Introduction

Efficient cell type-specific knockout of genes is crucial for unraveling the intricate molecular mechanisms underlying cell function, development, and transformation. Zebrafish (Danio rerio) have emerged as a powerful genetic model organism for studying these processes. The large number of animals that can be easily generated make it advantageous (compared to other systems like mice) for querying large numbers of genetic variants that may have links to human diseases. However, most studies in this system use germline knockout, which makes it difficult to determine whether the identified gene has a specific function in a particular lineage, or whether the phenotype reflects a more general developmental defect. Furthermore, many genes which might have lineage specific functions cannot be studied using germline knockout, since embryonic lethality often accompanies global knockout of essential genes¹. To overcome these limitations, several groups have explored tissue-specific and conditional knockout strategies in zebrafish^2–10. For example, the MAZERATI system uses tissue-specific promoter transgenes to drive Cas9 expression, which are introduced via plasmid injection into one-cell stage embryos^3,11. Despite these advancements, efficiency remains of paramount importance. In normal development, low-efficiency knockout of a gene can mask important phenotypes, as the mutant cells can readily be outcompeted by wild-type or inframe mutations. Efficiency is even more important in the cancer context. Mutation of a tumor suppressor (i.e., PTEN) need not be highly efficient, since clones with a Darwinian advantage will inevitably outcompete other clones. However, in situations where we want to study necessity of a tumor promoting gene in vivo (i.e., an oncogene), cells which have mutations of a key driver gene will easily be outcompeted by nearby clones. Such efficiency is critical as sequencing projects such as All of Us, the UK Biobank and the Cancer DepMap continue to identify hundreds of new candidate genes on a regular basis, many of which have yet to be characterized in vivo^12–14.

Skin color pigmentation is amongst the most heterogeneous of genetic systems, with a large number of loci linked to melanocyte development. Mutations in pleiomorphic genes such as sox10, ednrb, jam3b, tuba8l3, meox1 and kit yield clear defects in melanocyte/melanophore development, but these mutants also have defects in other tissues^15–21. Many of these genes are also expressed in melanoma, but their specific in vivo function in tumorigenesis remains poorly understood.

In this study, we wished to develop a highly efficient and scalable system to understand in vivo genetic dependencies, using melanocytes and melanoma as an exemplar. By integrating the Cas9 nuclease into the endogenous melanocyte-specific mitfa locus, we achieved highly efficient and specific gene disruption across the melanocyte developmental lineage. We used our mitfa^Cas9 knockin animals to inactivate the pigmentation gene albino and the embryonic essential genes sox10, tuba1a, and ptena/b in melanocytes without additional developmental or off-target phenotypes. We also demonstrated that this system can be used to efficiently induce melanomas in wild-type fish by inactivating the tumor suppressors tp53 and ptena/b within melanocytes also expressing the melanoma oncogene BRAF^V600E. Inactivating the neural crest transcription factor sox10 in these tumors significantly decreased melanoma initiation but resulted in rare invasive tumors that highly expressed the sox9 transcription factor, highlighting that unexpected in vivo phenotypes can emerge that aren’t predicted from in vitro studies^22,23. Our system can thus be used to advance our understanding of melanocyte biology and the genetic underpinnings of phenotypic switching in melanoma, an approach that can readily be used for other cell types.

Results

Targeted integration of Cas9 to the mitfa locus

To selectively disrupt genes in a melanocyte lineage-specific manner, we established a transgenic zebrafish line harboring Cas9 at the endogenous mitfa locus. Mitfa expression is detectable by 18 hours postfertilization (hpf) and is highly expressed in pigment cells including melanocytes, xanthophores and their progenitors. To achieve knock-in, we utilized the GeneWeld method, which enables efficient integration of DNA cassettes using homology mediated end joining (HMEJ)^24,25. The knock-in cassette, containing Cas9 and BFP driven by the eye-specific γ-crystallin promoter, was targeted to exon 2 of the mitfa gene, which leads to disruption of the endogenous mitfa gene (Figure 1A). We strategically selected this genomic site for several reasons. First, this is the sole exon shared among the three mitfa protein-coding transcripts. Furthermore, targeting Cas9 to exon 2 offers the flexibility of replicating this knock-in strategy in casper zebrafish, a transparent fish widely used for melanoma studies that harbor a premature stop codon in mitfa exon 3²⁶.

Figure 1.

The knock-in vector, along with Cas9 protein and guide RNAs targeting either the mitfa integration site or the sequences flanking the 5’ and 3’ homology arms, were injected into 1-cell stage wild-type Tropical 5D (T5D) zebrafish embryos (Figure 1B)²⁷. An average of 19% of injected embryos had detectable ocular BFP expression at 3 days postfertilization (dpf), indicating successful integration (Figure S1A). Embryos were raised to adulthood then crossed with WT T5D fish. The resulting F1 embryos were screened for BFP+ eyes and tail clippings were sequenced to confirm precise integration of the knock-in cassette at the mitfa locus. We sequenced four F1 fish and detected on-target integration at the mitfa locus in all four fish (Figure S1B). In zebrafish, one copy of mitfa is sufficient for the normal development of neural crest-derived melanocytes, but loss of both copies of mitfa results in loss of the melanocyte lineage^26,28. To determine the effect of a single copy of mitfa^Cas9 on melanocytes, mitfa^Cas9 fish were in-crossed and the resulting siblings were imaged at various time points (Figure S1C). WT and mitfa^Cas9 fish exhibited indistinguishable melanocyte patterning during both embryonic and adult stages. However, as expected, fish carrying two copies of Cas9 (mitfa^Cas9/Cas9) resembled the nacre mutant (mitfa-/-), characterized by complete loss of melanocytes, indicating the endogenous mitfa gene is disrupted on knock-in alleles²⁸. We used zebrafish harboring one knock-in allele (mitfa^Cas9) for all subsequent experiments to allow us to test gene function in melanocytes.

