these conflicting views are easily reconciled: the body plan consists of some structures that are dependent on specific Hox codes for their determination and are therefore highly constrained, and others that show dissociation or uncoupling from the zootype and are therefore more easily modified during evolution. We use the term dissociation (uncoupling, disengagement or dissociability; Needham, 1950; Richardson, 1995) to imply that two mechanisms, acting on the same pattern, can be varied independently during evolution.

Examples of genes that may encode specific structures, rather than relative position, include the non-zootype homeobox genes tinman and Pax6. These may specify heart and eye, respectively (Azpiazu and Frasch, 1993; Scott, 1994; Schultheiss et al., 1995). tinman is expressed in cardiac precursors in Drosophila and vertebrates, even though the hearts of these animals are not traditionally considered homologous. Specific structures can also be encoded by zootypic genes. For example orthodenticle homologues encode for the rostral neural tube region and vision-associated cells in a range of chordates (Williams and Holland, 1996). The position of the forelimb and associated structures is also specified by characteristic Hox codes (Gaunt, 1994; Burke et al., 1995).

Somite identity is controlled independently of somite number (Cooke, 1975). Cooke predicted that there were unlikely to be genes specifying the identity of a particular numbered somite. This is now thought to be true. It is likely that axial identity along the vertebral column is specified according to a Hox code (Kessel and Gruss, 1991). Treatment of embryos with retinoic acid causes an anterior shift both in the boundaries of Hox gene expression, and in the axial identity of vertebrae (Kessel and Gruss, 1991). Experimental transpositions^* of this type support the view that segment identity can be varied independently of segment number.

Transpositions are also seen during evolution. In a landmark paper, Burke and colleagues compared the axial level of Hoxc-6 expression in species with different vertebral formulae (Burke et al., 1995). The anterior boundary of Hoxc-6 expression in the paraxial mesoderm lay opposite different somite numbers in different species (Fig. 5). Yet it always corresponded to the axial level of the brachial plexus, even though this varies widely between species (Fig. 5). Thus evolutionary changes in the axial level of the forelimb are associated with a corresponding shift in the axial level of Hox gene expression. Gaunt (1994) reached similar conclusions. These studies indicate that the specification of positional values in the mesoderm is independent of the specification of boundaries between somites. Therefore, because somitogenesis is not tied to particular Hox codes, it is freed from the evolutionary constraint of the zootype.

Somitogenesis can be uncoupled from the zootype. We need to know if this is an isolated phenomenon or part of a wider pattern in vertebrate evolution. We argue that there are other examples of repeating patterns uncoupled from the genes that encode segment identity.

Early models of limb development proposed that a morphogen gradient specified not only the pattern of digits, but also digit identity (Tickle et al., 1975). Thus a single mechanism accounted for the whole pattern and no dissociation was possible. But it later became apparent that there may be two distinct mechanisms that interact to give the final pattern: a repeating pattern to generate a ground plan of skeletal elements and a positional field to make the digits non-equivalent (Wolpert and Stein, 1984; Wolpert, 1989; Tabin, 1992).

The two mechanisms can be uncoupled experimentally in reaggregates. These are made by placing a pellet, reaggregated from a single-cell suspension of limb-bud mesenchyme, into an ectodermal limb-bud jacket. When grafted to a host, the reaggregates give rise typically to 1-3 identical digit-like structures (Ros et al., 1994; Hardy et al., 1995). When Hoxd gene expression was examined in the reaggregate, a uniform pattern across the digits was seen, instead of a series of nested domains (Ros et al., 1994; Hardy et al., 1995). Thus a pattern of digits had been generated in the absence of a gradient of positional information (Fig. 6).

Like the somites and the limb, feather patterns (and the associated feather pigment patterns) probably involve a repeating pattern and a mechanism for specifying positional information (Richardson et al., 1989, 1990, 1991). Experiments suggest that feathers are laid down according to a spacing mechanism (McLachlan, 1980; Davidson, 1983). A separate mechanism makes the feather germs different from one another.

