Zebrafish pigmentation mutations and the processes of neural crest development

The questions of how cells become specified and patterned within the embryo are fundamental to developmental biology. The vertebrate neural crest is an attractive model system for studying these questions because it gives rise to a great diversity of different cell fates which after migration form highly patterned structures, including much of the craniofacial skeleton, peripheral nervous system and pigment patterns (Hörstadius, 1950; Weston, 1970, 1991; Le Douarin, 1982; Eisen and Weston, 1993).

Neural crest cell development involves at least five general processes, which overlap in their timing. (1) Specified precursors must be generated from initially multipotent neural crest cells (Bronner-Fraser and Fraser, 1988). (2) These precursors must be patterned within the embryo. Neural crest patterning involves two steps: choice of migration pathway and choice of site of localisation. Two major migration pathways are used by neural crest cells (Weston, 1963; Le Douarin, 1982). The lateral migration pathway lies between the skin and the somites, and is used predominantly by pigment cells, whereas the medial pathway lies between the neural tube and the somite, and is used by most or all derivatives (Le Douarin, 1982). Cells of each derivative migrate to characteristic locations at which they form stereotypical arrangements of cells. (3) Neural crest cells are usually highly proliferative (e.g. Fraser and Bronner-Fraser, 1991; but compare Raible and Eisen, 1994). (4) Neural crest cells are dependent on various trophic factors for their survival (e.g. Stemple et al., 1988; Morrison-Graham and Weston, 1993). (5) Cells of each derivative must express characteristic differentiation products.

Although in vivo and in vitro studies have revealed much about the biology of these different processes (see reviews by Le Douarin, 1982; Weston, 1991; Anderson, 1993; Selleck et al., 1993; Stemple and Anderson, 1993), key molecular regula-tors involved are largely unknown. We have undertaken a genetic approach to the dissection of these processes. Studies of mouse mutations that affect coat pigmentation have identified several genes crucial for certain steps of pigment cell development (Silvers, 1979; Lyon and Searle, 1989; Jackson, 1994).

Thus, Dominant spotting (W) and Steel (Sl) have been shown to have multiple effects on proliferation (Reid et al., 1993), survival (Morrison-Graham and Weston, 1993), and probably also the choice of migration pathway (Wehrle-Haller and Weston, 1995) of melanoblasts. In addition, mouse pigmentation mutations have been important in understanding melanocyte differentiation (Jackson, 1994). For example, in dilute (d) mutant mice, melanocytes appear less dendritic (Silvers, 1979). d encodes a novel myosin protein that may be involved in dendrite formation or in the transport of melanosomes, the membrane-bound pigment-containing organelles of melanocytes, into dendrites (Mercer et al., 1991; Jackson, 1994). These studies suggest that extensive screening for mutations disrupting pigmentation would be a powerful approach to the identification of key regulators of neural crest development.

The advantages of zebrafish as a genetic model organism have been discussed elsewhere (Kimmel, 1989; Mullins and Nüsslein-Volhard, 1993; Driever et al., 1994; Henion et al., 1995). Analysis of zebrafish neural crest has revealed many parallels with other vertebrates, in terms of the range and types of derivatives, their migration routes, and their embryonic distribution (Raible et al., 1992; Schilling and Kimmel, 1994). Furthermore, the ability to follow individual cells in living embryos has been elegantly exploited to characterise neural crest fate specification (Raible et al., 1992; Eisen and Weston, 1993; Raible and Eisen, 1994; Schilling and Kimmel, 1994).

Zebrafish offer unique opportunities for addressing the mechanisms underlying neural crest cell development. Their larvae display a rich, reproducibly patterned arrangement of three distinct pigment cell types (chromatophores): melanophores, xanthophores and iridophores (Milos and Dingle, 1978; Kimmel et al., 1995; this work). Since these cells are naturally labelled, readily observed under the dissecting microscope, and complete development of the larval pigment pattern within 6 days post-fertilisation, zebrafish provide a useful system for the identification of mutations affecting all aspects of their development. A number of larval pigmentation mutations have been described previously (Streisinger et al., 1981, 1986), and the genetic basis of adult stripe formation is being studied using these and other mutations (Johnson et al., 1995). A small-scale screen for zebrafish neural crest mutations has recently been described (Henion et al., 1995).

As part of a large-scale screen for embryonic/early larval zebrafish mutations (Haffter et al., 1996), we isolated many mutations affecting pigment cell development. In this paper we present detailed descriptions of pigment mutant phenotypes. We relate these to aspects of neural crest development that may be involved, and distinguish mutations affecting early processes of crest development (e.g. specification) from those affecting later processes (e.g. differentiation). Many of the advantages of zebrafish pigmentation as a neural crest model system apply also to the zebrafish craniofacial skeleton, and many mutations affecting these structures were isolated in the same screen (Piotrowski et al., 1996; Schilling et al., 1996). This work forms the foundation for a detailed analysis of the development of particular neural crest derivatives.

Fish breeding and maintenance is described by Mullins et al. (1994), and identification, isolation and complementation analysis of mutations is described in an accompanying paper (Haffter et al., 1996). Stocks of previously described pigmentation mutations (albino (alb), golden (gol), brass (brs), sparse (spa) and rose (rse); Streisinger et al., 1981, 1986; Johnson et al., 1995) were maintained similarly.

We kept all mutations that affected pigmentation by changing chromatophore cell number, pattern, morphology or intensity of pigmen-tation, but which were without severe defects in early development. Mutations with pigment defects believed to be secondary effects of another aspect of the phenotype were not kept unless they fitted criteria described for the other phenotype (see accompanying papers). Many mutations with severe melanophore degeneration phenotypes are necrotic by the sixth day (all stages as defined by Kimmel et al., 1995) and were also not kept.

