Novel CRISPR-Cas9 BAP1 knockout pre-clinical tumor model recapitulates human melanoma tumorigenesis and immune evolution Authors: Mona M. Wang, Yuanhan Li, Candice E. H. Ho, Willie S. S. Yu, Sarah E. Coupland, Gayoung Park, Anita S. Y. Chan, Carlos R. Figueiredo Published: 2026 ABSTRACT BAP1-deficient melanocytic tumors exhibit strong immunosuppressive features and poor prognosis. Currently, no immune-competent preclinical models exist to study their tumor-immune interactions or test new immunotherapies. This limitation hinders progress in understanding how BAP1 loss drives tumor aggressiveness and immune evasion. To address this, we generate a syngeneic BAP1 knockout melanocyte tumor line using CRISPR-Cas9. We then evaluate its functional and immunological impact in immune-competent mice, including its ability to recapitulate metabolic and immunosuppressive features of human BAP1-deficient melanomas. The selected knockout clone exhibits hallmarks of aggressive skin and intraocular melanomas, including epithelioid morphology, in vivo tumorigenic potential, rapid growth, and key immunosuppressive features, mirroring those observed in human BAP1-deficient melanomas. Cross-species single-cell transcriptome analysis demonstrates strong molecular overlap between BAP1 knockout mouse tumors and high-risk (class 2) human uveal melanomas, highlighting shared pathways in lipid metabolism, transmembrane receptor signaling, and immune modulation. Gene Set Enrichment Analysis confirms that lipid metabolic reprogramming, previously described in human tumors, is also a key feature of our model, validating its ability to recapitulate human disease biology. This study introduces a syngeneic preclinical model that mimics the immunosuppressive landscape of BAP1-deficient melanocytic tumors, enabling the development and optimization of new combination immunotherapies. Introduction BRCA1-associated protein 1 (BAP1) is a tumor suppressor gene encoding a nuclear deubiquitinase that governs chromatin remodeling, DNA damage repair, cell cycle regulation, and apoptosis1. Loss of BAP1 triggers a critical tumorigenic phenotype that reshapes tumor behavior by promoting aggressive growth, immune suppression, and therapeutic resistance2,3, and is strongly linked to poor prognosis and immune evasion in malignant mesothelioma (MM)4, uveal melanoma (UM)5,6, cutaneous melanoma (CM)7, renal cell carcinoma8, and potentially other malignancies9,10. Melanocytic tumors, primarily UM and CM, share genetic characteristics yet diverge in clinical behavior. UM, the most common intraocular malignancy in adults, is highly aggressive with a strong predilection for liver metastases11,12. CM, while more prevalent, frequently demonstrates metastatic progression in its advanced stages, resulting in significant morbidity and mortality13,14. However, targeted therapies and immune checkpoint inhibitors (ICIs) have shown effectiveness in specific CM subtypes, especially those with a high tumor mutational burden15–17. UM remains largely resistant to immunotherapy because unique immune suppression mechanisms create an “immune desert” phenotype18,19. These differences underscore the need for tailored therapeutic approaches addressing the unique tumor microenvironments and resistance mechanisms in UM and CM. BAP1 loss plays a central role in melanocytic tumorigenesis and immune modulation20. Our earlier findings confirmed that the loss of BAP1 drives tumorigenesis and reshapes the tumor immune microenvironment5. We further demonstrated that the absence of BAP1 expression is tightly linked to metastatic progression and profound resistance to immunotherapy5,21. These insights emphasize the critical role of BAP1 in immune evasion and highlight an urgent need to target BAP1-related pathways to develop effective therapeutic strategies. They also point toward unexplored therapeutic opportunities for melanocytic tumors based on the BAP1 loss phenotype, emphasizing the importance of unraveling immune evasion strategies mediated by BAP1 loss. Understanding the immunological consequences of BAP1 loss in melanocytic tumors, particularly in UM, and advancing immunotherapy approaches tailored to metastatic progression are therefore imperative. Moreover, since BAP1 loss is implicated in several other cancer types1,10, studying BAP1-related mechanisms could unveil broader insights into molecular processes and open new therapeutic avenues across various malignancies. Existing preclinical models for BAP1 loss melanocytic tumors have critical limitations. Widely used xenograft models rely on immunodeficient mice22–26, rendering them unsuitable for immunotherapy studies due to the absence of a functional immune system. These models also complicate understanding BAP1-dependent biological processes because of additional genetic alterations and mutations that can obscure the specific impact of BAP1 loss. Similarly, immune-competent models are based on murine CM cells, such as the B16 cell lines27–30, which inherit multiple mutations and genetic transformations affecting tumor evolution and microenvironment. Their lack of adaptive immunity and HLA incompatibility between human tumors and mouse hosts further limit their use in immunotherapy development29,30. The absence of an immunocompetent preclinical model capturing the specific impact of BAP1 loss represents a significant obstacle to advancing immunotherapy development. To address this challenge, we developed a syngeneic preclinical model using a CRISPR-engineered BAP1−/− mouse tumor model focused exclusively on BAP1 loss, which has been identified as fundamentally crucial in tumors such as UM and mesothelioma. This model, established in a mouse with a fully functional immune system, enables the study of the role of BAP1 loss in tumorigenesis and immune suppression, independent of other confounding genetic alterations. This proof-of-concept model provides a unique platform for in vitro and in vivo studies to dissect immune responses and evaluate novel therapeutic strategies. By advancing our understanding of BAP1 loss, this study sets the foundation for developing innovative therapies for BAP1 loss melanocytic tumors, with implications extending to other cancers driven by BAP1 dysfunction. Results Successful generation and validation of the CRISPR-engineered BAP1−/− mouse melanocytes Developing a preclinical model for melanocytic tumors requires a reliable, scalable cell system to study BAP1 loss. Primary ocular melanocytes would provide the most physiologically relevant context. Still, their limited proliferative capacity, batch-to-batch variability, and rapid senescence render them impractical for large-scale or long-term studies. Immortalization can overcome these constraints by enabling consistency and scalability, but may introduce phenotypic alterations that confound downstream analyses. For our proof-of-concept model, we selected the well-characterized Melan-a line, a murine immortalized melanocyte population devoid of key oncogenic driver mutations, allowing us to isolate the specific effects of BAP1 loss with minimal additional confounding mutations. Although Melan-a cells are derived from cutaneous melanocytes of trunk neural crest origin, whereas uveal melanocytes arise from the cranial neural crest, both lineages originate from the neural crest and retain similar melanocytic morphology and functional characteristics31,32. Despite not fully recapitulating all aspects of primary ocular melanocyte biology, the genetic stability, consistency, and scalable growth of Melan-a cells make them a practical and reproducible platform for studying the functional consequences of BAP1 loss in tumorigenesis and immune suppression. To validate a minimal-mutation background for our BAP1 knockout (KO) studies, we performed whole-genome sequencing (WGS) and compared the mutational profiles of primary mouse ocular melanocytes, Melan-a cells, and B16F10 cells (Supplementary Data 1–3). B16F10 cells, a common murine CM model, exhibited characteristic mutations associated with cutaneous melanoma. As expected, B16F10 cells harbored multiple nonsynonymous mutations commonly reported in melanoma models, including variants in Braf and Pten (Supplementary Table 1). While the specific mutations identified in B16F10 (Braf C246R; Pten A39V, T131P) are not established oncogenic hotspot drivers comparable to Braf V600E or truncating PTEN loss-of-function mutations, their presence reflects the overall genetically altered and transformed background of this widely used melanoma line. In contrast, Melan-a cells display a relatively minimal mutational landscape, supporting their use as a low-background model to isolate the specific functional consequences of BAP1 loss. Melan-a cells did not harbor these skin melanoma-associated mutations and lacked the defining uveal melanoma mutations in GNAQ (Q209L and Q209P), GNA11 (Q209L and Q209P), and BAP1. Although they harbor mutations in SF3B1, these differ from the specific mutations observed in human UM, providing a low mutational profile and histological characteristics closer to nontumorous melanocytes. This supports their use as a surrogate for investigating UM immunobiology and treatment responses under BAP1 loss conditions. We emphasize, however, that the absence of known driver mutations does not by itself imply full functional or transcriptomic equivalence to primary ocular melanocytes. Rather, it provides a genetically “cleaner” background in which to isolate the effects of BAP1 loss. Downstream in vitro phenotyping and cross-species transcriptomic analyses in this study further investigate the extent to which our CRISPR-engineered Melan-a model replicates key features of ocular melanocyte biology. To generate BAP1 KO cells, gene expression was silenced using CRISPR-Cas9 gene editing. Small guide RNAs (sgRNAs) were designed to target mouse BAP1 (NM_027088) in exon 1 and exon 14 regions (Supplementary Fig. 1). Introducing indels in exon 1, close to the start of the coding sequence, created frameshift mutations that disrupted the reading frame, leading to premature stop codons and subsequent nonsense-mediated decay of the mRNA, ensuring effective gene inactivation. Targeting exon 1 also minimizes the risk of generating functional proteins from alternative transcripts. Additionally, we targeted exon 14 based on UniProtKB/Swiss-Prot (Q99PU7.1) annotations, indicating that this region interacts with BRCA1. Disrupting exon 14 thus interferes with this critical interaction, offering insights into BAP1’s functional role relative to BRCA1. This dual-targeting strategy ensured comprehensive gene silencing by disrupting translation and essential protein interactions. The CRISPR-Cas9 workflow involved plasmid construction, transfection, sorting of GFP-positive cells, single-cell culture, and clone screening (Fig. 1a). Transfection efficiency was first assessed by image cytometry and then confirmed by flow cytometry (Supplementary Fig. 2a). No GFP-positive cell was detected in the control group, while cells transfected with the sgRNA plasmid targeting BAP1 showed a GFP-positive population of 3.52% at 48 h, indicating successful plasmid delivery. These GFP-positive cells were then sorted for single-cell clone