GNAQ Induces Melanomagenesis in Mitfa-Independent Melanocyte Progenitors in a Zebrafish Model of Uveal Melanoma

Choroid-targeted GNAQ^Q209L expression induces anatomically correct uveal melanoma in adult zebrafish, with germline mitfa deletion expanding mitfa-independent melanocyte progenitors with enhanced susceptibility that are transcriptionally distinct from the subpopulation transformed by BRAF^V600E.

Abstract

Melanocytes reside in diverse microenvironments that influence their susceptibility to oncogenic transformation; however, investigation of rare melanoma subsets has been limited by the lack of suitable preclinical animal models. In this study, we developed a primary, immunocompetent zebrafish model to study uveal melanoma using choroidal melanocyte–targeted injection and electroporation of plasmids encoding human GNAQ^Q209L together with CRISPR/Cas9 cassettes for somatic tumor-suppressor gene deletion. Single-cell transcriptional profiling of primary melanocytes and melanoma derived from the eye and skin revealed distinct transcriptional programs, with epithelial-to-mesenchymal transition pathways enriched in ocular tumors. In addition, choroidal fibroblasts from tumor-bearing eyes exhibited marked transcriptional changes, including increased fibronectin and collagen expression, consistent with stromal remodeling. Given prior associations between mitfa loss and accelerated GNAQ^Q209L tumor onset, the model was applied to determine whether melanocyte differentiation state contributes to the emergence of GNAQ-driven tumors. The increased susceptibility resulted from expansion of Mitfa-independent melanocyte progenitor populations in germline mitfa-mutant zebrafish, rather than somatic mitfa loss in differentiated melanocytes, as conditional, melanocyte-specific mitfa deletion in adult zebrafish did not accelerate tumor growth. Furthermore, pax3a-positive melanocyte progenitor cells in mitfa-deficient zebrafish embryos and adult eyes and skin were highly susceptible to transformation induced by GNAQ^Q209L but not BRAF^V600E. Analogous PAX3 positive populations were also identified in mouse and human single-cell transcriptomic datasets. Collectively, these findings establish a critical role for Mitfa-independent melanocyte progenitors in uveal melanoma pathogenesis.

Significance:

Choroid-targeted GNAQ^Q209L expression induces anatomically correct uveal melanoma in adult zebrafish, with germline mitfa deletion expanding mitfa-independent melanocyte progenitors with enhanced susceptibility that are transcriptionally distinct from the subpopulation transformed by BRAF^V600E.

Introduction

Uveal melanoma is a rare cancer of the melanocytes in the uveal tract. The majority of uveal melanomas (∼90%) occur in the choroid, the vascular structure supplying the outer retina, with the remainder arising from the ciliary body or the iris (1). Oncogenic mutations in GNAQ or its paralog GNA11 are observed in up to 90% of uveal melanomas but are rare in cutaneous melanoma (CM; ref. 2). Such mutations reduce GTPase activity of the heterotrimeric G protein α subunit, resulting in constitutive activation that drives cellular proliferation through downstream signaling pathways, including MAPK, YAP/TAZ, and PI3K (3, 4). The transcriptional mechanisms of uveal melanoma onset and progression remain understudied, in part, because existing preclinical models inadequately capture the earliest events of ocular melanocyte transformation within the correct microenvironment (ME).

A recent study by Phelps and colleagues (5) showed that loss of mitfa, which is considered the master regulator of the melanocyte lineage, accelerates tumorigenesis in transgenic Tg(mitfa:GNAQ^Q209L); mitfa^−/−; tp53^−/− zebrafish compared with Tg(mitfa:GNAQ^Q209L); tp53^−/− controls. These findings indicate that, in the context of tumor-suppressor loss, oncogenic GNAQ can drive tumor formation independently of differentiated melanocytes. This contrasts with BRAF^V600E-induced melanoma, in which Mitfa is required for tumor formation (6). Such observations imply that melanocyte progenitor cells, which persist in mitfa-mutant zebrafish, may serve as the cells of origin for GNAQ-induced transformation. Whereas transgenic models have been incredibly informative for studying spontaneous tumor formation, germline loss of mitfa or tumor-suppressor genes, together with melanocyte lineage expression of oncogenes throughout development, may unintentionally perturb melanocyte lineage specification. Consequently, these models may capture tumorigenesis in a lineage already altered during development rather than melanoma initiated by somatic mutations in normal melanocytes. Conditional knockout (KO) approaches as well as lineage restricted expression of oncogenes in the appropriate ME will improve our understanding of uveal melanoma pathogenesis.

Zebrafish are a well-established model for studying melanocytes and melanoma due to conserved melanocyte differentiation pathways, efficient genetic manipulation, and the availability of transparent lines that facilitate high-resolution live-cell imaging. Zebrafish have 3 types of pigment cells—dark melanocytes, iridescent iridophores, and yellow xanthophores—that arise from multipotent, neural crest (NC)-derived pigment progenitors (7). To study early events in uveal melanoma, we developed a novel zebrafish model that uses Transgene Electroporation of Adult Zebrafish (TEAZ; ref. 8) to express the human GNAQ^Q209L oncogene and conditionally delete tumor-suppressor genes in cells of the melanocyte lineage within the skin (TEAZ-Skin) and eye (TEAZ-Eye). Choroidal targeted injection and electroporation drive primary uveal melanoma onset from the same anatomical site as human uveal melanoma. In this study, we demonstrate the existence of a melanocyte progenitor population in zebrafish eyes with increased sensitivity to transformation by GNAQ^Q209L. Such progenitor cells are expanded in mitfa-deficient zebrafish and result in reduced GNAQ^Q209L-induced tumor latency compared with wild-type (WT) animals.

Materials and Methods

Plasmid constructs and zebrafish generated in this study may be obtained by a request to the lead contact.

Human donor eye procurement and histology

Human donor eyes were obtained from the Iowa Lions Eye Bank. Written informed consent was obtained from patients. Studies were conducted in accordance with the Declaration of Helsinki. Studies were approved by the Institutional Review Board of the University of Iowa. Globes were processed for histologic analysis in the laboratory of Dr. Mullins at the University of Iowa. Tissues were fixed, paraffin-embedded, sectioned, and stained with hematoxylin and eosin (H&E) using standard protocols.

Zebrafish husbandry

Adult zebrafish used in this study were bred and maintained at the University of Iowa Animal Care Facility. The zebrafish are kept consistently at 28°C, 7.4 pH, and controlled salt concentrations and cyclically receive 14 hours of light followed by 10 hours of darkness. Animal studies were approved by an Institutional Animal Care and Use Committee (IACUC) at the University of Iowa. Zebrafish embryos were maintained at 28.5°C and staged by hours after fertilization (hpf) or days after fertilization.

Zebrafish mutant lines

WT zebrafish in this study include the AB line (RRID: ZFIN_ZDBGENO-960809-7) and the WIK line (RRID: ZIRC_ZL84). Transgenic lines include casper (mitfa^w2/w2; mpv17^a9/a9; RRID: ZIRC_ZL1714; refs. 9, 10), nacre (mitfa^w2/w2; RRID: ZIRC_ZL2104; ref. 11), Tg(mitfa:GFP; ref. 12), and Tg(mitfa:GFP); mitfa^w2/w2. TEAZ-Eye and TEAZ-Skin was performed on adult zebrafish (inclusion criteria: >4 months after fertilization and <1.5 years old). Similar proportions of male and female zebrafish were used within each experiment.

Induction of melanoma using strategies A and B

To model GNAQ-driven melanoma, we generated mitfa promoter–driven plasmid systems expressing human oncogenic GNAQ^Q209L, CRISPR/Cas9, and guide RNAs (gRNA) targeting tp53, ptena, and ptenb. Two delivery strategies were used. In strategy A, a single-plasmid contained 2 expression cassettes under the mitfa promoter: one encoding GNAQ^Q209L fused to a T2A self-cleaving peptide followed by eGFP, and a second encoding Cas9, together with 3 U6-driven gRNAs targeting tp53 and ptena/b. Because T2A-mediated cleavage leaves residual C-terminal amino acids that may alter GNAQ oncogenic activity, we also used an alternative multiplasmid approach (strategy B). This system consisted of 5 separate plasmids: mitfa:GNAQ^Q209L, mitfa:GFP, mitfa:Cas9, U6:gRNA-tp53, U6:gRNA-ptena, and U6:gRNA-ptenb. Plasmids expressing mitfa:GNAQ^Q209L and mitfa:GFP were generously provided by Dr. Jacqueline Lees (MIT), and plasmids encoding Cas9 and gRNAs were provided by Dr. Richard White (University of Oxford). Strategy B was designed to avoid unintended C-terminal modification of GNAQ and to improve electroporation efficiency due to reduced plasmid size. Fish were monitored weekly for tumor formation at the electroporation site. Tumor onset was defined as the first appearance of a visible, expanding pigmented or translucent mass. Tumor latency was calculated as days after electroporation (dpe) to first detection. Kaplan–Meier analyses were used to compare tumor onset between experimental groups.

