Research Article

A zebrafish melanoma model reveals emergence of neural crest identity during melanoma initiation

See allHide authors and affiliations

Science  29 Jan 2016:
Vol. 351, Issue 6272, aad2197
DOI: 10.1126/science.aad2197

Visualizing the beginnings of melanoma

In cancer biology, a tumor begins from a single cell within a group of precancerous cells that share genetic mutations. Kaufman et al. used a zebrafish melanoma model to visualize cancer initiation (see the Perspective by Boumahdi and Blanpain). They used a fluorescent reporter that specifically lit up neural crest progenitors that are only present during embryogenesis or during adult melanoma tumor formation. The appearance of this tumor correlated with a set of gene regulatory elements, called super-enhancers, whose identification and manipulation may prove beneficial in detecting and preventing melanoma initiation.

Science, this issue p. 10.1126/science.aad2197; see also p. 453

Structured Abstract


The “cancerized field” concept posits that cells in a given tissue sharing an oncogenic mutation are cancer-prone, yet only discreet clones within the field initiate tumors. Studying the process of cancer initiation has remained challenging because of (i) the rarity of these events, (ii) the difficulty of visiualizing initiating clones in living organisms, and (iii) the transient nature of a newly transformed clone emerging before it expands to form an early tumor. A more complete understanding of the molecular processes that regulate cancer initiation could provide important prognostic information about which precancerous lesions are most prone to becoming cancer and also implicate druggable molecular pathways that, when inhibited, may prevent the cancer from ever starting.


The majority of benign nevi carry oncogenic BRAFV600E mutations and can be considered a cancerized field of melanocytes, but they only rarely convert to melanoma. In an effort to define events that initiate cancer, we used a melanoma model in the zebrafish in which the human BRAFV600E oncogene is driven by the melanocyte-specific mitfa promoter. When bred into a p53 mutant background, these fish develop melanoma tumors over the course of many months. The zebrafish crestin gene is expressed embryonically in neural crest progenitors (NCPs) and is specifically reexpressed only in melanoma tumors, making it an ideal candidate for tracking melanoma from initiation onward.


We developed a crestin:EGFP reporter that recapitulates the embryonic neural crest expression pattern of crestin and its expression in melanoma tumors. We show through live imaging of transgenic zebrafish crestin reporters that within a cancerized field (BRAFV600E-mutant; p53-deficient), a single melanocyte reactivates the NCP state, and this establishes that a fate change occurs at melanoma initiation in this model. Early crestin+ patches of cells expand and are transplantable in a manner consistent with their possessing tumorigenic activity, and they exhibit a gene expression pattern consistent with the NCP identity readout by the crestin reporter. The crestin element is regulated by NCP transcription factors, including sox10. Forced sox10 overexpression in melanocytes accelerated melanoma formation, whereas CRISPR/Cas9 targeting of sox10 delayed melanoma onset. We show activation of super-enhancers at NCP genes in both zebrafish and human melanomas, identifying an epigenetic mechanism for control of this NCP signature leading to melanoma.


This work using our zebrafish melanoma model and in vivo reporter of NCP identity allows us to see cancer from its birth as a single cell and shows the importance of NCP-state reemergence as a key event in melanoma initiation from a field of cancer-prone melanocytes. Thus, in addition to the typical fixed genetic alterations in oncogenes and tumor supressors that are required for cancer development, the reemergence of progenitor identity may be an additional rate-limiting step in the formation of melanoma. Preventing NCP reemergence in a field of cancer-prone melanocytes may thus prove therapeutically useful, and the association of NCP genes with super-enhancer regulatory elements implicates the associated druggable epigenetic machinery in this process.

Neural crest reporter expression in melanoma.

The crestin:EGFP transgene is specifically expressed in melanoma in BRAFV600E/p53 mutant melanoma-prone zebrafish. (Top) A single cell expressing crestin:EGFP expands into a small patch of cells over the course of 2 weeks, capturing the initiation of melanoma formation (bracket). (Bottom) A fully formed melanoma specifically expresses crestin:EGFP, whereas the rest of the fish remains EGFP-negative.


The “cancerized field” concept posits that cancer-prone cells in a given tissue share an oncogenic mutation, but only discreet clones within the field initiate tumors. Most benign nevi carry oncogenic BRAFV600E mutations but rarely become melanoma. The zebrafish crestin gene is expressed embryonically in neural crest progenitors (NCPs) and specifically reexpressed in melanoma. Live imaging of transgenic zebrafish crestin reporters shows that within a cancerized field (BRAFV600E-mutant; p53-deficient), a single melanocyte reactivates the NCP state, revealing a fate change at melanoma initiation in this model. NCP transcription factors, including sox10, regulate crestin expression. Forced sox10 overexpression in melanocytes accelerated melanoma formation, which is consistent with activation of NCP genes and super-enhancers leading to melanoma. Our work highlights NCP state reemergence as a key event in melanoma initiation.

Understanding the earliest events in cancer formation remains an incompletely fulfilled goal in biology, with important implications for human health. In cancer initiation, an activated oncogene or inactivated tumor suppressor can trigger tumor formation. However, it is unclear as to why only sporadic cells with these genetic alterations complete the conversion to a malignant state when they are present in a large group of cancer-prone cells, sometimes described as a “cancerized field” (1). Better characterizing initiating events would help identify targets for early therapeutic interventions and also provide prognostic information about which precancerous lesions are most likely to progress to cancer.

Melanoma is a cancer of transformed melanocytes, which are pigment-producing cells derived from the embryonic neural crest lineage. It is frequently driven by BRAF or RAS mutations (~80% of cases) (2, 3). Melanoma is treatable and curable when it is localized and resected completely but remains largely incurable once it has spread, even when treated with new kinase- and immune checkpoint–targeted therapies (4). Our laboratory previously developed an animal model of a BRAFV600E-driven cancer by placing the human BRAFV600E gene under the control of the melanocyte-specific mitfa-promoter in transgenic zebrafish (5). When crossed into a p53 mutant loss-of-function background, these zebrafish (referred to here as p53/BRAF) invariably develop nevi and, after several months, invasive melanoma (5). Despite creating this extensive “cancerized field” in which all melanocytes harbor both oncogenic BRAFV600E and p53 loss throughout their life span, these p53/BRAF melanoma–prone zebrafish all develop one to three melanoma tumors after several months of age, indicating that other molecular alterations are important for tumor initiation.

