To investigate the relationship between keratinocytes and melanoma cells in vivo, we created a transgenic zebrafish line in which GFP is expressed under the krt4 promoter (Gong et al., 2002). This line faithfully marks all adult keratinocytes present throughout the fish epidermis and scales, similar to previous lines using this promoter (Figure 1A). We then initiated melanomas in this background using the TEAZ method (Transgene Electroporation of Adult Zebrafish) (Callahan et al., 2018; Montal et al., 2024), in which plasmids containing oncogenes or sgRNAs against tumor suppressors can be introduced directly into the skin (Figure 1B). The major advantage of this method is that we can visualize melanoma initiation when the tumor is in its early stages, consisting of a small number of cells. We initiated tumors with a combination of BRAFV600E, sgRNAs against PTEN (ptena/b), and germline loss of p53 (Figure 1C). To account for skin wounding from electroporation in assessing changes in TAKs, we also performed TEAZ using a control vector that labels melanocyte-precursors but does not induce melanoma formation. Fluorescent imaging 8 weeks post-electroporation with the control vector demonstrated an injury-free epidermis, in contrast to the pronounced melanoma development in zebrafish administered with the oncogenic vectors (Figure 1D).

To better investigate the changes in the keratinocytes, we used confocal microscopy on fish 8 weeks after TEAZ. This revealed a marked disruption of keratinocyte morphology in the tumor-bearing fish, which was not seen in control fish. We specifically noted disrupted cell–cell junctions, a disorganized pattern of keratinocytes, and loss of the normal hexagonal cell layer (Figure 1E). These changes were reminiscent of keratinocyte EMT, which has been previously noted to occur in wounded epidermis (Haensel and Dai, 2018; Leopold et al., 2012; Moreno-Bueno et al., 2009). To further assess this possibility, we excised tissues from both tumor and control skin and used FACS to isolate keratinocytes (GFP+) and melanoma cells (tdTomato+) and performed qPCR. As expected, we found enrichment of mitfa in melanoma cells and krt4 in keratinocytes, validating the successful cell-type isolation (Figure 1F). Comparative analyses between TAKs and normal keratinocytes (NKCs) from tissue without melanoma revealed upregulation of EMT markers vimentin and N-cadherin, consistent with our imaging results (Figure 1G).

We next asked whether these changes were also seen in human samples. To address this, we performed co-culture experiments between keratinocytes and melanoma cells. We grew GFP-labeled HaCaT keratinocytes either alone or with A375 melanoma cells for 21 days, followed by isolation by FACS for bulk RNA-sequencing (Figure 1H; Tagore et al., 2023). Consistent with our in vivo results in the fish, the top pathway altered in the co-cultured HaCaT cells was enrichment of EMT (Figure 1I). Differential gene expression analysis showed notable upregulation of the mesenchymal markers vimentin and N-cadherin in co-cultured keratinocytes compared to monocultured control (Figure 1J), similar to what was found in vivo. Collectively, our data indicate that melanoma cells induce morphological and molecular markers of EMT in nearby keratinocytes.

EMT is usually driven by upstream transcription factors, which then act on downstream targets to repress adhesion molecules such as E-cadherin or activate other adhesion molecules such as N-cadherin (Moreno-Bueno et al., 2009). We next wanted to understand which of these transcription factors was responsible for the EMT-like behavior in our keratinocytes. To address this, we utilized an existing scRNA-sequencing dataset of a BRAFV600E-driven zebrafish melanoma (Figure 2A). Dimensionality reduction with UMAP and subsequent clustering revealed two keratinocyte populations as indicated by module scoring for genes enriched in keratinocyte populations (Figure 2B, C). Subsequent differential gene expression and Gene Set Enrichment Analysis (GSEA) of the two keratinocyte clusters revealed one cluster with enrichment for EMT, similar to what we observed above (Figure 2D, E). We refer to this EMT cluster as TAKs, and the other cluster as NKCs. We focused on three EMT-transcription factors expressed in zebrafish keratinocytes, snai1a, snai2, twist1a, zebrafish homologs of human SNAIL, SLUG, and TWIST. Differential expression showed significant enrichment in snai1a and twist1a in TAK vs. NKC clusters (Figure 2F). The significant enrichment of twist1a in the TAK cluster, coupled with its rare expression in NKCs, positioned twist1a as a promising candidate for further investigation into its potential role in driving EMT-like changes in keratinocytes and, consequently, its impact on melanoma progression.