Cloning

The MiniCoopR (6) vector (MiniCoopR 2xU6:gRNA, mitfa:Cas9 was a gift from Leonard Zon Addgene plasmid # 118844) was modified to replace the mitfa coding sequence with GNAQ^Q209L by removing mitfa using StuI and SmaI restriction enzymes (NEB). Gibson assembly was used to insert GNAQ^Q209L:T2A:GFP DNA as a gBlock hifi (IDT) with sequence on the 3′ (GAA​GCT​AAC​ACA​TAG​TTG​AAC) and 5′ (CAA​TGC​CAA​CTA​AAT​TTC​ATG) as overhangs. Gibson assemblies were transformed using NEB 5-α–competent Escherichia coli cells via heat shock. GNAQ^Q209L:T2A:GFP insertion was confirmed using Sanger sequencing (primers: CAA​GGA​AGC​CCG​GCG​GAT​CAA​CGA, CAT​ACT​TGT​ATG​GGA​TCT​TGA​GTG​TGT​CCA, GTG​GAG​TCA​GAC​AAT​GAG​AAC​CGA​ATG​GAG, and CAA​GCT​GGA​GTA​CAA​CTA​CAA).

mitfa gRNA was designed using IDT’s CRISPR-Cas9 gRNA design tool (sequence: ATGGACAAAGCTGGACCATG). U6:gRNA-ptena was linearized by PCR (forward primer: GTTTAAGAGCTATGCTGGAAACAGCAT, reverse primer: GAACAAAGAGCTGGAGGGAG). Linear plasmid was purified from a 2% agarose gel using the Monarch GNA Gel Extraction Kit (#T1020S). ssDNA sequences for gRNA-mitfa were annealed together in NE buffer r2.1. Gibson assembly was used to insert gRNA-mitfa into linearized U6 plasmid. Assembled product was transformed into NEB 5-α–competent E. coli cells via heat shock and selected with 50 ug/mL streptomycin. Insertion of gRNA-mitfa was confirmed using PCR amplification of lysed bacteria from single colonies (forward primer: ATGGACAAAGCTGGACCATG, reverse primer: CTTAGCTGGATAACGCCAC). Positive products were isolated using the QIAGEN Plasmid Maxi Kit. Entire plasmid was sequenced using Plasmidsaurus sequencing service.

TEAZ-Skin

Adult zebrafish were anesthetized in 0.2% tricaine prior to upright placement on a prewetted Kimwipe. Each injection consisted of 1,000 ng total DNA: 100 ng of Tol2 plasmid and an equal division of the remaining 900 ng for the remaining constructs. Injection stocks contained a small volume (<1 µL, 1:50 dilution) of green food coloring to visualize localization of injection. Injection volumes ranged from 1 to 1.5 µL depending on the injection mix, keeping 1,000 ng total constant. Needles were pulled from 10 cm long filamented borosilicate glass capillaries at a pressure of 500, heat of 650, pull strength of 100, velocity of 200, and time delay of 40 in a Model P-97 Flaming/Brown micropipette puller from Sutter Instrument. Following injection, zebrafish were immediately electroporated (<1 minute) by placing the cathode probe of ultrasound gel-coated Platnium Tweezertrode, 3MM, on the side of the injection site and the anode probe on the other side of the fish body. Electroporation was achieved with five 60-millisecond pulses of 40 V with a 1-second pulse interval from an ECM 830 Electro Square Porator from BTX Harvard Apparatus. Following electroporation, zebrafish were placed into a recovery tank with fresh, gently stirred system water and monitored until normal swimming resumed. Electroporated zebrafish were imaged weekly. mitfa:GNAQ^Q209L; mitfa:GFP plasmid constructs were kindly provided by Dr. Jacqueline A. Lees (Massachusetts Institute of Technology). mitfa:Cas9, mitfa:tdTomato, PCS2FA Tol2, U6:gRNA-tp53, U6:gRNA-ptena, and U6:gRNA-ptenb plasmids were kindly provided by Dr. Richard White (Sloan Kettering Institute).

TEAZ-Eye

Preinjection preparation for eye injections is the same as for skin injections including: 0.2% tricaine anesthesia, needle size, and injection concentrations. To inject the eye, a blunt spherical head of a pin was gently pressed against the bottom half of the zebrafish eyeball to reveal the dorsal posterior regions of the eye. The needle was gently pressed into this location, where the cornea is softer, just through the sclera, keeping the needle as superficial as possible as to not damage the back of the eye. Injection volumes were around 0.25 µL and were terminated with a slight visual swelling of the eye. Following injection, zebrafish were immediately electroporated (<1 minute) by the cathode probe of ultrasound gel-coated Platnium Tweezertrode, 1MM, on the superior side of the eye nearest the injection and the anode probe on the inferior side of the eye. Electroporation was achieved with three 50-millisecond pulses of 75 V with a 1-second pulse interval from an ECM 830 Electro Square Porator from BTX Harvard Apparatus. Recovery and imaging procedures for these fish did not differ from the skin injections.

Imaging and image processing

Injected zebrafish were imaged using an upright Leica M205 FCA microscope with brightfield and GFP filters. Zebrafish were briefly anesthetized with 0.2% tricaine and positioned on a Kimwipe mound for upright orientation. Images were taken and processed through the LAS X software. All histology slides were imaged using brightfield, DAPI, and GFP filters on an EVOS M5000 microscope from Invitrogen for individual images or on a Leica DMi8 microscope and stitched together by LAS X Thunder Imaging software for full tissue images.

H&E and immunofluorescence – zebrafish tissue

Whole zebrafish or zebrafish eyes were fixed in 4% paraformaldehyde for 48 hours at 4°C and paraffin embedded in the University of Iowa Pathology Core. Fish were sectioned at 7 µmol/L thickness and placed on Superfrost Plus Microscope slides from Fisher Scientific, baked at 60°C, stained with H&E, and mounted with Cytoseal 60 from Epredia. For immunofluorescence (IF), sections were deparaffinized and hydrated followed by heat-induced antigen retrieval with 10 mmol/L Na-citrate buffer and blocking with 3% milk in 0.1% Tween 20.

The slides were stained with primary antibodies against GFP (NB600-308 rabbit anti-GFP polyclonal from Novus Biologicals, RRID: AB_10003058, at 1:250) overnight at 4°C and then fluorescent secondary antibodies against rabbit IgG (Alexa Fluor 488 goat anti-rabbit IgG, RRID: AB_2576217, at 1:500 from a 2 mg/mL stock) for 1 hour at room temperature. Retinal pigmented epithelium (RPE) autofluorescence was minimized using the Vector TrueVIEW Autofluorescence Quenching Kit. Sections were counterstained with DAPI, mounted using VECTASHIELD Vibrance Antifade Mounting Medium, and imaged the same day.

IF - mouse eye tissue

C57BL/6 mice (RRID: MGI:2159769) eyes were collected, formalin-fixed, paraffin-embedded, and sectioned into 7-µm–thick sections. Sections were deparaffinized and hydrated followed by heat-induced antigen retrieval with citrate buffer (Vector Laboratories, Inc.), permeabilized with 1% horse serum in 0.2% Triton-X, blocked with 5% horse serum in 0.2% Triton-X, and incubated with PAX3 primary antibody (Invitrogen 38-1801 rabbit anti-PAX3 polyclonal, RRID: AB_2533359, at 1:200) overnight at 4°C. Samples were incubated with DyLightTM 488-labeled secondary antibody (Vector Laboratories, DI-1088-1.5, horse antirabbit, RRID: AB_2336403, at 1:2,000) for 1 hour at room temperature, treated using an autofluorescence quenching kit (Vector Laboratories, Inc., SP-8400-15), and coverslips were fitted using Vectashield Antifade Mounting Medium with DAPI (Vector Laboratories, Inc., H-1000-10).

Kaplan–Meier analysis

Zebrafish were followed for up to 70 weeks, and tumor-free survival was analyzed with the Kaplan–Meier method. Tumors were defined as the first presence of a GFP-positive nodular mass at the injection site. Zebrafish were excluded from curves if they died or were euthanized without any visible signs of tumors during the study. Differences between injection groups were analyzed using log-rank statistics.

Single-cell suspension and single-cell RNA sequencing analysis

Zebrafish were euthanized with 0.4% tricaine for 10 minutes followed by immersion in ice water until no opercular movement was observed for at least 30 minutes as detailed in our IACUC protocol. Tumors were harvested and minced into small pieces. Minced tumors were digested for 1 hour at 37°C with collagenase and hyaluronidase in Hank’s Balanced Salt Solution (HBSS) with 2% FBS (HF). Suspensions were washed with 1:4 HF:NH₄Cl and then further digested for 5 minutes with prewarmed trypsin. Following trypsin inactivation, cell suspensions were washed twice with 10% FBS in HBSS.

Cellular suspensions were loaded on a 10x Genomics Chromium instrument to generate single-cell gel beads in emulsion (GEM). Approximately 10,000 to 20,000 cells were loaded per channel depending on the 10x kit used. See supplementary figures for kit information per sample.

Next GEM kit: targeted cell number = 10,000. Single-cell RNA sequencing (scRNA-seq) libraries were prepared using Single Cell 3′ Reagent Kits v2: Chromium Single Cell 3′ Library & Gel Bead Kit v2, PN-120237; Single Cell 3′ Chip Kit v2, PN-120236; and i7 Multiplex Kit, PN-120262 (10x Genomics) and following the Single Cell 3′ Reagent Kits v2 User Guide (Manual Part # CG00052 Rev A).

GEM-X kit: targeted cell number = 20,000. scRNA-seq libraries were prepared using Chromium GEM-X Single Cell 3′ Reagent Kits v4: Library Construction Kit C, PN-1000694; Single Cell 3′ GEM Kit v4, PN-1000693; Single Cell 3′ Gel Bead Kit v4, PN-2001128; and Dual Index Kit TT Set A, PN-3000431; and following the Chromium GEM-X Single Cell 3’ Reagent Kits v4 User Guide (Manual Part # CG000731 Rev B). Libraries were sequenced on an Illumina HiSeq 4,000 as 2 × 150 paired-end reads. Sequencing results were demultiplexed and converted to FASTQ format using Illumina bcl2fastq software.