crestin transgenics mark neural crest

To investigate the dynamics and mechanism of sporadic melanoma formation, we visualized and characterized melanoma lesion initiation. The functionally uncharacterized zebrafish crestin gene marks the neural crest during embryonic development, becomes undetectable by ~72 hours after fertilization (6, 7), and is specifically reexpressed in melanoma tumors in adult zebrafish (8). We reasoned that a crestin-based reporter transgene would allow us to track embryonic neural crest cells as well as melanoma tumors in vivo, potentially from their earliest onset. We amplified by means of polyermase chain reaction (PCR) a 4.5-kb upstream region common to multiple crestin insertions in the zebrafish genome and cloned this element upstream of an enhanced green fluorescent protein (EGFP) reporter (Fig. 1A, crestin:EGFP). In stable transgenic zebrafish embryos, this construct reproduced crestin mRNA expression through EGFP fluorescence (Fig. 1, B and C, and fig. S1A), and time-lapse videos demonstrated the dorsal emergence and wide migration of these crestin-expressing putative neural crest progenitor cells (movies S1 and S2). Neural crest expression was reproducible in multiple independent lines and with additional reporter genes (creERT2 and mCherry) (Fig. 1, D to G, and fig. S5, A to C). As with endogenous crestin expression, transgenic crestin:EGFP expression was not detectable after 3 days after fertilization and did not come back on in wild-type juvenile or adult zebrafish.

Fig. 1 The crestin promoter/enhancer drives neural crest-specific gene expression.

(A) Prototypical crestin retrotransposon locus with predicted ORF, LTR-like, and U3-like promoter regions. Shown are locations of 4.5- and 1-kb segments used for crestin:EGFP constructs (white box/promoter arrow indicate β-globin gene minimal promoter). (B) Endogenous expression pattern of crestin transcript by means of ISH (purple staining) at 24 hours after fertilization marks developing and migrating neural crest cells. (C) This expression pattern (green) is recapitulated by a stable Tg(crestin:EGFP) embryo at 24 hours after fertilization. (D to G) Genetic lineage tracing of cells that express crestin [Tg(crestin:creERt2;crystallin:YFP) X Tg(–3.5ubi:loxP-GFP-loxP-mCherry)] marks multiple neural crest lineages (red cells), including melanocytes (bracket) on (D) the dorsum and (E) the eye (72 hours after fertilization), (F) jaw cartilage (ventral view, 5 days after fertilization), and (G) glial cells of the lateral line (arrows, dorsal view posterior to the yolk, 72 hours after fertilization). (H) Tg(crestin:EGFP) expression overlaps substantially with a sox10:mCh transgene (confocal image, dorsal view over yolk, 24 hours after fertilization).

To confirm that the crestin transgenes target neural crest progenitors, we also generated transgenics for crestin:creERT2 to genetically mark crestin-expressing embryonic cells using a Cre/lox–dependent EGFP-to-mCherry switching line (“ubi:switch”) (9) and genetically labeled neural crest-derived cells, including melanocytes/pigment cells (Fig. 1, D and E, red cells), jaw cartilage (Fig. 1F), and lateral line glia (Fig. 1G). Because the crestin gene is specific to zebrafish, we wanted to ensure that crestin reporter embryonic expression is consistent with another conserved early neural crest marker, the transcription factor sox10. Confocal analysis of double-transgenic Tg(crestin_1kb:EGFP) and Tg(sox10:mCh) (10) zebrafish embryos showed a high degree of overlap in reporter gene expression (Fig. 1H), with any differences matching published in situ hybridization (ISH) data (11). Thus, our crestin transgenic lines recapitulate crestin expression and specifically mark the embryonic neural crest stem/progenitor cell population.

crestin transgenics visualize melanoma initiation

We next determined whether crestin:EGFP is reexpressed in melanoma tumors, as noted previously by ISH (8). We found cingstin:EGFP is expressed in tumors arising on triple transgenic p53/BRAF/crestin:EGFP adult zebrafish but is absent in the remainder of the animal, highlighting its specificity to the tumor (Fig. 2A). We next followed developing zebrafish in order to observe the onset of crestin:EGFP+ expression. Before EGFP-expressing patches of cells formed raised melanoma lesions on a given fish (Fig. 2B), we were able to detect single isolated EGFP+ cells in p53/BRAF/crestin:EGFP zebrafish (Fig. 2C). We could track their persistence and enlargement (fig. S2, A and B). Small patches of EGFP+ cells, containing <50 cells, are readily tractable as they enlarge (fig. S2C). Analysis of single scales with discrete crestin:EGFP+ patches demonstrated that transgene expression detectable with fluorescence microscopy overlaps with crestin mRNA detected with ISH (Fig. 2D). Together, these observations reveal that after pan-neural crest expression confined to the embryo, our crestin reporter expresses specifically and reproducibly in melanoma tumors, thus providing an in vivo genetic label for melanoma cells that is earlier than with previous detection methods (5, 12).

Fig. 2 Tg(crestin:EGFP) specifically marks melanoma tumors and precursor lesions.

(A) Spontaneously arising tumors (outlined) in p53/BRAF/crestin:EGFP zebrafish express EGFP (brackets), whereas the remainder of the animal is negative. (B) crestin:EGFP expression is also visible in precursor, nonraised lesions. (C) Example of a single crestin:EGFP+ cell in p53/BRAF background. (D) Scales expressing crestin:EGFP from precursor, nonraised regions [(B), bottom, arrow] were plucked, photographed [(D), left and middle], and subjected to ISH for crestin transcript [(D), right]. There is a concordance of EGFP (green) and crestin transcript (purple, dotted outlines, scales curl during ISH procedure, indicated by the curved arrow, observed in 5 of 5 scales). (Bottom right) crestin:EGFP scales are negative for crestin ISH staining (observed in 7 of 7 tested scales). (E) Cohorts of p53/BRAF/crestin:EGFP zebrafish were tracked over time for the appearance of crestin:EGFP+ patches and tumors, with crestin:EGFP+ cells/patches (green line) identifiable before raised melanoma tumors (black line). (F) Example of an EGFP+ preclinical patch tracked over time (6, 9, 11.5, and 17 weeks) as it expands into a clinically apparent melanoma tumor. (G) Scale autotransplant and expansion of crestin:EGFP+ patch of cells. At day 0, the recipient site is free of crestin:EGFP+ cells (pre–scale transplant), but immediately after transplant of a single scale (post–scale transplant), the patch of EGFP+ cells is apparent (white circle). This patch expands outward, and even upon removal of the original transplanted scale after the day 33 photograph, EGFP+ cells remain in place and continue to expand. The magnification and size of white circle is the same in each image.