Having identified Twist as a potential driver of EMT-like changes in TAKs, we next asked how it affected melanoma phenotypes. We created new transgenic zebrafish in which the krt4 promoter was used to drive the two zebrafish TWIST paralogs (twist1a and twist1b) in the context of BRAF-driven melanomas. In this experiment, we used standard 1-cell injection of the plasmids rather than TEAZ to ensure that the results were more generalizable to different melanoma initiation assays. Injected embryos were sorted at 5 days post-fertilization for tdTomato (melanocyte) and GFP (keratinocyte) expression, indicating successful melanocyte and keratinocyte transformation (Figure 3A). These were mosaic animals, rather than stable lines, so we could directly assess the effect of TWIST on melanoma growth (although this precluded us from examining the effect of TWIST in keratinocytes without a nearby melanoma). We monitored the fish for tumor-free survival as well as overall survival over the ensuing 26 weeks. Interestingly, we found no difference in melanoma initiation rate in the TWIST overexpression condition compared to empty vector (Figure 3B). Unexpectedly, however, we noted that overall survival was improved in the transgenic animals expressing TWIST in the keratinocytes (Figure 3B).

This discrepancy between melanoma-free and overall survival suggested that the tumors in the TWIST condition should be phenotypically distinct from the control tumors. To assess this, we performed immunohistochemistry on the tumors (n = 3 in each condition) and surrounding tissues from the control and TWIST conditions (Figure 3C). The oncogenic driver in this melanoma model is hBRAFV600E and serves as an IHC marker for the tumor cells. Comparison of hBRAFV600E staining revealed significant melanoma infiltration into the zebrafish body in the CTRL condition, as opposed to a nearly non-invasive tumor in the TWIST condition (Figure 3D). The lack of melanoma invasion was also observed when the melanoma developed in other anatomical locations as well (Figure 3—figure supplement 1), although because survival started to decrease around 26 weeks, we could not let these fish grow longer to see if they eventually became invasive at very late time points. These findings suggest that twist1a/twist1b overexpression in keratinocytes does not impair tumor initiation, but instead impairs melanoma invasion and improves survival when expressed in the microenvironment.

Because this was a somewhat unexpected finding, we wanted to ensure that this was not an artifact of the zebrafish system. We therefore developed an assay to measure interactions between human keratinocytes and melanoma cells. To simulate melanoma invasion into neighboring cells, we developed a novel cell culture-based invasion assay using HaCaT lines and multiple melanoma cell lines. Melanoma cells cultured on poly-l-lysine coated coverslips were placed atop keratinocytes, allowing assessment of melanoma cell infiltration after 20 hr (Figure 4A). Due to the migratory nature of melanoma cells, they will migrate off the coverslip and infiltrate the layer of keratinocytes, allowing us to assess relative differences between culture conditions.