During reidentification from outcross families, pigmentation phenotypes were reexamined to get consistent descriptions. Embryos were examined, during the hatching period and on the sixth day, with a Wild stereo microscope with 0.8×, 2×, 4× and 8× objectives, using both transmitted and incident light. The latter was vital for examining iridophore and xanthophore phenotypes. Observations of pigment cell morphologies were made with a Zeiss Axiophot during the photography of selected phenotypes.

Embryos were generally kept in the light on a white background. Due to background adaptation, melanophores in 6-day-old larvae show their pigment organelles (melanosomes) clustered tightly around the cell centre, and so appear smaller than at earlier stages. Hence, mutants in which this process of background adaptation was hindered were easily found, by the much larger appearance of their melanophores.

To count crest cells migrating on the lateral migration pathway (Raible et al., 1992), living embryos were mounted in embryo medium (Mullins et al., 1994) under a No. 1 coverslip, using No.1 coverslips as spacers. Migrating neural crest cells were identified by their characteristic morphology and location immediately below the periderm (Raible et al., 1992). Cells were counted on one side over somites 5-15 in each embryo, between the 26-somite and prim-24 stages of development (22-35 hours of development at 28.5°C; stages as defined by Kimmel et al., 1995). Individually identified embryos were then raised overnight in embryo medium and their phenotype identified.

Embryos/larvae were mounted in 2% methyl cellulose without a coverslip or between No. 1 coverslips, using 2 (pharyngula -hatching periods) or 3 (sixth day) No. 1 coverslips as spacers. Phenotypes were photographed in living fish with a Zeiss Axiophot microscope with 5×, 10×, 20× and 40× (water immersion) objectives, using Ektachrome 64T film. Embryos were illuminated with incident light from a fibre-optic tungsten light source to photograph iridophores, and the orientation of larvae with respect to the light source, as well as the exposure time, were comparable.

The larval zebrafish pigment pattern consists of an array of three neural crest-derived pigment cell types that is essentially complete by the sixth day (Figs 1, 2A,C). The pattern consists of four longitudinal stripes of black, melanin-expressing melanophores (Milos and Dingle, 1978). The dorsal stripe and the ventral stripe extend from the head to the tail tip; the yolk sac stripe extends from under the heart to the anus; and the lateral stripe runs in the horizontal myoseptum of somites 6-26 out of the normal 30-34 somites (with some variation in the boundaries between individuals). Iridophores appear silver under incident light, or gold when viewed through overlying xanthophores (Fig. 2C). They are associated with three of the melanophore stripes, and have a characteristic pattern in each of them: a medial array in the dorsal stripe, a series of bilateral pairs and a pair of lateral patches (the large patches above the yolk sac) in the ventral stripe, and a dense band in the yolk sac stripe (Fig. 2L). Individual xanthophores are difficult to distinguish (Fig. 2J,K), even on the sixth day, but together they give a strong yellow cast to the larval body, most prominent dorsally, and rather weaker ventrally. They are not visible ventral to the ventral stripe.

Zebrafish embryos begin to develop pigmentation visible under the stereo microscope at around 25 hours of development at 28.5°C (Kimmel et al., 1995; see also Milos and Dingle, 1978). Melanin is visible first in the retinal epithelium, and within 2 hours is visible within migrating melanophores in the anterior trunk and posterior head regions. This pigmentation becomes stronger throughout the second day. Xanthophores and iridophores become visible during the hatching period. Xan-thophores appear as a faint yellow cast to the head, especially dorsally. Iridophores become visible as tiny silver spots on the eye and in the dorsal tail region. Over the next 3 days all three chromatophore types continue to appear and become more strongly pigmented, and the larval pattern matures.

We identified 285 mutations with pigment phenotypes which promise to include mutations in key genes of all processes of chromatophore development. 250 (88%) of these mutations have been assigned to 94 complementation groups and these all fall within seven major phenotypic classes (Table 1). We later use this classification to suggest testable hypotheses regarding the process of neural crest development that may be affected in each class. A few genes affect pigment cell number, either all three pigment cell types (Class I), or just a single type (Classes II and III). Chromatophore distribution is affected in two ways: (1) part of the wild-type pattern is absent (Class IV), or (2) the pattern is complete but ectopic chromatophores are also present (Class V). The majority of mutations affect the degree of pig-mentation of one or more cell types (Class VI). Mutations altering chromatophore morphology (Class VII) were identified by their effects on melanophores; two of these also affect xanthophore morphology. No mutations affecting iridophore morphology were found.

All mutations are described briefly in Tables 2-6, and in more detail in the next sections and Figs 2-13. Many mutations show additional phenotypes, and affected structures are noted in the tables and the papers referenced. The genes thus defined have from 1-12 alleles; 54 (58%) have only a single allele, although some of the unresolved mutations (Table 1) may be allelic with these complementation groups.

Mutations in two genes result in dramatic reductions of all chromatophore cell types (Table 2; Fig. 2). white tail (wit) mutants show melanophores with normal morphology and pig-mentation, but their number is reduced and they are absent posterior to the midtrunk (see Jiang et al., 1996). wit affects neurogenesis and mutant embryos die during the hatching period (Jiang et al., 1996). Hence defects in xanthophores and iridophores cannot be evaluated.