generation, and microscopy images further validated the transfection outcomes (Supplementary Fig. 2b). Genetic and protein-level analyses confirmed the introduction of targeted mutations. Sanger sequencing revealed large frameshift deletions in KO#1 and #3 and single-nucleotide frameshift mutations in KO#2 and #4 (Supplementary Fig. 2c, d). To assess allelic editing efficiency, Variant Allele Frequency (VAF) analysis was performed using whole-exome sequencing (Supplementary Fig. 2e), revealing that KO#1 harbored a mixed allele population consistent with partial disruption (36.6% deletion), whereas KO#2, KO#3, and KO#4 displayed near-complete loss (97.6%, 91.7%, and 99.1% deletion, respectively), indicative of bi-allelic knockout. Quantitative PCR (qPCR) showed significant downregulation of BAP1 expression across both exon 1 and exon 14 in all KO clones (#1–#4) compared to wild-type controls (WT) cells (Fig. 1b). Western blot analysis further confirmed the absence of BAP1 protein, with KO#1 exhibiting partial loss and KO#2-#4 demonstrating complete knockout (Fig. 1c). Notably, all validation results were consistent across early (P4-6) and late (P15 and above) cell passages.Fig. 1Successful construction and characterization of a CRISPR-engineered BAP1−/− cell line model.a Schematic workflow illustrating the construction of the CRISPR/Cas9 system targeting BAP1. Plasmid map of pGuide-it-ZsGreen1 ligated with BAP1-targeted sgRNA generated using SnapGene; graphical elements created with BioRender.com (license ID: SW29KU8LBJ). b qPCR analysis showing significant downregulation of BAP1 expression in CRISPR-engineered BAP1−/− knockout clones (n = 6) compared to WT cells (n = 6), assessed at exon 1 (left) and exon 14 (right). Data are presented as mean ± SEM. Statistical significance was determined using an unpaired two-tailed t-test. c Western blot showing loss of BAP1 protein expression in mouse immortalized melanocyte BAP1 knockout clones. Successful construction and characterization of a CRISPR-engineered BAP1−/− cell line model. a Schematic workflow illustrating the construction of the CRISPR/Cas9 system targeting BAP1. Plasmid map of pGuide-it-ZsGreen1 ligated with BAP1-targeted sgRNA generated using SnapGene; graphical elements created with BioRender.com (license ID: SW29KU8LBJ). b qPCR analysis showing significant downregulation of BAP1 expression in CRISPR-engineered BAP1−/− knockout clones (n = 6) compared to WT cells (n = 6), assessed at exon 1 (left) and exon 14 (right). Data are presented as mean ± SEM. Statistical significance was determined using an unpaired two-tailed t-test. c Western blot showing loss of BAP1 protein expression in mouse immortalized melanocyte BAP1 knockout clones. CRISPR-engineered BAP1−/− cell clones exhibit major morphological and tumorigenic features in vitro and in vivo Compared to the immortalized melanocytes without CRISPR treatment (WT controls), BAP1 KO cell clones exhibit a more aggressive epithelioid histology (Supplementary Fig. 3). H&E staining revealed distinct alterations in cellular architecture across the KO clones, with larger, irregular cell sizes and reduced cell-cell adhesion indicative of a shift toward a more aggressive phenotype (Supplementary Fig. 3, top panel). Despite these changes, all BAP1 KO clones retained positive gp100 IHC staining (Supplementary Fig. 3, bottom panel), confirming their melanocytic lineage. Loss of BAP1 significantly influenced cell growth and proliferation, with varying effects across clones. Adherent colony formation assays, performed two weeks after seeding 2000 cells per well, showed enhanced proliferation in BAP1 KO#1, #2, and #3 compared to WT, while KO#4 exhibited markedly reduced proliferation (Fig. 2a, left panel). Quantitative analysis using the RTCA Xcelligence doubling rate assay confirmed these distinct proliferation dynamics (Fig. 2a, right panel). In soft agar colony formation assays, WT cells formed a moderate number of large colonies. KO#3 cells formed a significantly higher number of colonies compared to WT, although the average colony size was smaller (Fig. 2b). This phenotype reflects an altered pattern of anchorage-independent growth in KO#3, characterized by enhanced clonogenic initiation and survival under non-adherent conditions (indicative of increased anoikic resistance), yet limited per-colony expansion. Such behavior aligns with prior observations in uveal melanoma showing that BAP1 loss promotes cohesive, cluster-like growth through upregulation of cell–cell adhesion molecules33, suggesting BAP1-dependent remodeling of adhesion and anchorage behavior rather than a simple increase in proliferative vigor. In contrast, KO#1, KO#2, and KO#4 exhibited fewer colonies overall, with variable sizes, underscoring clone-specific effects of BAP1 loss on anchorage-independent growth (Fig. 2b). Moreover, we observed that KO clone #1, and more intensely, KO clone #2 showed improved proliferation in the adherent colony formation assay but reduced proliferation in soft agar. The metastatic potential of BAP1 KO clones was assessed by their migration and invasion abilities. A wound-healing assay monitoring scratch closure over time revealed a significantly increased migration rate in all BAP1 KO clones compared to WT (Supplementary Fig. 4a), as quantified by the percentage of recovery area (Fig. 2c). In a transwell invasion assay, KO#2, #3, and #4, with complete knockout, showed a significantly higher number of invaded cells compared to WT, while KO #1, which exhibits a partial knockout as indicated by western blot, displayed no significant difference in invasion (Supplementary Fig. 4b and Fig. 2d). This suggests a potential dose-dependent relationship between BAP1 expression and regulation of metastatic properties, where the partial retention of BAP1 in clone 1 maintains normal invasion regulation.Fig. 2BAP1 knockout promotes increased proliferation, clonogenicity, migration, invasion, and in vivo tumor progression.a Adherent colony formation assay (n = 3) in monolayer cultures stained with crystal violet after two weeks (left panel). Clones KO#1, KO#2, and KO#3 show elevated proliferation compared with WT (whereas KO#4 displays lower proliferation. Doubling rates were quantified by RTCA Xcelligence (right panel). b Anchorage-independent colony growth assay conducted in soft agar (n = 16). Colonies were stained with 0.05% crystal violet and observed under an inverted microscope at ×4 magnification (left panel). c Wound healing assay over 78 h showing accelerated closure in BAP1 KO clones compared with WT (n = 8). Data represent the percentage of recovery area at each time point. d Invasion assay (n = 6) showing increased invasive capacity in KO#2, KO#3, and KO#4 relative to WT and KO#1. e Tumor volume was significantly elevated in ocular injections with BAP1 KO#1-3 clones (n = 12) compared to WT (n = 12). f Post 5-week optical coherence tomography (OCT) of eyes injected with WT or KO#2 clones. The KO#2 clone disrupts typical iris architecture, unlike the WT clone. g Elevated intraocular pressure (IOP) was observed in eyes injected with BAP1 KO#1-3 clones (n = 12) compared to those injected with WT (n = 12) and non-injected controls (n = 12). h Tumor volume was significantly elevated in subcutaneous injections with BAP1 KO#1-3 clones (n = 12) compared to WT (n = 7). i Representative images of subcutaneous tumor development at 1 and 8 weeks postinjection. BAP1 KO#2 injected mice display more aggressive tumor growth than WT. Data quantification is presented as mean values ± SEM. All p-values were determined by an unpaired 2-tailed t-test. BAP1 knockout promotes increased proliferation, clonogenicity, migration, invasion, and in vivo tumor progression. a Adherent colony formation assay (n = 3) in monolayer cultures stained with crystal violet after two weeks (left panel). Clones KO#1, KO#2, and KO#3 show elevated proliferation compared with WT (whereas KO#4 displays lower proliferation. Doubling rates were quantified by RTCA Xcelligence (right panel). b Anchorage-independent colony growth assay conducted in soft agar (n = 16). Colonies were stained with 0.05% crystal violet and observed under an inverted microscope at ×4 magnification (left panel). c Wound healing assay over 78 h showing accelerated closure in BAP1 KO clones compared with WT (n = 8). Data represent the percentage of recovery area at each time point. d Invasion assay (n = 6) showing increased invasive capacity in KO#2, KO#3, and KO#4 relative to WT and KO#1. e Tumor volume was significantly elevated in ocular injections with BAP1 KO#1-3 clones (n = 12) compared to WT (n = 12). f Post 5-week optical coherence tomography (OCT) of eyes injected with WT or KO#2 clones. The KO#2 clone disrupts typical iris architecture, unlike the WT clone. g Elevated intraocular pressure (IOP) was observed in eyes injected with BAP1 KO#1-3 clones (n = 12) compared to those injected with WT (n = 12) and non-injected controls (n = 12). h Tumor volume was significantly elevated in subcutaneous injections with BAP1 KO#1-3 clones (n = 12) compared to WT (n = 7). i Representative images of subcutaneous tumor development at 1 and 8 weeks postinjection. BAP1 KO#2 injected mice display more aggressive tumor growth than WT. Data quantification is presented as mean values ± SEM. All p-values were determined by an unpaired 2-tailed t-test. In vivo tumorigenic potential was further investigated using intraocular injections in immunocompetent C57BL/6 mice, followed by longitudinal tumor monitoring. WT clones exhibited minimal tumor growth over five weeks, evidenced by slit-lamp imaging and optical coherence tomography (OCT) scanning (Supplementary Fig. 5a, b). In contrast, BAP1 KO#1–3 showed rapid and substantial tumor growth, with KO#2 displaying the most aggressive progression, extensive anterior chamber invasion, and severe iris structural disruption (Fig. 2e, f). Intraocular pressure (IOP) increased significantly in the KO#1–3 groups, correlating with tumor volume and aggressive ocular involvement (Fig. 2g). KO#4, however, exhibited negligible tumor development, similar to WT controls, highlighting the variability in tumorigenic potential among clones. Subcutaneous injections over eight weeks further evaluated tumor growth in a non-ocular environment. BAP1 KO#1–3 produced significantly larger tumors compared to WT, with KO#2 again demonstrating the most aggressive growth trajectory (Supplementary Fig. 5c). While clones KO#1 and KO#3 also formed sizable tumors, their growth rates and extents were lower than KO#2 (Fig. 2h, i). WT clones and KO#4 failed to produce notable subcutaneous tumors, suggesting that, in addition to BAP1 loss, clonal background or secondary molecular factors may influence tumorigenic potential. To explore this, we performed whole-exome sequencing (WES) comparing KO clones with WT normal controls, which revealed additional somatic variants introduced during CRISPR-Cas9 gene editing34. Shortlisted exonic mutations, including nonsynonymous SNVs, frameshift, and premature stop codon events, were subjected to Gene Ontology (GO) analysis (Supplementary Data 4–9). KO#4 uniquely carried 666 mutated genes, with enriched terms related to intercellular anchoring junction (FDR = 2.09E–04), negative regulation of cell communication and signaling (FDR = 4.09E–03), indicating potential disruption of cell–cell interactions and cytoskeletal remodeling. These findings highlight possible molecular contributors to the divergent tumorigenic behavior observed among KO clones. Quantification of coding variants revealed that both WT and BAP1-KO Melan-a clones carry a substantial immunogenic mutation burden, with KO clones exhibiting levels comparable to WT and markedly higher than normal ocular melanocytes (Supplementary Table 2). These data indicate that the Melan-a background possesses a coding mutational load consistent with the potential to generate neoantigens. Collectively, our data reinforce that BAP1 loss promotes tumorigenic conversion in Melan-a cells, with the KO#2 clone demonstrating the most robust and reproducible tumorigenic phenotype, making it the most suitable model for further in vivo investigations. Transcriptomic analysis of tumor samples through bulk RNA sequencing further revealed distinct gene expression profiles and correlation patterns between WT and different BAP1 KO clones. Principal Component Analysis (PCA) and unsupervised hierarchical clustering (HC) revealed distinct clustering patterns between WT and KO clones. Notably, the biological replicates for each clone, derived from different cell passages, consistently clustered within their respective groups, thereby underscoring both the reproducibility of the experiments and the unique transcriptional identities of the clones (Fig. 3a). The PCA plot distinctly separated WT and KO clones into discrete populations, with the partial BAP1 KO clone (KO#1) clustering closer to WT cells, while fully silenced KO clones (KO#2, KO#3, and KO#4) formed more divergent groups. PC1 primarily separated the fully BAP1-deficient clones from WT and KO#1, indicating that BAP1 status was the dominant source of transcriptional variance, whereas PC2 reflected clone-specific adaptive divergence among the fully KO lines. To further contextualize this pattern, WES was performed across all KO clones, revealing a limited number of additional exonic mutations beyond BAP1 (Supplementary Data 4–7). These secondary alterations are consistent with the modest genomic variability expected in CRISPR-Cas9 edited populations35 and likely contribute to the dispersion along PC2 rather than driving the primary PC1 separation. Volcano plots of differentially expressed genes (DEGs) further highlighted the substantial differences in gene expression between WT and KO clones (Fig. 3b). KO#1 exhibited minimal changes, with relatively few DEGs, whereas KO#2, KO#3, and KO#4 showed widespread transcriptional alterations characterized by a high number of upregulated and downregulated genes. Overlap analysis using Venn diagrams demonstrated that full BAP1 KO clones exhibited a larger proportion of unique upregulated genes, while downregulated genes were more consistently shared across all KO clones (Fig. 3c). This suggests that BAP1 loss results in both conserved and clone-specific transcriptional changes, reflecting the diverse phenotypic consequences of BAP1 loss.Fig. 3Transcriptomic and pathway analysis of BAP1 WT and KO clones.a Principal Component Analysis (PCA) and unsupervised hierarchical clustering (HC) of BAP1 WT and KO clones based on transcriptomic data. b Volcano plots displaying differentially expressed genes (DEGs) in BAP1 KO clones compared to WT. Blue dots represent significantly downregulated genes, and brown dots represent significantly upregulated genes (log2 fold change > 2, adjusted p < 0.05). KO clones #2–4 show more DEGs than KO#1, consistent with their complete BAP1 silencing. c Venn diagrams illustrating the overlap of DEGs among BAP1 KO clones. The left panel shows shared and unique upregulated genes, while the right panel displays downregulated genes, emphasizing both conserved and clone-specific transcriptional responses to BAP1 loss. d Unsupervised hierarchical clustering of Pearson’s correlation scores for upregulated (left) and downregulated (right) genes in BAP1 WT versus KO clones. Heatmaps reveal distinct clusters in WT clones (C1, C2, and C3) and a more diverse clustering pattern in KO clones (C1–C6), reflecting extensive reorganization of gene networks following BAP1 knockout. e Gene Set Enrichment Analysis (GSEA) for upregulated gene sets in fully silenced BAP1 KO clones (KO#2, KO#3, and KO#4). Enrichment plots highlight pathways associated with cellular communication, immune response, proliferation, and immunosuppression. KO#2 is enriched in pathways related to immune signaling and cell proliferation, KO#3 predominantly in proliferation pathways, and KO#4 in pathways regulating cell growth and differentiation. Transcriptomic and pathway analysis of BAP1 WT and KO clones. a Principal Component Analysis (PCA) and unsupervised hierarchical clustering (HC) of BAP1 WT and KO clones based on transcriptomic data. b Volcano plots displaying differentially expressed genes (DEGs) in BAP1 KO clones compared to WT. Blue dots represent significantly downregulated genes, and brown dots represent significantly upregulated genes (log2 fold change > 2, adjusted p < 0.05). KO clones #2–4 show more DEGs than KO#1, consistent with their complete BAP1 silencing. c Venn diagrams illustrating the overlap of DEGs among BAP1 KO clones. The left panel shows shared and unique upregulated genes, while the right panel displays downregulated genes, emphasizing both conserved and clone-specific transcriptional responses to BAP1 loss. d Unsupervised hierarchical clustering of Pearson’s correlation scores for upregulated (left) and downregulated (right) genes in BAP1 WT versus KO clones. Heatmaps reveal distinct clusters in WT clones (C1, C2, and C3) and a more diverse clustering pattern in KO clones (C1–C6), reflecting extensive reorganization of gene networks following BAP1 knockout. e Gene Set Enrichment Analysis (GSEA) for upregulated gene sets in fully silenced BAP1 KO clones (KO#2, KO#3, and KO#4). Enrichment plots highlight pathways associated with cellular communication, immune response, proliferation, and immunosuppression. KO#2 is enriched in pathways related to immune signaling and cell proliferation, KO#3 predominantly in proliferation pathways, and KO#4 in pathways regulating cell growth and differentiation. To explore the broader impact of BAP1 loss on gene regulatory networks, we evaluated the correlation pattern of the expression levels of these differently expressed genes across all KO clones with high (WT) and low (full KO) BAP1 expression and performed an unsupervised hierarchical clustering (HC) of the correlation scores, which indicates the clustering pattern of cells with high and low BAP1 levels (clusters C1, C2, C3, etc.). This analysis revealed distinct clustering patterns that depended on BAP1 expression levels. As observed in the correlation matrix of Fig. 3d, different BAP1-associated genes show a range of correlation scores from −1 to +1. These scores are useful to provide sufficient complexity to the matrix for further unsupervised clustering analysis, the purpose of which is not to evaluate the DEGs with higher or lower correlation but ultimately to observe how the cluster identity of all these DEGs fluctuates in the context of tumors expressing high and low BAP1. WT clones exhibited limited clustering diversity, represented by clusters C1, C2, and C3, whereas KO clones showed an expanded cluster repertoire (C1-C6), indicating increased heterogeneity and complexity in gene expression networks following BAP1 knockout (Fig. 3d). Generating more clusters in the KO group suggests that BAP1 loss drives a broader and more dynamic reorganization of gene regulatory networks, potentially underpinning the aggressive phenotypes observed in BAP1 loss tumors. Gene Set Enrichment Analysis (GSEA) provided further insights into the functional pathways enriched in full BAP1 KO clones (Fig. 3e). KO#2 was enriched in pathways associated with cell proliferation, immune signaling, and immunosuppression, highlighting its potential role in promoting tumor-immune escape (Fig. 3e, left panel). KO#3 primarily showed enrichment in cell proliferation pathways without immune-related functions (Fig. 3e, middle panel), aligning with its aggressive in vitro growth characteristics (Fig. 2a, b). Conversely, KO#4 was enriched in pathways regulating cell growth and differentiation (Fig. 3e, right panel), resembling features of primary cells with limited tumorigenic potential (Fig. 2f–i). Although KO#1 exhibited in vivo tumorigenicity, its transcriptomic profile revealed minimal differential gene expression compared to WT, and no pathways reached statistical significance by GSEA. This limited transcriptional divergence is consistent with KO#1 clustering near WT in the PCA, explaining its exclusion from Fig. 3e. Immunoprofiling and integrative single-cell transcriptomic analysis of BAP1 knockout tumors in ocular and subcutaneous models for translational validation Single-cell RNA sequencing (scRNA-seq) was performed to characterize the tumor microenvironment (TME) following subcutaneous and intraocular injections of WT and BAP1 KO#2 Melan-a cells. Tissue samples were harvested at standardized time points (8 weeks post-injection for subcutaneous sites and 5 weeks post-injection for intraocular sites) to ensure comparability. Although WT cells did not form progressively growing tumors, small, stable nodular lesions consistently developed at the injection sites. These Melan-a–derived nodular tissues were collected and analyzed in parallel with the larger, aggressive tumors generated by BAP1 KO#2 clones, thereby enabling direct comparison of cellular states and microenvironmental responses across distinct tumorigenic contexts. scRNA-seq revealed pronounced differences in cellular composition and transcriptional programs between BAP1-deficient and WT-derived tissues. Melanocytic lineage cells were identified in both conditions by canonical marker expression, providing a framework for comparing how BAP1 loss reprograms melanocyte-derived cell populations and their surrounding TME. In subcutaneous BAP1−/− tumor models, UMAP analysis identified distinct cell populations in WT tumors, including non-tumorous melanocytes, fibroblasts, macrophages, neutrophils, endothelial cells, and B cells (Fig. 4a and Supplementary Fig. 6). A heatmap of marker gene expression confirmed these identities, illustrating a relatively stable and organized TME in the absence of BAP1 mutation. In the BAP1 KO#2 subcutaneous tumors, highly proliferative melanocytes dominated the cellular composition, consistent with the tumor-promoting effect of BAP1 loss. Compared with WT tumors, KO tumors contained a similar or slightly increased proportion of macrophages, which showed clear M2-like transcriptional features, along with a modest increase in T cells (Fig. 4b). The intraocular TME displayed a more complex cellular landscape, reflecting the eye’s unique immune environment36. WT tumors contained melanocytes, fibroblasts, and ocular cell types such as photoreceptors, progenitor cells, and basal cells, accompanied by CD68− macrophages consistent with a non-activated or pre-M2 macrophage