HaCaT keratinocytes were transformed via lentiviral infection to overexpress TWIST1 (HaCaT-TWIST) or an empty vector control (HaCaT-CTRL) (Figure 4B). Western blot analysis confirmed robust Twist protein overexpression in HaCaT-TWIST compared to HaCaT-CTRL (Figure 4C), and immunofluorescence imaging revealed nuclear localization of Twist (Figure 4D). Co-culture of HS294T melanoma cells with HaCaT-TWIST resulted in significantly reduced melanoma cell invasion into keratinocytes compared to HaCaT-CTRL (Figure 4E, F), similar to what we had observed above in our zebrafish system. This finding was recapitulated using SKMEL2, demonstrating the inhibitory effect of TWIST1 overexpression in keratinocytes on melanoma invasion across different cell lines (Figure 4G, H). However, it was also possible that it was something lacking in the TWIST keratinocytes that led to less melanoma invasion (rather than actively suppressing invasion). To test this, we also performed a control assay in which we allowed tdTomato melanoma cells to invade into unlabeled melanomas of the same line. Using both HS294T and SKMEL2 cells, these cells do have some migratory capacity in this setting (Figure 4—figure supplement 1). This assay cannot discern whether the TWIST keratinocytes actively suppress invasion vs. lack of something that promotes invasion, which will be an area for future studies. Collectively, our results demonstrate that induction of EMT in keratinocytes is associated with reduced melanoma invasion and improvement in animal survival.

While EMT within the tumor cell is well recognized to promote invasion, our data suggest that EMT in the microenvironment paradoxically restrains tumor invasion. We wanted to better understand the downstream mechanisms accounting for this result. We therefore analyzed our control vs. TWIST tumors using single-cell RNA-sequencing, which would allow us to understand potential mechanisms by which melanoma cells were interacting with these keratinocytes. Melanoma and adjacent skin from three fish per condition (CTRL or TWIST) were dissociated into single-cell suspensions and FACS-sorted for eGFP+ keratinocytes and tdTomato+ melanoma cells. To enrich the keratinocyte population, keratinocyte and melanoma cell suspensions were combined at a 7:3 ratio for each condition and prepared for scRNA-sequencing (Figure 5A). The resulting CTRL and TWIST datasets were filtered for quality control as described (Figure 5—figure supplement 1A). UMAP dimensional reduction of the integrated dataset shows distinct clusters of melanoma and keratinocytes, with high eGFP expression in keratinocyte clusters and high tdTomato expression in melanoma clusters (Figure 5B). Furthermore, we identified two KC clusters in the present in both the CTRL and TWIST conditions and a third cluster of KCs in the TWIST condition. We performed differential gene expression analysis between clusters and GSEA on differentially expressed genes between the two clusters present in both conditions showed enrichment in EMT as seen in Figure 2, identifying EMT-enriched cluster as TAK and other as NKC (Figure 5—figure supplement 1B). As expected, we also identified a unique cluster of KCs only present in the TWIST dataset that highly expresses twist1a/b (Figure 5—figure supplement 1C), the genes that we overexpressed in the TWIST condition, and labeled this cluster Twist-High (Figure 5C).

We first analyzed the melanomas that arose in the control vs. TWIST animals by comparing their gene signature to well-annotated signatures of human melanoma cell states. It is now widely recognized that melanomas exist in at least four different transcriptional states, ranging from undifferentiated/invasive to neural crest, to intermediate, to melanocytic/proliferative (Tsoi et al., 2018). These distinct states are mutually exclusive, yet exhibit a high degree of plasticity, with interconversion between the states (Tsoi et al., 2018). Interestingly, we found a cluster of melanoma cells that developed only in the TWIST condition was enriched for the melanocytic/proliferative state but not undifferentiated/invasive gene markers (Figure 5E). Consistent with this, GSEA of the TWIST condition also showed significant enrichment for oxidative phosphorylation (Figure 5—figure supplement 1D–F), which we previously showed is representative of the melanocytic state (Lumaquin-Yin et al., 2023). Because this state is among the least invasive ones, this is consistent with our in vivo observations that these melanomas are phenotypically less invasive.

We hypothesized that this change in cell state might be induced by physical interactions between the TWIST keratinocytes and the melanoma cells. To address this, we analyzed potential cell–cell interactions using CellChat, a software tool that allows us to quantitatively characterize and visualize cell–cell communications using a curated zebrafish ligand–receptor interaction database (Jin et al., 2021; Figure 5F). Two unique ligand–receptor pairings were identified that only occur between TWIST keratinocytes and the melanomas that arose in these animals: a homotypic jam3b–jam3b interaction and a pgrn–sort1a (progranulin–sortilin) interaction (Figure 5G).