To assess whether Cas9 expression was restricted to melanocytes, we performed fluorescence in situ hybridization (FISH) on 3dpf WT and mitfa^Cas9 embryos to measure mitfa and Cas9 mRNA expression (Figure 1C and S1D). In both WT and mitfa^Cas9 fish, mitfa was detected within the embryonic melanocyte stripe regions as expected. In mitfa^Cas9 fish, Cas9 mRNA was detected only within melanocyte stripe regions also expressing mitfa. These results provide functional validation for on-target integration and tissue-restricted expression of Cas9 to the melanocyte lineage.

Melanocyte-specific loss of the human skin color associated gene slc45a2 results in robust loss of pigmentation

To assess the effectiveness of our stable mitfa^Cas9 line, we first targeted the albino gene (slc45a2). Large scale GWAS studies have repeatedly identified SNPs in this gene as tightly linked to human skin color variation, and germline disruption leads to a complete loss of pigmentation in melanocytes^29–31. This distinct phenotype enables the visual observation of Cas9 activity in zebrafish injected with albino gRNA. To test our system, we designed plasmids, denoted MG-gRNA, containing a zU6:gRNA cassette followed by mitfa:GFP to enable visualization of cells expressing the albino gRNA. MG-albino and MG-NT were injected into one-cell stage embryos from crosses between mitfa^Cas9 and WT fish, which express Cas9 in all melanocytes but exhibit mosaic gRNA expression (Figure 2A).

Figure 2.

Embryos were screened for BFP+ eyes to identify those harboring mitfa^Cas9. Upon adulthood, 92% of F0 MG-albino fish had mosaic loss of pigmentation within the melanocyte stripe regions (Figure 2B and C). Unpigmented regions maintained GFP expression (Figure 2B), indicating that the breaks in melanocyte stripes arise from loss of pigmentation in melanocytes rather than loss of the melanocyte lineage (Figure 2B). As expected, control MG-NT clutch mates displayed normal pigmentation and GFP+ melanocytes were indistinguishable from GFP-regions on the same fish (Figure 2B and C).

We next aimed to determine the efficiency of mitfa^Cas9 in the F1 generation, in which every melanocyte stably expresses the albino gRNA and mitfa^Cas9. To generate F1 lines, we outcrossed the F0 MG-gRNA fish to WT and sorted embryos for BFP+ eyes and GFP+ melanocytes. F1 MG-albino fish exhibited a near complete loss of pigmentation, demonstrating that our mitfa^Cas9 system is highly efficient in F1 fish (Figure 2D and E). To study the effect of albino inactivation across various stages of development, F1 MG-gRNA fish were once again outcrossed to WT, generating F2 MG-gRNA fish expressing one copy of mitfa^Cas9. Imaging revealed a measurable pigmentation loss in F2 MG-albino melanocytes as early as 3 days post-fertilization (dpf) (Figure 2F and G). F2 MG-albino fish maintained loss of pigmentation throughout development, indicating the sustained activity of mitfa^Cas9 (Figure S2A-D).

To confirm on-target cutting of the albino gene in the melanocyte lineage (and not other tissues), the skin from F1 MG-albino fish was dissected and GFP+ and GFP-cells were isolated using FACS (Figure 2H). GFP+ cells encompass mitfa-expressing melanocytes, xanthophores, and their progenitors, while GFP-cells represent other non-melanocytic, non-mitfa expressing skin cells, such as keratinocytes, fibroblasts, and immune cells. We performed CRISPR-seq on DNA from both populations to directly measure Cas9 efficiency and specificity, quantifying the proportion of indels at the albino locus. Although we have previously shown that FACS sorting melanocytic cells is possible, it is important to note that we often observe high levels of contaminating keratinocytes in FACS sorted melanocyte populations, which may lead to an underestimation of the true allelic frequency in our CRISPR-seq³². Despite this, robust inactivation of the albino gene was observed in GFP+ (mitfa-expressing) cells, confirming somatic inactivation (Figure 2I). The majority of indels were frame-shift inducing mutations. A likely PCR or sequencing artifact resulting in a “T” insertion was excluded from the frameshift indel percentage as it was observed in all conditions including WT fish with no pigmentation phenotype (Figure S2E). These results confirm that our mitfa^Cas9 system allows us to inactivate genes within the melanocytic lineage in vivo in a highly efficient and specific manner.

The neural crest-related gene sox10 has specific function in adult melanocyte patterning and regeneration

Having confirmed that our mitfa^Cas9 system allows us to rapidly, efficiently and specifically manipulate melanocyte gene expression, we next wanted to target embryonic essential genes. Detecting “negative” gene knockout phenotypes such as cell death can be challenging in vivo due to the selective advantage that WT cells have over mutants. Current tissue specific knockout systems lack the sensitivity to detect negative phenotypes.

To assess the effectiveness of our mitfa^Cas9 model in detecting negative phenotypes, we utilized it to target sox10, a transcription factor vital to the neural crest lineage and indispensable for overall survival of both zebrafish and mice^15,33. Consequently, a global knockout approach is not feasible for studying the function of sox10 in adult melanocytes in vivo. Germline knockout of sox10 in zebrafish leads to severe defects in peripheral nerve development and the complete absence of melanocytes, and the animals die around day 14^15,34,35. Due to the complete absence of melanocytes and their precursor cells, melanoblasts, in sox10^-/- fish, it has not been possible to study the specific role of sox10 on these more differentiated cell types in vivo. One of the key unanswered questions is how sox10 balances its role in maintaining melanoblast stemness with promoting terminal differentiation, particularly in the context of adult melanocyte regeneration.