state (Fig. 4c). In contrast, BAP1 KO#2 intraocular tumors showed increased CD68⁺ macrophages and monocytes, together with a modest rise in T-cell infiltration. These macrophages exhibited strong M2/tumor-associated macrophages (TAM)-associated signatures (Fig. 4d), indicating a shift toward an immunoregulatory phenotype. Consistently, the single-cell transcriptomic analysis revealed a consistent enrichment of M2 macrophage populations in BAP1 KO tumors compared with WT. In parallel, expression of antigen-presentation–associated genes was markedly reduced across macrophage clusters in both ocular and cutaneous tumor models (Supplementary Fig. 7). Together, these proportional and transcriptional changes support that BAP1 loss promotes not only enhanced tumor proliferation but also a shift toward an immunosuppressive, M2-enriched macrophage landscape within the tumor microenvironment.Fig. 4Single-cell RNA sequencing analysis of TME in WT and BAP1-loss (KO#2) models following subcutaneous and intraocular injections.a UMAP plot and heatmap depicting the TME in WT clones after subcutaneous injection, with the heatmap illustrating marker gene expression for cell-type confirmation. b UMAP plot and heatmap of the TME in BAP1-loss KO#2 clones after subcutaneous injection. The heatmap highlights the differential expression of marker genes across cell clusters. c UMAP plot showing the TME in WT clones after intraocular injection with corresponding marker gene expression displayed in the heatmap. d UMAP plot and heatmap representing the TME in BAP1-loss KO#2 clones after intraocular injection. Single-cell RNA sequencing analysis of TME in WT and BAP1-loss (KO#2) models following subcutaneous and intraocular injections. a UMAP plot and heatmap depicting the TME in WT clones after subcutaneous injection, with the heatmap illustrating marker gene expression for cell-type confirmation. b UMAP plot and heatmap of the TME in BAP1-loss KO#2 clones after subcutaneous injection. The heatmap highlights the differential expression of marker genes across cell clusters. c UMAP plot showing the TME in WT clones after intraocular injection with corresponding marker gene expression displayed in the heatmap. d UMAP plot and heatmap representing the TME in BAP1-loss KO#2 clones after intraocular injection. To elucidate transcriptional changes induced by BAP1 loss in the TME, we conducted GSEA on upregulated and downregulated pathways across all cell populations in subcutaneous and intraocular BAP1−/− tumor models (Fig. 5 and Supplementary Data 10). In the intraocular injection model, BAP1 KO tumor cells (KO#2 clone) populations exhibited significant enrichment in pathways related to lipid metabolism, cytokine binding, G protein-coupled receptor signaling, and transmembrane receptor activity (Fig. 5a, left panel), suggesting a metabolically active, signaling-driven phenotype that contributes to aggressive behavior. In contrast, BAP1−/− tumor cells in the subcutaneous model were enriched in pathways associated with cell cycle progression, mitotic spindle organization, and chromosome segregation (Fig. 5a, right panel), underscoring a proliferative phenotype compared with the signaling- and metabolic-focused profile in intraocular tumors.Fig. 5GSEA of upregulated and downregulated pathways across tumor and non-tumor stromal cell populations in subcutaneous and ocular tumor models.a GSEA for upregulated pathways in the BAP1−/− Melan-a clone#2 population of BAP1 KO#2 intraocular model (left panel) and BAP1 KO#2 subcutaneous model (right panel). b Upregulated (left panel) and downregulated (right panel) GSEA in subcutaneous injection model macrophage populations. c Upregulated (left panel) and downregulated (right panel) GSEA in ocular injection model macrophage populations. d Upregulated (left panel) and downregulated (right panel) GSEA in ocular injection model neutrophil populations. e Downregulated GSEA in fibroblast populations in the subcutaneous injection model. f Upregulated GSEA in fibroblast populations in the ocular injection model. g Downregulated GSEA in endothelial cell populations in the subcutaneous injection model. h Upregulated GSEA in endothelial cell populations in the ocular injection model. GSEA of upregulated and downregulated pathways across tumor and non-tumor stromal cell populations in subcutaneous and ocular tumor models. a GSEA for upregulated pathways in the BAP1−/− Melan-a clone#2 population of BAP1 KO#2 intraocular model (left panel) and BAP1 KO#2 subcutaneous model (right panel). b Upregulated (left panel) and downregulated (right panel) GSEA in subcutaneous injection model macrophage populations. c Upregulated (left panel) and downregulated (right panel) GSEA in ocular injection model macrophage populations. d Upregulated (left panel) and downregulated (right panel) GSEA in ocular injection model neutrophil populations. e Downregulated GSEA in fibroblast populations in the subcutaneous injection model. f Upregulated GSEA in fibroblast populations in the ocular injection model. g Downregulated GSEA in endothelial cell populations in the subcutaneous injection model. h Upregulated GSEA in endothelial cell populations in the ocular injection model. This analysis also identified immune-suppressive pathways among upregulated gene sets and immune-supportive pathways among downregulated gene sets, highlighting the immunosuppressive effects of BAP1 loss. In the subcutaneous model, macrophages displayed significant enrichment of upregulated pathways related to the cell periphery, molecular transducer activity, signaling receptor activity, and transmembrane signaling receptor activity (Fig. 5b, left panel), suggesting that these macrophages are adapted to respond to external signals and participate in tumor-associated signaling. Conversely, downregulated pathways in these macrophages included inflammatory responses, cytokine activity, defense responses, leukocyte and neutrophil chemotaxis, and regulation of cell killing (Fig. 5b, right panel), reflecting diminished immune activation and cytotoxicity within the subcutaneous TME of BAP1−/− tumors. In the intraocular model, macrophages exhibited upregulated pathways related to the negative regulation of cytokine production, interleukin-6 production, immune effector processes, and lymphocyte-mediated immunity (Fig. 5c, left panel), suggesting a potentially suppressive phenotype. Future functional studies are required to confirm whether such macrophages with an immunosuppressive phenotype actively inhibit immune responses. Their downregulated pathways included ribosome biogenesis, nucleolar activity, protein ubiquitination, mitochondrial activity, and mRNA processing (Fig. 5c, right panel), suggesting reduced protein synthesis and altered metabolic activity. Neutrophils in the intraocular model displayed significant enrichment in upregulated pathways associated with immune suppression and regulation, including negative regulation of cytokine production, immune effector processes, lymphocyte proliferation, and inflammatory responses (Fig. 5d, left panel). Additionally, pathways for neutrophil chemotaxis, migration, regulation of interleukin-6 production, T-cell activation, and tumor necrosis factor production were enriched, indicating that neutrophils contribute to an immunosuppressive phenotype by dampening pro-inflammatory cytokine signaling and T-cell activation while retaining migratory functions. Conversely, neutrophils showed downregulated pathways primarily related to protein synthesis and metabolic activity (Fig. 5d, right panel). Notably, no significant pathways were enriched for neutrophils in the BAP1−/− subcutaneous model, emphasizing that their transcriptional reprogramming in response to BAP1 loss is specific to the intraocular context. Fibroblasts in the BAP1−/− subcutaneous model exhibited significant downregulation of pathways related to locomotion, extracellular space, and responses to biotic and external stimuli, as well as interspecies interactions (Fig. 5e), reflecting a reduced capacity for structural remodeling. In contrast, fibroblasts in the intraocular model demonstrated significant upregulation of pathways related to extracellular signaling, immune modulation, including cytokine activity and binding, and the external side of the plasma membrane (Fig. 5f). Additional upregulated pathways included extracellular exosome activity, plasma membrane regions, and cell-cell adhesion via adhesion molecules, indicating active roles in extracellular communication and structural remodeling, along with enriched immune-related pathways that suppress cell and lymphocyte activation. Endothelial cells in the BAP1−/− subcutaneous model demonstrated significant downregulation of pathways associated with anatomical structure morphogenesis, the extracellular region, and space (Fig. 5g), suggesting a loss of structural and functional engagement in the vasculature under BAP1 loss37. Conversely, in the intraocular model, endothelial cells exhibited significant upregulation of immune-related and structural pathways (Fig. 5h), including adaptive immune responses, lymphocyte-mediated immunity, leukocyte recruitment, and regulation of T-cell-mediated immunity. These cells also showed enrichment in structural pathways such as cell adhesion, homotypic cell-cell adhesion, and extracellular receptor complex organization, reflecting a dual role in promoting immune-cell interactions and maintaining vascular integrity. Altogether, BAP1 loss promotes metabolic reprogramming and shapes the immune landscape by inducing transcriptional changes associated with T cell dysfunction and immune-regulatory macrophage phenotypes, despite increased T cell infiltration. These findings suggest a complex immune environment characterized by potential immune suppression, which warrants further functional validation. Importantly, these results provide initial evidence that such mechanisms are recapitulated in a syngeneic system using our novel tumorigenic model. This model provides a relevant platform for studying metabolic and immune dynamics in the TME and investigating therapeutic strategies to restore anti-tumor immunity. To further validate the relevance of the BAP1−/− ocular injection model in recapitulating the clinical and molecular features of UM observed in patients, we compared the transcriptomic profiles of our BAP1−/− ocular injection model with those of high-risk (class 2) human UM tumors. To achieve this, we performed a retrospective analysis of scRNA-seq data from 11 UM patients (GSE139829)38. Following stringent quality control and batch effect correction39, we retained single cells for dimensionality reduction and unsupervised clustering, identifying seven major cell clusters (Fig. 6a). Among these, tumor cells were distinctly clustered for downstream comparative analysis and segregated tumors from class 1 (low-risk) and class 2 (high-risk) UM patients40. A volcano plot of DEGs highlighted the transcriptional upregulation of 9425 genes and downregulation of 1250 genes in class 2 tumors compared to class 1 tumors (Fig. 6b). Similarly, DEGs from the BAP1−/− mouse model revealed 8266 upregulated genes and 3,663 downregulated genes in tumor cells compared to WT tumors (Fig. 