This phenotype is very different to that of colourless (cls), a gene originally defined by a spontaneous mutant allele, cls^t3, found in our rose stock, and represented in this screen by two alleles. cls mutants have a normal eye and a fully pigmented retinal epithelium, but are otherwise without pigmentation (Fig. 2B). Close examination reveals a few chromatophores of each type in some mutant individuals: isolated xanthophores in the head, patches of iridophores on the eye and/or in the lateral patches or dorsal stripe, and tiny melanophores in the dorsal stripe (Fig. 2F-H). These few remaining cells are all in normal positions. Mutant embryos survive until the eleventh day with normal morphology.

To test whether cls mutants affect chromatophore precursors at an early stage we examined crest cells on the lateral migration pathway. Crest cells using this route are fated to become chromatophores (Raible and Eisen, 1994). Hence we wanted to know if the number of such cells was decreased in cls mutants. Counts of lateral pathway neural crest cells revealed a 95% reduction in the number of such cells in cls embryos compared with their wild-type siblings, consistent with the general effect on chromatophore number that we observe. Hence, cls affects the development of neural crest-derived chromatophores prior to their migration on the lateral pathway. In addition to the chromatophore phenotype, cls has an early effect on ear development (Whitfield et al., 1996).

13 mutations, defining two complementation groups, shady (shd) and stolen pearls (snp), result in reduced numbers of iridophores, with no effect on the intensity of pigmentation (Table 2; Fig. 2). The shd complementation group form a clear phenotypic series. Mutants in the strongest alleles have no more than a few iridophores on the inside face of the eye only (Fig. 2E). The weakest alleles show an approximately 50% reduction in the extent of iridophores in the yolk sac stripe, but a much weaker reduction elsewhere. Intermediate phenotypes show iridophores in the lateral patches and on all surfaces of the eye (Fig. 2D,I), but relatively few in the yolk sac stripe. Some alleles are homozygous adult viable (see Odenthal et al., 1996a). Unlike shd mutants, snp mutants cannot be identified until the sixth day (Fig. 2M). The phenotype is weaker than that of any shd allele and is most evident as a variable reduction in the number of iridophore spots in the ventral stripe and in the completeness of the yolk sac stripe.

Mutants of salz (sal) and pfeffer (pfe) show partial absence of xanthophores in the dorsal head and have spaces between remaining xanthophores (Fig. 2K; see Odenthal et al., 1996b). We find an allelic series with homozygotes for strong alleles showing almost no xanthophores. These mutations seem to be distinguishable from those affecting xanthophore differen-tiation (Class VI.J), since mutants in the latter process show pale yellow, whitish, or simply granular cells in xanthophore positions, usually without normal-looking cells interspersed (Odenthal et al., 1996b; see also Figs 9,12). Like the analogous iridophore mutations, all sal and pfe alleles are homozygous adult viable and show a phenotypic series of adult phenotypes (see Odenthal et al., 1996a).

We found mutants in which essentially all xanthophores or iridophores are absent (Class II). Surprisingly, however, we found no mutants showing an analogous absence of melanophores, despite their being the most obvious cell type we examined. Although a number of mutants with reduced numbers of morphologically normal melanophores were found, even the strongest alleles show no more than 50% reduction in cell number on the second day.

We found ten mutants with a phenotype similar to sparse (spa) (Table 2; Fig. 3) (Streisinger et al., 1986). The embryonic phenotype consists of a striking reduction in melanophore number, most pronounced in the anterior head, in more ventral stripes, and, within any stripe, in more posterior regions (Fig. 3D,E). During the hatching period melanophores accumulate behind the otic vesicle, but disappear by the sixth day. Melanophores initially look normal, but by the sixth day they are abnormal in shape and size, and some contain just an isolated spot of melanin (Fig. 3F). Nine of these are new sparse alleles and show a range of phenotypic strengths. Alleles that have been tested are homozygous adult viable, as is the original allele. The tenth mutation complements spa, but is phenotypi-cally indistinguishable from a strong spa allele; we have named this locus sparse-like (slk).

Mutations in five genes cause defects in pigment pattern. In these, chromatophores are morphologically normal and appear at normal times, but part of the wild-type pattern is absent or forms incorrectly (Table 3; Fig. 4). Most striking is the choker (cho) mutant allele, which causes two pigment pattern defects: deletion of the lateral stripe (Fig. 4E) and an ectopic band of melanophores forming a partial collar in the posterior hindbrain region (Fig. 4D). cho mutants show a somite defect (Fig. 4E). Other mutants with more severe somite defects e.g. no tail (ntl) (Halpern et al., 1993; Odenthal et al., 1996a), also have no lateral stripe. Together, these observations suggest that the horizontal myoseptum is important in lateral stripe formation, and that the pigment pattern defect in cho may be more direct than in other somite mutations (see Fig. 4E and van Eeden et al., 1996a).

Mutants of the single fullbrain (ful) allele show fewer lateral stripe melanophores. Also melanophores in the dorsal and ventral stripes of the anterior trunk are displaced laterally to leave the dorsal midline devoid of pigment cells (Fig. 4F; see also Jiang et al., 1996). ful mutants have a complex phenotype and die during the hatching period. It is not clear to what extent these defects are merely secondary effects of an axial defect.

Part of the ventral stripe directly overlying the swimbladder is absent in mutant unsaddled (uns) alleles (Fig. 4H). Both melanophores and some iridophores of the lateral patches are affected. Initially, individ-ual melanophores may be seen in this region; presumably these subsequently migrate away or die.

The variable absence of melanophores in the posterior yolk sac stripe characterises the single mutant allele of lost trail (los).

Finally, both mutant mercedes (mes) alleles result in a lateral duplication of the ventral part of the tail fin (see Hammerschmidt et al., 1996), and also in duplication of the ventral stripe associated with this fin.