We leveraged our mitfa^Cas9 system to specifically inactivate sox10 in the melanocyte lineage. mitfa^Cas9 fish injected with a MG-sox10 plasmid exhibited noticeable gaps in their melanocyte stripes (Figure 3A-B). Germline deletion of a sox10 enhancer region similarly affects adult stripe patterning in zebrafish³⁶. In contrast to the albino F0 fish, the gaps in the stripes of the sox10 F0 fish were not occupied by unpigmented GFP+ melanocytes, indicating a potential defect in the survival or differentiation of melanocytes upon loss of sox10. This phenotype became even more pronounced in the stable F1 lines, where several gaps are visible (Figure 3C). Zebrafish have four dark stripes occupied by melanocytes, which alternate with light colored interstripes occupied by yellow pigmented xanthophores³⁷. Typically, melanocyte stripe 1D (dorsal) is wider than the interstripe X0 directly below it, as observed in MG-NT F1 fish (Figure 3C and D). However, the opposite is true for MG-sox10 F1 fish, where we observed an apparent widening of the interstripe region and narrowing of the melanocyte stripes (Figure 3C and D). Melanocytes were counted in fish treated with epinephrine to aggregate melanosomes, revealing a significant reduction in the number of melanocytes in the F1 sox10 KO fish compared to F1 NT fish (Figure 3E). Reduction in GFP+ cells within the xanthophore interstripes (X0) suggested that sox10 deficiency also negatively affected xanthophore populations, in line with previous work indicating a role for sox10 in specifying both xanthophores and melanocytes in zebrafish embryos³⁸.

Figure 3.

Melanocytes continually regenerate throughout life, typically from a progenitor population of melanoblasts or melanocyte stem cells. We next assessed whether sox10 was required for this. We treated adult fish with neocuproine, a copper chelator previously shown to kill mature, pigmented melanocytes³⁹. We then compared regeneration from these progenitors in MG-sox10 and MG-NT F1 fish by counting the percentage of melanocytes that emerged after 7, 15, and 70 days (Figure 3F). At days 7 and 15, when the melanocytes were still in the process of regenerating from melanoblasts, there were no significant differences between the two groups. However, by the time regeneration was completed we found a significant disparity between the regenerative potential of MG-NT and MG-sox10 fish. At 70 days post-neocuproine treatment, MG-NT fish had regenerated an average of 82% of their melanocytes, whereas only 53% of melanocytes had regenerated in MG-sox10 fish (Figures 3G-H). These findings demonstrate that sox10 is required for adult melanocyte regeneration, highlighting its requirement within melanoblast populations outside of neural crest specification.

Non-autonomous functions of tuba1a on melanocytes

One of the significant challenges of conventional global knockout methods is the inability to differentiate between cell-autonomous and non-autonomous phenotypes. This distinction is particularly crucial in melanocyte studies due to their direct and indirect interactions with diverse cell types. A notable example of genes with pleiotropic effects is the tubulin gene family, whose α/β tubulin heterodimers form microtubules essential for various cellular processes such as cell division, motility, and intracellular transport across various cell types⁴⁰. Microtubules are also critical for the transport of melanosomes in melanocytic cells⁴¹.

Mutation of tubulin genes such as tuba8l3a in zebrafish leads to highly pleiotropic effects including defects in melanocyte patterning, CNS abnormalities, and altered craniofacial morphology⁴². We chose to focus on the closely related α tubulin gene tuba1a, which is highly expressed in many cell types of the developing zebrafish including melanocytes, but whose specific function in these cells remains unexplored⁴³. Non-cell type specific knockout of tuba1a with the Alt-R CRISPR Cas9 system results in extensive embryo abnormalities including a curved tail phenotype, pericardial edema, and melanocytes with more dispersed pigmentation (Figure 4A-B, S3A-B). We found that compared to 71% survival of NT Alt-R embryos, only 10% of embryos injected with tuba1a Alt-R survive to 14dpf (Figure 4C).

Figure 4.

Similar to sox10, the embryonic lethality of global tuba1a knockout prevents the study of knockout phenotypes in adult fish. To overcome this limitation, we implemented our mitfa^Cas9 system to specifically knock out tuba1a in melanocytes. Unlike global knockout of tuba1a, melanocyte specific loss of tuba1a resulted in viable adult fish with no obvious abnormalities (Figure S3C-D). Surprisingly, we did not observe the dispersed melanocyte phenotype in any mitfa^Cas9 MG-tuba1a embryos (Figure 4D, E). Based on this observation, we hypothesized that tuba1a may be functioning in a cell non-autonomous manner on melanocytes.

The dispersed pigmentation observed in tuba1a Alt-R embryos resembles that seen in blind zebrafish, which fail to contract their melanosomes in response to light⁴⁴. This light-mediated camouflage involves retinal ganglion cells signaling hypothalamic neurons to release melanin-concentrating hormone (MCH), which then binds to MCH receptors on melanocytes, triggering melanosome aggregation via microtubules⁴⁴. The normal melanosome dispersion observed in mitfa^Cas9 MG-tuba1a fish suggests that rather than a cell-intrinsic defect in melanosome aggregation, tuba1a may indirectly influence melanocytes through potential defects in retinal and/or central nervous system cells crucial for light-mediated camouflage. Interestingly, TUBA1A mutation in humans can result in both ophthalmologic and brain abnormalities and morpholino-based knockdown of tuba1a in zebrafish inhibits CNS development^45,46.

To further test this idea, we treated tuba1a Alt-R embryos with epinephrine to assess their ability to contract melanosomes in response to a non-visual stimulus. Tuba1a Alt-R embryos exposed to epinephrine had robust aggregation of melanosomes, demonstrating that the loss of tuba1a did not impair melanosome transport (Figure 4F, G). This finding further supports a non-autonomous role for tuba1a in melanocytes. Utilizing the mitfa^Cas9 system distinguish cell-autonomous from non-autonomous gene functions provides valuable insights into the complex dynamics of cell-cell interactions and how genes function within cellular and organismal networks.