6c). A significant overlap in DEGs between the BAP1−/− mouse model and class 2 human UM tumors was observed. Among the upregulated genes, 63% (5239 genes) were shared, while 237 downregulated genes overlapped between the two datasets (Fig. 6d). This substantial overlap underscores the translational relevance of the BAP1−/− tumor model for high-risk UM.Fig. 6Integrative single-cell transcriptomic analysis reveals a high overlap of differentially expressed genes (DEGs) in BAP1 KO mouse tumor models and high-risk (class 2) UM.a Single-cell RNA sequencing (scRNA-seq) analysis identifies major cell types within TME of UM patients, including T cells, NK cells, fibroblasts, endothelial cells, and tumor cells, as visualized using UMAP clustering. b Volcano plot of DEGs in human UM, comparing Class 2 tumors (high metastatic risk) to Class 1 tumors (low metastatic risk). Significantly upregulated genes in Class 2 tumors are shown in red, while significantly downregulated genes are shown in blue, reflecting transcriptional changes associated with high metastatic risk. c Volcano plot of DEGs in BAP1−/− tumor cells from the BAP1-loss (KO#2) mouse ocular injection model compared to BAP1 WT tumor population. Red dots indicate significantly upregulated genes in BAP1 KO tumors, and blue dots indicate significantly downregulated genes, highlighting the transcriptional consequences of BAP1 loss in the mouse model. d Venn diagrams illustrate the overlap of DEGs between the BAP1 KO mouse model and human Class 2 UM tumors. The top diagram shows a high overlap of 5239 shared upregulated genes, while the bottom diagram highlights 237 shared downregulated genes, underscoring the translational relevance of the BAP1 KO model in reflecting high-risk human UM. e GSEA of overlapping DEGs from BAP1 KO mouse tumors and human high-risk Class 2 UM tumors. The top panel displays pathways enriched among downregulated genes. The bottom panel highlights pathways enriched among upregulated genes. Integrative single-cell transcriptomic analysis reveals a high overlap of differentially expressed genes (DEGs) in BAP1 KO mouse tumor models and high-risk (class 2) UM. a Single-cell RNA sequencing (scRNA-seq) analysis identifies major cell types within TME of UM patients, including T cells, NK cells, fibroblasts, endothelial cells, and tumor cells, as visualized using UMAP clustering. b Volcano plot of DEGs in human UM, comparing Class 2 tumors (high metastatic risk) to Class 1 tumors (low metastatic risk). Significantly upregulated genes in Class 2 tumors are shown in red, while significantly downregulated genes are shown in blue, reflecting transcriptional changes associated with high metastatic risk. c Volcano plot of DEGs in BAP1−/− tumor cells from the BAP1-loss (KO#2) mouse ocular injection model compared to BAP1 WT tumor population. Red dots indicate significantly upregulated genes in BAP1 KO tumors, and blue dots indicate significantly downregulated genes, highlighting the transcriptional consequences of BAP1 loss in the mouse model. d Venn diagrams illustrate the overlap of DEGs between the BAP1 KO mouse model and human Class 2 UM tumors. The top diagram shows a high overlap of 5239 shared upregulated genes, while the bottom diagram highlights 237 shared downregulated genes, underscoring the translational relevance of the BAP1 KO model in reflecting high-risk human UM. e GSEA of overlapping DEGs from BAP1 KO mouse tumors and human high-risk Class 2 UM tumors. The top panel displays pathways enriched among downregulated genes. The bottom panel highlights pathways enriched among upregulated genes. To further explore the functional implications of these overlapping DEGs, we performed GSEA on the shared upregulated and downregulated genes (Fig. 6e and Supplementary Data 11). The analysis of downregulated genes revealed significant suppression of pathways associated with ribosome biogenesis, RNA metabolic processes, and protein biosynthesis, indicating impaired translational and metabolic functions in the BAP1 loss tumors. Key pathways such as nucleolar activity, macromolecular biosynthesis, and post-translational protein modification were also significantly downregulated, reflecting broader disruptions in cellular homeostasis (Fig. 6etop panel). Conversely, the GSEA of upregulated genes highlighted pathways associated with lipid metabolism, transmembrane signaling receptor activity, and molecular transport processes. Specific enrichment was observed in pathways related to plasma membrane activity, glycoprotein metabolism, lipase activity, and signaling transduction, emphasizing enhanced signaling and metabolic adaptations in the tumor microenvironment (Fig. 6ebottom panel). These upregulated pathways align with the aggressive phenotype and metastatic potential observed in high-risk UM21,41,42. These findings demonstrate that the BAP1−/− mouse model recapitulates important molecular hallmarks of class 2 UM tumors, including alterations in key signaling and lipid metabolic pathways. The high degree of overlap in transcriptional programs between the mouse model and human high-risk UM highlights the translational utility of this model for understanding BAP1-driven tumor and immune biology, as well as for developing immune therapeutic strategies. Discussion The BAP1 gene plays a critical tumor suppressor role in melanocytic tumors, particularly in aggressive forms such as UM and certain subtypes of CM, where its loss is associated with poor prognosis, tumor progression, and resistance to therapy5,43,44. Existing preclinical models, however, fall short of fully capturing the transcriptomic and microenvironmental complexities of melanocytic tumors with BAP1 loss22–25,29. This study bridges these gaps by developing CRISPR-engineered BAP1−/− models that closely mimic the transcriptomic features of advanced melanoma tumors. These models highlight the critical role of BAP1 loss in driving aggressive tumor phenotypes and reshaping the TME in ways that cannot be studied in existing systems. Importantly, they provide a platform to explore molecular and cellular mechanisms of BAP1 loss and identify therapeutic interventions that can be further tested in syngeneic systems. The syngeneic BAP1−/− models offer a significant advantage in studying tumor-host interactions in an immune-competent environment, filling an essential gap in preclinical research. It is important to note that Melan-a cells are classically 12-O-tetradecanoylphorbol-13-acetate (TPA)-dependent and non-tumorigenic. In our study, BAP1 loss introduced by CRISPR-Cas9 editing resulted in TPA-independent growth and robust tumor formation. However, we acknowledge that prolonged in vitro culture can lead to spontaneous transformation in Melan-a cells45,46, which may contribute to the observed tumorigenic phenotype. Notably, the BAP1 KO cells retained visible pigmentation in vitro and in tumors, indicating preservation of their melanocytic identity. This caveat underscores the need for future studies using early-passage cells and additional controls to isolate the specific contributions of BAP1 loss to tumorigenesis. The observed tumorigenic transformation in BAP1 KO cells, encompassing morphological changes, increased proliferation, and enhanced tumorigenic potential, aligns with existing literature linking BAP1 loss to aggressive phenotypes in human melanocytic tumors, especially UM47. The epithelial-like transformation of BAP1 KO melanocytes mirrors clinical findings of aggressive histopathological features in BAP1 loss tumors, such as epithelioid cell dominance in UM48. Similarly, the retention of melanocyte lineage markers, such as gp100, underscores the duality of BAP1 loss, promoting tumor aggressiveness while maintaining lineage-specific traits. These findings suggest that therapies targeting melanocyte antigens, such as gp100-based vaccines49,50, may hold promise in BAP1 loss tumors when combined with approaches to overcome their immunosuppressive TME. BAP1 loss is associated with significant immunological and metabolic changes contributing to tumor progression and immune evasion51,52. Our findings are consistent with reports of BAP1 loss creating an immunosuppressive TME characterized by altered immune cell phenotypes and stromal remodeling5. Specifically, BAP1 loss has been associated with reduced cytotoxic T-cell infiltration53, increased M2-polarized macrophages54, and impaired antigen presentation3. These effects have been observed in various cancers, including UM3,5,54, mesothelioma55, pancreatic ductal adenocarcinoma53, renal cell carcinoma8, etc. Such findings highlight the broad implications of BAP1 loss across diverse cancer types and underscore its critical role in shaping the TME toward an immunosuppressive state. These features are particularly relevant for immunotherapy, as the immunosuppressive TME in BAP1 loss tumors may limit the efficacy of ICIs8,55. Additionally, our transcriptomic analyses revealed the enrichment of pathways associated with lipid metabolism, immune modulation, and signaling reprogramming in BAP1 loss tumors. While our scRNA-seq analysis revealed key transcriptomic and microenvironmental differences between BAP1-KO and WT conditions, we acknowledge limitations in comparing tissues with differing tumorigenic potential. WT Melan-a cells did not form progressively growing tumors; however, they consistently produced small, palpable nodules at the site of injection. These nodules were collected at the same timepoints as KO tumors and contained a population of melanocytic lineage cells. While these cells enabled lineage-specific transcriptomic comparisons, they may represent a less transformed state than their KO counterparts. Consequently, some observed differences in gene expression may reflect both BAP1-driven changes and context-dependent features of tumor progression. We have taken this into account in our interpretation and highlighted the importance of cautious extrapolation when comparing tumor-like tissues to fully transformed tumors. Nonetheless, this approach allowed us to profile early in vivo responses and evaluate how BAP1 loss reshapes the immune and stromal landscape even in the presence of minimal tumor mass. This aligns with our previous studies identifying lipid metabolic reprogramming as a hallmark of BAP1 loss tumors, supporting their energy demands and aggressive growth21. Lipid metabolism reprogramming, including lipolysis, has been implicated in shaping the TME by fostering tolerogenic behaviors in immune cells56,57. For instance, TAMs can undergo lipid metabolism alterations that promote an immunosuppressive phenotype, aiding tumor progression58. Lipid accumulation within the TME can impair dendritic cell function, reducing T cell activation, affecting Natural Killer T (NKT) cell functions, and contributing to immune tolerance56,57,59. These findings highlight the importance of developing syngeneic preclinical models that enable the testing of therapies aimed at reversing these processes in the context of BAP1 loss, serving as a critical first step toward translating novel immunotherapy approaches60. The phenotypic divergence observed among our shortlisted BAP1 KO clones (Fig. 3 and Supplementary Data 9), including differences