The jam3b interaction was of particular interest to us, as this protein has been recently identified as one required for melanophore survival in zebrafish (Eom et al., 2021) and for human melanoma metastasis (Arcangeli et al., 2012; Langer et al., 2011). In normal human skin, JAM3 is expressed in keratinocytes of the superficial epidermis, whereas its heterotypic partner JAM1 is only expressed in basal keratinocytes (Uhlén et al., 2015). Since melanomas largely arise in this basal area, this suggests that TWIST expression in the keratinocytes was leading to aberrant expression of jam3b, forming stronger homotypic jam3b–jam3b attachment between these keratinocytes and the melanoma cells and preventing their invasion.

Finally, we wished to determine if the TWIST^hi state in our zebrafish models was also present in human melanomas. We use the GeoMX spatial transcriptomics platform to interrogate a series of early melanoma precursor lesions, a subset of a larger study using this method to look at spatial organization of melanoma stages (Vallius et al., 2026). We first stained sections with antibodies to cytokeratin (to mark keratinocytes) and SOX10/MART1 (to mark melanocytes and melanoma cells). This allows for selection of microregions of interest (MR) which are then subject to RNA-seq (Figure 6A–D). From this, we then queried for the TAK/TWIST gene sets found in the zebrafish. Whereas normal skin melanocytes showed no such enrichment, there was an enrichment of these signatures as the lesions progressed from early to late precursor (Figure 6E). This suggests that EMT-like alterations in keratinocytes are an early event in both zebrafish and human melanoma.

Zebrafish were maintained in a dedicated facility with controlled temperature (28.5°C) and salinity. The fish were kept on a 14-hr light/10-hr dark cycle and fed a standard zebrafish diet consisting of brine shrimp followed by Zeigler pellets. Embryos were obtained through natural mating and incubated in E3 buffer (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄) at 28.5°C. For procedures requiring immobilization, zebrafish were anesthetized using Tricaine-S (MS-222, Syndel) prepared as a 4 g/l stock solution with a pH of 7.0. The stock solution was protected from light exposure and diluted to the appropriate concentration to achieve fish immobilization. All experimental procedures and animal protocols described in this manuscript were conducted in compliance with the Institutional Animal Care and Use Committee (IACUC) protocol #12-05-008, approved by the Memorial Sloan Kettering Cancer Center (MSKCC).

Embryos at the one-cell stage from the Casper Triple zebrafish line (mitfa:BRAFV600E;p53−/−;mitfa−/−;mpv17−/−) (Callahan et al., 2018; White et al., 2008) were injected with a krt4:eGFP expression cassette in the 394 vector of the Tol2Kit (Kwan et al., 2007) with tol2 mRNA. Larvae were sorted for positive GFP fluorescence at day 3 and raised to adult for breeding. F0 fish were in-crossed and resulting F1 were outcrossed with Casper Triple zebrafish for consistent GFP expression. Starting from F2, the krt4:eGFP zebrafish line was maintained by out-crossing with Casper Triple zebrafish and sorting for GFP expression.

TEAZ was utilized to generate melanoma as previously described (Callahan et al., 2018; Montal et al., 2023). Krt4:eGFP zebrafish (krt4:eGFP Casper Triple) were anesthetized with tricaine and injected with a plasmid solution containing miniCoopR-tdT (250 ng/μl), mitfa:Cas9 (250 ng/μl); zU6:sgptena (23 ng/μl) with target sequence GAATAAGCGGAGGTACCAGG, zU6:sgptenb (23 ng/μl) with target sequence GAGACAGTGCCTATGTTCAG, and the tol2 plasmid (55 ng/μl). For control electroporation without generating melanoma, zebrafish were injected with mitfa:tdTomato (250 ng/μl), mitfa:Cas9 (250 ng/μl), zU6:non-targeting (46 ng/μl), and tol2 plasmid (55 ng/μl). All fish were injected on the left flank below the dorsal fin and electroporated with the BTX ECM 830 electroporator using 3 mm platinum Tweezertrodes (BTX Harvard Apparatus; #45-0487). Electroporator settings used: LV Mode, 40 V, 5 pulses, 60-ms pulse length, and 1-s pulse interval. Electroporated zebrafish were screened for successful electroporation 7 days post-electroporation by tdTomato expression using fluorescence microscopy and melanoma tracked by imaging once per week. All live zebrafish imaging were performed with the Zeiss AxioZoom V16 fluorescence microscope.