We next addressed the dynamics of reemerging crestin expression in cohorts of p53/BRAF/crestin:EGFP zebrafish. At the population level, crestin:EGFP+ patches of cells (fig. S1, B and C) were visible before the appearance of grossly raised melanoma lesions (Fig. 2E, fig. S1D, and movie S3). The crestin:EGFP expression is undetectable in the p53/BRAF fish from 3 to >21 days after fertilization, which is again consistent with previous in situ analyses for endogenous crestin. We tracked individual small patches of crestin:EGFP+ cells over time as they progressed into fully formed raised melanoma lesions (Fig. 2E) and found that all melanomas tracked in this manner initiated from crestin:EGFP+ patches of cells (30 out of 30). Thus, if a patch is seen in the p53/BRAF background, it will become an overt melanoma. These data demonstrate that reemergence of crestin:EGFP expression, and a neural crest progenitor state, correlates with melanoma initiation in an in vivo model of de novo melanoma formation.

To establish that pretumor patches of crestin:EGFP+ cells are tumorigenic and can autonomously expand locally after transplant, we performed scale auto-transplants on p53/BRAF/crestin:EGFP zebrafish (13). After transplant, patches of crestin:EGFP+ cells survive and expand at the new site. The EGFP+ cells persist and further expand when later removing the transplanted scale, suggesting that the cells have invaded the hypodermis (Fig. 2G, representative example). We achieved similar results with isolated scales placed in tissue culture but on a shorter time scale (figs. S3A, and S4, A and B) and with allotransplants to sublethally irradiated recipient zebrafish (fig. S3B) (8). Thus, early patches of crestin:EGFP–expressing cells are transplantable in a manner suggesting that they are already tumorigenic.

Transcriptional regulators of crestin expression

As the crestin element proved to be a highly specific and distinct tool for monitoring neural crest and melanoma development, we aimed to identify (i) a minimal element within the 4.5-kb crestin promoter/enhancer that could drive this expression pattern and (ii) key transcriptional regulators within the element. Sequence analysis of the crestin locus, which is replicated throughout the zebrafish genome >40 times, is similar to another retroelement called bhikari that is expressed in early mesendoderm (fig. S6A) (6, 7, 14). Both a 1-kb segment from the putative retroelement promoter region and a smaller 296–base pair (bp) subregion fully reproduced the neural crest- and melanoma-specfic expression pattern of the full 4.5-kb crestin element (Fig. 1H and fig. S5, A to G), with slightly weaker expression for the 296-bp element. Hence, key neural crest regulatory elements are contained in this 296-bp of DNA, although additional contributory binding sites may also be functional in the context of the larger crestin element.

Database searches identified multiple predicted transcription factor binding sites for important neural crest developmental regulators within the 296-bp segment, including two sox10, one pax3, one E-box (myc or mitf-binding site), and one tfap2 site (Fig. 3A and table S1) (15). To determine which sites are functionally required for crestin transgene expression, we individually mutated the core consensus for each site (Fig. 3A and fig. S6B) and tested expression at 24 hours after fertilization in F0 embryos injected with the different crestin_296bp:EGFP constructs at the one-cell stage, examining >80 successfully injected F0 embryos per construct. Whereas mutation of the predicted pax3 site left the expression pattern largely unchanged from that of the wild type (~55% of embryos with neural crest predominant expression in both), mutation of either sox10 site drastically reduced neural crest expression (≤20%), as did mutation of the tfap2 site or the E-box site (<10% and none, respectively) (Fig. 3A and fig. S6, C and D). These functional transcription factor binding sites provide an explanation for the neural crest specificity of crestin transgene expression, which integrates regulatory signals of multiple neural crest transcription factors, including sox10.

Fig. 3 Reemergence of neural crest progenitor identity in melanoma initiation.

(A) Mutation of key neural crest transcription factor binding sites in the 296-bp crestin element, including sox10, tfap2, and an E-box for myc or mitf, substantially reduces neural crest EGFP expression at 24 hours after fertilization, whereas mutation of the predicted pax3 site does not alter expression. Coinjection of a ubiquitous ubi:mCh transgene confirmed successful injection for the >80 independently injected F0 embryos analyzed for each construct. Scales from p53/BRAF/Na/MiniCoopR/crestin:EGFP adult zebrafish (B) with and (C) without EGFP+ cells were collected, and total RNA was isolated for microarray analysis. (D) Quantitative RT-PCR of crestin:EGFP+ versus crestin:EGFP scales reveals enrichment of neural crest (crestin, dlx2a, sox10) and melanoma marker expression (crestin, mia, sox10). (E) GSEA analysis shows a positive association between crestin:EGFP+ patch-enriched genes and neural crest–expressed genes in zebrafish (left) and in human ES–derived neural crest cells (right). (F) Model for the importance of reemergence of NCP state through SE activation as an essential step in melanoma initiation. The acquisition of genetic lesions in normal tissue leads to oncogene activation (BRAFV600E) and tumor suppressor loss (p53−/−) and represents an initial barrier that generates a cancerized field from which rare clones (green) overcome the additional barrier of achieving a NCP state to initiate melanoma formation and then tumor expansion. Favoring reemergence of the neural crest progenitor state would then increase melanoma formation, and strengthening this barrier to inhibit adoption of the crestin+ NCP state would block melanoma initiation. (G) Misexpression of the NCP transcription factor sox10 accelerates melanoma onset as compared with controls in p53/BRAF/Na zebrafish rescued with the MiniCoopR construct.