Another group of mutants also have chromatophores in normal positions, but show additional chromatophores in ectopic locations (Table 3; Fig. 5).

Mutants in one large complementation group, moon-shine (mon), show overabun-dant iridophores in the dorsal stripe, post-anal ventral stripe and the yolk sac stripe, where they extend into the adjacent median fins (Fig. 5C,D). mon mutations cause no defect in parts of the pigment pattern that are not associated with median fins. Moreover, all mutant alleles show distinct reductions in the median fins (Fig. 5D), and the strongest seven alleles result in reductions in the number of circulating blood cells (see Ransom et al., 1996).

Three mutations, defining two different genes, tiger (tig) and parade (pde), cause transient accumulations of chro-matophores on the medial migration pathway. tig mutants combine a strong ventral curvature of the tail with collected melanophores oriented dorsoventrally on this pathway in the trunk and tail. Both phenotypes are transient, but the pigment phenotype disappears more rapidly than the tail curvature. The two pde alleles cause a pigmentation phenotype like tig, but it appears one day later than in tig and is not associated with tail defects. Ectopic cells in pde mutants seem to contain pigment organelles characteristic of both iridophores (iridosomes) and melanophores (melanosomes) (Fig. 5H-K).

Finally two complementation groups, floating head (flh) and momo (mom), mutants of which show dramatic axial defects (see Odenthal et al., 1996a), also have minor pig-mentation defects. These mutations cause temporary accu-mulations of ectopic melanophores around the neural tube during the pharyngula period (Fig. 5G). In contrast, wild-type embryos show pigmented melanophores adjacent to the neural tube only in a dorsal position (Fig. 5F). The fusion of the somites below the neural tube in these mutants, and the consequent obstruction of the medial migration pathway, fully explains the phenotype. These cells subsequently disappear, suggesting that they dedifferentiate, die in this position, or are able to migrate away on the lateral migration pathway.

Mutants showing reduced levels of pigmentation within each cell account for the majority of the mutations we found. This large group can be readily subdivided into ten subclasses (VI.A-J) according to the combination of pigment cell types that are affected. Three classes of melanophore differentiation phenotypes could be distinguished by the precise appearance of the melanophores; incomplete melanogenesis (Class VI.A), delayed melanogenesis (Class VI.B), or melanophore degeneration (Class VI.C). Effects on iridophores and xanthophores (Classes VI.G, VI.I and VI.J) could not be subdivided at this level of analysis and await further study. Where mutations affect melanophores in addition to another pigment cell-type (Classes VI.D, VI.E, VI.F and VI.H), melanophore morphology is used as a pre-liminary evaluation of the mutations’ effect on all the affected pigment cell types.

Mutations affecting only melanophore pigmentation cause one of three general phenotypes (Classes VI.A-C). The first is char-acteristic of mutations in melanin-synthesis genes and appears to be directly analogous to the albino (c) locus of mice and humans. Seven such genes are defined in our collection: albino (alb), sandy (sdy), mustard (mrd), golden (gol), nickel (nkl), lead (led) and pewter (pew) (Table 4; Fig. 6). Two of these loci, alb and gol, have been described previously (Streisinger et al., 1986). In the embryo, both the pigmented retinal epithe-lium and melanophores themselves are evenly pale or unpigmented, and melanophore size and number are normal. Fur-thermore, for alb and mrd mutants melanin is synthesised upon addition of DOPA; in contrast, sdy mutants do not make melanin in this assay (Odenthal et al., 1996a). Hence, sdy shows no tyrosinase activity, and is thus a strong candidate for a fish tyrosinase locus (Odenthal et al., 1996a). The adult phenotypes of these zebrafish mutations are described elsewhere (Odenthal et al., 1996a).

Mutants for strong alleles of alb, sdy, and mrd show no trace of melanin even on the sixth day (or later), but even these can be distinguished from mutants with a reduced number of melanophores (Class III). Melanin synthesis mutants show unpigmented gaps in the dorsal head xanthophore pattern that mimic the characteristic wild-type melanophore pattern and suggest the presence of normally patterned, unpigmented melanophores (Fig. 6). In contrast, mutations affecting melanophore number cause no such gaps: xanthophores surround the remaining (pigmented) melanophores, but ‘fill in’ the positions where melanophores are absent (see spa in Fig. 3; also Fig. 8D).

Two mutations [defining two loci, landing (lnd) and slow tan (sln)] do not affect the pigmented retinal epithelium, but result in most melanophores during the pharyngula period appearing only as tiny black dots (Table 4; Fig. 7C). Other melanophores are very pale, but of normal size (Fig. 7F). Melanophore number appears reduced, but by the sixth day melanophore number, size and pigmentation are almost indistinguishable from wild type (Fig. 7L). These mutations thus appear to cause delayed melanophore pigmentation.

Eleven touch-down (tdo) alleles cause a similar phenotype, but there is a reduced touch response (Granato et al., 1996) and the melanophore number seems to be reduced. Mutants in strong alleles initially show almost all melanophores as tiny spots and the apparent number is highly reduced (Fig. 7B,E). They are similar in size and appearance to the few melanophores seen in cls mutants, but are clearly more abundant, and more widely distributed. Unlike lnd and sln, the phenotype is only partially recovered by the sixth day. Xanthophores ‘fill-in’ the gaps in the melanophore pattern (Fig. 7K), thus suggesting that, in contrast to melanin synthesis (Class VI.A) mutations, melanophore number is decreased at this stage. In mutants of a weaker allele (tdo^tp8), most melanophores are initially merely pale and by the sixth day the number of normally pigmented melanophores is almost wild type. These mutations seem therefore to combine delayed pigmentation with a reduced survival of melanophores. It should also be noted that the touch-response defect shown by tdo mutations is a transient phenotype and may be explained by delayed differentiation of the touch-response circuitry.