Targeting tumor suppressors with mitfa^Cas9 induces melanoma

We next turned our attention to melanoma. In humans, tumor suppressors such as PTEN and TP53 are somatically inactivated in melanoma^47,48. In contrast, the most commonly used zebrafish models of melanoma use a germline tp53 mutation, which is not typically seen in humans with the disease and precludes our ability to discern the role of tp53 inactivation specifically in melanocytes and melanoma⁴⁹. While global tp53 loss is not lethal, it does lead to a wide range of non-melanoma tumors, which can deteriorate fish health and confound melanoma studies⁵⁰. Furthermore, current zebrafish melanoma models typically involve the utilization of MiniCoopR mitfa rescue cassettes into casper or nacre zebrafish, which are devoid of normal pigment cells. Although this method is highly efficient, it does not accurately represent melanoma initiation in the context of normal skin architecture and relies on artificial overexpression of mitfa. To mimic human melanoma initiation more accurately, we used our mitfa^Cas9 model to assess tumor initiation using melanoma-specific gene knockout in animals with wild-type skin. Previous studies have shown that expression of BRAF^V600E and loss of tp53 in WT (non-casper) fish results in relatively low tumor burden, so we decided to also target ptena and ptenb, the zebrafish orthologs of human PTEN⁵¹. Ptena/ptenb loss in combination with tp53 has been shown to accelerate melanoma formation in zebrafish^52,53.

To first address the effect of ptena/b loss on normal melanocytes, we generated melanocyte-specific knockouts of ptena and ptenb (Figure 5A). Like sox10 and tuba1a, germline knockout of ptena/b is embryonic lethal^54,55. In our ptena/ptenb F0 melanocyte-specific KO fish, survival was normal. We observed an aberrant expansion of the melanocytes outside of the stripe regions they normally are restrained to, suggesting that inactivation of ptena and ptenb induces defects in melanocyte patterning (Figure 5A and B). However, loss of ptena/b alone was not sufficient to induce melanoma.

Figure 5.

To generate tumors, we co-injected mitfa^Cas9 embryos with plasmids encoding mitfa:BRAF^V600E and sgRNAs targeting tp53 and ptena/b (Figure 5C). Larvae were sorted for BFP+ eyes, indicating mitfa-Cas9 integration, and monitored for tumors over 30 weeks. No tumors were detected in BFP-fish from any of the injection conditions (n=136). We also did not observe any tumors in BFP+ fish injected with either BRAF^V600E alone (n=35) or ptena/b; tp53 guides in the absence of BRAF^V600E (n=17). In contrast, tumors developed in all BFP+ groups injected with both mitfa:BRAF^V600E and gRNA plasmids. Approximately 5% of fish from the BRAF;tp53 group developed tumors (Figure 5D). This aligns with a previous study where 6% of tp53-/- fish developed tumors when injected with a BRAF^V600E construct ⁵¹. Targeting of ptena and ptenb in mitfa^Cas9 fish co-injected with mitfa:BRAF^V600E resulted in a higher tumor incidence of 29% (Figure 5D).

When all three tumor suppressors (tp53, ptena, ptenb) were targeted in combination with BRAF^V600E, 59% of fish developed tumors by 30 weeks post fertilization (wpf) (Figure 5D). Immunohistochemistry (IHC) revealed that all tumor samples expressed BRAF^V600E, while p-AKT, a hallmark of PTEN inactivation, was only detected in samples with ptena/b KO (Figure 5E). To confirm on-target cutting, DNA was extracted from tumors and CRISPR sequencing was performed on PCR amplicons for each tumor suppressor gene. A large proportion of reads contained indels at the target locus for each of the target genes (ptena, ptenb, and tp53) (Figure 5F and S4). Some tumors had an array of mutations, while others appeared to be composed largely of a one dominant clone (Figure S4A to C). Our system thus allows us to rapidly dissect the melanocyte-specific function of both oncogenes (BRAF^V600E), as well as putative tumor suppressors (tp53, ptena/b), while avoiding pleiotropic effects of global tumor suppressor loss.

Targeting of lineage-specific oncogenes decreases tumor initiation but promotes progression

Our next objective was to leverage the mitfa^Cas9 fish as a tool for identifying or validating potential genetic dependencies within melanoma. Genetic dependencies can be difficult to study in vivo since inactivation of a required gene inevitably leads to selection for non-mutant alleles, as has been previously shown⁵⁶. In melanoma, the DepMap has identified SOX10 as the top in vitro genetic dependency necessary for tumor cell proliferation, suggesting it acts as a lineage-specific oncogene^14,57. In prior work in both zebrafish and mice, loss of sox10 is clearly associated with a loss of tumor initiating potential and cell proliferation^56,57. Yet its specific role in vivo remains unclear, since it is required for neural crest and melanocyte specification, making it difficult to know whether its effects reflect loss of the entire lineage⁵⁷. Given our data above showing that sox10 is required specifically in melanocyte patterning and regeneration, we wished to investigate its function in melanoma.

To assess this, we generated plasmids containing multiple sgRNAs driven by three distinct zU6 promoters (zU6A, zU6B, and zU6C) to simultaneously target both tumor suppressors (tp53, ptena/ptenb) and tumor promoting genes (sox10) in the same cells, which reduces the chance of selecting for non-mutant alleles². We co-injected these plasmids with mitfa:BRAF^V600E, sorted fish for BFP+ eyes, and tracked tumor free survival over the course of 50 weeks (Figure 6A and B). Consistent with the DepMap prediction, sox10 inactivation markedly reduced tumor incidence, which likely reflects loss of proliferation. Only 3/96 (3.1%) fish with sox10 knockouts initiated melanomas compared to the 24/94 fish (25.5%) from the NT condition that developed tumors (Figure 6B). Using IHC, we validated that sox10 KO tumors expressed lower levels of Sox10 protein compared to NT tumors (Figure 6C and S5A). Quantification of Sox10 intensity with immunofluorescence confirmed this finding (Figure S5B).

Figure 6.