in tumorigenicity and immunosuppressive features (e.g., KO#4 and KO#2), likely reflects a combination of BAP1-dependent biological effects, CRISPR-associated genomic remodeling, and clonal adaptation during single-cell selection35,61,62. Although prior studies have shown that BAP1 loss can reduce proliferation and induce metabolic or transcriptional stress63,64, Melan-a BAP1 KO clones exhibited heterogeneous proliferative and clonogenic behaviors, consistent with context-dependent adaptation to the different levels of BAP1 deficiency. Clone-specific pathway enrichment, including terms related to intercellular junctional organization and adhesion programs in selected clones, is also consistent with reports that BAP1 loss can be associated with upregulation of genes encoding adhesion molecules such as CDH1, CADM1, and SDC2, facilitating adhesion-dependent proliferation33. Collectively, these findings underscore the importance of careful clone selection and comprehensive molecular characterization when interpreting the functional effects of BAP1 loss in preclinical models63–67. While human tumors with BAP1 loss also exhibit heterogeneity in aggressiveness and immune modulation, they involve additional molecular alterations beyond BAP165–67. Thus, the differences in phenotypes among BAP1 KO clones in our model provide an opportunity to dissect the specific contributions of BAP1 loss and the underlying mechanisms driving this diversity. The consistent enrichment of G-protein-coupled receptor signaling and transmembrane receptor activity across both in vitro and in vivo models highlights these pathways as intrinsic features of BAP1 loss tumors and potential therapeutic targets. The analysis of BAP1 KO clones highlights how loss of BAP1 alters cellular growth patterns in variable ways. These findings suggest that the tumorigenic potential following BAP1 loss may be influenced by intrinsic clonal differences, possibly reflecting distinct transcriptional or epigenetic states among edited cell lines. We speculate that, in an in vivo setting, additional microenvironmental cues, including stromal and immune cell interactions, could further modulate these phenotypes and will be the focus of future studies. Consequently, selecting the most representative BAP1 KO clone is critical for downstream functional studies and preclinical applications. The comparison of our selected BAP1 KO#2 models with human transcriptomic datasets further validates their relevance for studying high-risk melanocytic tumors. Upregulation of lipid metabolic and signaling pathways, and notably, the infiltration of T cells only in BAP1 KO tumors, emphasizes the ability of these pre-clinical models to recapitulate the molecular and cellular features of human melanoma tumors with BAP1 loss21,68,69. Moreover, the observed context-specific adaptations, such as the prioritization of immune evasion in intraocular tumors versus proliferation in subcutaneous tumors, underscore the influence of the TME on the biology of tumors lacking BAP1 expression. These insights highlight the importance of integrating TME-targeted therapies with systemic approaches to overcome both intrinsic and extrinsic tumor resistance mechanisms. A consistent macrophage phenotype emerged across both tumor models, pointing toward a shared immunologic consequence of BAP1 loss. In both the subcutaneous and intraocular settings, BAP1-deficient tumors displayed a clear shift from M1-like to M2-like macrophage states, accompanied by a reduction in antigen-presentation programs. This pattern was evident at multiple levels: UMAP clustering showed expansion of CD68+, TAM-like macrophages in KO tumors; dot-plots confirmed increased expression of canonical M2-associated genes with parallel suppression of MHC class I/II and proteasome components; and cell-fraction analyses revealed an overall increase in M2-polarized macrophages relative to WT. These findings indicate that BAP1 loss biases the tumor microenvironment toward an M2-enriched macrophage landscape as described in human uveal melanoma5,38,69. These observations motivate future functional studies to test whether M2 polarization and reduced antigen presentation directly contribute to T cell dysfunction in BAP1 KO tumors. While this study provides a suitable platform for studying BAP1 loss melanocytic tumors, it is not without limitations. The use of immortalized melanocyte Melan-a enabled controlled investigation of BAP1 loss with minimal confounding driver mutations. However, the absence of oncogenic mutations alone does not equate to functional or transcriptomic similarity to primary ocular melanocytes. Indeed, Melan-a cells primarily reflect the gene expression and epigenetic features of skin-derived melanocytes, which may limit the model’s applicability for faithfully recapitulating the biology of ocular melanocytic tumors11,12,70. Although intraocular tumor formation was observed following Melan-a cell injection, this does not fully overcome the biological divergence between cutaneous and ocular melanocytes. Immortalized ocular melanocytes would offer a more representative model, but primary ocular melanocytes were excluded due to their slow growth and fragility in culture71. We recognize that transcriptomic and phenotypic analyses (e.g., Figs. 3–6) are essential to assess how closely our model simulates ocular melanocyte biology. In the future, this model will be a foundation for developing improved models that immortalize primary melanocytes from different tissues so that melanocytic tumors lacking BAP1 can be studied orthotopically. Additionally, our model provides the first clear insights into the impact of BAP1 loss on the microenvironment of subcutaneous tumors and offers valuable intraocular data. While not based on immortalized ocular melanocytes with BAP1 loss injected into the eye, it still recapitulates the transcriptomic features of BAP1 loss in human UM tumors. Expanding this approach to other melanocytic tumors could further illuminate BAP1’s role across melanocyte-derived malignancies. Addressing these limitations through longitudinal studies and more comprehensive modeling systems will be crucial for advancing our understanding of BAP1-driven tumor biology. Our WES analysis confirmed successful BAP1 knockout across all CRISPR-Cas9 edited Melan-a clones (Supplementary Data 4–7). Yet, we recognize that clonal adaptation may have contributed to the variable growth phenotypes and tolerance to the BAP1 loss observed among them. BAP1 depletion is known to reduce cellular fitness, which could probably be mitigated by prolonged culture or selecting single-cell clones for adaptive variants63,72. Comparative WES of all four KO clones revealed certain recurrent mutations shared across clones, indicating that most secondary alterations likely represent common consequences of the editing and selection process rather than clone-specific events or targeted BAP1 KO by CRISPR-Cas9. Nevertheless, subtle clonal adaptations cannot be excluded entirely and may underlie phenotypic heterogeneity in proliferation or tumorigenic potential. Although off-target contributions are also possible, the WES analysis detected only a small number of additional exotic mutations, and the consistent phenotypic features across multiple KO clones suggest that the major effects are attributable to BAP1 loss. The immunogenic mutation burden observed in the Melan-a–derived lines suggests that these models retain the capacity to generate neoantigens, providing a reasonable molecular basis for studying immune recognition in the context of BAP1 loss. Nonetheless, determining whether these predicted mutations translate into functional immunogenicity will require dedicated in vivo validation. A key distinction between our model and human UM lies in the sequence of oncogenic events. In human UM, activating mutations in GNAQ, GNA11, CYSLTR2, or PLCB4 typically initiate tumorigenesis, while biallelic BAP1 loss occurs later to drive progression and metastasis. In contrast, our study deliberately focuses on BAP1 loss in the absence of these initiating mutations to isolate and functionally characterize its specific contribution to tumorigenic behavior and immune modulation. Unlike germline models73 that assess spontaneous tumor initiation, our approach begins with immortalized but non-tumorigenic melanocytes manually implanted in vivo, thereby modeling tumor development and immune modulation rather than de novo initiation. This reductionist design builds upon our previous findings that BAP1 loss alone is a potent determinant of immune and metabolic reprogramming independent of other oncogenic drivers5. Using the immortalized but non-tumorigenic Melan-a melanocyte line, we selectively introduced BAP1 deletion via CRISPR-Cas9 to examine how it shapes intrinsic cell states and the immune microenvironment in an immune-competent, syngeneic setting. While this approach does not replicate the full mutational hierarchy of human UM, it provides a controlled preclinical platform to dissect BAP1-driven immuno-metabolic pathways and their therapeutic vulnerabilities. Future iterations of this model, incorporating canonical UM driver mutations (GNAQ, GNA11, CYSLTR2, and PLCB4), will further refine its translational relevance and enable exploration of cooperative genetic dynamics in UM progression. Overall, several limitations have been recognized when interpreting functional outcomes in this model. While the consistent phenotypic features observed across multiple BAP1-deficient clones suggest that the major effects are attributable to BAP1 loss, subtle clonal adaptations and/or off-target contributions associated with CRISPR-Cas9 editing cannot be completely eliminated. In addition, the absence of non-targeting guide controls limits higher-level separation of BAP1-specific effects from background alterations introduced during editing and single-cell selection. Finally, because scRNA-seq and in vivo profiling were conducted from pre-clinical models developed using a single representative BAP1 knockout clone, some transcriptional features may reflect clone-specific cellular states despite overall consistency across edited clones. Future refinements incorporating multiple independent BAP1-KO clones, non-targeting controls, and parallel KO/rescue validation will further delineate BAP1-dependent programs from editing- or clonal-associated variability and strengthen the interpretive power of this preclinical system. In conclusion, this study demonstrates the multifaceted impact of BAP1 loss on melanocytic tumor biology, encompassing aggressive tumor phenotypes, immunosuppression, and metabolic reprogramming. The developed CRISPR-engineered models provide a valuable tool for exploring the interplay between intrinsic tumor characteristics and TME dynamics. This study focuses on whether BAP1 loss recapitulates the tumorigenicity and microenvironment of melanocytic tumors rather than its role in metastatic dissemination, particularly to the liver, a hallmark of advanced UM11,12,70. While metastatic modeling is beyond the scope of the current phase of this study, the syngeneic nature of the model offers a unique opportunity for future research into metastatic processes in UM and