One group of melanophore mutations (13 alleles, representing four distinct genes; Table 4; Fig. 8) causes a third melanophore phenotype, progressive deterioration of melanophores.

Mutations in these genes result in additional phenotypes; in fade out (fad), fading vision (fdv) and quasimodo (qam) mutants, melanin in the pigmented retinal epithelium also becomes paler, and retinal degeneration occurs in fad and fdv. Remarkably, one such allele, fdv^th236a, is homozygous adult viable, and results in fish with pale melanophore stripes and lacking eyes. As these fish age, large tumours grow from the eye sockets (Odenthal et al., 1996a).

Unlike mutants with melanin synthesis (Class VI.A) or delayed pigmentation (Class VI.B) phenotypes, melanophores are initially normal in these melanophore degeneration mutants (Class VI.C). Then, at a time specific to particular alleles, but ranging between the second and fourth days, melanophore pigmentation changes. Some paler melanophores, or others having only small, black melanin spots (larger than the tiny spots in ton), become visible (Fig. 8B). These phenotypes become progressively more severe: by the sixth day mutants in strong alleles show melanophores that are almost all small and spot-like in appearance (Fig. 8D). These small, spot-like melanophores collect in abnormal locations and arrangements: ventrolateral to the ear, on the hindgut, and in clusters in the dorsal stripe (Fig. 8D,F,H).

Mutations in two genes [polished (pol) and cold-light (cot)] cause the melanophore degeneration phenotype exhibited by Class VI.C mutants, but also result in pale xan-thophores (Table 4; Fig. 9). We cannot examine xanthophore morphology in the same detail as that of melanophores. Hence we are uncertain whether xanthophores also degenerate.

Eight genes defined by 23 mutations affect the pigmentation of melanophores and iridophores (this and the following section; Table 4; Fig. 10). Mutations in seven complementation groups combine melanophore degeneration phenotypes (as in Class VI.C) with dull iridophores (Fig. 10E-H). Mutations in six additionally cause a pale pigmented retinal epithelium, and mutants in two of these show retinal degeneration by the sixth day (see Heisenberg et al., 1996). Mutants in the single allele defining freckles (frk) have normally pigmented retinas (Fig. 10E), and may affect only the survival of the neural crest-derived melanophores them-selves.

Mutants in one complementation group, brassy (bss), have a melanophore phenotype (Fig. 10D) similar to the melanin-synthesis mutants (as in Class VI.A). Additionally, iridophores are duller and whiter in appearance than in wild types, but occupy approximately the normal extent (see, for example, the eye and lateral patches in Fig. 10C). Thus, both melanophore and iridophore numbers appear normal, but pigment synthesis or processing is abnormal. The phenotype of bss is similar to that of brass (Streisinger et al., 1986); no brass allele was found in our screen.

Five genes, defined by 13 mutations, affect only xanthophore and iridophore pigmentation; mutants in these genes all share at least a slight reduction in eye size. It is striking that mutations in two of these genes, mlk and pio, show jaw/arch defects (Table 5; Fig. 11; see Schilling et al., 1996). The com-bination of defects in distinct neural crest derivatives (chromatophores and craniofacial skeleton) is particularly interesting since it suggests roles for these genes in the development of multiple neural crest fates.

We suspect that some of these mutations may cause xanthophore and iridophore degeneration in a manner analogous to the melanophore degeneration mutations (Class VI.C), but xanthophore and iridophore phenotypes vary in severity between alleles, and are difficult to evaluate (see above). However, there is no sign of the normal-looking cells characteristic of mutations with decreased xanthophore or iridophore cell number (Class II). Iridophores are duller than wild type in all these mutants, but whether the number is reduced is difficult to say without better markers; in pio mutants there are fewer by the sixth day (e.g. in the lateral patches, Fig. 11H). Xan-thophore defects in pistachio (pio), milky (mlk), cookie (coo) and choco (cco) mutants may be due to poor xanthophore differentiation, since in most alleles some faintly pigmented xanthophores are visible in the dorsal head. Thus, whereas there may by the sixth day be an effect on xanthophore and iri-dophore cell number, there is also always a defect in pigmentation itself in the remaining cells.

The xanthophore and iridophore mutations (Class VI.G), like the melanophore degeneration mutations (Class VI.C), also cause an eye defect. However, whereas in the melanophore degeneration mutants the eye degenerates, losing pigmentation as it becomes smaller (Fig. 8D,F), in the xanthophore and iridophore mutants there is no sign of degeneration; the eye simply fails to grow as large as in the wild type (Fig. 11D,G; see Heisenberg et al., 1996).

12 genes (17 alleles) affect intensity of pigmentation of all three pigment cell types (Table 5; Fig. 12). Mutants in ten of these genes show the melanophore degeneration phenotype described in Class VI.C. The xanthophore and iridophore phenotypes, as in Classes VI.D-F, vary in strength, and are difficult to evaluate. We have been unable to correlate the severity of the xanthophore and iridophore defects with the melanophore defect.