Neural crest signature in melanoma initiation

To test whether early precursor melanoma lesions express other melanoma and neural crest progenitor markers in addition to crestin, we isolated individual scales from p53/BRAF/Na/MiniCoopR/crestin:EGFP zebrafish with early crestin:EGFP+ patches and compared them with adjacent individual scales without crestin:EGFP expression (Fig. 3, B and C) and performed Affymetrix microarrays (12). Such scales appear well-matched in their cell make-up, particularly in regards to melanocytes, as shown through mitf:mCh co-expression (marking melanocytes), and crestin:EGFP (fig. S7). Crestin:EGFP+ scale-enriched genes include neural crest– (such as crestin, mitf, and dlx2a) and melanoma-expressed [such as mia and mt (metallothionein)] genes (table S2) (1618). We confirmed enrichment by means of quantitative reverse transcription PCR (RT-PCR) on independent crestin:EGFP+ and crestin:EGFP scales (including crestin, dlx2a, and mia) (Fig. 3D). We also found sox10 expression enriched in the crestin:EGFP+ samples (Fig. 3D). Sox10 is a known marker and key regulator of neural crest identity (Fig. 1H) (19) and, from our data, of crestin expression (Fig. 3A); it is also sufficient to direct reprogramming of human fibroblasts to induced neural crest cells (20) and can be highly expressed in melanoma, where it is involved in growth control (2123). We used gene set enrichment analysis (GSEA) to query a rank-ordered list of the crestin:EGFP+ scale enriched genes for an association with all neural crest–expressed genes in the Zebrafish Information Network (ZFIN) database, and we found a significant correlation [false discovery rate (FDR) Q = 0.019, and familywise error rate (FWER) P = 0.019] (Fig. 3E, left, and table S6). Similarly, we used GSEA to compare a rank-ordered list of genes enriched in embryonic stem (ES)–derived human neural crest cells (24) with genes enriched ≥2-fold in crestin:EGFP+ scales, and we found a positive correlation, detectable even across species (FDR Q = 0.089, FWER P = 0.089) (Fig. 3E, right, and table S6). These data collectively support the concept that key aspects of NCP state reemerge at the time of melanoma initiation, as read out by the crestin:EGFP reporter.

Neural crest progenitor identity and melanomagenesis

On the basis of our analysis of crestin expression, which provides an in vivo readout of NCP identity at the time of melanoma initiation, we reasoned that favoring entry into or inappropriately maintaining the NCP state in a cancerized field of melanocytes would accelerate the onset of melanoma formation (Fig. 3F). The neural crest master transcription factor sox10 has been shown to increase crestin mRNA in embryos when overexpressed (25), and we found similar results for our crestin:EGFP reporter (fig. S8). Misexpression of SOX10 in postnatal fibroblasts also generates multipotent neural crest cells in culture (20). We therefore overexpressed sox10 in melanocytes using the transgenic MiniCoopR system (12) and found that sox10 overexpression in melanocytes accelerated melanoma onset significantly versus controls (Fig. 3G). To examine the consequence of sox10 inactivation, we used a melanocyte-specific CRISPR/Cas9 system to target sox10 in the p53/BRAF/Na background (fig. S9A). As compared with controls in which p53 is redundantly targeted (already mutated in our system) using an analogous vector, we found a significant slowing of median tumor onset in the sox10 CRISPR/Cas9 setting [133 days (p53) versus 180 days (sox10), P < 0.0001] (fig. S9B). When melanomas developed in the sox10-targeted background, the sox10 target genomic locus exhibited a propensity for mutations that preserve predicted sox10 function (for example, point mutations or in-frame deletions, ~60% of sequenced genomes) as opposed to inactivating mutations (for example, frame-shifts, ~40% of sequenced genomes), suggesting a selective pressure for retention of sox10 function (fig. S9, C and D). These gain- and loss-of-function results together strongly support our hypothesis that reemergence of NCP state is an important event in melanoma tumor initiation.

Neural crest super-enhancers and melanoma

In order to understand how the expression of neural crest genes such as sox10 may be regulated in zebrafish and human melanoma, we used a combination of chromatin immunoprecipitation sequencing (ChIP-seq) and assay for transposase-accessible chromatin using sequencing (ATAC-seq). Chromatin regions with high levels of H3K27Ac histone marks have been referred to as super-enhancers (SEs), or stretch-enhancers (26, 27), and have been identified as key transcriptional regulatory elements that modulate cell type–specific and cancer-related gene expression (2629). We used ChIP-seq to identify H3K27Ac-enriched regions in a zebrafish crestin:EGFP+ melanoma cell line (zcrest1) that we derived from p53/BRAF/Na/MiniCoopR/crestin:EGFP zebrafish and noted substantial regions of H3K27Ac enrichment at crestin loci, identified as SEs (representative locus shown in Fig. 4A, red bar indicates SE). We also identified Sox10 binding by means of ChIP-seq across the crestin locus (Fig. 4A, bottom track), which is consistent with our promoter analysis linking sox10 to crestin transcriptional regulation (Fig. 3A). We examined the sox10 locus in the zcrest 1 zebrafish melanoma cell line and also identified H3K27Ac SE marks (Fig. 4B). These SEs were similarly found at sox10 and crestin via H3K27Ac ChIP-seq performed on a freshly isolated primary zebrafish melanoma tumor (fig. S10, A and B, red bars), supporting our findings on the cell lines as being representative of the in vivo landscape. ATAC-seq identified open and accessible chromatin corresponding to the SEs at crestin and sox10 (Fig. 4, A and B) and other SE-associated loci in two zebrafish melanoma cell lines (fig. S10, E and F) (30). These data suggest a molecular basis for the epigenetic state readout by crestin of NCP identity in initiating melanoma cells.

Fig. 4 A SE signature in zebrafish and human melanoma.

(A) ChIP-seq for the H3K27Ac histone mark (top row) in a crestin:EGFP+ zebrafish melanoma cell line (zcrest 1) reveals enriched peaks, identified as a SE (red bar), at a representative crestin locus. Sequences of crestin_4.5kb, crestin_1kb, and crestin_296bp shown with blue horizontal bars. ATAC-seq on two zebrafish melanoma lines (zcrest 1 and zcrest 2) identifies open chromatin coincident with the H3K27Ac marks at crestin loci. ChIP-seq for Sox10 shows enrichment across the crestin locus (bottom row) in zcrest 1 cells. (B) ChIP-seq for the H3K27Ac histone mark (top row) on the zcrest 1 line identifies robust enrichment and a SE at sox10 (red bar). ATAC-seq identifies corresponding regions of open chromatin (rows 2 and 3). (C) ChIP-seq for the H3K27Ac mark on multiple SOX10-expressing melanoma lines (A375, CJM, COLO679, SKMEL2, SKMEL30, and UACC257) and a rare SOX10-negative melanoma line (LOXIMVI). Robust peaks corresponding to SEs (red bars) are identified in all lines, except the SOX10-negative LOXIMVI line. Published H3K27Ac ChIP-seq data from hNCCs reveal a similar SE pattern. ChIP-seq for H3K4Me1, an enhancer mark, on a representative melanoma line, A375 (row 2), identifies regions corresponding to the H3K27Ac marks. (D) H3K27Ac signal is robust at the DLX2 locus in melanoma cell lines not expressing the melanocyte differentiation genes TYR and DCT (blue box), in hNCCs, and in the SOX10-negative LOXIMVI melanoma line. Human genomic track images were generated at (E) High relative H3K27Ac signal at SOX10 (top) and DLX2 (bottom) identifies SEs (presence = red/orange bar, absence = blue bars) and is largely enriched in melanomas and hNCCs compared with 66 normal and 18 cancer cell types.