A custom reference genome for the Zebrafish was constructed with GRCz11 primary assembly (Ensembl, RRID: SCR_002344) using Cell Ranger (Cell Ranger mkref function). 10x Genomics scRNA-seq reads were then processed and aligned to this reference with Cell Ranger (Cell Ranger Count function). Further analysis and visualization were performed using Seurat (v4.1.0; ref. 13), and cells with fewer than 200 RNA feature counts and greater than 5% mitochondrial contamination were removed with filtering (>15% for WT eye samples). RNA counts were normalized, and FindVariableFeatures was run with the following parameters: selection.method = “vst” and nfeatures = 2,000. The cells were originally clustered in a Uniform Manifold Approximation and Projection (UMAP) by RunPCA then RunUMAP with dims 1:30. FindNeighbors was run with dims 1:30 followed by FindClusters using a resolution of 0.4. These methods identified up to 20 clusters in individual samples.

For comparative analysis, RNA-seq datasets were integrated together using IntegrateData with dims 1:30. New clusters were generated as before. Clusters were annotated using FindAllMarkers, and the top 25 enriched genes in each cluster were compared with the Daniocell dataset (14) and Spectacle (15). Cluster names were given according to the closest matched cellular population between these 2 datasets. To determine enriched pathways, danio rerio gene names were converted into human Ensembl gene codes with g:Profiler (16), and then data were analyzed with the use of QIAGEN IPA (QIAGEN, Inc.; ref. 17) and gene set enrichment analysis (GSEA). (18) Pseudotime was performed using Monocle3 (v0.2.3.0; ref. 19).

Verification of gRNAs

Zebrafish were euthanized with 0.4% tricaine for 10 minutes followed by immersion in ice water until no opercular movement was observed for at least 10 minutes as detailed in the IACUC protocol. Tumors were harvested and minced into small pieces. DNA was harvested according to the DNeasy Blood and Tissue Kit from Qiagen. Regions surrounding cut sites were PCR-amplified, and the purity of the PCR amplicon was confirmed by agarose gel electrophoresis. Samples were cleaned up using the QIAquick PCR Purification Kit from Qiagen and sent for Sanger sequencing at the Iowa Institute of Human Genetics. Gene editing was confirmed with DECODR INDEL analysis by comparing with WT sequence (20). PCR primer sequences were as follows:

• mitfa PCR F: CCG​GTA​TGT​ATT​CAC​ATT​GTC​TTG

• mitfa PCR R: TCT​TGC​TTA​GGA​TGC​CTA​TGT​ATT

• ptena PCR F: CAT​CCC​ACC​AAG​TGA​GGT​TAA​AC

• ptena PCR R: CAC​ATA​CAC​AGT​CAA​GGG​TGA​G

• ptenb PCR F: CAG​TTC​TGT​TGC​ACC​CAA​TAA​G

• ptenb PCR R: CTG​GTG​GTG​TTG​AGG​CTA​TAA​AG

• tp53 PCR F: AAG​TAT​TCA​GCC​CCC​AGG​TG

• tp53 PCR R: CGC​TTT​TGA​CTC​ACA​GTG​CAA​G

Results

Generation of GNAQ-driven melanoma in adult zebrafish

To model GNAQ-driven melanoma in adult zebrafish, we generated mitfa promoter–driven plasmid systems expressing oncogenic GNAQ^Q209L, Cas9, and gRNAs targeting tp53 and ptena/b (Fig. 1A). We tested 2 delivery strategies: a single-plasmid GNAQ^Q209L-T2A-GFP construct incorporating all expression cassettes (strategy A), and a multiplasmid approach designed to avoid C-terminal modification of GNAQ by the T2A cleavage site while improving electroporation efficiency through the use of smaller individual plasmids (strategy B; see “Materials and Methods”). TEAZ applied to adult zebrafish skin (TEAZ-Skin) induced melanoma in WT animals using either strategy (Fig. 1B), with strategy B significantly reducing tumor latency (Fig. 1C and D; Supplementary Fig. S1A). Melanomas developed exclusively from clonal expansion of cells harboring edits introduced by all gRNAs. Neither oncogenic GNAQ expression nor individual tumor-suppressor loss alone was sufficient to drive tumorigenesis, demonstrating that combined oncogenic signaling and inactivation of all gRNA-targeted tumor-suppressors were required for transformation (Supplementary Fig. S1A). Consistent with prior reports, loss of mitfa markedly accelerated tumor onset following TEAZ-Skin, with tumors arising within ∼30 days in nacre (mitfa^w2/w2) and casper (mitfa^w2/w2, mpv17^a9/a9) zebrafish (Fig. 1C and D; Supplementary Fig. S1B–S1D). In these backgrounds (see “Materials and Methods”), oncogenic GNAQ alone was sufficient to drive melanoma, consistent with increased susceptibility of Mitfa-independent cells. Tumor identity was distinguished from xanthomas by gross morphology and pigmentation (Supplementary Fig. S1E), and CRISPR-mediated editing of target genes was confirmed by Sanger sequencing (Supplementary Fig. S2). Together, these data show that TEAZ-Skin efficiently induces GNAQ-driven melanoma in adult zebrafish, with strategy B providing faster tumor onset than strategy A and mitfa loss substantially accelerating disease compared with TEAZ-Skin in WT zebrafish (Fig. 1C and D).

Figure 1.

GNAQ ^Q209L-induced melanoma by TEAZ-Skin and TEAZ-Eye. A, TEAZ-Skin was used to induce GNAQ-positive melanoma using 2 strategies. Strategy A involves a single-plasmid construct containing expression cassettes driven by the mitfa promoter for GNAQ^Q209L-T2A-eGFP and Cas9, as well as a U6 promoter driving the expression of 3 separate gRNAs (targeting tp53, ptena, and ptenb, or a nontargeting control). Strategy B uses a multiplasmid approach with 5 plasmids for TEAZ-Skin. Expression cassettes include (a) mitfa:GNAQ^Q209L-mitfa:eGFP, (b) mitfa:Cas9, (c) U6:gRNA (tp53), (d) U6:gRNA (ptena), and (e) U6:gRNA (ptenb). B, Schematic of TEAZ-Skin using WT (AB), nacre (mitfa^w2/w2), and casper (mitfa^w2/w2: mpv17^a9/a9) zebrafish lines. C, TEAZ-Skin using strategy B plasmids in adult WT (AB) and casper zebrafish. Representative brightfield and fluorescent images shown at 4, 8, and 12 weeks after electroporation for WT and at 2, 4, and 6 weeks for casper. Scale bar, 0.5 cm. D, Kaplan–Meier curves were used to estimate tumor-free survival, defined as the time from electroporation until the first visible appearance of GFP+ nodular tumors, in adult zebrafish via TEAZ-Skin. Strategy B WT (n = 9), nacre (n = 5), and casper (n = 24) curves and TEAZ-Skin (GNAQ^Q209L:GFP) curves (n = 3 per condition) compared pairwise with log-rank (Mantel–Cox) test. E, Schematic of the zebrafish eye with dashed box indicating the retina and choroid. Comparative H&E of the retina and choroid from a WT zebrafish (left) and human donor (right; 20×). Choroidal melanocytes indicated with red arrows. Scale bar, 20 μm. F, IF analysis of transgenic Tg(mitfa:GFP) zebrafish using an anti-GFP antibody. Sections were counterstained with DAPI to visualize nuclei. Representative image showing GFP expression in mitfa-positive cells. Secondary-only antibody was used as background control. Scale bar, 50 μm. G, Schematic of TEAZ-Eye using choroidal targeted injection. H, Live imaging of adult WT and casper zebrafish following TEAZ-Eye with strategy B plasmids to induce uveal melanoma (UM). Brightfield imaging and the corresponding GFP fluorescence images highlighting tumor progression. Time points in dpe as labeled. Scale bars, 250 μm and 0.5 cm (whole zebrafish images). I, Kaplan–Meier curves comparing tumor-free survival among WT (n = 5) and casper (n = 13) zebrafish following TEAZ-Eye injection using strategy B plasmids. Curves compared pairwise with log-rank (Mantel–Cox) test. J, H&E analysis of UM in casper zebrafish induced by TEAZ-Eye. Images were acquired at 20× magnification and stitched from 85 fields using Thunder imaging software. Normal eye tissue and UM as labeled. Scale bar, 100 μm. Upper right is a representative 40× magnification image showing GFP expression in tumors cells after anti-GFP IF analysis of TEAZ-Eye–induced UM in casper zebrafish. B, Created in BioRender. Kenny, C. (2026) https://BioRender.com/e6ui09e; G, Created in BioRender. Kenny, C. (2026) https://BioRender.com/p23ej4l. CH, choroid; GCL, ganglion cell layer; ILM, internal limiting membrane; INL, inner nuclear layer; IPL, inner plexiform layer; ONL, outer nuclear layer; POS, photoreceptor outer segments; SCL, sclera; VIT, vitreous. *, P < 0.05; **, P < 0.01; ****, P < 0.0001; ns, not significant.

The mitfa promoter is active in choroidal melanocytes

We next sought to adapt the TEAZ approach to induce melanocyte transformation in the zebrafish eye (TEAZ-Eye). The zebrafish and human eyes share highly conserved overall organization and cellular composition [reviewed in (21)], including distinct retinal layers, RPE, and a melanocyte-containing choroidal layer. Most major human ocular compartments have clear zebrafish counterparts, although the zebrafish lacks a fovea and has a proportionally thinner choroid (Fig. 1E). We first assessed specificity of the mitfa promoter in driving transgene expression in choroidal melanocytes by performing IF imaging with an anti-GFP antibody in eyes harvested from transgenic Tg(mitfa:GFP) zebrafish. Choroidal melanocytes exhibited strong GFP positivity compared with the secondary antibody-only control (Fig. 1F). Quenching autofluorescence signal from RPE improved the signal-to-noise ratio (Supplementary Fig. S3A and S3B). We next applied TEAZ-Eye by injecting and electroporating plasmids containing mitfa promoter-driven GFP expression (mitfa:GFP) into the choroid. Eyes were harvested 2 weeks after electroporation to assess GFP expression by IF. Cells at the injection site exhibited strong GFP signal compared with the secondary antibody-only control (Supplementary Fig. S3C), demonstrating that the mitfa promoter is sufficient to drive transgene expression in choroidal melanocytes using TEAZ-Eye.