sgRNA sequences for TEAZ listed below:

Zebrafish with or without melanoma were anesthetized in tricaine (MS-222, Syndel) as described above. Site of injection is visually identified by the presence of melanoma or the area below the dorsal fin. Scales were removed with tweezers and fixed in 4% PFA in PBS (Santa Cruz 281692) in a 96-well plate for 15 min. Fixed scales were washed three times with PBS and permeabilized with 0.1% Triton X-100 (Thermo Scientific 85111) in PBS, then blocked with 10% goat serum (Thermo Fisher 50062Z). Scales were incubated with 1:250 GFP polyclonal antibody, Alexa Fluor 488 (Thermo Fisher A21311) overnight at 4°C. The next day, scales were washed three times with PBS, incubated with 1:1000 Hoechst 33342 (Thermo Fisher H3570) for 1 hr and mounted onto slides with VECTASHIELD Vibrance Antifade Mounting Media (Vector Laboratories H-1700). Samples were imaged on the Zeiss LSM880 inverted confocal microscope, and images were processed using FIJI v1.53.

Zebrafish were euthanized using ice-cold water. Melanoma and adjacent skin were dissected from fish with melanoma, and skin alone was dissected from below the dorsal fin. Subsequently, samples were cut into 1-mm strips using a clean scalpel and placed into 15-ml conical tubes (Falcon 352099) with 3 ml of DPBS (Gibco 14190250) and 187.5 μl of 2.5 mg/ml Liberase TL (Roche 5401020001). Samples were incubated in dissociation solution at room temperature for 30 min on a shaker with gentle movement to prevent tissue from settling at the bottom of the tube. At 15 min of incubation, a wide bore p1000 pipette tip (Thermo Scientific 2079G) was used to gently pipette the sample up and down for 90 s. After 30 min, 250 μl of FBS (Gemini Bio) was added to stop the enzymatic activity of Liberase TL, and samples were pipetted up and down using a wide bore p1000 pipette tip for 90 s. Dissociated cells were then filtered through a 70-μm cell strainer (Falcon 352350) into a 50-ml conical tube (Falcon 352098) placed on ice. Samples were centrifuged at 500 × g at 4°C for 5 min and supernatant was removed by pipetting. The cell pellet was resuspended in 500 μl of PBS with 5% FBS and filtered again through 40 μm tip filters (Bel-Art H136800040) into 5 ml polypropylene tubes (Falcon 352063). For subsequent FACS analysis, 0.5 μl of 1000x DAPI (Sigma-Aldrich D9542) was added to each sample. Samples were FACS sorted (BD FACSAria) at 4°C for GFP-positive keratinocytes and tdTomato-positive melanoma gated using fluorophore-negative zebrafish controls.

FACS sorted zebrafish cells were deposited directly into 750 μl TRIzol LS Reagent (Invitrogen 10296010) in Eppendorf DNA LoBind Tubes (Eppendorf 022431021). After collection, samples were snap-frozen using dry ice and stored at −80°C. RNA extraction was performed per TRIzol LS manufacturer protocols. For precipitation of RNA, 10 µg supplemental glycogen (Roche 10901393001) was used per sample to account for low cell numbers. Resulting RNA was resuspended in Nuclease-free water (Fisher Scientific AM9937). 25 ng RNA per sample was transcribed to cDNA using Superscript III First-Strand Synthesis System (Invitrogen 18080051). cDNA mix was diluted 1:10 with Nuclease-free water for real-time quantitative PCR using Power SYBR Green PCR Master Mix (Applied Biosystems 4368708) and the Bio-Rad CFX384 Touch Real-Time PCR System (Bio-Rad 1855484). Resulting Cq values were normalized to hatn10 as previously described.