To compare our fish studies with human melanoma, for which SE analysis is limited, we examined the Cancer Cell Line Encyclopedia (CCLE) database (31) and found that most human melanoma lines (51 of 60) express SOX10 according to Affymetrix microarray data (fig. S11A). As with zebrafish melanomas, ChIP-seq showed enriched H3K27Ac marks near the SOX10 locus in six SOX10-expressing human melanoma lines tested but not in a rare SOX10-negative human melanoma cell line (LOXIMVI) (Fig. 4, C and E, and fig. S11A). The SOX10 SE’s were ranked 3 and 6 out of 842 SEs from ~15,000 total enhancers in the A375 line (fig. S11B and table S3). Clustering based on the SE landscape yielded two distinct groups of SOX10-expressing lines that correlated with the presence or absence of expression of melanocyte differentiation markers, TYR and DCT (fig. S11, C and D). H3K4me1, a histone modification typically at active enhancers, was also enriched at SOX10 in the representative A375 melanoma line (Fig. 4C). Remarkably, these SOX10 SE peaks were also found in published H3K27Ac data from human ES-derived neural crest cells (hNCCs) (Fig. 4, C and E) (32). Examining multiple normal and cancer cell types (66 and 18 types, respectively), the enrichment of H3K27Ac signal at SOX10 was evident and specific to melanoma cells, hNCCs, and brain tissue, which contains SOX10-expressing oligodendrocytes (Fig. 4E) (26). Beyond SOX10, a similar SE epigenetic signature was shared for the neural crest transcription factor DLX2 among melanomas across species and was enriched in melanomas and hNCCs (Fig. 4, D and E, and fig. S10, C and E). DLX2 expression is enriched in sorted crestin+ embryonic neural crest cells (table S4), in crestin+ precursor melanoma patches (Fig. 3D), and in the less differentiated, TYR/DCT melanomas relative to cultured normal human melanocytes (fig. S11, C, D, and F; and table S5). SEs were also found at TFAP2 family members, whose functional binding site was identified in the 296-bp crestin element (Fig. 3A), in A375 human melanoma (TFAP2C) (fig. S12A), and in the zebrafish melanoma cell line and primary melanoma (tfap2a) (fig. S10, D and F). Such SEs are also found near other crestin+ cell-enriched genes MIA and MT2A (metallothionein genes) in A375 melanoma cells (fig. S12, B and C).


This work shows that melanoma precursor cells reinitiate an embryonic neural crest signature and activate a melanoma gene program. Although the expression of stem cell factors or embryonic genes has been noted previously in advanced malignancies including melanoma (8, 23), it remained uncertain whether this was simply due to aberrant misexpression of these genes or whether these genes were present from tumor outset (Fig. 3F). Our data support a model in which the stem/progenitor cell gene programs are an integral part of cancer initiation and not reacquired later (Fig. 3F). Analogously, this theme was also recently described in basal cell carcinoma (BCC), a cancer arising from a different cell lineage, in which careful analysis of gene expression from tissue collected during early BCC formation in a mouse model of the disease exhibited an embryonic hair follicle progenitor signature, indicating that this may be a general feature of cancer initiation (33). There also is a conceptual similarity between the reprogramming-induced pluripotent stem process and the tumorigenic reemergence of NCP state seen in our system as melanocytes with the required, but insufficient, p53/BRAF genetic mutations stochastically reenter a NCP state (as read out by crestin expression) early at the initiation of melanoma. As the initiating stages of more cancers are analyzed, stem cell/progenitor phenotype reacquisition may be a generally observed phenomenon in most cancers (33).

Our work establishes a method to live-image cancer development when a tumor starts, potentially as a single cell, providing a singular view of the initiating events. NCP identity arises, and is likely a necessary step, early in tumor development. In our p53/BRAF/crestin:EGFP model, all EGFP+ patches of crestin:EGFP cells that we track go on to enlarge and form tumors; given the rarity of capturing single cells in this background, we cannot rule out that clones may rarely fail to progress to patches. The crestin gene has the distinct characteristic of being expressed in neural crest cells in embryos, but not being expressed in adult tissues except when melanoma arises. Super-enhancers shared between melanoma and neural crest are specifically activated. The neural crest signature encompasses a combinatorial code including SOX10, DLX2, and the TFAP2 family. It is unlikely that any one of these genes is sufficient for the reprogramming event, but it is the combination of multiple transcription factors that participate. In contrast to SOX10, DLX2 expression is characteristic of genes that are expressed by neural crest and melanoma, but not expressed in melanocytes. As such, this gene set may have diagnostic importance in the initiating cancer cell. A reevaluation of published expression data for nevi, primary melanoma, and metastatic melanoma does reveal increased SOX10 expression in malignant melanomas (34). Nevi show a wide range and consistently lower amount of SOX10 expression than that of melanoma, raising the possibility that the histologically defined category of nevus is capturing a range of melanoma-initiating capacity (those nevi expressing higher SOX10 levels may have initiated or may be more prone to initiating melanoma).

Several major questions remain about an initiating cancer cell that may now be more accessible with our live visualization tool. The niche environment must participate in the process of initiation, perhaps through the activation of neural crest signaling pathways akin to the development of the normal neural crest or by stress pathways related to irradiation or oxidative damage. Further genetic mutations, as opposed to isolated epigenetic changes, may be required for tumor initiation from the cancerized field; however, the absence of identifiable functionally relevant exomic mutations in a study of 53 zebrafish melanoma tumors to which we contributed would tend to favor that the key genetic drivers (BRAFV600E and mutant p53) are already present in our model (35). Work on this question will provide information on how cancer initiates. The reprogramming event appears to occur in one melanocyte or progenitor in a cancerized field, and defining why the process initiates in that single cell rather than an adjacent cell will provide an understanding of protective mechanisms in cancer formation.