TEAZ-Eye induces uveal melanoma in the choroid of WT and casper zebrafish

A major limitation in the study of uveal melanoma onset and progression is the lack of anatomically accurate in vivo models. To address this gap, we used the TEAZ-Eye technique to introduce oncogenic GNAQ and tumor-suppressor gRNAs into the choroidal space of zebrafish (Fig. 1G). Remarkably, TEAZ-Eye in WT zebrafish promoted transformation of cells within the choroid. Live in vivo imaging of TEAZ-Eye enabled identification of individual GFP+ cells at the injection site, serving as a proxy for GNAQ^Q209L expression (Fig. 1H). We next applied TEAZ-Eye to mitfa-deficient casper zebrafish (Fig. 1H). As with TEAZ-Skin, tumor latency of TEAZ-Eye injected casper zebrafish (n = 13) was significantly reduced compared with WT (n = 5) (Fig. 1I). H&E analysis confirmed that TEAZ-Eye induced uveal melanoma within the choroid, progressing into surrounding ocular structures (Fig. 1J). Notably, 100% (n = 10) of tumors formed between the sclera and RPE, highlighting the model's ability to induce anatomically accurate uveal melanoma. Finally, anti-GFP IF analysis confirmed expression of plasmid constructs following TEAZ-Eye (Fig. 1J).

Uveal melanoma transformation involves activation of NC transcriptional programs

During embryonic development, choroidal melanocytes are derived from the cranial NC (CNC; ref. 22). As our model can detect the earliest events in uveal melanoma transformation, we asked whether reactivation of CNC programs is one of the initial steps in uveal melanoma development. To address this point, tumors were derived in WT zebrafish using TEAZ-Eye with strategy B plasmids. At 50 to 90 dpe, when gross nodular tumors were identified, surgical enucleation was performed, and dissociated cells from the entire eye were subjected to scRNA-seq. We compared pooled WT tumors (2 sequencing replicates) with tumor-free paired eyes or sibling-matched control eyes (2 and 4 pooled eyes, respectively, 2 sequencing replicates) (Fig. 2A; Supplementary Fig. S4). Given that whole eyes were used as control samples and melanocytes are most abundant in the choroid, melanocytes identified under control conditions are hereafter referred to as choroidal melanocytes. Tumor cells were identified by expression of GFP and oncogenic GNAQ (Fig. 2B), whereas primary choroidal melanocytes were marked by expression of mlana (Supplementary Fig. S5A and S4B). Annotation of additional cell types within the tumor microenvironment (TME) was accomplished by comparing with previously published human choroid and retina scRNA-seq datasets, (15, 23) the Zebrafish Information Network (24), and from the zebrafish embryo single cell atlas “Daniocell” (Fig. 2A; ref. 14). Interestingly, tumor cells aligned most closely with choroidal melanocytes, ruling out the RPE as an origin of our TEAZ-Eye–induced tumors (Fig. 2C). In addition, profiling GFP/GNAQ^Q209L+ uveal melanoma cells revealed that tumors exhibited strong expression of NC markers (tfap2a, tfap2c, pax7b, sox10, and foxd3) and melanocytic markers (mitfa, tfap2e, kita, mlana, pmela, slc22a7a, mlpha, and mtbl; Fig. 2D). Unexpectedly, CNC marker genes twist1 or dlx1 (25, 26) were not enriched in uveal melanoma tumors compared with the TME. Pathway analysis showed that genes enriched in TEAZ-Eye–derived tumors were strongly associated with MITF-M–dependent gene expression, RAF/MAP kinase, mTOR and PI3K/AKT signaling, NF-κB, RHO GTPase, and NOTCH activity, as well as fatty acid oxidation pathways (Supplementary Fig. S6A and S6B).

Figure 2.

Single-cell profiling of uveal melanoma (UM) and the TME in adult zebrafish. A, UMAP representation of scRNA-seq from 3 dissociated WT eyes harboring UM. UM tumors were generated by TEAZ-Eye (sequencing replicates, n = 2). Control eyes represent pools of 2 tumor-free paired WT eyes and 4 sibling-matched WT eyes (sequencing replicates, n = 2). Annotated cell clusters as labeled. B, UMAP and feature plot showing eGFP and GNAQ^Q209L expression in control eyes and UM tumor. Red dotted line highlights the melanocyte and UM clusters. C, Comparison of UM tumor marker genes with the markers of cell types within the control eye ME. Only genes with a log₂FC > 1 and FDR < 0.01 were included in the analysis. D, Dot plot representing enriched genes (log₂FC > 1; FDR < 0.01) in choroidal cell populations, UM tumors cells, immune cell populations, fibroblasts, Schwann cells, pericytes, and vascular endothelial cells as shown. Dot size indicates the percentage of cells expressing each gene; color intensity (gray to blue) reflects normalized average expression levels (low to high). E, UMAP clustering of cells in the choroidal ME. Annotated cell clusters as labeled. F, Comparison of relative cluster sizes between the 2 conditions, control and TEAZ-Eye, shown in E. Schwann cells serve as a control cluster, displaying no change in frequency between conditions. Gray = TEAZ-Eye; blue = control. G, Venn diagram representing the overlap of genes between UM tumors and choroidal melanocytes compared with the cells in their respective MEs. Enriched genes for each condition were identified by differential expression analysis comparing melanocytes with the ME or tumor cells to the TME using a threshold of log₂FC > 1 and FDR < 0.01. Genes of interest have been labeled in their respective subsets; bolded genes represent genes associated with worse overall survival or disease-free survival in human UM TCGA datasets. H, Dot plot showing enriched genes (log₂FC > 1; FDR < 0.01) in UM tumor cells, choroidal melanocytes, or the TME. Dot size indicates the percentage of cells expressing each gene; color intensity (gray to blue) reflects normalized average expression levels (low to high). I, HOMER promoter motif enrichment analysis of genes enriched in UM tumors vs the TME in G. The top ranked motif for “known” and “de novo” motifs are shown with their corresponding P values. J–L, Kaplan–Meier disease-free survival and overall survival curves generated using UM TCGA data. Patients stratified by relative expression levels of KIT, FABP3, or LGALS2. Gene expression thresholds were defined by median expression within each cohort. Log-rank P values as shown. Dotted lines represent 95% confidence interval. M, Dot plot showing enriched genes (log₂FC > 1; FDR < 0.01) in CAFs, normal fibroblasts (NF), or the TME without CAFs. Dot size indicates the percentage of cells expressing each gene; color intensity (gray to blue) reflects normalized average expression levels (low to high). Genes have been grouped according to their labeled functions.

Reclustering of choroidal cells from control and tumor conditions revealed a marked expansion of tumor cells relative to melanocytes, accompanied by increases in immune cells and fibroblasts (Fig. 2E and F). Immune cell populations within the TME included B cells (igic1s1, pax5, cd79a, and zgc:153659), T cells (traf1, si:ch211-67e16.3, cd27, zap70, cd8a, and sla2), a population of cytokine producing cells, likely representing T cells (il4, il13, il11, ca2, il34, and il6), as well as macrophages (marco, ccl34a.4, mrc1b, c1qa/b/c, and grn1) and neutrophils (cpa5, lect2l, mmp9, npsn, and scpp8). Additional cell types within the choroidal TME included fibroblasts (fn1a, col1a1a, col1a1b, col1a2, col5a1, dcn, and serpinf1), pericytes (pdgfrb, notch3, foxf2b, and cd248a), Schwann cells (dlx5a, apoda.1, mpz, plp1b, cd59, s100b, and mbap), and vasculature endothelial cells (vwf, flt4, epas1b, ecscr, and fabp11a; Fig. 2D–F; Supplementary Table S1).

We next compared choroidal melanocytes and uveal melanoma tumors, revealing a significant overlap in gene expression (n = 644; hypergeometric P < 0.0001) alongside a large set of tumor-enriched genes (n = 1,177; Fig. 2G). Shared genes included melanocytic regulators (sox4a, ednrb, sox10, foxd3, mitfa, mlpha, kita, mlana, and pax7b), supporting a melanocytic origin, with many transcripts further upregulated in uveal melanoma tumors compared with choroidal melanocytes (Fig. 2H). HOMER promoter motif enrichment analysis of genes enriched in uveal melanoma tumors revealed binding motifs for the MiT/TFE family of transcription factors, as well as PAX3/7 and HOXB13, suggesting that these regulators may contribute to uveal melanoma–specific transcriptional programs (Fig. 2I).

To assess the relevance of our model to human disease, we examined whether genes enriched in zebrafish uveal melanoma tumors were associated with clinical outcomes in patients. Differential expression analysis identified a gene signature, including LGALS2, FABP3/5, GCH, MET, and KIT, whose high expression was associated with reduced overall survival, disease-free survival, and with The Cancer Genome Atlas (TCGA) molecular subtypes 3 and 4 (27), corresponding to aggressive uveal melanoma (Fig. 2G, and J–L; Supplementary Figs. S7 and S8, Supplementary Table S2).