qPCR primer sequences:

Zebrafish scRNA-seq data from Hunter et al., 2021 was reanalyzed using R 4.2.0 and Seurat 4.3.0 (Butler et al., 2018; Hao et al., 2021). Cluster identities were maintained as published. Keratinocyte Module Scores were calculated using the AddModuleScore function with default parameters using published gene lists. Differential gene expression analyses between clusters were performed using FindMarkers. Differentially expressed gene lists were converted from zebrafish genes to human orthologs using DIOPT as previously described (Hu et al., 2011; Campbell et al., 2021). GSEA analysis on differentially expressed genes between keratinocyte clusters was performed using fgsea 1.22.0 and the Hallmark pathways set from MSigDB (Subramanian et al., 2005; Liberzon et al., 2015).

Twist1a (ENSDART00000043595.5) and twist1b (ENSDART00000052927.7) were TOPO cloned into the attL1-L2 Gateway pME vector and LR cloned into the pDestTol2pA2 vector (Tol2Kit 394) with p5E-krt4 promoter and p3E-polyA (Tol2Kit 302). To generate the zebrafish melanoma model as previously described, one-cell stage Casper Triple zebrafish embryos (mitfa:BRAFV600E;p53−/−;mitfa−/−;mpv17−/−) were injected with miniCoopR-tdTomato, krt4-eGFP, either krt4-twist1a and krt4-twist1b for Twist overexpression condition or empty vector for control condition, tol2 mRNA, and phenyl red. Injections were performed three times on different days with parents from the same clutch. Embryos were grown at standard conditions and sorted at 5 days post-injection for eGFP and tdTomato expression using the Zeiss AxioZoom V16 fluorescence microscope. eGFP+/tdTomato+ fish in the CTRL (n = 135) and TWIST (n = 118) conditions were maintained to adulthood.

Zebrafish were regularly monitored for melanoma formation and survival every 4 weeks, beginning at 10 weeks post-fertilization. Melanoma formation was screened visually using the Zeiss AxioZoom V16 fluorescence microscope under ×20 magnification. Kaplan–Meier curves and corresponding statistics were generated using GraphPad Prism 9. Statistical differences in survival between conditions were determined by the Mantel–Cox log-rank test.

Zebrafish were euthanized in tricaine (MS222, Syndel). Each fish was dissected in three sections consisting of head, body, and tail. Samples were placed in 4% PFA in PBS (Santa Cruz 281692) for 72 hr on a shaker at 4°C, then paraffin embedded. Histology was performed by HistoWiz Inc (histowiz.com) using a Standard Operating Procedure and fully automated workflow. Samples were processed, embedded in paraffin, and sectioned at 5 μm. Immunohistochemistry was performed on a Bond Rx autostainer (Leica Biosystems) with enzyme treatment (1:1000) using standard protocols. Sections were stained with H&E or IHC with antibodies including BRAFV600E (ab228461) and GFP (ab183734). Bond Polymer Refine Detection (Leica Biosystems) was used according to the manufacturer’s protocol. After staining, sections were dehydrated and film coverslipped using a TissueTek-Prisma and Coverslipper (Sakura). Whole slide scanning (40x) was performed on an Aperio AT2 (Leica Biosystems).

Human melanoma lines A375, HS294T, and SKMEL2 were obtained from ATCC (CRL-1619, HTB-140, HTB-68). Human keratinocyte line HaCaT was obtained from AddexBio. All cells were routinely tested and confirmed to be free from mycoplasma. Cells were maintained in a humidified incubator at 37°C and 5% CO₂. Cells were maintained in DMEM (Gibco 11965) supplemented with 10% FBS (Gemini Bio) and split when confluent, approximately 2–3 times per week.