CM driven solely by BAP1 changes. These findings lay the groundwork for future studies leveraging the syngeneic aspect of our model to explore targeted therapies that counteract BAP1 loss. Additionally, this model offers a unique platform to test combinatorial immunotherapies aimed at reversing the immunosuppressive status associated with BAP1 loss, aiming at restoring responses to currently refractory immune checkpoint therapies in resistant melanocytic tumors. Methods Cell line and mice Melan-a, a murine melanocyte cell line (Sigma-Aldrich SCC202)31, was cultured in RPMI 1640 medium supplemented with 10% (v/v) fetal bovine serum (FBS, Cytiva SV30160.03HI) and 1% (v/v) penicillin/streptomycin (Gibco). Cells were maintained at 37 °C in a 5% CO₂ humidified incubator. For in vivo experiments, C57BL/6 mice were bred and maintained at the SingHealth Experimental Medicine Centre. Mice were housed under specific pathogen-free conditions with a 12-h light/12-h dark cycle and provided food and water ad libitum. Both male and female mice aged 6–8 weeks were used, with no sex-based exclusion. All animal experiments were approved by the SingHealth Institutional Animal Care and Use Committee. Animals were assigned to experimental groups according to the study design, without formal randomization. Experiments were performed under standardized conditions using age-matched mice housed in controlled environments, and all procedures were conducted consistently across groups to minimize potential confounding effects74,75. Blinding was not performed during allocation, experimentation, outcome assessment, or data analysis. No predefined inclusion or exclusion criteria were established a priori, and no animals or data points were excluded from the analyses. The experimental unit for in vivo studies was an individual animal. Sample sizes (n) represent the number of animals per group and are reported in the corresponding figure legends. Animal experiments BAP1+/+ (Class 1 UM) and BAP1−/− (Class 2 UM) clones were injected subcutaneously74 and intraocularly into C57BL/6 mice using established anterior chamber and ciliary body injection methods to mimic human uveal and cutaneous melanoma tumors76. Animals were monitored weekly using slit-lamp examination and anterior segment optical coherence tomography (OCT) imaging to assess tumor development and progression. Tumor growth was defined as the primary outcome measure. Tumor volume (TV) was calculated as: TV = mean largest height × mean basal diameter, and growth rates were determined by normalizing tumor volume to time post-injection. Mice were monitored throughout the study for general health, including body weight, activity, and signs of distress. Humane endpoints were applied in accordance with institutional guidelines, including tumor progression causing ocular perforation, severe intraocular pressure elevation, or evidence of significant discomfort or distress. For subcutaneous tumors, animals were euthanized when tumor volume reached 2000 mm3, or earlier if humane endpoints were met. Euthanasia was performed in accordance with institutional animal welfare guidelines. Cas9 sgRNA plasmid and BAP1 KO clones generation BAP1 KO clones were generated using the pGuide-it CRISPR-Cas9 System (Green) (Cat. No. 632601, Clontech Laboratories). Small guide RNAs (sgRNAs) targeting exon 1 and exon 14 of the BAP1 gene were selected from the designs screened using CHOPCHOP77. sgRNAs were ligated into the pGuide-it plasmid following the manufacturer’s protocol. Plasmids were transfected into Melan-a cells using Effectene Transfection Reagent (Qiagen 301425). Transfected GFP-positive cells were isolated by Fluorescence-Activated Cell Sorting (FACS). Initial gating excluded debris (FSC-A vs. SSC-A), followed by doublet discrimination (FSC-H vs. FSC-W and SSC-H vs. SSC-W). Single GFP-positive cells were seeded into 96-well plates and cultured in the conditioned medium (10% (v/v) fetal bovine serum (FBS) and 1% (v/v) penicillin/streptomycin) for clonal expansion. PCR and qPCR To assess whether the CRISPR-Cas9 system successfully introduced mutations in the BAP1 gene, a PCR assay was performed to amplify the targeted regions around exon 1 and exon 14. As previously described78, the PCR reaction was initiated with a pre-denaturation step at 95 °C for 10 min. Thermal cycling consisted of 50 cycles of denaturation at 95 °C for 15 s, annealing at 58 °C for 30 s, and extension at 72 °C for 30 s, followed by a final extension at 72 °C for 10 min. Primers were custom-designed and synthesized by Integrated DNA Technologies. To measure BAP1 expression levels, total RNA extracted from various passages of the BAP1 KO clones was reverse-transcribed to cDNA using the iScript RT reagent Kit (Bio-Rad 1708891). Quantitative PCR (qPCR) was conducted using the Roche LC480 qPCR system with SYBR Green Master Mix (Roche KAPA KK4618). Relative gene expression was calculated using the 2−ΔΔCt method, normalized against housekeeping genes (GAPDH and ACTB), and presented as fold changes compared to WT cells. Western blot For protein extraction, cells were lysed in RIPA buffer with protease and phosphatase inhibitors (Santa Cruz sc-24948) on ice, followed by centrifugation at 14,000 g for 15 min at 4 °C. Equal amounts of protein (25 µg), quantified by Bradford assay (Bio-Rad 5000002), were separated on a 10% SDS-PAGE gel (Bio-Rad) and transferred to nitrocellulose membranes (Amersham 10600001). Membranes were blocked with 5% (w/v) non-fat milk (Santa Cruz sc-2324) and 5% (w/v) BSA (Sigma-Aldrich) in Tris-buffered saline with 0.1% Tween-20 (TBS-T) for 1 h at room temperature and incubated overnight at 4 °C with primary antibodies (BAP1 Rabbit mAb #13271, Cell Signaling Technology, 1:1000 dilution). After washing with TBS-T, membranes were incubated with HRP-conjugated secondary antibody (Cell Signaling Technology #7074) and fluorescence-labeled GAPDH and β-actin antibodies (Santa Cruz sc-47724 AF790 and sc-47778 AF680) as loading controls for 1 h at room temperature. Protein bands images were acquired using chemiluminescent substrate (Thermo Scientific #34580) on the LI-COR Odyssey Fc imaging system and analyzed with Image Studio software. For figure presentation, non-adjacent lanes from the same original blot were assembled where appropriate. Sanger sequencing PCR products were purified using the previously published method78. The amplified PCR products were purified using the QIAquick Gel Extraction Kit (Qiagen) for Sanger sequencing. Post-sequencing, the quality of the DNA traces was assessed using FinchTV software (Geospiza), ensuring high-fidelity sequencing reads. Multiple sequence alignments were conducted using CLC Sequence Viewer (Qiagen) to compare the sequenced data with reference sequences. A sequence homology search was performed using the NCBI blast algorithm (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to confirm sequence identity and analyze potential sequence variations. Cytology and IHC For cytological analysis and immunohistochemistry (IHC), cell pellets or suspensions were pre-fixed in PreservCyt fixative as previously outlined79 and submitted to the Department of Anatomical Pathology and Cytology at Singapore General Hospital (SGH). Cell samples were centrifuged using a Thermo Scientific Shandon Cytospin (Shandon Scientific Ltd, Cheshire, UK)80 to prepare cytospin slides. These slides were subjected to Diff-Quik (DQ, Baxter, McGraw Park, IL) and Papanicolaou (PAP) staining following standard diagnostic procedures81 and previously established IHC protocols82. For IHC, single staining was performed using antibodies against gp100 (Abcam, #ab137078), employing the Bond Polymer Refine Detection Kit (Leica Biosystems) with diaminobenzidine (DAB) as the chromogen. Paraffin-embedded tissue sections were stained with hematoxylin and eosin (H&E) to assess histopathological features. Cytospin slides and tissue sections were visualized under an Olympus BX42 microscope at 60x magnification (field diameter: 0.55 mm). RNA and DNA extraction Cell suspensions were centrifuged at 500 × g for 3 min, washed three times with PBS, and lysed using TRIzol reagent for RNA isolation. RNA was extracted with the Direct-zol RNA Kit (Zymo Research), including on-column DNase I treatment to remove genomic DNA contamination. Purified RNA was eluted in 20 µL RNase-free water and stored at −80 °C. Genomic DNA was extracted from separate aliquots of cell suspensions using the DNeasy Blood & Tissue Kit (Qiagen) and stored in 50 µL RNase-free water at −20 °C. RNA and DNA concentrations and purity were assessed using a NanoDrop spectrophotometer (A260/A280 ratios). Cell proliferation assay Cell proliferation was monitored in real-time using the xCELLigence Real-Time Cell Analysis (RTCA) system (ACEA Biosciences)83, which measures electrical impedance proportional to cell number, viability, and morphology. BAP1+/+ and BAP1-/- clones were evaluated for proliferation differences. E-Plates were prepared with 50 µL of RPMI-1640 medium containing 10% FBS, baseline impedance recorded, and equilibrated at 37 °C with 5% CO2 for 30 min. Cells (5 × 103 per well) were seeded in 150 µL of medium, with triplicate wells for each clone and negative controls. Impedance (cell index, CI) was measured every 15 min for 6 h and then every 30 min thereafter. RTCA software analyzed proliferation kinetics. Data from three independent replicates confirmed reproducibility. Wound healing assay 1 × 106 cells/well of BAP1+/+ and BAP1−/− clones were seeded in a 6-well plate and cultured for 24 h. A sterile 1000 µL pipette tip created a scratch (wound) on the cell monolayer. After the wound was made, the cells were cultured in fresh medium at 37 °C with 5% CO2. Wound width measurements were taken at 0, 6, 24, 30, 48, 54, 72, and 78 h using a Nikon ECLIPSE TS100 inverted microscope at 4× magnification. Six fields per well were imaged, and the percentage of wound closure was calculated using the ImageJ MRI Wound Healing Tool. The experiment was repeated independently three times. Cell invasion assay For the Matrigel Invasion Assay, Transwell inserts (8 μm pore size; Corning 3422) were coated with 50 μL of Matrigel (BD Biosciences) diluted 1:3 in cold serum-free medium and incubated at 37 °C for 1 h. BAP1+/+ and BAP1−/− clones (2 × 105 cells/well) were seeded into the upper chamber in a serum-free medium, while the lower chamber contained a medium with 10% FBS as a chemoattractant. After 24 h at 37 °C with 5% CO2, non-invading cells were removed, and invaded cells on the lower membrane surface were fixed with 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. Cells were stained with DAPI (1 μg/mL in PBS) for 10 min in the dark and visualized under a fluorescence microscope (×10 magnification). Images were taken from at least five random fields per well to quantify invaded cells using ImageJ software. The assay was performed in triplicate, and results were presented as the mean number of DAPI-positive cells per field. Colony formation (clonogenic) assay BAP1+/+ and BAP1−/− clones (1000 cells/well) were seeded into 6-well plates in 3 mL of complete growth medium and cultured at 37 °C with 5% CO₂ for 10–14 days until visible colonies (≥50 cells) were formed. The