Cancer-associated fibroblasts within the uveal melanoma TME upregulate components of the extracellular matrix

Recent studies have significantly advanced our understanding of the TME as an active driver of melanoma progression (28, 29). Although it is well-established that components of the TME, particularly cancer-associated fibroblasts (CAF), play a key role in malignancy (30–32), in vivo models to study such interactions in uveal melanoma remain limited. Our model allows for direct comparison of the ME in TEAZ-Eye–injected and control-injected eyes within the same animals (i.e., left and right eyes). We compared the transcriptome of CAFs and normal fibroblasts in WT zebrafish. Upon reclustering and differential gene expression analysis, we found that several ECM proteins were upregulated in CAFs, indicating ECM remodeling in the presence of uveal melanoma. Enriched genes broadly fell into 4 categories: ECM, ECM regulators, TNFs, and collagens. Notably, fibronectin (fn1a) was among the most significantly upregulated genes in CAFs compared with normal fibroblasts (q < 0.0001). Several collagens were also found to have higher expression in CAFs, including collagen I (col1a1b and col1a2), collagen V (col5a1, col5a2a, and col5a3a), and collagen XII (col12a1a; Fig. 2M; Supplementary Table S3). Consistently, cell–cell communication analysis using CellChat (33) identified FN1 and collagen signaling from CAFs to tumor cells as significantly enriched pathways and predicted the transmembrane heparan sulfate proteoglycan SDC4 as a mediator of these interactions in uveal melanoma cells (Supplementary Fig. S9).

To control for injection-related fibrosis potentially causing fibroblast activation and upregulation of ECM components, we mock-injected 3 WT eyes with plasmids containing mitfa:GFP and mitfa:Cas9 cassettes followed by electroporation (mock-TEAZ) and performed scRNA-seq. Fibroblasts from mock-injected sibling controls (3 pooled eyes), noninjected sibling controls (4 pooled eyes), and noninjected tumor-matched control eyes (2 pooled eyes) were equally distributed on the UMAP, with no alteration in fn1a or lgals2a expression or other changes to ECM genes upon mock-TEAZ (Supplementary Fig. S10). These findings indicate that TEAZ-Eye tumor induction depends on genetic modification of target cells, not simply physical injection, triggering a CAF response that may be critical to melanoma progression.

Skin and eye primary melanocytes, and their resulting tumors, have unique gene expression profiles

Having established a system to induce genetically identical tumors in the eye and skin of adult zebrafish, we asked whether tumor transcriptional programs differed by anatomic site. Such differences could reflect site-specific tumor–stromal interactions or intrinsic differences between uveal and cutaneous melanocytes. We generated TEAZ-Eye and TEAZ-Skin tumors using strategy B plasmids and performed scRNA-seq with matched control eye and skin tissues (Fig. 3A). We compared tumors with a combined TME and performed a parallel analysis of eye and skin primary melanocytes relative to the combined ME (Fig. 3B; Supplementary Table S4). GNAQ^Q209L expression levels were comparable between eye and skin WT tumors (Supplementary Fig. S11A), and differentially expressed genes (DEG) showed substantial overlap between sites, with 41% of 2,192 genes shared, consistent with common transcriptional programs driven by oncogenic GNAQ during transformation (Fig. 3B). Notably, only 25.8% of DEGs overlapped between eye and skin melanocytes, indicating substantial transcriptional differences in melanocytes originating from distinct anatomic regions (Fig. 3B).

Figure 3.

Comparison of GNAQ-driven tumors and primary melanocytes in the skin and eyes of adult zebrafish. A, UMAP representation of GNAQ-positive tumors generated using TEAZ-Eye and TEAZ-Skin, along with their respective control tissues using scRNA-seq. Tumor samples include 3 dissociated WT eyes harboring uveal melanoma (UM) tumors generated via TEAZ-Eye and 2 dissociated WT skin UM tumors generated via TEAZ-Skin (n = 2 sequencing replicates each). Control eye samples consist of a pool of 2 uninjected tumor paired WT eyes, 3 mock-injected sibling-matched WT eyes, and 4 uninjected sibling-matched WT eyes (n = 3 sequencing replicates). Control skin samples consist of dissociated normal skin from 2 WT zebrafish (n = 2 sequencing replicates). Cell clusters are annotated as tumor cells (red = eye; blue = skin), TME cells shared by both eye and skin samples (light gray), and TME cells specific to the eye samples (dark gray). B, Venn diagram showing overlapping genes between TEAZ-Eye, TEAZ-Skin, choroidal melanocytes, and skin melanocytes compared with the cells in their respective MEs. Enriched genes for each condition were identified by differential expression analysis comparing melanocytes with the ME or tumor cells to the tumor TME, using a threshold of log₂FC > 0.5 and FDR < 0.01. Bar chart displaying overlaps between enriched gene sets from B. Percentages represented as the number of overlapping genes over the total unique genes in both gene sets. C, UMAP representation of primary melanocytes from both eye (salmon) and skin (teal). Volcano plot of DEGs between eye and skin primary melanocytes (log₂FC > |0.5| and FDR < 0.01). D, GSEA enrichment plots of pathways that were enriched in skin melanocytes and of pathways enriched in eye melanocytes from C. P values and FDRs as labeled. E, UMAP representation of GNAQ^Q209L-induced melanoma from both eye (salmon) and skin (teal). Volcano plot of DEGs between eye and skin tumors (log₂FC > |0.5| and FDR < 0.01). F, GSEA enrichment plots of pathways that were enriched in skin melanoma and of pathways enriched in eye melanoma from E P values and FDRs as labeled.

We hypothesized that DEGs shared by only the 2 tumor sites and not the primary melanocytes would represent essential mechanisms for uveal melanoma growth and survival. Consistent with this hypothesis, pathway analysis revealed enrichment of pathways involved in cancer proliferation and metabolism (Supplementary Fig. S12A). Additional enriched pathways included WNT signaling, PTEN regulation, and RAF/MAPK signaling, which have been previously implicated in uveal melanoma pathogenesis (34–42). In contrast, IPA analysis of gene sets shared between eye and skin melanocytes represented homeostatic pathways, including enrichment of MITF-M–dependent programs, as well as pathways related to nucleotide biosynthesis and mTORC1-regulated amino acid metabolism (Supplementary Fig. S12B).

We next reclustered normal melanocytes from eye and skin for direct comparison, revealing distinct UMAP clusters (Fig. 3C). We identified 186 DEGs enriched in eye melanocytes and 229 DEGs enriched in skin melanocytes (log₂FC > |0.5|, FDR < 0.01), whereas 3,387 genes were not differentially expressed (FDR > 0.01, expressed in ≥10% of cells, Supplementary Table S5). GSEA revealed that skin melanocytes were enriched for DNA repair pathways and UV response–associated genes (Fig. 3D). In contrast, eye melanocytes showed enrichment for pathways involved in UV response downregulation and oxidative phosphorylation. We next reclustered and directly compared eye and skin tumors, which also separated by tissue of origin on the UMAP (Fig. 3E). We identified 230 DEGs enriched in eye tumors and 375 DEGs enriched in skin tumors (log₂FC > 0.5, FDR < 0.01), whereas 5,754 genes were not differentially expressed between tumor types. Eye tumors were enriched for epithelial-to-mesenchymal transition (EMT) and selenoamino acid metabolism pathways (Fig. 3F). In contrast, skin tumors were enriched for PI3K–AKT–mTOR signaling, RHO GTPase–mediated activation, and YAP1 activity. Together, these analyses demonstrate that both primary melanocytes and GNAQ^Q209L-driven melanomas exhibit site-specific transcriptional programs.

Melanocyte progenitor cells are highly susceptible to oncogenic GNAQ transformation in mitfa-deficient zebrafish

H&E analysis of nacre and casper zebrafish eyes showed loss of choroidal melanocytes (Fig. 4A), whereas melanin in the RPE remained less affected, as previously demonstrated (43). These observations imply that an mitfa-independent melanocyte progenitor cell persists in nacre and casper eyes and that they are susceptible to oncogenic GNAQ-driven transformation. To identify such cells, we crossed nacre (mitfa^w2/w2) with transgenic Tg(mitfa:GFP) lines to create Tg(mitfa:GFP); mitfa^w2/w2 zebrafish. We then harvested adult eyes at 5 months and performed anti-GFP IF. Whereas GFP+ melanocytes were readily detected in the choroid of Tg(mitfa:GFP) zebrafish (Fig. 1F, Supplementary Fig. S3), staining of Tg(mitfa:GFP); mitfa^w2/w2 eyes revealed rare cells with GFP positivity (Fig. 4B). Such GFP+ cells were primarily located in the ciliary body, with isolated cells also found in the choroid, suggesting the persistence of an mitfa-independent progenitor population in adult mitfa-deficient zebrafish eyes (Fig. 4B). We next asked whether such cells retain the ability to undergo melanocyte differentiation by using TEAZ-Eye to deliver plasmids containing the mitfa promoter driving expression of mitfa (i.e., to overexpress mitfa mRNA). Interestingly, reexpression of mitfa transcript resulted in expansion of choroidal melanocytes in WT animals and rescue of choroidal melanocytes in mitfa-deficient (nacre) zebrafish as shown by H&E analysis (Supplementary Fig S13A and S13B). To assess whether GFP-positive cells are sensitive to oncogenic transformation by GNAQ, we expressed mitfa:GNAQ^Q209L in Tg(mitfa:GFP); mitfa^w2/w2 zebrafish using TEAZ-Eye. In all injected fish (n = 3), uveal melanoma tumors were GFP-positive, suggesting they originated from GFP+ progenitor cells within the eye of Tg(mitfa:GFP); mitfa^w2/w2 zebrafish (Supplementary Fig. S13C).