Feather number may evolve independently of the positional field that determines feather identity. Thus although the same types of feathers (small covets or large flight feathers for example) may be present in different species, the number of rows of feathers of each type can vary between species (Lucas and Stettenheim, 1972). This could be achieved by a change in the spacing frequency of the feathers, or the size of the pteryla, with no need to invoke changes in Hox gene regulation. Again, because of dissociation, patterns can vary even though one mechanism involved in their specification is a conserved positional field.

In the vertebrate hindbrain, Hox genes may provide a positional field, which is superimposed upon a segmental ground plan (Wilkinson et al., 1989; Zhang et al., 1994). Studies on the murine segmentation mutant kreisler suggest that Hox codes can be established independently of segmentation (Frohman et al., 1993). This view is supported by the fact that retinoic acid has different effects on segmentation and the specification of segment identity in the hindbrain (Wood and Thorogood, 1994).

However, there is also evidence that hindbrain segmentation and positional specification are not independent mechanisms (Krumlauf, 1992; Lumsden and Krumlauf, 1996). Both Krox-20 and kreisler/valentino probably regulate Hox gene expression, suggesting that they are related to the control of segment identity as well as to segmentation itself (Sham et al., 1993; Moens et al., 1996). This interdependence of functions might explain why rhombomere number is far more stable during evolution than somite number. Another factor may be that rhombomere segmentation lacks the open-ended properties of somitogenesis in the tailbud.

The pattern of Hox genes expressed in the hindbrain appears to specify positional values of cells in the pharyngeal arches (Hunt and Krumlauf, 1992). We believe that there is good evidence of independent variation of segmentation and positional specification in the pharyngeal apparatus. In lampreys, the simple, segmental pattern of pharyngeal pouches laid down in the embryo remains largely unaltered through later stages and the embryo therefore resembles the adult (Fig. 7B,C). By contrast, in the chicken and other amniote embryos, expression patterns of bmp-7 reveal that the anterior pharyngeal clefts have become remodelled before the posterior ones have even formed (Fig. 7D-F). Because of this, chick embryos never display the complete, undifferentiated series of arches depicted by Haeckel (1874).

Variations in somite number suggest to us that patterns of homeobox gene expression do not always impose rigid evolutionary constraint on embryonic development. Homeobox genes appear to function in two distinct ways. They can encode non-repeating structures such as eyes or heart; or they can encode position value in a series of repeating elements. In the latter case, the final pattern is determined by two distinct mechanisms. If these are uncoupled from each other, as they appear to be in the development of somites, limb, feathers and pharyngeal arches, the system can show considerable evolutionary lability. Thus the body plan consists of different components with different degrees of evolutionary constraint. It is possible that at least some of the highly constrained components correspond to Cooke’s (1975) non-repeating elements.

It is widely recognised that all vertebrate embryos share a common and highly conserved developmental programme. However, dissociation of developmental mechanisms is at odds with the widespread view that certain embryonic stages remain almost unchanged during evolution (Haeckel, 1874; Elinson, 1987; Wolpert, 1991; Duboule, 1994; Collins, 1995, Gerhart and Kirschner, 1997). Some authors have suggested that developmental mechanisms in embryos are so tightly coupled that any evolutionary change in embryonic development is likely to be lethal (Collins, 1995). We believe that, although developmental mechanisms are highly conserved, dissociation allows at least some evolutionary modifications to be made, even at phylotypic stages. Modifications of this type have been important in the evolution of body form and vertebral number in several vertebrate groups.

Special thanks are due to Michael Akam for critical contributions. Many other scientists have shared ideas with us, including Mike Coates, Ron Conlon, Jonathan Cooke, Jacqueline Deschamps, Brian Hall, Peter Holland, Kirsty Lawson, Andrew Lumsden, Alessandro Minelli, Bill Perrin, Jose de la Pompa, Sabine Renous, Jonathan Slack, Claudio Stern and David Wilkinson. Brian Houston kindly provided the chick bmp-7 probe and Philippa Francis-West commented on the manuscript. John Bluemink and the staff of the Netherlands Institute for Developmental Biology made embryos available for study. Richard Tibbitts prepared the line drawings. Financial support was provided by the British Heart Foundation (M. K. R. and S. P. A.), The Wellcome Trust and British Council (M. K. R.), and The NSF, IBN 94-19407 (J. H.).