medium was refreshed every 3–4 days. Colonies were fixed with 4% paraformaldehyde for 15 min and stained with 0.5% crystal violet solution (in 20% methanol) for 30 min. Excess stain was washed with distilled water, and plates were air-dried. Colonies were imaged using a Nikon Eclipse microscope and quantified using NIS-Elements Advanced Software (V4.20.23), applying an automated method based on marker intensity, density, and object size. Only colonies with ≥50 cells were included. Three independent experiments were performed, with results expressed as the mean colony count per condition. Soft agar colony formation assay The soft agar colony formation assay evaluated the anchorage-independent growth of BAP1+/+ and BAP1−/− clones. A base layer (1 mL) of 1% agarose in RPMI-1640 medium with 10% FBS and 1% penicillin/streptomycin was poured into 6-well plates and solidified at room temperature for 30 min. A second layer containing 0.5% agarose and 10 K cells in the same medium was added and allowed to solidify. To maintain hydration, 1 mL of complete RPMI-1640 medium was added to each well, and plates were incubated at 37 °C with 5% CO2 for 4–6 weeks, with medium changes every 3–4 days. Colonies were stained with 0.005% crystal violet, imaged using a Nikon Eclipse microscope, and quantified with NIS-Elements Advanced Software (V4.20.23). Each experiment was performed in triplicate, and clone colony counts were compared. Single-cell RNA-seq and data analysis For subcutaneous tumors, tissues were harvested at 8 weeks post-injection, and for intraocular tumors, tissues were harvested at 5 weeks post-injection. In WT Melan-a–injected mice, although they did not develop progressively growing tumors, we consistently observed small, stable nodular masses at the injection site. These palpable nodules were collected and processed for single-cell RNA sequencing in parallel with the larger tumors formed by BAP1 KO clones. scRNA-seq was performed using the 10x Genomics Chromium platform, explicitly utilizing the 10x Genomics Flex assay for fixed cells. Following the manufacturer’s guidelines, ocular and cutaneous tumor tissues were first dissected from mice and fixed according to the Flex 10x Genomics protocol, allowing for sample preservation before processing. After fixation and disassociation, the cells were stored at −80 °C until further processing. Upon thawing, cell viability was assessed to ensure adequate preservation of cell integrity, and the cells were subsequently washed and counted. Approximately 10,000 cells per sample were used for GEM (gel bead-in-emulsion) generation, leveraging the 10x Chromium Flex 3’ Library & Gel Bead Kit. Following GEM formation, reverse transcription was performed inside individual GEMs to capture mRNA transcripts and synthesize cDNA. This was followed by cDNA amplification, library construction, and sequencing using an Illumina NovaSeq 6000 platform, targeting a depth of at least 10,000 reads per cell. Raw sequencing data were processed using the Cell Ranger (version 6.0, 10x Genomics) pipeline, which aligned reads to the mouse reference genome (GRCm38) and generated a gene-barcode matrix for each sample. Quality control and filtering steps were performed using Seurat (version 5.0) in R84. Cells with high mitochondrial content (>10%) and low unique molecular identifiers (UMIs) were excluded to ensure data quality. Data normalization, scaling, and clustering were conducted within the Seurat framework84. Significant principal components were identified through PCA, and cell clustering was performed using the first 50 principal components at a resolution of 0.3. Dimensionality reduction was visualized using UMAP plots, facilitating the exploration of distinct cellular subpopulations within the tumor microenvironment. Cell type annotation was based on canonical lineage markers. T cells were identified by Cd3d, Cd3e, Cd3g, and Trac expression, whereas NK cells were distinguished by Ncr1 and Klrb1c. A Pcdha-enriched cluster lacking neuronal markers (Snap25, Map2, Tubb3, Nefl) was annotated separately. Melanocyte and proliferative melanocyte subsets were defined by the co-expression of melanocytic markers (Pmel, Tyrp1, Mlana, Tyr) with or without cell-cycle markers (Mki67, Pcna). Genes with an adjusted p < 0.05 and a log2 fold change >1 were considered significant. Further pathway analysis was conducted using clusterProfiler to identify key biological processes enriched within specific cell clusters. Immune cell deconvolution was performed using both SingleR and CIBERSORTx. Visualization tools such as UMAP, heatmaps, and violin plots were employed to depict gene expression patterns, and the results were cross-validated with published datasets to ensure the robustness of the findings. To address batch effects in single-cell RNA sequencing data, the R package Harmony39 was utilized, specifically designed to handle batch variation by integrating datasets from different experimental conditions while maintaining the biological variation of interest. Since dataset integration across different tissue locations or dominant cell populations may result in loss of true biological differences following batch correction, as previously described85, UMAPs were generated and analyzed separately for normal and tumor samples, as well as for distinct tissue locations (skin and ocular). This approach preserves distinct transcriptional states of normal and tumor tissues, thereby maintaining biologically meaningful structure86. Cross-condition comparisons were subsequently performed by subsetting equivalent cell populations for downstream analyses. For cross-species comparative analysis, publicly available single-cell RNA-seq data from 11 human uveal melanoma patients (GSE139829)38 were obtained from the Gene Expression Omnibus (GEO). The dataset was processed using the same Seurat-based workflow, including quality control, normalization, dimensionality reduction, and clustering. Tumor cells were identified using canonical melanoma markers and further stratified into class 1 (low-risk) and class 2 (high-risk) transcriptional signatures, as previously described. Differential gene expression analysis was performed to compare tumor cell populations between groups, and overlapping differentially expressed genes between the BAP1 KO mouse tumor model and class 2 human UM tumors were identified for downstream comparative analysis. Bulk transcriptome and data analysis RNA quality and concentration were assessed using a Nanodrop 2000 spectrophotometer (Thermo Scientific) and an Agilent 2100 Bioanalyzer (Agilent Technologies), with only samples showing RNA integrity number (RIN) values above 8.0 being used for further analysis. Library preparation was performed using the NEBNext Ultra II RNA Library Prep Kit for Illumina (New England Biolabs), and sequencing was conducted on the Illumina NovaSeq 6000 platform to achieve a read depth of at least 30 million paired-end reads per sample. Before alignment, the raw sequencing data were subjected to quality control using FastQC (version 0.11.9) to check for read quality. Reads were aligned to the mouse reference genome (GRCm38) using the STAR aligner (version 2.7.9a). The aligned reads were then quantified at the gene level using featureCounts (version 2.0.1) from the Subread package. Reads were aligned to the mouse reference genome (GRCm38) using the STAR aligner (version 2.7.9a). Data visualization, including heatmaps, volcano plots, and principal component analysis (PCA), was performed using ggplot2 and pheatmap in RStudio. Whole-genome sequencing (WGS) and whole-exome sequencing (WES) analysis Whole-genome sequencing (WGS) data were generated for primary melanocytes, B16F10 cells, and WT Melan-a cells to assess global genomic integrity and copy number alterations. Raw sequencing reads were quality-checked and aligned to the mouse reference genome (mm10) using BWA-MEM. Aligned reads were subsequently processed using GATK (v4), including sorting and duplicate marking with MarkDuplicatesSpark. Whole-exome sequencing (WES) was performed for WT Melan-a cells and shortlisted BAP1 knockout (KO#1-4) Melan-a clones to characterize coding-region mutations associated with BAP1 loss. Raw sequencing reads were subjected to quality control and aligned to the mouse reference genome (mm10) using BWA-MEM. Aligned reads were processed according to GATK best practices, including sorting, duplicate marking, and base quality recalibration. Somatic variant calling was performed using GATK Mutect2, with each BAP1 KO clone analyzed against WT Melan-a cells as the matched normal control. This approach enabled the identification of clone-specific somatic single-nucleotide variants (SNVs) and small insertions/deletions (indels) within coding regions. Variant filtering was applied using standard Mutect2 filtering steps to remove sequencing artifacts and low-confidence calls. Variants were annotated using a reference gene annotation database, and downstream analyses were restricted to protein-coding regions and canonical splice sites. This WES-based strategy was used to assess coding-region mutational landscapes and overall genomic stability of BAP1 KO clones relative to WT controls. Statistics and reproducibility Statistical analyses were performed to evaluate experimental data. Sample sizes (n) represent biologically independent samples, animals, or experiments, as indicated in the relevant figure legends. Sample sizes for animal experiments were determined based on previously established tumor growth and single-cell sequencing studies using similar models, consistent with standard practice in exploratory preclinical research74,75. Technical replicates were averaged prior to statistical analysis, where applicable. Data normality was assessed using the Shapiro–Wilk normality test to guide the selection of appropriate statistical tests. For normally distributed data, differences between two groups were analyzed using a two-tailed unpaired t-test. Non-normally distributed data were analyzed using non-parametric tests, including the Mann–Whitney U test. Correlations between variables were assessed using Pearson’s correlation test for normally distributed data and Spearman’s correlation test for non-normally distributed data. For experiments involving three or more groups, statistical significance was assessed using two-way ANOVA followed by Dunnett’s multiple comparisons test. For tumor growth and IOP measurements, the reported p-values correspond to comparisons at the final time point unless otherwise stated. Results are presented as mean ± standard error of the mean for normally distributed data or median with interquartile range for non-normally distributed data. Exact sample sizes, statistical tests, and p-values are provided in the corresponding figure legends. A p < 0.05 was considered statistically significant. Ethics declarations All animal experiments were conducted with approval from SingHealth Institutional Animal Care and Use Committee (IACUC) under reference number 2022/SHS/1756. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Reporting Summary Transparent Peer Review File Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Reporting Summary Transparent Peer Review File