Figure 4.

mitfa-independent melanocyte progenitor cells are expanded in mitfa-deficient zebrafish and are susceptible to GNAQ^Q209L transformation. A, H&E analysis representation of the retina and choroid in WT (left), nacre (middle), and casper (right) zebrafish (20× magnification). Note the loss of melanocytes (red arrows) in the choroid of nacre and casper eyes. Scale bar, 50 μm. B, Immunofluorescent analysis of transgenic Tg(mitfa:GFP); mitfa^w2/w2 (nacre) zebrafish using an anti-GFP antibody. Sections were counterstained with DAPI to visualize nuclei. Representative images showing GFP-positive cells in the ciliary body and choroid are shown by yellow arrowheads. Secondary-only antibody was used as background control. Scale bar, 20 μm. C, UMAP representation of scRNA-seq data from 3 paired dissociated casper eyes and from 3 casper eyes harboring uveal melanoma (UM) tumors induced using strategy B via TEAZ-Eye (n = 1, sequencing replicate). Annotated cell clusters are labeled. Progenitor cells and UM tumor clusters are outlined with dashed red lines. Representative brightfield images of a normal eye and a UM tumor in casper zebrafish are shown in the top left corner of the UMAP. Scale bar, 250 μm. D, Reclustered UMAP representing progenitor cells from control eyes and UM tumor cells from TEAZ-Eye–injected casper zebrafish. Heterogeneous cell populations as labeled. E, Reclustered UMAP from D showing melanocyte progenitor cells from control eyes in red and UM tumor cells from TEAZ-Eye in blue. Feature plots for pax3a, foxd3, and sox10 expression are shown; dashed line marks the progenitor cell population. F, Dot plot illustrating DEGs (log₂FC > 0.5; FDR < 0.05) between heterogeneous cell populations in D. Dot size indicates the percentage of cells expressing each gene; color intensity (blue to red) reflects normalized average expression levels (low to high). G, Kaplan–Meier curves comparing tumor-free survival in WT and casper zebrafish following TEAZ-Eye injection using strategy B plasmids for mitfa conditional KO. WT zebrafish were injected with either strategy B plasmids plus nontargeting gRNAs (n = 9) or strategy B plasmids plus mitfa-targeting gRNAs (n = 9). Casper zebrafish were injected with strategy B plasmids and mitfa-targeting gRNAs (n = 9). H, Representative brightfield and GFP-overlay images of tumors that formed in casper (9 of 9) and WT zebrafish injected with mitfa-targeting gRNAs (8 of 9). Metastatic cells are indicated by red arrows. Scale bar, 0.5 cm. I, UMAP obtained after clustering GFP-positive cells sorted from Tg(mitfa:GFP) and Tg(mitfa:GFP); mitfa^w2/w2 (nacre) zebrafish embryos at 28 hpf. Annotated cell clusters as labeled. J, Comparison of relative cluster size for the 2 genotypes; Tg(mitfa:GFP) and Tg(mitfa:GFP); mitfa^w2/w2 (nacre) in I. Scale bar, 0.5 mm. K, Dot plot illustrating DEGs (log₂FC > 0.5; FDR < 0.05) in melanophores and progenitor cells in mitfa^w2/w2 and sibling embryos at 28 hpf. Dot size indicates the percentage of cells expressing each gene; color intensity (gray to red) reflects normalized average expression levels (low to high). CH, choroid; GCL, ganglion cell layer; INL, inner nuclear layer; ONL, outer nuclear layer; POS, photoreceptor outer segment; VIT, vitreous. **, P < 0.01; ***, P < 0.001; ns, not significant.

Melanocyte progenitor cells express NC marker genes in mitfa-deficient zebrafish

We performed scRNA-seq and pseudotime analysis on 3 pooled uveal melanoma tumors from the eyes of casper zebrafish and compared them with the pooled tissue from 3 normal casper eyes (Fig. 4C; Supplementary Fig. S14). Within the casper eyes, we identified a population of cells expressing the mitfa^w2/w2 allele that clustered closely with uveal melanoma tumors under TEAZ-Eye conditions, termed progenitors (Supplementary Fig S15A and SB; Supplementary Tables S6–S8). We reclustered the progenitor and tumor cells (Fig. 4D), revealing 7 distinct clusters consisting of NC-like, melanoblast-like, proliferating, progenitor, angiogenic, stress-like, as well as an unidentified cluster marked by tmsb1, krt91, and vim (Fig. 4D; Supplementary Table S9). Unsupervised pseudotime analysis supports a lineage trajectory from the progenitor cells to heterogeneous uveal melanoma tumor cells (Supplementary Fig. S15C). Notably, the progenitor cell population in normal casper eyes expressed neural crest and melanocyte stem cell marker genes, including tfap2a, foxd3, sox10, pax3a, vim, and yap1 (Fig. 4E and F). Uveal melanoma tumor cells retained expression of these genes but activated additional genes associated with pigment progenitor cell function, including tfec and kita (Fig. 4F). Notably, mitfa expression in tumors and progenitor cells in casper zebrafish reflects the expression of the mutant mitfa^w2/w2 allele, which functions as a lineage tracer to mark cells that would normally express mitfa under WT conditions (Supplementary Fig. S15B). Interestingly, despite the lack of Mitfa activity in casper zebrafish, GNAQ^Q209L-positive uveal melanoma tumors included a melanoblast-like cluster that exhibited expression of pigmentation genes, including pmela, mlpha, and mlana, likely explaining the occasional emergence of melanin within tumors as they progress in casper zebrafish.

Conditional deletion of mitfa in adult zebrafish fails to recapitulate germline loss-of-function effects on tumor-free survival in GNAQ^Q209L-driven melanoma

We next asked whether the accelerated growth of GNAQ^Q209L-driven tumors in mitfa-deficient zebrafish was attributable to increased oncogene expression or cell-intrinsic loss of Mitfa in adult melanocytes. scRNA-seq analysis showed higher GNAQ^Q209L expression in WT tumors than in casper tumors (Supplementary Fig. S11B), thus excluding oncogene overexpression as the basis for the accelerated phenotype. To directly test cell-intrinsic effects of mitfa loss in adult melanocytes, we performed conditional somatic deletion of mitfa using strategy B plasmids with either nontargeting gRNAs or gRNAs targeting exon 5 of mitfa. TEAZ-Skin was conducted in 3 groups: (i) casper zebrafish with mitfa-targeting gRNAs (n = 9; which controls for off-target effects of the gRNA and for potential function of the mitfa^w2/w2 allele); (ii) WT zebrafish with nontargeting gRNAs (n = 9); and (iii) WT zebrafish with mitfa-targeting gRNAs (n = 9). GFP expression was confirmed in all injected animals. As expected, by 100 dpe tumors developed in 100% (9 of 9) of casper zebrafish. Interestingly, there was no significant difference in tumor latency between WT and mitfa conditional KO, with both indicating significantly longer latency than in casper animals (Fig. 4G and H). Thus, somatic loss of mitfa in adult melanocytes is insufficient to reproduce the reduced tumor-free survival observed with germline mitfa deficiency. Notably, mitfa-deficient tumors arising in either background developed distal metastases, a feature not observed in mitfa WT TEAZ-Skin tumors (Fig. 4H; Supplementary Fig. S16). Together, these findings argue against Mitfa as a repressor of GNAQ-induced transformation in adult melanocytes and indicate that somatic loss in the adult is insufficient to drive tumor initiation.

Mitfa-deficient zebrafish exhibit expanded pax3a- and tfec-positive melanocyte progenitors during embryogenesis

We next asked whether melanocyte progenitor cells are expanded in mitfa-mutant zebrafish relative to WT zebrafish. To test this prediction, we performed scRNA-seq on GFP-positive cells sorted from Tg(mitfa:GFP);mitfa^w2/w2 and mitfa^+/w2 sibling-matched transgenic zebrafish embryos at 28 hpf, when melanocytes (zebrafish melanophores) begin to differentiate (Fig. 4I; Supplementary Fig. S14). We assigned cell-type annotations based on our previously annotated GFP-positive cells from Tg(mitfa:GFP) embryos (44). The 10 clusters included 6 main cell types: NC cells (sox10 and foxd3), a tripotent precursor of melanoblasts (M), iridoblasts (I), and xanthoblasts (X; tfap2a, cdkn1ca, slc15a2, ino80e, id3, mycn, and tfec), termed MIX cells, a cluster that expressed high levels of melanoblast/xanthoblast markers (mitfa, erbb3b, impdh1b, gch2, and id3), termed MX cells, a melanoblast cluster (mitfa, dct, pmel, and tyr), as well as 2 additional clusters corresponding to xanthoblasts and xanthophores (Supplementary Tables S10 and S11). For this analysis, we referred to MIX and MX cells as melanophore progenitors (Fig. 4I). As expected, melanophore cells were reduced in Tg(mitfa:GFP): mitfa^w2/w2 zebrafish (Fig. 4I and J). However, melanophore progenitors were expanded in Tg(mitfa:GFP); mitfa^w2/w2, whereas NC cells remained unaffected (Fig. 4I and J). Melanophore markers were downregulated in GFP-positive cells sorted from Tg(mitfa:GFP); mitfa^w2/w2, whereas melanocyte progenitor markers (pax3a, sox10, and foxd3), as we identified in adult zebrafish eyes, were upregulated in embryonic progenitors from Tg(mitfa:GFP); mitfa^w2/w2 embryos (Fig. 4K; Supplementary Table S12). A role for pax3a in adult fish eye tissue has not been previously characterized; however, recent studies have identified PAX3 expression in ocular surface melanocytes (45), and another study reports an association between PAX3 and stem cell markers in uveal melanoma progression (46).