Although sox10 clearly reduced tumor initiation, we noted “escapers”. These tumors appeared morphologically very distinct compared to the sox10 intact tumors. Specifically, these tumors were highly invasive with a mesenchymal morphology compared to the typical BRAF^V600E melanoma (Figure 6C). This observation raised the possibility that these tumors had undergone “phenotype switching”, a form of cell plasticity in which cells reversibly move between opposite extremes of proliferation versus invasive states^22,23,57,58. In melanoma, the proliferative state is thought to be characterized by high expression of SOX10, whereas the mesenchymal, invasive state is characterized by high expression of SOX9²³. These two highly related transcription factors have seemingly opposite and possibly antagonistic roles in melanoma. In vitro, knockout of SOX10 in a variety of human cell lines is associated with acquisition of a SOX9^hi state, which has been suggested to be linked to the invasive phenotype²³. Re-analysis of publicly available RNA-seq data using 3 patient-derived human melanoma cell lines treated with siRNA targeting SOX10 confirmed this observation, with the SOX10 lines now becoming SOX9^hi (Figure 6E to J, S5C to L)²³. These data raised the hypothesis that the more invasive tumors we see in our escaper fish was due to gain of sox9 expression. To test this, we stained our sox10 CRISPR tumors for sox9 and discovered that 2 of the 3 sox10 KO tumors markedly upregulated sox9 relative to control tumors expressing NT gRNA (Figure 6C and D). This data highlights that in vitro genetic studies such as the DepMap, which rely on proliferation as the readout, can mask more complex in vivo phenotypes.

Discussion

The ability to model genetic dependencies in vivo is one of the most important uses of model organisms. Forward genetic mutagenesis screens using compounds such as ENU or EMS led to the identification of many key developmental genes in organisms such as Drosophila, C. elegans, Saccharomyces, zebrafish, and many others^59,60. These approaches have rapidly accelerated in the era of TALENs and CRISPR, readily allowing for knockout of nearly any gene in a wide variety of non-model organisms. Cre/Lox approaches have further augmented our ability to discern the cell-type specific effects of these genes, although these are still laborious and time consuming, especially in models such as mice¹⁰.

Alongside these approaches have been remarkable advances in large scale sequencing of human tissues. GWAS-like efforts such as the UK Biobank and the US All of Us programs have identified thousands of germline variants (SNPs) that may be linked to interesting phenotype variation^12,13. A conceptually similar approach is also happening in diseases such as cancer, in which efforts such as the TCGA, ICGC and DepMap have continued to unveil new potential somatic variants (SNVs) linked to cancer^14,61,62.

Across these efforts, in both development and disease, the large number of candidate variants makes it challenging to connect these genotypes to phenotypes. This is especially acute in melanoma, where the large number of somatic variants generated by UV radiation leads to a high number of background mutations that may or may not have pathogenic function. Thus, there is a substantial need for in vivo models that allow for rapid and efficient modeling of candidate genes. Our development of the mitfa-Cas9 knockin line provides such a method. One major advantage is that phenotypes can be readily screened either in the F0 generation (as mosaics) or in the F1 generation (3 months later). The F1 fish in particular are advantageous, since they allow for highly efficient biallelic knockout of the gene of interest, allowing for very quick identification of genes that only have recessive phenotypes (i.e., albino). On a per gene basis, this saves well over 3 months of work compared to germline knockouts and uses vastly fewer animals since all F1 animals (even those for recessive alleles) are informative. Additionally, this model eliminates the issue of inbreeding, promoting healthier genetic lines and facilitating the potential for outcrossing with any available fish line. Because we demonstrate that the Cas9 is completely restricted to the melanocyte lineage, with no detectable off-target expression, this easily allows us to discern the melanocyte-specific effect of pleiotropic genes very efficiently.

One interesting observation of our melanoma studies is evidence that loss of melanoma genetic dependencies like SOX10 (as suggested by the human DepMap data) can still lead to tumors, albeit with very different biological behaviors. In both the fish and in the human samples we analyzed, loss of SOX10 leads to acquisition of a SOX9^hi state instead²³. SOX9 is increasingly recognized to be a potent mediator of tumor cell invasiveness, which is what we observe in our fish^63,64. One implication of this finding is that clinical targeting of transcription factors like SOX10, long considered an ideal therapeutic approach, could lead to unexpected outcomes in which tumors might proliferate less but become more invasive or resistant to therapy. Whether this phenomenon of SOX10/SOX9 switching is a general phenomenon, or specific to that class of TFs, remains to be determined. However, a similar phenomenon has been observed for MITF, another melanocytic TF, suggesting this idea of TF switching needs to be more thoroughly explored in the future^58,65,66.

Finally, while our studies clearly focused on the melanocyte lineage, highly similar approaches could be taken for virtually any cellular lineage in which master regulators or markers are known. Given the large number of already identified promoter/enhancer fragments in the zebrafish, this would be an important extension of our system to study gene variants in the context of development and other diseases⁶⁷.

Our approach has several limitations that may need to be addressed in future studies. Due to the constitutive nature of the Cas9, we have little control over the timing of its expression. Because it is knocked into the endogenous mitfa locus, the expression of Cas9 will vary with the endogenous regulation of that gene. In future iterations of this general approach, swapping the Cas9 for an inducible version would allow for better timed knockout of genes⁶⁸. Because mitfa itself is expressed in melanoblasts, melanophores and xanthophores during different stages of development, choosing a promoter (e.g. PMEL) that is more tightly restricted to melanocytes may be advantageous. Finally, although our system was relatively efficient, we still did not see 100% knockout. This may reflect that it is relatively easy to bypass the CRISPR lesion in an exon with an in-frame mutation. The use of multiple sgRNAs, or sgRNAs that target promoter regions, rather than exons, could further augment efficiency of knockout. Alternatively, instead of targeting DNA, we could consider using RNA targeting CRISPR enzymes. Recent work has shown that RNA degradation induced by Cas13 works in zebrafish and may allow for more nuanced knockdown rather than knockout of a given candidate gene^69,70. This may especially have advantages in the context of melanoma, where genes are commonly dysregulated rather than completely knocked out. We envision it would be relatively straightforward to knock Cas13 into the locus of interest using an analogous approach to the one we have taken here.