Materials and methods

Cloning of crestin promoter/enhancer

The transcript for crestin, originally described as AF195881, was used with Basic Local Alignment Search Tool (BLAST) on the zebrafish genome and, as has been previously noted, identified many partial or complete highly similar (>90% identical) sequences spread throughout the genome (6, 7). Multiple insertions are present on chromosome 4, and a ~4.5-kb sequence located upstream of the predicted crestin open reading frame (ORF) was noted to be present in multiple instances. Reasoning that this segment may contain the relevant regulatory elements of crestin, primers were designed to amplify via PCR upstream sequences of the crestin locus (LOC796814) on chromosome 4 in the TU background (primers 299 and 302; primers supplemental table). The 1-kb fragment and 296-bp fragment were isolated by using PCR primers 302/340 and 517/516, respectively. Fragments were cloned into pENTR5′ (Life Technologies, Grand Island, NY) per manufacturers instructions. Expression vectors were derived by using Multisite Gateway technology per manufacturer’s specifications and the Tol2 Kit (36) [EGFP, vector 383; mCh, vector 386; SV40 poly A, vector 302; EGFP with mouse minimal ß-globin promoter (37); destination vectors, vector 394 alone and modified with addition of crystallin:YFP marker; zmitfa middle entry vector, gift of Craig Ceol; MiniCoopR from (15)].

Production of transgenic zebrafish and lineage tracing

One-cell-stage embryos from the AB strain grown under standard, Institutional Animal Care and Use Committee (IACUC)–approved conditions were injected with the given DNA construct at 25 ng/μl with Tol2 mRNA at 20 ng/μ (36). Embryos were screened at 24 hours after fertilization for neural crest expression of EGFP or mCh, or in the case of Tg(crestin:CreERt2;crystallin:YFP), screened for yellow fluorescent protein (YFP)–positive lenses at 4 days after fertilization. These were grown to adulthood and outcrossed so as to identify founders that gave germline transmission. For each DNA construct, ≥2 independent lines were generated to confirm the expression pattern. For lineage tracing of crestin-expressing cells, a stable transgenic line of Tg(crestin:creERt2;crystalline:YFP) was crossed to the Tg(-3.5ubi:loxP-GFP-loxP-mCherry) line (ubi:Switch) (9); embryos were collected and treated with 10 μM 4-hydroxytamoxifen (4-OHT) at 50% epiboly and 24 hours after fertilization while grown at 28°C in E3 medium, as per standard protocol.

Embryo and adult imaging

Transmitted light and fluorescence images of adult and nonconfocal images of embryos were collected on a Zeiss Discovery V.8 Stereoscope with an Axiocam HRc. Static confocal images were collected on a Nikon C2si Laser Scanning Confocal using 25× objective on embryos mounted in 1% low melt agarose. Maximum intensity projections of Z stacks or three-dimensional reconstructions are presented here. Movies of developing embryos were collected on a Nikon Eclipse Ti Spinning Disk Confocal with a 10× objective, with tiled images collected every 6 to 7 min. Images were processed by using Photoshop, ImageJ, or Imaris. Multiple tiled images of adult zebrafish were stitched together by using the automated Photomerge function in Photoshop or manually aligned.

Melanoma model and MiniCoopR system

Experiments were performed as outlined in (15). Briefly, p53/BRAF/Na embryos were injected with equal amounts of MiniCoopR alone or MiniCoopR;mitf:sox10 and selected for melanocyte rescue at 48 hours. Equal numbers of melanocyte-rescued embryos were grown to adulthood (n = 14 for control and n = 13 for mitf:sox10) and scored for the emergence of raised melanoma lesions as per (15). Survival curves and statistics were generated in Prism.

Quantitative PCR, microarray, and in situ hybridization

Adult p53/BRAF/Na/MiniCoopR/crestin:EGFP fish were anesthetized with Tricaine, viewed under the fluorescent dissecting scope, and precursor patches of crestin:EGFP+ cells were identified and associated single scales removed. These fish contained stable (germ-line transmitted) alleles of crestin:EGFP and control MiniCoopR (no test gene). Neighboring crestin:EGFP–negative scales from the same zebrafish were selected for controls. Scales were immediately placed in Trizol, and total RNA was purified following the manufacturer’s instructions, with the additional use of Sigma GenElute LPA carrier. RNA was analyzed on a BioAnalyzer, and high-quality samples were chosen for microarray libraries generated by using the Ovation Pico WTA System V2 and Encore Biotin labeling system for hybridization on the Zebrafish Gene 1.0 ST Affymetrix Array or used for cDNA preparation using SuperScript III. RMA-normalized Affymetrix Array results were sorted for maximum fold (log2) increase in gene expression in GFP-positive samples versus negative samples. Quantitative PCR reactions were run on triplicate biological samples with triplicate technical replicates and normalized to β-actin expression, with representative results presented. For in situ hybridizations, we used the methods of (38) and the crestin probe from (7).

ChIP-seq and SE analysis

Human melanoma cells were grown to confluence in Dulbecco’s Modified Eagle medium (DMEM) + 10% fetal calf serum (FCS), and ~ 1 × 108 cells were formaldehyde cross-linked and collected. ChIP was performed by using the methods of (30, 39) with antibodies against H3K27Ac (ab4729) (Abcam), H3K4Me1 (ab8895) (Abcam), and SOX10 [Santa Cruz Biotechnology (Dallas, TX), sc-17342x]. Libraries were prepared by using the NEBNext Multiplex Oligos for Illumina kit (NEB) and run on an Illumina HiSeq 2000. Data analysis, including enhancer and SE calling, was performed as described in (26). Genomic track images were generated by using the IGV package (40) and the University of California, Santa Cruz Browser (41).

All human ChIP-seq data sets were aligned to build version NCBI37/HG19 of the human genome using Bowtie (version 0.12.9) (42) with the following parameters: –n2, –e70, –m2, –k2, –best. We used the MACS version 1.4.1 (Model based analysis of ChIP-seq) (43) peak finding algorithm to identify regions of ChIP-seq enrichment over background. A P value threshold of enrichment of 1 × 10–9 was used for all data sets. Wiggle files for gene tracks were created by using MACS with options –w –S –space = 50 to count reads in 50-bp bins. They were normalized to the total number (in millions) of mapped reads producing the final tracks in units of reads per million mapped reads per base pair (rpm/bp).