Transcriptional programs enriched in oncogenic BRAF- versus GNAQ-driven melanoma overlap with specific melanocyte progenitor populations

Oncogenic BRAF and NRAS are unable to promote transformation in the absence of mitfa, (6, 47) suggesting that distinct transcriptional programs, potentially reflecting unique cellular origins, are required for uveal melanoma and CM transformation. To address this hypothesis, we compared previously published scRNA-seq datasets in transgenic Tg(mitfa:BRAF^V600E): tp53^−/−: mitfa^w2/w2 zebrafish in which TEAZ-Skin was used to rescue mitfa and induce melanoma (47) with our GNAQ^Q209L-induced TEAZ-Skin tumors (Fig. 5A). We predicted that by analyzing primary tumors, the transcriptional signature of the origin cell will be retained and can be identified by overlapping gene signatures identified in the embryonic melanocyte lineage (44). Interestingly, whereas the expression of mitfa was significantly higher in GNAQ-positive tumors, the expression of pigmentation genes, including tyrp1b, dct, scl45a2, sox10, and pmela, were significantly higher in BRAF-positive melanoma (Supplementary Fig. S17). Genes enriched in GNAQ–uveal melanoma and BRAF-CM were defined as those with a log₂FC > 1 and adjusted P < 0.0001 between the 2 tumor types (Fig. 5B; Supplementary Table S13). We next overlapped this differential gene signature with genes expressed by cells within the melanocyte lineage isolated from zebrafish embryos (Fig. 5C). As expected, BRAF-driven tumors strongly mapped to melanoblast and melanophore clusters (Fig. 5D), supporting recent findings that BRAF-driven melanoma originates from melanoblasts, a melanocyte progenitor population (47, 48). When comparing GNAQ-enriched genes with the melanocyte lineage, we found that signatures were not confined to a single cluster but were distributed among NC cells, melanocyte progenitors, and xanthophore clusters (Fig. 5D; Supplementary Fig. S18). Such results support the hypothesis that GNAQ- and BRAF-driven melanomas depend on distinct transcriptional programs, reflecting less and more differentiated melanocytic states, respectively.

Figure 5.

Oncogenic GNAQ and BRAF transform transcriptionally distinct cells within the melanocyte lineage. A, UMAP obtained after clustering tumor cells from Tg(mitfa:BRAF^V600E): tp53^−/−: mitfa^w2/w2 zebrafish where TEAZ-Skin was used to rescue mitfa expression and KO ptena and ptenb to induce CM (47), along with our GNAQ-induced tumors by TEAZ-Skin (strategy B: mitfa:Cas9, U6:gRNA-tp53, U6:gRNA-ptena, U6:gRNA-ptenb, mitfa:GNAQ^Q209L-mitfa:GFP). BRAF-induced tumors are shown in red, and GNAQ-induced tumors in blue. B, Violin plot showing DEGs (FDR < 0.05) between BRAF-induced and GNAQ-induced tumors from A. The plot displays the distribution of gene expression across individual cells, with the width indicating cell density at specific expression levels and the height representing the range of expression values. C, Integrated UMAP after clustering GFP-positive cells sorted from Tg(mitfa:GFP) and Tg(mitfa:GFP); mitfa^w2/w2 (nacre) zebrafish embryos at 28 hours after fertilization (hpf). Annotated cell clusters as labeled. D, Average enrichment of signature genes from BRAF- and GNAQ-driven tumors projected onto the UMAP in C using UCell with ‘FeaturePlot’ function. Signature genes represent the top 50 enriched genes from each tumor genotype that are also expressed in the embryonic melanocyte lineage. E, UMAP of primary melanocytes and melanocyte progenitors from 9 control WT eyes, 2 control WT skin sections, and 3 control casper eyes. Cells are labeled according to their genotype (casper and WT) and tissue origin (eye and skin). F, Feature plots representing the relative mRNA expression of pax3a, tfec, sox4a, yap1, sox10, foxd3, and ednrba in melanocytes described in E. Color intensity (gray to red) reflects normalized average expression levels (low to high). The red dotted line indicates cells that are clustering closely with the melanocyte progenitors identified in casper eyes. G, Schematic illustrating GNAQ^Q209L-driven uveal melanoma (UM) in the context of tumor-suppressor loss. Melanocyte progenitor cells (pax3a/b, tfec, and sox4a) can directly give rise to UM when oncogenic GNAQ^Q209L is present (black arrow) or differentiate into mitfa-high melanocytes that preferentially form nevi (black arrow). These nevi may subsequently progress to UM following reduced Mitfa activity (gray dashed arrow). In contrast, germline mitfa loss blocks melanocyte differentiation and expands melanocyte progenitors, enabling GNAQ^Q209L to drive UM directly from this progenitor population. G, Created in BioRender. Kenny, C. (2026) https://BioRender.com/wrl0qoc.

We next asked whether less differentiated melanocytes, similar to the mitfa-deficient progenitors in casper eyes, are also present in adult WT eyes. As melanocyte stem cells are known to reside in the zebrafish skin (49), we included skin tissue as a control and reclustered melanocytes from WT skin and eyes with progenitor cells from casper eyes. Interestingly, a subset of WT melanocytes from both eye and skin samples clustered closely with casper progenitors (Fig. 5E; Supplementary Fig. S19) and expressed pigment cell progenitor marker genes, including pax3a (50), tfec (43), sox10 (51), sox4a (bioRxiv 2025.12.23.695681; ref. 52), foxd3 (53), and yap1 (Fig. 5F). These findings suggest either melanocyte plasticity within the eye or the presence of an eye-resident melanocyte progenitor cell population in adult WT zebrafish. In support of this notion, reanalysis of published scRNA-seq datasets from human (23) and mouse (54) eyes identified PAX3-positive melanocyte subpopulations (Supplementary Fig. S20). Moreover, we confirmed the presence of PAX3-positive cell populations in adult C57BL/6 mouse eyes by immunohistochemistry (Supplementary Fig. S21). Together, these data support the existence of a PAX3-expressing melanocyte precursor population in both fish and mammalian eyes, which may mirror PAX3 expression in skin melanocyte stem cells (50, 55). Such cells may be more susceptible to GNAQ^Q209L-driven transformation and represent candidate cells of origin for uveal melanoma in this context (Fig. 5G).

Discussion

Our uveal melanoma model enables analysis of the earliest events in uveal melanoma initiation and progression. Compared with transgenic approaches, this system produces anatomically correct tumors with reduced tumor latency. Importantly, genes identified in this model are associated with survival differences in human uveal melanoma, offering an efficient and clinically relevant platform for studying uveal melanoma pathogenesis.

We found that fatty acid–binding proteins, particularly fabp3, are enriched in uveal melanoma tumors. Whereas lipid droplets have been identified as vulnerable targets in CM and regulators of melanoma plasticity (47, 56), lipid signaling in uveal melanoma has received limited attention (57–60). Interestingly, high expression of FABP3 and FABP5 are associated with worse disease-free survival in patients with uveal melanoma but not CM. Our analysis also identified LGALS2 and GCH, two fatty acid metabolism–related genes whose expression was strongly associated with worse overall survival in uveal melanoma but showed an inverse relationship in CM. We also found increased kita expression in GNAQ-driven tumors compared with choroidal melanocytes, which is particularly notable given recent studies linking KIT-positive uveal melanoma tumors with poor prognosis and chromosome 3 monosomy (61). In addition, met expression was elevated in uveal melanoma tumors relative to choroidal melanocytes. MET signaling is a key pathway in uveal melanoma pathogenesis, with established roles in tumor invasion, metastatic progression, and responsiveness to hepatocyte growth factor, particularly in the context of liver metastasis (62, 63). Notably, MET activation has also been described in other cancer types through ligand-independent fibronectin–integrin α5β1–mediated signaling (64). Given the marked increase of fn1a in choroidal CAFs, it is plausible that fibroblast-driven ECM remodeling enhances MET signaling in uveal melanoma. Furthermore, expression of these genes was highest in molecular subclasses 3 and 4 of the uveal melanoma TCGA dataset which represents the subset of patients with the highest rates of BAP1 loss and metastasis (27). Such observations demonstrate the utility of our model in uncovering key genes with disease-specific roles in uveal melanoma whose function could be further explored through genetic perturbations in zebrafish.

To explore how anatomical context influences transcriptional states, we compared tumors induced by identical genetic perturbations in the skin and eye. We find that transcriptional profiles of uveal melanoma tumors arising from melanocytes in the choroid and skin are influenced by more than the intrinsic differences of their cells of origin. For example, eye tumors were enriched for EMT programs compared with skin tumors. In contrast, eye melanocytes did not exhibit these signatures relative to skin melanocytes, suggesting that signals from the ocular ME may promote a more mesenchymal state. Alternatively, these differences may represent a unique cellular origin of uveal melanoma in the eye and skin. Further investigation is needed to determine whether these differences are driven by the ME or by intrinsic properties of melanocytes or precursors cells from distinct anatomic sites.