Whereas no other obvious phenotypes (except an uninflated swimbladder on the sixth day) are caused by mutations in two genes, washed-out (wsh) and bleich (bli), all other mutations in this class cause further defects. In several cases, e.g. blanched (bch) and sallow (sll), these other defects are believed to be characteristic of a general retardation of larval development. All these mutants show severe melanophore degeneration phenotypes and may also have relatively severe defects in xanthophore and iridophore pigmentation. These mutants show a range of eye defects from severe eye degeneration (e.g. bleached (blc) and pech (pch); Fig. 12L) to slightly reduced size (e.g. bch). There is some correlation between severity of eye defect and that of the pigmentation phenotype, as in other mutants with melanophore degeneration (Classes VI.C and E).

An eleventh gene, puzzle (puz), has a single mutant allele. Mutants have slowly differentiating melanophores but, unlike delayed melanophore differentiation mutants (Class VI.B), they never appear as tiny spots. Unlike the melanin-synthesis mutants (Class VI.A), however, there appears to be an initially reduced number of melanophores in ventral stripes (compare spa, Fig. 3), but by the fifth day this defect is essentially recovered. Mutant individuals are too necrotic to examine xanthophore and iridophore phenotypes carefully, but iridophores appear reduced in number by the fifth day.

Finally, u-boot (ubo), also with a single allele, has a melanophore phenotype like puz, but additionally lacks lateral stripe melanophores. The latter aspect is a secondary effect of the somite defect shown by this mutation (see van Eeden et al., 1996a). In ubo mutants xanthophores and iridophores, like melanophores, are reduced in apparent number early, but have largely recovered by the sixth day. Pigmentation in the head appears normal even on the second day, so that the areas showing pigment and somite defects correlate well. However, this pigment defect is not shown by other mutants with a similar somite phenotype (see van Eeden et al., 1996a).

A group of mutations cause a dull iridophore phenotype, but no other defects (Table 5; Fig. 6E,F). Further study will be required to evaluate the basis of the phenotype.

This large class of mutations (Table 5) are discussed in detail elsewhere (Odenthal et al., 1996b).

Finally, a small set of mutations cause defects in chromatophore size and/or shape (Table 6; Fig. 13). Two of these, tinte and petroglyph, are of further interest because they show semi-dominance.

Three mutations, defining three genes, result in pale, spindly melanophores. The tinte allele, (tin^tq266a) (Fig. 13B,C) is semi-dominant: homozygous mutants have small, spot-like melanophores with short single or double projections, whilst presumed heterozygous individuals have larger melanophores of intermediate appearance between those of homozygotes and wild types. In homozygotes, xanthophores are also paler. Whereas homozygotes die by the fifth day, heterozygotes are viable, and are indistinguishable from wild type by the sixth day. The single petroglyph (pet) allele is also semi-dominant. Homozygous mutants have paler, more spindly, but multipo-lar, melanophores; melanophores of heterozygotes are initially paler (but normal in shape) than in normal embryos (Fig. 13E,F). All embryos look normal by the sixth day. The mutation gossamer (ger^tc11) has a similar phenotype to tin, except that there is no semi-dominant phenotype and no effect on the xanthophores; the phenotype is less severe, and homozygotes survive until the sixth day.

One mutation, defining the gene obscure (obs), has a subtle embryonic phenotype, with melanophores being slightly smaller than in wild types.

A large set of mutations all cause a consistent phenotype of small, rather stellate melanophores in which a bare central area presumably reflects the position of the nucleus (Fig. 13D,H,I). All except one of these have a similar effect on the xanthophores (Fig. 13H). These 11 alleles define the gene stars- and-stripes (sas). These alleles complement the remaining mutation, in which mutants have no xanthophore phenotype, and which defines a second gene union jack (uni; Fig. 13I).

Around the fifth day of development, zebrafish larval melanophores respond to the general tone of the background surface by altering the distribution of melanosomes within the cell. This process, known as background adaptation, is apparently defective in a group of nine alleles, defining seven genes: noir (nir), zwart (zwa), melancholic (mac), submarine (sum), fata morgana (fam), dropje (drp) and lakritz (lak) (Table 6). Mutants in these genes show fully expanded melanophores, even on the sixth day. Four of these mutations show other defects, in three cases a motility defect (see Granato et al., 1996).

The development of neural crest derivatives is a highly complex process and is expected to involve many genes. We have focussed here on the identification of mutations affecting development of chromatophores, three crest derivatives that are easily scored. This is reflected in the large proportion of mutations in this collection that affect pigmentation (see Haffter et al., 1996). These mutations may identify most genes involved in chromatophore development that can mutate to a phenotype visible under the conditions of the screen (see Haffter et al., 1996).

The wide range of phenotypes seen, plus the readiness with which most mutations (and genes) could be classified into particular classes, suggested that these classes might reflect the process of neural crest development that is affected in the mutants. A summary of these processes, the predicted phenotypes of mutations affecting each, and the pigmentation mutation classes that might fit these predictions are given in Table 7 and discussed below. As expected, the majority of genes identified are clearly involved in the differentiation or survival of pigment cells. A small number of complementation groups have the phenotypic characteristics expected for genes involved in specification, proliferation and patterning, and are candidates for mutations affecting these processes. These functions need not be mutually exclusive: there are several examples in which the phenotype suggests at least two distinct roles for a particular gene.