Acknowledgements

We thank members of the White lab for valuable discussions and feedback on this project. We thank Nelly Cruz Esteves for helpful feedback on the manuscript. We also thank the MSKCC flow cytometry core for assistance with cell sorting, Integrated Genomics Operation Core, funded by an NCI Cancer Center Support Grant (P30 CA08748) for sequencing assistance, and Molecular Cytology Core for their help with imaging. We thank the MSKCC aquatics core facility for their support of this work. Schematic figures were all made using Biorender.com with a paid Biorender license.

Funding

RMW was funded through the NIH/NCI Cancer Center Support Grant P30 CA008748, the Melanoma Research Alliance, The Debra and Leon Black Family Foundation, NIH Research Program Grants R01CA229215 and R01CA238317, NIH Director’s New Innovator Award DP2CA186572, The Pershing Square Sohn Foundation, The Mark Foundation for Cancer Research, The American Cancer Society, The Alan and Sandra Gerry Metastasis Research Initiative at the Memorial Sloan Kettering Cancer Center, The Harry J. Lloyd Foundation, Consano and the Starr Cancer Consortium. This work was also supported by NIH ORIP R24OD020166 (MM). YM was supported by a Medical Scientist Training Program grant from the NIH under award number T32GM007739 to the Weill Cornell/Rockefeller/Sloan Kettering Tri-Institutional MD-PhD Program and Kirschstein-National Research Service Award (NRSA) predoctoral fellowship under award number F30CA265124. MVH was funded by a K99/R00 Pathway to Independence Award from the National Cancer Institute (1K99CA266931). SP was funded by a Molecular Imaging in Cancer Biology Training Grant at Memorial Sloan Kettering Cancer Center from the National Cancer institute (T32CA254875) and the Ruth L. Kirschstein National Research Service Award for Individual Predoctoral Fellows from the National Cancer Institute (F31CA271518).

Author Contributions

Conceptualization: SP, RMW

Investigation: SP, YM, MVH, JBS, ZM, JX

Visualization: SP, MVH, JBS

Supervision: TL, MM, RMW

Writing—original draft: SP, RMW

Writing—review & editing: SP, MVH, RMW

Declaration of Interests

MM has competing interests with Recombinetics Inc., LifEngine Technologies, and LifEngine Animal Health Laboratories, Inc. RMW is a paid consultant to N-of-One Therapeutics, a subsidiary of Qiagen and is on the scientific advisory board of Consano but receives no income for this. RMW receives royalty payments for the use of the casper zebrafish line from Carolina Biologicals. SP, YM, MVH, JBS, ZM, JX, and TL declare no competing interests.

Methods

Materials Availability

All data are available in the main text or the supplementary materials, or upon request. Plasmids, fish lines and other materials generated from this study are available upon request.

Zebrafish husbandry and ethics statement

All zebrafish experiments adhered to institutional animal protocols and were conducted in compliance with approved procedures. Fish stocks were maintained at a temperature of 28.5 °C, under 14:10 light:dark cycles, with pH set at 7.4, and salinity-controlled conditions. The zebrafish were fed a standard diet comprising brine shrimp followed by Zeigler pellets. Approval for the animal protocols outlined in this manuscript was obtained from the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee (IACUC), under protocol number 12-05-008. Anesthesia was conducted using Tricaine (4 g/l, Syndel, Ferndale, WA, USA) from a stock of 4 g/l and diluted to 0.16 mg/ml. Adult zebrafish of both sexes were equally employed in all experiments. Embryos, collected through natural mating, were incubated in E3 buffer (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl2, 0.33 mM MgSO4) at 28.5 °C. The wild-type strain used was Tropical 5D Zebrafish (T5D)²⁷.

GeneWeld plasmid construction

A combination of restriction enzyme digestion and HiFi cloning was used to construct the GeneWeld pPRISM-nCas9n, γcry1:BFP vector for targeted integration to isolate an nls-Cas9-nls knockin. A 765 bp fragment containing the Porcine teschovirus-1 polyprotein 2A peptide sequence was amplified from the pPRISM-2A-Cre, γcry1:BFP vector (Addgene #117789) with primers Ori-F and 2A-R^24,25. The nCas9n cDNA was amplified from the expression vector pT3TS-nCas9n (Addgene #46757) with primers Cas9-F and Cas9-R⁷¹. A KpnI/SpeI fragment containing the pPRISM-2A-Cre, gcry1:BFP vector backbone was assembled with the 2A and nCas9n PCR amplicons using the NEBuilder HiFi DNA Assembly Cloning Kit (NEB # E5520S) following the manufacturer’s instructions.

Homology arms were designed as previously described^24,25. Briefly, 48bp homology arms complementary to the mitfa target site were designed using GTagHD (http://www.genesculpt.org/gtaghd/) for the pPRISM GeneWeld plasmid series. The two pairs of complementary oligos can be found in Table S1. 5’ and 3’ arms were cloned sequentially into the pPRISM-nCas9n, γcry1:BFP vector. The 3’ homology arm oligos were annealed and cloned into the vector using the BspQI restriction site as previously described. Due to the multiple BfuAI sites within Cas9, instead of using BfuAI restriction cloning, we alternatively used in-fusion cloning (Takara Bio) to insert a gBlock (IDT) containing the 5’ homology arm into the pPRISM-nCas9n, gcry1:BFP vector. Primers and gblock sequence used for cloning are listed in Table S1. The 5’ and 3’ homology arms were sequenced with primers 5’homologycheckF and R3’_pgtag_seq, respectively.