An interesting distinction between GNAQ and BRAF oncogenes is their differential dependency on Mitfa activity for tumorigenesis. Whereas BRAF^V600E requires Mitfa activity to induce melanoma (6), GNAQ^Q209L-driven tumor latency was significantly reduced in mitfa-deficient zebrafish compared with WT controls. Our results are consistent with prior studies showing that mitfa-deficient zebrafish activate distinct signaling pathways, favoring YAP over MAPK signaling, and give rise to tumors with reduced latency compared with WT tumors (5). We observed yap1 expression in melanocyte progenitor cells of mitfa-deficient and WT zebrafish, suggesting that Yap1 activity may be carried over from the progenitor cell during transformation. Whereas germline mitfa loss accelerates tumor onset, our data suggest that this effect is not solely due to the deletion of mitfa from differentiated melanocytes. Instead, we propose that germline mitfa deficiency expands a population of pax3a-positive melanocyte progenitor cells, which are more susceptible to GNAQ-induced transformation. Alternatively, mitfa loss may impair the differentiation of melanocyte progenitors, preventing the formation of GNAQ^Q209L-positive nevi and promoting direct progression to malignancy. Another explanation for decreased tumor latency in casper zebrafish could be that germline loss of mitfa has effects on other cell types, such as immune cells, which may influence tumor behavior.

Although Mitfa-independent melanocytic cells are unlikely to be the cellular origin of CM in zebrafish models, studies have shown that MITF-independent BRAF-positive melanoma cells are central to disease recurrence (65). GNAQ-driven transformation may be facilitated by Mitfa paralogs such as Tfec and Tfeb. The Mitfa paralog Tfec has been shown to activate pigmentation genes in both the retinal pigment epithelium and melanocyte progenitor cells. Tfec can also rescue ectopic melanocytes in nacre zebrafish embryos (43, 66). Moreover, we have recently shown that the MITF paralog TFE3 promotes cellular plasticity in MITF-low melanoma cells (67), suggesting that MITF paralogs are compelling candidate transcription factors to drive uveal melanoma onset in casper zebrafish.

In our zebrafish study, melanocyte progenitors inherently express NC-associated programs, which are retained during GNAQ^Q209L-driven transformation. Reanalysis of mouse and human single-cell datasets reveals a subpopulation of PAX3-positive melanocytes, suggesting that similar progenitor-like states may be conserved in mammals. Together, these findings highlight how lineage plasticity contribute to uveal melanoma initiation and progression across species. Such observations are supported by an elegant study by Xu and colleagues (68), who established an immunocompetent mouse uveal melanoma model showing that transcriptional plasticity from melanocytic to NC-like states drives tumor progression.

Identifying the transcriptional and signaling mechanisms within mitfa-independent melanocyte progenitor cells that facilitate uveal melanoma onset and progression will have significant clinical implications, particularly for the differential diagnosis of high-risk lesions and the design of targeted therapies.

Limitation of study

The use of tp53 and ptena/b mutations in our uveal melanoma model does not fully recapitulate the genetics of human uveal melanoma. Mutations in TP53 and PTEN are rarely observed in uveal melanoma, in which loss-of-function mutations in BAP1 represent the most frequent tumor-suppressor alteration (40, 69, 70). However, both TP53 and PTEN signaling pathways have been shown to be altered in uveal melanoma pathogenesis. MDM2 overexpression, an upstream negative regulator of p53, and downregulation of PERP, a downstream effector of p53, have been demonstrated in many patients with uveal melanoma and are associated with worse outcome (71–75). Studies have also shown that more than half of uveal melanoma tumors have decreased PTEN immunostaining, and its loss is correlated with decreased survival (40–42). Whether genetic KO of these tumor-suppressors recapitulate the dysregulation of these pathways in human uveal melanoma remains an open question. Another limitation to our approach when examining the differences between genetically identical skin- and eye-derived tumors is the inability to distinguish between microenvironmental effects and differences in cell of origin (melanocytes in the skin vs. in the eye). To address this question, a syngeneic or immunodeficient model could be applied to transplant cells from a primary uveal melanoma tumor into both anatomic sites. In addition, our use of the mitfa promoter to drive oncogene expression presents a challenge in defining the exact cell of origin, as this promoter is active in both melanocyte progenitors and differentiated melanocytes. As such, it remains unclear whether tumors in mitfa WT fish (or those with conditional mitfa deletion) arise from a progenitor or a mature melanocyte. Lineage tracing studies will be essential to determine whether differentiated melanocytes possess the capacity to initiate primary uveal melanoma in mitfa-competent or conditional mitfa-KO zebrafish.

Supplementary Material

Figure S1

Mitfa-deficient zebrafish have decreased tumor latency compared to wild-types by TEAZ-Skin

Figure S2

Targeted Sanger sequencing to confirm CRISPR gene editing following TEAZ-Skin injection using Strategy B plasmids

Figure S3

Immunofluorescence of the choroid using an anti-GFP antibody

Figure S4

Gross images of tissue and tumors used for single cell RNA-Seq in wild-type zebrafish

Figure S5

UMAP of normal eyes and TEAZ-Eye uveal melanoma in wild-type zebrafish (Strategy B)

Figure S6

Ingenuity Pathway Analysis of genes enriched in TEAZ-Eye compared to the tumor microenvironment

Figure S7

Overall and disease-free survival analysis

Figure S8

TCGA group analysis.

Figure S9

CellChat analysis of intercellular communication between zebrafish fibroblast, CAF, and UM cell clusters

Figure S10

scRNA-Seq on control wild-type zebrafish eyes

Figure S11

Oncogenic GNAQ expression in wild-type and casper eye tumors

Figure S12

Ingenuity Pathway Analysis of enriched genes in TEAZ-Eye and TEAZ-Skin compared to the combined tumor microenvironment (eye and skin)

Figure S13

Melanocyte rescue experiments in nacre zebrafish by TEAZ-Eye using plasmids containing mitfa:mitfa; mitfa:GFP (miniCoopR)

Figure S14

Sample overview of adult casper zebrafish and nacre embryos

Figure S15

UMAP representation of UM tumors derived from casper zebrafish, with accompanying pseudotime analysis

Figure S16

Representative images of TEAZ-Skin induced tumors (Strategy B)

Figure S17

Violin plots representing mitfa expression as well as pigmentation genes in BRAFV600E- and GNAQQ209L-positive tumors by single cell RNA-Seq analysis

Figure S18

Chromatoblast gene signatures

Figure S19

UMAP analysis of melanocytes and melanocyte progenitors in casper eyes and in the three sequencing replicates of wild-type eyes

Figure S20

PAX3 positive melanocytes are observed in mouse and human eyes by scRNA-Seq analysis

Figure S21

Immunofluorescent staining for Pax3 in the mouse limbus

Table S1

Findallmarkers of the integrated object between wild-type control eyes and wild-type TEAZ-Eye

Table S2

RNA expression data of FABP3, FABP5, LGALS2, LGALS1, BSCL2, GCH1, and KIT across the 4 TCGA molecular subclasses in uveal melanoma with corresponding sample IDs

Table S3

Differentially expressed genes between cancer associated fibroblasts from wild-type TEAZ-Eye and normal fibroblasts from wild-type normal eyes

Table S4

Findmarkers for choroidal melanocytes, skin melanocytes, TEAZ-Eye, and TEAZ-Skin versus their respective microenvironments; gene names in every intersection of the Venn Diagram comparison of all four groups

Table S5

Differentially expressed genes between eye and skin melanocytes and between TEAZ-Eye and TEAZ-Skin, respectively

Table S6

Findallmarkers of the casper TEAZ-Eye single-cell object

Table S7

Findallmarkers of the casper normal eye single-cell object

Table S8

Findallmarkers of the integrated object between casper TEAZ-Eye and casper normal eye with cluster names labelled

Table S9

Findallmarkers of the subset of GFP-positive clusters from the integrated object between casper TEAZ-Eye and casper normal eye

Table S10

Findallmarkers of the integrated object of neural crest and neural crest-derived cells from Tg(mitfa:GFP) mitfa+/w2 and Tg(mitfa:GFP) mitfaw2/w2 embryos

Table S11

Differentially expressed genes between neural crest and neural crest-derived clusters from Tg(mitfa:GFP) mitfa+/w2 and Tg(mitfa:GFP) mitfaw2/w2 embryos

Table S12

Injection and tumor penetrance of wild-type and casper zebrafish injected with Strategy B plasmids and either U6:gRNA-nontargeting or U6:gRNA-mitfa

Table S13

Differentially expressed genes between BRAFV600E-driven tumors and GNAQQ209L-driven tumors in adult zebrafish

Data Availability

Datasets generated in this article are available from NCBI Gene Expression Omnibus (GEO) accession number GEO: GSE307227. Any additional information required to reanalyze the data reported in this work article is available from the lead contact upon request.

Authors’ Disclosures

J.I. Yevdash reports support from the Predoctoral Training in the Pharmacological Sciences T32 Training Grant from the National Institute of General Medical Sciences (No. T32GM144636) during the conduct of the study. E.M. Binkley reports other support from Castle Biosciences outside the submitted work. C. Kenny reports grants from the National Institutes of Health, Melanoma Research Alliance, Outrun the Sun, and American Cancer Society during the conduct of the study. No disclosures were reported by the other authors.

Authors’ Contributions

J.I. Yevdash: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. D. Robinson: Investigation, visualization, methodology, writing–review and editing. R. Moore: Investigation, visualization, methodology. Z. Li: Investigation, visualization, methodology. K.R. Campbell-Hanson: Investigation, visualization, methodology. D. Gutelius: Investigation, visualization, methodology. S.P.G. Moore: Investigation, visualization, methodology. D. Friend: Investigation, visualization, methodology. I. O’Toole: Investigation, visualization, methodology. B.E. Harman: Methodology. C. Montgomery: Writing–review and editing. R.A. Cornell: Funding acquisition. A. Birbrair: Data curation, methodology, writing–review and editing. J.D. Riordan: Investigation, methodology. A.J. Dupuy: Investigation. E.M. Binkley: Writing–review and editing. R.F. Mullins: Methodology, writing–review and editing. D. Lang: Funding acquisition, visualization, writing–review and editing. R.J. Weigel: Visualization, writing–review and editing. C. Kenny: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.