Identifying SEs

The identification of SEs has previously been described in detail (26). Briefly, H3K27Ac peaks were used to identify constituent enhancers. These were stitched if within 12.5 kb, and peaks fully contained within ±2kb from a TSS were excluded from stitching. H3K27Ac signal (less input control) was used to rank enhancers by their enrichment. Super-enhancers were assigned to active genes by using the ROSE software package (


Zebrafish melanoma cell lines were grown to 80% confluence, trypsinized, and counted, and 50,000 cells were lysed and subjected to “tagmentation” reaction and library construction as described in (30). Libraries were run on an Illumina HiSeq 2000. All zebrafish ATAC-seq data sets were aligned to build version Zv9 of the zebrafish genome by using Bowtie2 (version 2.2.1) (42) with the following parameters: –end-to-end, –N0, –L20. We used the MACS2 version 2.1.0 (43) peak-finding algorithm to identify regions of ATAC-seq peaks, with the following parameter–nomodel–shift –100–extsize 200. A Q value threshold of enrichment of 0.05 was used for all data sets.

Scale transplants

Adult p53/BRAF/Na/MiniCoopR/crestin:EGFP were anesthetized, viewed under the fluorescent dissecting scope; precursor patches of crestin:EGFP+ cells were identified (not from raised melanoma lesions); and single associated scales were removed and placed in 50-μl drops of E3 buffer on a petri dish lid. Anesthetized recipient zebrafish (in the case of allotransplants) or the same zebrafish (in the case of autotransplants) were gently placed on a wet sponge and a recipient site selected (free of crestin:EGFP–expressing cells). One scale was removed at the donor site, and the previously selected donor scale from the drop of E3 was placed on the zebrafish and slid posterior to anterior into place by using surrounding scales to hold it in place. Recipients were quickly placed in fresh zebrafish water and monitored for recovery from anesthesia. Transplants were monitored frequently with some loss of transplanted scales (~20%) occurring quickly within 1 day because of simple dislodgement. Once in place for ~4 days, scales were firmly incorporated and could be monitored over time and photographed at the same magnification under the dissecting scope after mild Tricaine anesthesia of the recipient zebrafish. In single-scale autotransplants in a representative cohort of 10 fish, three scales were lost in the first week (we consider a technical failure with the scales falling out), five showed expansion of the crestin:EGFP cells, and three showed no change or loss of crestin:EGFP cells. In single-scale allotransplants onto sublethally irradiated casper recipient fish, eight EGFP+ scales were transplanted onto different recipients with one scale lost in the first week, five scales showing expansion of the crestin:EGFP cells for ≥2.5 weeks, and two showing stable appearance for ≥ 2 weeks. Six of six EGFP scales showed loss of pigmented cells in this time frame.

Promoter analysis

Transcription factor binding sites were predicted by using JASPAR (14). We used the Q5 Site-Directed Mutagenesis Kit from NEB to introduce mutations to destroy the chosen transcription factor binding site (primers supplemental table). To analyze expression in vivo, equal volumes of a mixture of the crestin_296bp:EGFP construct variant at 20 ng/μl mixed with 5 ng/μl of ubi:mCh (9, 36) (for an injection control) were injected with 20 ng/μl of Tol2 mRNA into single-cell AB embryos. At 24 hours after fertilization, embryos were fixed in paraformaldehyde (PFA), washed and stored in phosphate-buffered saline (PBS), and scored for mCh expression to identify successfully injected embryos and for EGFP to bin based on the predominant expression pattern. More than 80 transgenic F0 embryos for each construct were scored.

Investigating SOX10 and DLX2 SEs in healthy normal cells and melanoma cells

To investigate whether the two SOX10-asscociated SEs identified in A375 cells are also present in other cell types, they were compared with the SEs identified in other melanoma cell types, neural crest cells, and 84 additional cell types from normal or cancer cells described in (26).

First, the SOX10-asscociated SEs were operationally called “present” in a cell type if the SEs identified in the cell-type overlap with the SOX10 SEs in A375 cells by at least 1 bp. Second, the H3K27Ac signal density of the SOX10-asscociated SEs identified in A375 cells was compared with the mean H3K27Ac signal density of all enhancer clusters in each cell type. For each cell type, the average H3K27Ac ChIP-seq read density was calculated in rpm/bp for the two SOX10 SEs identified in A375 cells as well as all enhancers clusters identified by using the ROSE software package in the cell type ( The fold difference of H3K27Ac ChIP-seq signal at SOX10 SEs over the mean H3K27Ac ChIP-seq signal at all enhancers in each cell type was plotted in Fig. 4E. The same process was undertaken for the SEs at the DLX2 locus.

Pairwise comparison of SEs between different melanoma cell lines and neural crest cells

The set of SE regions in each cell type (SKMEL2, SKMEL30, UACC257, LOXIMVI, A375, CJM, COLO679, and neural crest cells) were merged together if overlapping by 1 bp, resulting in a total of 3407 merged SE regions. The neural crest cell data were previously published in (32). The average H3K27Ac ChIP-seq read density was calculated in rpm/bp for each of the merged regions. The pair-wise comparisons by Pearson correlation were performed on all data sets by using the average read density at the merged regions. The average linkage hierarchical clustering of the Pearson correlation was shown in the heatmap (fig. S11C).

Zebrafish melanoma cell lines and in vitro scale imaging

A single EGFP+ melanoma tumor arising in a p53/BRAF/Na fish injected with MiniCoopR plasmid and crestin:EGFP plasmid (zcrest 1 line) or a p53/BRAF/Na/crestin:EGFP fish injected with MiniCoopR plasmid (zcrest 2 line) was removed after killing the adult fish. Briefly, tumors were dissociated with a razor blade and trypsin, filtered, and plated on a fibronectin-coated well and grown in rich media supplemented with FBS and zebrafish embryo extract as described (44). After several passages, the zcrest 1 line was sorted for EGFP+ cells, which were continued as the line. For the zcrest 2 line, after several passages, most if not all cells remaining were EGFP+. Both lines continue to be EGFP+ and have been grown for >50 passages on plastic in standard DMEM + 10% FBS with 1X GlutaMAX supplement and penicillin/streptomycin antibiotics.

Scales from p53/BRAF/Na/crestin:EGFP/MiniCoopR and p53/BRAF/crestin:EGFP fish with and without EGFP+ patches of cells were placed in zebrafish melanoma growth medium in fibronectin coated wells in 384-well format, flat-bottomed plates and imaged daily on a Yokogawa CV7000 confocal imager with brightfield and z-stack image projections collected.