The HaCaT cell line was labeled with eBFP to allow for identification during co-culture. 293T (ATCC CRL-11268) was transfected with the pLV-Azurite plasmid (Addgene 36086) with pMD2.5 (Addgene 12259) and psPAX2 (Addgene 12260) using Invitrogen Lipofectamine 3000 Transfection Reagent (Invitrogen L3000015) according to the manufacturer’s protocol. HaCaTs were infected with lentivirus containing CMV:eBFP and selected for eBFP positivity using ampicillin and FACS. Subsequently, HaCaT-eBFP was infected with lentivirus containing CMV:TWIST1 (Horizon Precision LentiORF Human TWIST1 OHS5898-202622685) or CMV:empty control created by removing ORF of CMV:TWIST1 plasmid. HaCaT line overexpressing TWIST1 was labeled as HaCaT-TWIST and HaCaT line with empty vector was labeled as HaCaT-CTRL. HaCaT lines were subsequently sorted for nuclear eGFP expression present in the plasmid as part of the Precision LentiORF system. HaCaT-CTRL and HaCaT-TWIST were cultured as previously described.

Cells were washed with DPBS (Gibco 14190250) and lysed in RIPA buffer (Thermo Scientific 89901) with the addition of protease and phosphatase inhibitors (Thermo Scientific 78440). Lysates were centrifuged at 13,000 × g at 4°C and quantified using the Pierce BCA Protein Assay Kit (Pierce 23227). Samples were reduced with the Laemmli SDS-sample buffer (Boston BioProducts BP111R) and boiled for 10 min. For gel electrophoresis, samples were loaded into 4–15% precast protein gels (Bio-Rad 4561084), then transferred to 0.2-µm nitrocellulose membranes (Bio-Rad 1704158). Membranes were washed in TBST and blocked with 5% milk in TBST (Boston BioProducts P1400) for 1 hr at RT. Membranes were washed and incubated with primary antibodies overnight. Antibodies used include Twist (Abcam ab50887) and beta-actin (CST 3700S). On the next day, membranes were washed with TBST and incubated with appropriate secondary antibodies for 1 hr. Blots were incubated with the Immobilon Western Chemiluminescent HRP Substrate (Millipore WBKLS0500) and imaged with the Amersham ImageQuant 800.

Cells were cultured on chamber slides (Thermo Scientific 154739) overnight at standard cell culture conditions. Culture media was washed with DPBS (Gibco 14190250) and fixed with 2% PFA in PBS (Santa Cruz 281692) for 15 min at RT. Cells were washed with DPBS and permeabilized with 0.1% Triton X-100 (Thermo Scientific 85111) in PBS, then blocked with 10% goat serum (Thermo Fisher 50062Z) for 1 hr at RT. Primary antibodies used include Twist (Abcam ab50887). Cells were incubated with primary antibody in 10% goat serum overnight at 4°C and washed with DPBS the next day, before incubation with 1:1000 Hoechst 33342 (Thermo Fisher H3570) for 1 hr and mounted with VECTASHIELD PLUS Antifade Mounting Medium (Vector Laboratories H1900). Slides were imaged on the Zeiss LSM880 inverted confocal microscope and images were processed using FIJI v1.53.

RFP-labeled melanoma cell lines, including A375, SKMEL2, HS294T, were plated on poly-L-lysine coated round glass coverslips (Corning 354085) placed in 24-well plates at 150–200k cells per coverslip. HaCaT cell lines were plated in 6-well plates at 250–300k cells per well. Cells were allowed to attach overnight and the coverslip containing melanoma cells was transferred to 6-wells containing HaCaT cell lines using tweezers. All coverslips were placed in the center of the well. KC–melanoma co-cultures were incubated in standard cell culture conditions for 24 hr. Co-cultures were imaged by fluorescence microscopy at four locations of each coverslip: top, right, bottom, and left, to capture variations in melanoma cell infiltration into KC lines. FIJI v1.53 was used to count the number of infiltrating melanoma cells per image and average infiltrating melanoma cells were calculated per well. All experiments were performed in 3 sets, with 3 replicates per set per condition. Average infiltrating cell numbers per well were normalized to the average infiltrating cell number per well in the HaCaT-CTRL condition. scRNA-sequencing analysis of zebrafish melanoma.