We have identified several candidates for genes involved in specification of pigment cells from the neural crest (Classes I and II). Mutations causing defects in chromatophore specification were expected to cause very severe pigment phenotypes, affecting one or more chromatophore types, and in mutants for strong alleles, to show a total absence of that cell-type(s). Furthermore, we expected no effect on melanin synthesis in the non-neural crest-derived pigmented retinal epithelium in these mutants. Although wit mutants have a very severe pigment defect, they also have clear central nervous system defects and may disrupt neural crest formation rather than pigment cell specification (see Jiang et al., 1996).

cls is the only candidate for a gene involved in the specifi-cation of all pigment cell types. cls mutants have the most severe pigmentation defect, with essentially no pigment cells, but the pigmented retinal epithelium is normal and other neural crest derivatives (e.g. craniofacial skeleton) are unaffected. The interpretation of cls as a gene important in chromatophore specification is supported by our counts of migrating neural crest cells in mutant embryos. The 95% reduction in their number is consistent with the loss of chromatophores and suggests a defect at some early stage of crest development, prior to migration.

By similar criteria, sal and pfe are the best candidates for genes important in xanthophore specification, and shd may be necessary for iridophore specification. None of these putative specification mutations show an increase in one chromatophore type corresponding to a decrease in another. This suggests that no one chromatophore fate is a default state, and implies that each pigment cell-type has to be actively specified. We found no melanophore equivalent of the shd or sal and pfe genes that specifically disrupts melanophore fate. Melanophore specifica-tion could be a further role of cls itself.

The phenotype of cls mutants suggests that cls has multiple roles in melanophore development. Residual melanophores in cls mutants show almost no melanin. In contrast, remaining xanthophores and iridophores are phenotypically normal. The similarity of the early melanophore phenotype in tdo mutants (Class VI.B), which we attribute to delayed differentiation, with that of cls mutants is consistent with an inability of cls melanophores to differentiate fully. Hence, cls has an addi-tional role in melanophore differentiation. Analogous multiple functions are already assigned to W (c-kit) and Sl (SCF) genes in mice, which have roles in the proliferation, survival and perhaps also patterning of melanocytes (Reid et al., 1993; Morrison-Graham and Weston, 1993; Wehrle-Haller and Weston, 1995).

Neural crest cell patterning involves control of cell migration and localisation. Mutants in genes affecting pigment cell localisation were expected to have phenotypically normal chromatophores in normal numbers, except in discrete regions. Mutations in five genes show defects fitting these criteria (Class IV). Whereas each of these has effects on a different part of the wild-type pattern, only one mutation results in the total absence of a stripe. Assuming no interactions between them, these genes alone cannot account for all of the wild-type pattern: the sum of all the pattern defects would not remove all of the wild-type pattern. It seems likely that mutations causing similar pattern defects affecting other aspects of the pattern were overlooked (see below).

These pattern mutants and other phenotypes we describe suggest a working hypothesis of zebrafish larval pigment pattern formation, in which localised cues direct melanophores and iridophores to particular positions, and then xanthophores occupy remaining sites in the skin. Pattern mutations disrupt parts of an individual stripe, and imply localised cues in the environment to which migrating pigment precursors respond. For example, in the case of the lateral stripe it is clear that any mutation with a severe effect on the horizontal myoseptum results in loss of that stripe, and hence we believe that some component(s) of that myoseptum is required for localisation of melanophores to the lateral stripe. chk may affect that component more directly: it certainly has the least severe somite defect of all mutants that lack the lateral stripe. In addition, local patterning influences between chromatophore cell types are suggested by observations of xanthophore distribution in wild types and mutants. The exclusion of xanthophores from dorsal regions occupied by melanophores is seen most clearly in melanin synthesis mutations (Class VI.A). Where melanophore number is decreased, e.g. in spa, xanthophores are found throughout these regions as if they are filling in spaces normally occupied by melanophores. Possible inter-actions between xanthophores and melanophores have been described in adult zebrafish tail regeneration and in axolotl body pigment patterning (Goodrich and Nichols, 1931; Goodrich et al., 1954; Epperlein and Löfberg, 1990). In axolotl, melanophores migrate first and spread evenly throughout the skin, while xanthophores form spaced clusters in the premigratory crest. As the latter cells then migrate adjacent melanophores are repelled, thus generating the pattern of alternating dorsoventral bars of melanophores and xanthophores. Furthermore, formation of the wild-type iridophore stripes in the adult may be dependent on correct patterning of the melanophores (Johnson et al., 1995). Our mutations give no evidence for interactions between other combinations of chromatophore types being important in larval pigment pattern formation.

Pigment pattern formation must include control of neural crest cell migration. Two mutations producing ectopic chromatophores are candidates for genes affecting this aspect of neural crest development. Phenotypes of both tig and pde mutations (Class V) suggest the temporary interruption of chromatophore migration on the medial pathway. Although in tig this phenotype is associated with a strong tail curvature, the two phenotypes may not be causally linked since (1) other mutations causing tail curvatures do not give pigment phenotypes (Brand et al., 1996) and (2) there is no such correlation in pde. Elucidation of the basis for these phenotypes might reveal mechanisms involved in controlling pigment cell migration in zebrafish.

We suggest that spa and slk (Class III) are candidates for mutations affecting the proliferation of melanoblasts. The degree of pigment cell proliferation that occurs in the first few days of development can be estimated from published studies of zebrafish neural crest clone sizes (Raible and Eisen, 1994). Premigratory melanophore precursors undergo on average just one division. Hence we predict a twofold reduction in melanophore number in mutations preventing melanoblast proliferation over this time period. This melanoblast proliferation estimate is consistent with the approximately twofold melanophore decrease by the third day in the strongest spa and slk alleles (Class III). A further role in melanophore survival or differentiation seems likely, since in both cases the origi-nally normal-looking melanophores become morphologically abnormal and decrease in number. The Raible and Eisen (1994) study also suggests that, in the absence of proliferation, the xanthophore/iridophore number should be reduced by up to 75%. cls and shd mutants have many fewer chromatophores, suggesting that they affect specification, rather than proliferation, of pigment cell precursors. Effects on the general proliferation of the early crest are more difficult to eliminate, but might be expected to affect other neural crest derivatives as well as chromatophores. We see no evidence of this. For example, the jaw and arch elements and fin mesenchyme are normal in these animals.