Injection of GeneWeld reagents

Injections of GeneWeld reagents were performed on one-cell stage embryos. We used the Alt-R CRISPR-Cas9 system (IDT) for the initial knock-in of nCas9n, γcry1:BFP to the mitfa locus. In brief, 100 μM tracrRNA (IDT 1075928) was mixed at a 1:1 ratio with either mitfa crRNA or universal crRNA then incubated at 95 °C for 5 min. Recombinant Cas9 (IDT 1081059) was incubated with both sgRNAs for 10 min at 37 °C to form the RNP complex. The GeneWeld pPRISM-5’mitfaHom-nCas9n, gcry1:BFP-3’mitfaHom plasmid was then added to the injection mix along with phenol red (Sigma-Aldrich P0290). The final concentrations injected into the embryos were: 75 pg/nl Cas9, 12.5 pg/nl mitfa sgRNA, 12.5 pg/nl Universal sgRNA, and 5 pg/nl pPRISM plasmid.

Genotyping

Tail clips were taken from 2 month postfertilization (mpf) zebrafish and DNA was extracted using the DNeasy Blood and tissue kit (Qiagen). DNA was PCR amplified with Q5 high fidelity DNA polymerase (NEB). The forward primer MitfaPCR-F binds to the mitfa genomic region upstream of the insertion site and the reverse primer 5’homology_amplify-R binds the p2A region of the insertion cassette. Expected PCR amplicon size is 269bp. Amplicons were sequenced using sanger sequencing (Azenta Life Sciences).

RNA-FISH HCR

We adapted the HCR protocol from Ibarra-Garcia-Padilla et al.⁷². Briefly, embryos were collected and treated with 0.0045% 1-phenyl 2-thiourea (PTU) (Sigma-Aldrich) after 24 hpf to block pigmentation. Embryos were harvested at 72 hpf and fixed in 4% PFA (Santa Cruz) for 24 hrs at 4 °C. Embryos were then washed with PBS and dehydrated/permeabilized with a series of MeOH (Millipore) washes. Embryos were rehydrated with a series of MeOH/PBST washes, permeabilized with acetone (Fisher Scientific) and incubated with Proteinase K (Millipore) for 30 min. This was followed by further fixation with 4% PFA. Embryos were pre-hybridized in probe hybridization buffer (PHB) before incubating them with HCR probes in PHB at concentration of 20nM. Probe sequences for mitfa and Cas9 can be found in Table S1. Embryos were washed with probe wash buffer before being pre-amplified with probe amplification buffer (PAB). Following pre-amplification, amplifier hairpins were thawed, snap-cooled, and mixed with PAB to a concentration of 36 nM. Embryos were incubated in the hairpin solution followed by washes with 5×SSCT and 1× PBST. Embryos were stained with 2 µg/mL DAPI in PBST and mounted on glass slides. Imaging was performed using a Zeiss LSM880 confocal microscope.

Guide RNA plasmids

The following plasmids were constructed using the Gateway Tol2kit:

zU6A:gRNA-NT;zU6B:gRNA-ptena;zU6C:gRNA-tp53/394,

zU6A:gRNA-NT;zU6B:gRNA-ptenb;zU6C:gRNA-tp53/394,

zU6A:gRNA-sox10;zU6B:gRNA-ptena;zU6C:gRNA-tp53/394,

zU6A:gRNA-sox10;zU6B:gRNA-ptenb;zU6C:gRNA-tp53/394.

In-fusion cloning (Takara Bio) was used to insert mitfa:GFP or mitfa:TdTomato into a zU6:gRNA/394 plasmid previously developed in the lab (See Table S1 for primers used). gRNAs were cloned into zU6:gRNA;mitfa:GFP/394 or zU6:gRNA;mitfa:TdTomato/394 using BsmBI sites. Other plasmids used in this study that were previously developed in the lab include zU6:gRNA-ptena/394, zU6:gRNA-ptenb/394, zU6:gRNA-tp53/394, mitfa-BRAF^V600E;cmlc2:eGFP. All guide RNAs used in this study besides tuba1a have previously been validated in zebrafish^24,32,56,73.

Validation of tuba1a sgRNA

The Alt-R CRISPR-Cas9 system (IDT) was used to achieve non-cell type specific knockout of tuba1a. Guide RNAs were designed using ChopChop⁷⁴. In brief, 100 μM tracrRNA (IDT 1075928) was mixed at a 1:1 ratio with tuba1a crRNA then incubated at 95 °C for 5 min. Recombinant Cas9 (IDT 1081059) was incubated with the tuba1a sgRNA for 10 min at 37 °C to form the RNP complex. Phenol red was added prior to injection into 1 cell-stage wild-type T5D embryos. To validate on-target cutting, 4 embryos were pooled and DNA was extracted using Quick Extract buffer (Fisher Scientific NC0302740). DNA was PCR amplified with Q5 high fidelity DNA polymerase (NEB) and primers tuba1a_seq-F and tuba1a_seq-R. Exosap IT for PCR Product Clean Up was added to PCR product prior to sanger sequencing. Primers tuba1a_seq-F and tuba1a_seq-R were also used for sequencing.

Transgenic lines

To generate MG-U6:sgRNA stable lines, one-cell-stage embryos from crossing WT and mitfa:Cas9 fish were collected and injected with the 30 pg indicated plasmid and 20 pg tol2 mRNA. Fish with GFP+ melanocytes and BFP+ eyes were selected and outcrossed with WT fish to produce the F1 generation. F1 fish were again sorted and outcrossed to generate F2 generation zebrafish. All fish were imaged on a Zeiss AxioZoom V16 with Zen 2.1 software.

Flow cytometry of melanocytes from adult zebrafish

Zebrafish were euthanized using ice-cold water and dissected using a clean scalpel and forceps. The epidermal and dermal layers of the skin as well as the fins were separated from the rest of the tissues and diced into 1-3mm pieces. Cells were dissociated with Liberase TL (Millipore Sigma 05401020001) and filtered for single cell suspensions as previously described³². Samples were then FACS sorted (BD FACSAria) for GFP+ and GFP-cells. WT fish were used as a GFP negative control. Genomic DNA was isolated from sorted cells using the DNeasy Blood and Tissue Kit (Qiagen).