Six zebrafish, three each from CTRL and TWIST conditions at 26 weeks post-injection were selected for scRNA-sequencing of melanoma tumors. To account for set differences, one fish from each of three injection sets was chosen in each condition. Melanoma and adjacent skin were dissected from the fish, then enzymatically and mechanically dissociated into single-cell solutions as described above. The samples were FACS sorted for GFP and RFP positivity, corresponding to eGFP expressed by keratinocytes and tdTomato expressed by melanoma. The sorted cells were placed in (Gibco 11965) supplemented with 10% FBS (Gemini Bio) and 1% penicillin–streptomycin–glutamine. To enrich for KCs, sorted KCs and melanoma cells from each fish were recombined at a 7:3 KC:melanoma ratio. Sorted cells were pelleted and resuspended in DPBS + 0.1% BSA. Samples were also combined based on their genetic perturbation condition. Droplet-based scRNA-seq was performed using the Chromium Single Cell 3′ Library and Gel Bead Kit v3 (10X Genomics) and Chromium Single Cell 3′ Chip G (10X Genomics). 10,000 cells were targeted for encapsulation. GEM generation and library preparation were performed according to kit instructions. Libraries were sequenced on a NovaSeq S4 flow cell. Resulting reads were aligned to the GRCz11 reference genome with the addition of eGFP and tdTomato sequences using CellRanger v5.0.1 (10x Genomics). scRNA-sequencing analysis was performed as detailed above. In addition, melanoma cells were scored using AddModuleScore to assess their enrichment of genes associated with the four main melanoma cell states and intermediate states (Tsoi et al., 2018). The highest scoring gene module for each cell was annotated as its cell state. CellChat (Jin et al., 2021) was used to analyze cell–cell communication between KC and melanoma clusters using its zebrafish L–R database.

The GeoMx DSP spatial transcriptomics data was reanalyzed from Vallius et al., 2026. First, the reanalyzed MRs were selected based on the availability of the keratinocyte compartment within the sequenced MRs. Only the MRs with keratinocytes present within the profiled region (referred to as the ‘segmented area’ per Bruker Spatial Biology (formerly NanoString); examples shown in Figure 6A–D) were reanalyzed and included in this manuscript. The H&E-stained serial sections were used for histopathological annotation of the samples, as described in Vallius et al. The biospecimen metadata and GeoMx MR annotations are shown in Supplementary file 1.

To account for the variable cell-type composition across MRs, we first computed the melanocyte fraction for each MR as the ratio of melanocyte count to total nuclei count. The melanocyte count was determined by the SOX10/MART1 immunofluorescence staining, by visually counting the nuclei staining positively for SOX10 and/or MART1. A non-melanocyte scaling factor was then defined as 1 − melanocyte fraction. With the top 100 upregulated genes in the TAK and TWIST programs, the raw gene-signature scores were calculated using Gene Set Variation Analysis and subsequently normalized by dividing each score by the corresponding non-melanocyte scaling factor, yielding keratinocyte-adjusted signature values. For visualization, keratinocyte-adjusted signature scores were assembled into a matrix (signatures × MRs) and standardized by row (z-score scaling across MRs) to emphasize the relative differences between MRs. Heatmaps were generated using the ComplexHeatmap package in R, with columns representing samples and rows representing gene signatures. Rows were hierarchically clustered, while columns were ordered using optimal leaf ordering based on pairwise sample distances.

Statistical analysis and figures were generated by GraphPad Prism 9, R Studio 4.2.0, and Biorender.com. Image processing was performed in FIJI v1.53. Statistical tests are described in figure legends and methods. Experiments were repeated at least three times unless otherwise noted. All animal and cell experiments were performed with a reasonable number of replicates by power calculations or feasibility of the experimental method.