GSEA analysis

Using our microarrays comparing crestin:EGFP+ versus crestin:EGFP scale gene expression, zebrafish genes were rank-ordered (10,705 genes) from high to low for enrichment (crestin scales rank.rnk). We generated a gene list of all neural-crest–expressed genes in the ZFIN database (317 genes, zfincomp.gmt) (45) and used the preranked GSEA analysis tool. Using a list of genes with ≥2-fold enrichment in crestin:EGFP+ scales by means of microarray (Crestin Scales.gmt), we queried a list of 17,575 genes rank-ordered for their enrichment in human neural crest cells versus parental ES cells from (32) (Rada Ranked.rnk), also using the preranked GSEA analysis tool (46).

RNA-seq analysis

Multiple human melanoma cell lines and adult human epidermal melanocytes (purchased from Life Technologies) were grown to near confluence, and total RNA was isolated by using the standard Trizol protocol. Illumina libraries were prepared by using Ribo-Zero Magnetic Gold Kit (epicenter) and NEBNext Ultra RNA Library Prep Kit (NEB) and run on a HiSeq 2500, reads aligned by using Tophat 2.0, and fragments per kilobase of exon per million fragments mapped values determined by using Cufflinks. For sorted crestin+ cells, transgenic crestin_1kb:EGFP adults were mated, and embryos collected and grown to the 15 somite stage. These were homogenized, filtered, and sorted by using fluorescence-activated cell sorting into PBS, collecting ~5500 EGFP+ cells and 100K EGFP cells. Total RNA was again collected by using Trizol and GenElute LPA carrier per manufacturer instructions. Libraries were prepared by using Ribogone kit (Clontech) and the SMARTer Universal Low RNA Kit (Clontech) and sequenced on the Illumina HiSeq 2500, with post-analysis performed as above with the zebrafish genome.

CRISPR/Cas9 experiment

Cas9 mRNA was produced by means of in vitro transcription from a pCS2 Cas9 vector (47) by using mMESSAGE mMACHINE SP6 kit (Invitrogen). Guide RNAs (gRNAs) were generated by following established methods (48). The Sox10 target sequence was GGCCGCGCGCAGGAAACTGG. Six hundred picograms of Cas9 mRNA and 25 pg of gRNA were injected into embryos of the AB strain. After microinjection, embryos were raised in E3 medium at 28.5°C. The T7E1 assay was performed as reported (49). Briefly, genomic DNA was extracted from 2-day-old embryos by using the hotSHOT method (50). A fragment of 434 bp was amplified from genomic DNA by using the following primers: GAAGTCCGACGAGGAAGAT and CTTGACTGAGTAAATAGTGCGT. The PCR amplicons were then purified on a 1% agarose gel. Two hundred nanograms of purified DNA were denatured at 95°C for 5 min and slowly reannealed before digestion with 10 units of T7E1 enzyme (NEB) for 1 hour at 37°C. The digestion product was finally run on a 2.5% agarose gel.

CRISPR/Cas9 tumor-free survival curves

In order to inactivate sox10 specifically in the melanocytes of our zebrafish melanoma model, the MiniCoopR vector was engineered to express Cas9 under the control of the melanocyte-specific mitfa promoter and a gRNA efficiently mutating sox10, described above, off a U6 promoter. A gRNA against p53 was used as a negative control (51). The two vectors were injected into one-cell stage, Tg(mitfa:BRAFV600E), p53−/−, mitfa−/− embryos, and tumor formation was monitored.

To sequence genomic DNA from tumors, tumor tissue was dissected carefully, digested in buffer with proteinase K (52), and after inactivation of proteinase K, PCR was performed by using the primers described above as for the T7E1 reaction. PCR fragments were cloned by using TopoTA cloning per manufacturer instructions, and colony PCR was performed on resulting individual clones and submitted for Sanger sequencing for fig. S9C. For next-generation sequencing in fig. S9D, nested PCR (primer sequences TGAACGGGTACGACTGGACGCT and TGTTGTAGCAGTGCGTTTA, yielding a 238-bp amplicon) was performed on the initially amplified genomic locus so as to bring the amplicon ends closer to the CRISPR target sequence in order to allow for coverage by using a MiSeq-based 150-bp paired-end Illumina run with pooled and barcoded samples at the MGH DNA Core. Both Sanger and compiled next-generation sequences were aligned to the wild-type locus by using Lasergene Seqman in order to identify changes at the CRISPR target sequence. Wild-type sox10 reads, although potentially from nontargeted sox10 loci in melanocytes and not necessarily from other tissue types in the sample, were excluded from the calculations of fractions of allele types. If included, these would only increase the fraction of active sox10 alleles and would further favor our interpretation of the results. When determining the fractions of alleles in fig. S9D, each tumor was weighted equally so as to avoid skewing from more reads from a given tumor.

Injection of sox10 mRNA

The sox10 cDNA from zebrafish was cloned into pENTR/D-TOPO and transferred into pCSDest (53) by using Gateway cloning, all per manufacturer instructions. mRNA was generated by using SP6 mMessage mMachine Kit (Ambion) per manufacturer, and 1 nl mRNA mix was injected into single-cell p53/BRAF/crestin:EGFP embryos at multiple concentrations, with 20 pg being the highest tolerated dose without substantial toxicity. Embryos were imaged on a Zeiss Discovery V.8 Stereoscope and scored for EGFP expression.

Supplementary Materials

Figs. S1 to S12

Tables S1 to S6

Movies S1 to S3

References and Notes

Acknowledgments: C.K.K. and research reported here is supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases of the National Institutes of Health under award K08 AR061071. C.M. received support from a European Molecular Biology Organization long-term fellowship, a Human Frontier Science Program long-term fellowship, and a Swiss National Science Foundation advanced postdoctoral fellowship. L.I.Z. is supported by R01 CA103846, the Ellison Foundation, the V Foundation, and the Melanoma Research Alliance and is a Howard Hughes Medical Institute Investigator. R.A.Y. is supported by NIH HG002668. Z.P.F. was supported by NIH HG002668. E.J.H. is a Howard Hughes Medical Institute Fellow of the Helen Hay Whitney Foundation. L.I.Z. is a founder and stock holder of Fate, Inc. and Scholar Rock. R.A.Y. is a founder of Syros Pharmaceuticals. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank the Massachusetts General Hospital DNA Core for providing technical assistance with deep sequencing. We thank J. Ablain for critical review of the manuscript, C. Ceol for the gift of the zebrafish mitfa middle entry clone, and G. Musso for helpful discussions on microarray data and GSEA. All microarray, Chip-seq, and RNA-seq are deposited at Gene Expression Omnibus, ID GSE75356.

Stay Connected to Science

Navigate This Article