The mon phenotype is particularly intriguing. Iridophore proliferation seems to be greater than normal, but only in regions adjacent to the median fin fold. The possible relation-ship between fin development and iridophore number remains to be elucidated.

Genes involved in survival of neural derivatives and melanocytes are known from mice (Stemple et al., 1988; Morrison-Graham and Weston, 1993). We propose that a large number of our mutations produce phenotypes consistent with affects on survival of all three chromatophore types. We distinguish three classes of melanophore phenotype. In the melanophore degeneration mutations (Class VI.C), the pro-gressive loss of pigment and frequent decrease in apparent cell number suggests melanophore death, although this inter-pretation must be tested by cytological studies. We have identified mutations in 24 genes that give this phenotype (Classes VI.C, VI.D, VI.E and most of VI.H). Often, mutations affecting melanophores and one or more other chro-matophores show such a melanophore phenotype, and in some cases, e.g. blr, degeneration of these other pigment cell types seems likely. We suggest, therefore, that this group rep-resents genes important in chromatophore survival. Since these mutations often produce additional phenotypes (e.g. sunbleached (sbl); Tables 4 and 5) some may have equiva-lent roles in other tissues, including other crest derivatives such as the branchial arches. However, some of these mutations might not be affecting a normal selection step in chromatophore development. Instead the degeneration phe-notypes might be caused by the abnormal accumulation of a cytotoxic gene product. For instance, the Light allele of the mouse gene brown, B^lt, causes melanocyte death due to a missense mutation in a melanin-synthesis gene (Johnson and Jackson, 1992).

The remaining genes in classes VI and VII are in many cases clearly affecting chromatophore differentiation, either the expression of pigment (e.g. melanin mutations, Class VI.A) or cell morphology (Class VII). Since proteins used in the synthesis of multiple organelles are known from a series of mouse mutations that show effects on both lysosome and melanosome synthesis (Jackson, 1994), there may be mutations affecting organelle formation among our mutations affecting the differentiation of multiple chromatophores (Classes VI.D-H). Their further characterisation should then reveal aspects of the genetic control of organelle synthesis. Melanophore morphology mutants (Classes VII.A-C) may be analogous to mouse mutations such as d, and will also be inter-esting in terms of the genetic control of cell morphology or organelle motility. The melanogenesis genes (Class VI.A) will be especially useful as cell markers e.g. in transgenic zebrafish (see Odenthal et al., 1996a). In contrast to melanin synthesis, pathways of pigment synthesis in xanthophores are poorly understood, although pteridine synthesis is assumed to utilise pathways similar to those characterised in Drosophila (Bagnara and Hadley, 1973). The xanthophore mutations (Class VI.J; Odenthal et al., 1996b) will be a useful resource for studying the biochemistry of vertebrate pteridine synthesis. Iridophore pigmentation is dependent on precise orientation of iridosomes; their shininess depends upon these organelles being oriented in parallel to each other. The dull iridophore mutations (Class VI.I) may therefore include genes affecting organelle orientation. Background adaptation of chro-matophores depends on hormonally controlled responses to ambient lighting (Bagnara and Hadley, 1973). Some compo-nents of this behavioural response should be revealed by study of the background adaptation mutations (Class VII.D) identi-fied in this collection.

As noted above, we believe that we have identified most of the genes important in chromatophore development that can mutate to a visible phenotype. We expect that further complementation testing of the unresolved mutations will show some to be allelic to the genes we have defined, and that the pro-portion of single hit genes will decrease. However, as expected in such a screen, certain phenotypic classes are probably not saturated. This is expected for a number of reasons. Firstly, mutant phenotypes may be too subtle to be readily found. This is especially true for two classes (pigment pattern (Class IV) and dull iridophore mutations (Class VI.I)). Secondly, mutant phenotypes may be very obvious, but fewer alleles will be found where genes are physically small, or where only unusual (e.g. neomorphic) alleles give the phenotype. This may be the case for cls and slk. Finally, in the class affecting pigmentation of all chromatophores (Class VI.H), the high proportion of single hits is a result of criteria used to decide which mutants to keep in the primary screen. Many similar mutants, but showing severe necrosis on the sixth day, were discarded. Hence mutations in this class may be relatively weak, hypo-morphic alleles. Whatever the explanation in particular cases, for all these phenotypic classes further genes await identification.

We expect that among this collection of pigmentation genes are homologues of at least some of the previously characterised mouse mutations. As mentioned above, sdy is likely to be a tyrosinase (c locus) homologue. Such ideas can be readily tested by sequencing candidate genes from mutant embryos (Schulte-Merker et al., 1994) and by locating the mutation and the candidate gene on the zebrafish genetic map (Postlethwait et al., 1994; Talbot et al., 1995).

The mutations described in this paper provide a wealth of material for the analysis of all stages of neural crest development, in an organism that can be manipulated and analysed at the single cell level. They should eventually lead to a molecular genetic understanding of many aspects of the development of this uniquely vertebrate tissue.

We wish to thank Drs Judith Eisen, Steve Johnson, Dave Raible and Jim Weston for valuable comments. R. N. K. was supported by a NATO Postdoctoral Fellowship.

Mutations in the genes colourless (this paper) and golas (Malicki et al., 1996) fail to complement each other and thus are to be considered allelic. Future references to this gene should use the name colourless.