Tumor Dissection

Fish with tumors were euthanized with ice-cold water and immediately dissected with a scalpel to isolate tumor tissue. Genomic DNA was purified from the tumor samples using the Dneasy Blood and tissue kit (Qiagen).

CRISPR sequencing

Primer pairs were designed to produce amplicons 200 to 280 bp in length with the mutation site within 100 bp from the beginning or end of amplicon (see Table S1 for primer sequences). Genomic DNA isolated from methods detailed above was PCR amplified with Q5 high fidelity DNA polymerase (NEB) and run on an agarose gel to visualize PCR products. The samples were then purified using the NucleoSpin Gel and PCR cleanup kit (Takara Bio). Deep sequencing was conducted using the CRISPR-seq platform. Sequencing data was analyzed with CRISPResso2⁷⁵. and indel charts were generated with CRISPRVariants R package⁷⁶.

Drug treatments

Adult zebrafish were treated for 24 hr with 750nM neocuproine (Sigma-Aldrich) dissolved in fish water with a final concentration of 0.0075% DMSO as previously described ³⁹. Cells from the middle melanocyte stripe were counted within a rectangular region of 4mm x 1.5mm. Epinephrine hydrochloride (Sigma-Aldrich) was used at 1 mg/ml in fish water. Fish were treated for 10 minutes prior to imaging.

Zebrafish tumor-free survival curves

To generate a transgenic melanoma model, one-cell-stage embryos from WT X mitfa:Cas9 fish crosses were injected with mitfa-BRAF^V600E;cmlc2:eGFP, 20 pg Tol2 mRNA, and the indicated gRNA plasmids for each condition. The total amount of plasmid per embryo did not exceed 30 pg. Embryos were screened for GFP positive hearts and sorted for BFP+ or BFP-eyes. Fish were screened for tumors using a microscope every 2 weeks starting at 4 weeks post fertilization. Tumors were only called if an elevated lesion was observed. Graphpad prism 9 was used to generate and analyze Kaplan-Meier survival curves. Statistical differences were determined using log-rank Mantel–Cox test. All screening and imaging was conducted on a Zeiss AxioZoom V16 using Zen 2.1 software.

Histology

Zebrafish were sacrificed using ice-cold water and placed in 4% paraformaldehyde (Santa Cruz 30525-89-4) in PBS for 72 h at 4°C on a rocker. Fish were then transferred to 70% EtOH for 24 h at 4°C on a rocker. Fish were sent to Histowiz, Inc (Brooklyn, NY, USA), where they were paraffin embedded, sectioned, stained and imaged. All the stainings were performed at Histowiz, Inc Brooklyn, using the Leica Bond RX automated stainer (Leica Microsystems) using a Standard Operating Procedure and fully automated workflow. Samples were processed, embedded in paraffin, and sectioned at 4μm. The slides were dewaxed using xylene and alcohol based dewaxing solutions. Epitope retrieval was performed by heat-induced epitope retrieval (HIER) of the formalin-fixed, paraffin-embedded tissue using Citrate based pH 6 solution (Leica Microsystems, AR9961) for 20 mins at 95 C. The tissues were first incubated with peroxide block buffer (Leica Microsystems), followed by incubation with the following antibodies for 30 min: BRAFV600E antibody (ab228461, 1:100), phospho-AKT antibody (CST4060, 1:50), SOX9 antibody (AB_185230, 1:1000), and SOX10 antibody (GTX128374, 1:500). This was followed by incubation with DAB secondary reagents: polymer, DAB refine and hematoxylin (Bond Polymer Refine Detection Kit, Leica Microsystems) according to the manufacturer’s protocol. The slides were dried, coverslipped (TissueTek-Prisma Coverslipper) and visualized using a Leica Aperio AT2 slide scanner (Leica Microsystems) at 40X.

Quantification of Sox10 intensity

FFPE slides were deparaffinized by consecutive incubations in xylene and 100-50% ethanol. For antigen retrieval, slides were placed in 10 mM sodium citrate pH 6.2 in a pressure cooker and heated to 95°C for 20 minutes. Slides were then cooled to room temperature and blocked in 5% donkey serum, 1% BSA, and 0.4% Triton-X100 in PBS for 1 hour at room temperature. The primary antibody targeting Sox10 (GeneTex, GTX128374) was added to the blocking buffer at 1:200 concentration and incubated at 4°C overnight. Slides were then washed in PBS before incubation with the secondary antibody (donkey anti-rabbit Alexa Fluor Plus 488, Thermo Fisher Scientific #AC32790; 1:250) and Hoescht (Thermo Fisher Scientific, #62249; 1:1000) in blocking buffer for 2 hours at room temperature. Slides were mounted in Vectashield Plus (Vector Labs, H-1900). Slides were imaged on a LSM 880 confocal microscope using a 40X oil immersion objective. 3 images were captured per sample. Quantification of Sox10 intensity per cell was performed using CellProfiler.⁷⁷ and MATLAB r2023b (Mathworks). Cells were segmented in CellProfiler to obtain a nuclear mask from the Hoescht signal and using this mask the mean intensity per nucleus was quantified.

Re-analysis of human melanoma scRNA-seq data

Human melanoma scRNA-seq data from Wouters et al. was accessed from GEO (GSE134432)²³. All analysis was done in R (version 4.3.1) using the Seurat package (version 5.0.2)⁷⁸. Counts matrices for each sample and cell line were imported into R and converted into Seurat objects before merging into one combined Seurat object. Counts were normalized using the Seurat function NormalizeData with default parameters. UMAP was calculated using the Seurat function RunUMAP with 10 dimensions.

Statistical analysis

Statistical significance was analyzed using t test, Log-rank (Mantel-Cox) test, Wilcoxon Rank Sum Test or as indicated in the figure legends. GraphPad Prism 9 software was used for data processing and statistical analysis. *P < 0.05; **P < 0.005; ***P < 0.001; ****P < 0.0001. Data are presented as means ± SD unless otherwise indicated.