Keratinocytes in the melanoma microenvironment undergo EMT-like changes.

(A) Generation of transparent zebrafish with GFP-labeling of keratinocytes. Casper Triples (mitfa-/-;mitfa:BRAFV600E;tp53-/-;mpv17-/-) were injected with the Tol2Kit 394 vector containing a krt4:eGFP cassette and tol2 mRNA. Brightfield shows a transparent zebrafish while fluorescence imaging shows eGFP-labeling of keratinocytes. (Scale bar = 5mm). Schematic in panel A created with BioRender.com. (B) Schematic of TEAZ (Transgene Electroporation of Adult Zebrafish). Plasmid mix containing miniCoopR:tdTomato, mitfa:Cas9, zU6:sgptena, zU6:sgptenb, and the tol2 plasmid was injected superficially in the flank of the zebrafish. Electroporation of the injection site results in rescue of melanocyte precursors and the generation of a localized melanoma that could be analyzed by microscopy and FACS. Schematic in panel B created with BioRender.com. (C) Brightfield and immunofluorescence of zebrafish 8-weeks post-TEAZ with localized and fluorescently labeled melanoma. (Scale bar = 5mm) (D) Immunofluorescence imaging of TEAZ region after 8-weeks, comparing empty vector control vs. miniCoopR:tdT conditions, with yellow dotted circles indicating general area of dissection for FACS. (Scale bar = 1mm) (E) Confocal imaging of zebrafish epidermis. Normal epidermis of Tg(krt4:eGFP) Casper Triple post-TEAZ with empty vector control shows eGFP-labeled, polygonal shaped keratinocytes regularly connected while epidermis with melanoma generated with miniCoopR-drived melanocyte rescue shows disrupted epidermis and irregularly shaped keratinocytes. Inset highlights melanoma cells with adjacent keratinocytes. (Scale bar = 50um) (F) qPCR of FACS sorted zebrafish epidermis with or without melanoma. tdTomato-labeled melanoma cells, eGFP-labeled keratinocytes and non-fluorescently labeled TME cells were isolated by dissection (as indicated in G) and FACS. Comparison of mitfa and krt4 expression of samples normalized to non-fluorescent cells, either TME-Other in tumor samples or Other in non-tumor samples, shows enrichment of mitfa in melanoma sample and krt4 in keratinocyte sample. ns = * = p<0.05, *** = p<0.001, **** = p<0.0001 by Tukey’s multiple comparisons test. (G) Comparison of the EMT-markers vim (vimentin) and cdh2 (N-cadherin) shows enrichment in TME keratinocytes vs. keratinocytes from epidermis without melanoma. * = p<0.05 by Welch’s t-test. (H) Schematic of keratinocyte-melanoma co-culture experiment. HaCaTs were cultured in monoculture or co-culture with A375 melanoma cells in triplicates for 21 days, followed by FACS isolation of keratinocytes for RNA-sequencing comparing co-culture vs. monoculture keratinocytes. Schematic in panel H created with BioRender.com. (I) Top 5 enriched Hallmark pathways in HaCaTs co-cultured with A375 melanoma cells compared with HaCaTs in monoculture. (J) Normalized counts of EMT biomarkers vimentin (VIM) and N-cadherin (CDH2).

Zebrafish scRNA-sequencing shows upregulation of EMT-TFs in tumor-associated keratinocytes.

(A) Schematic of scRNA-sequencing experiment. Embryo injection with miniCoopR:eGFP and tol2 mRNA in Casper Triple results in melanocyte rescue and subsequent melanoma formation. Melanoma was dissected and dissociated to single cell suspension for FACS isolation of eGFP+ melanoma cells and non-fluorescent TME cells for single cell RNA-sequencing. Schematic in panel A created with BioRender.com. (B) UMAP highlighting two keratinocyte clusters, Melanoma and other TME cells. (C) Violin plot of keratinocyte module scores comparing between keratinocyte clusters, TAK and NKC, versus melanoma and other TME cells. (D) Top 6 GSEA Hallmark analysis comparing TAK vs. NKC (NES=Normalized Enrichment Score). (E) Hallmark EMT pathway enrichment in TAK vs NKC. (F) Enrichment of EMT-transcription factors in TAK vs. NKC, Melanoma, and Other TME cells.

Overexpression of twist1a/b results in improved survival of fish with melanoma.

(A) Schematic of zebrafish melanoma model with labeling and perturbation of keratinocytes. Twist1a and twist1b are overexpressed under the keratinocyte-specific, krt4, promoter in the TWIST condition and an empty vector control was used in the CTRL condition. Fish were sorted at 5 days for eGFP and tdTomato positivity as marker of successful keratinocyte labeling and melanocyte rescue (n = 135 CTRL, n = 118 TWIST). Schematic in panel A created with BioRender.com. (B) Tumor-free survival and overall survival of A. ns = no significance, ** = p<0.005 by Log-rank (Mantel-Cox) test. (C) Sample images of zebrafish with melanoma at 26 weeks post-injection. Keratinocytes are labeled by GFP, melanoma are labeled by tdTomato. Melanomas are pigmented. Scale bar = 5mm. (D) H&E and IHC of cross-sections through zebrafish body and melanoma. Dotted yellow line demarcates border of body. % of tumor in body is calculated as tumor area within body border divided by total tumor area. Scale bar = 500um.

Zebrafish findings are recapitulated in human cell lines.

(A) Schematic of coverslip cell infiltration assay. Melanoma cells are plated on a coverslip and allowed to attach overnight. The coverslip is then transferred into a well of keratinocytes to assess melanoma infiltration into keratinocytes. Schematic in panel A created with BioRender.com. (B) Generation of a HaCaT cell line overexpressing TWIST1. HaCaT-BFP was infected with lentivirus containing cassette with nuclear localized GFP and CMV driven TWIST1 or no ORF. Infected cell lines were allowed to grow for a week before sorting for nuclear GFP as a marker of successful integration. Schematic in panel B created with BioRender.com. (C) Western blot for Twist expression in HaCaT-CTRL and HaCaT-TWIST. (D) Immunofluorescence imaging for Twist localization in HaCaT-CTRL and HaCaT-TWIST. TWIST staining is pseudo-colored in magenta, DAPI in blue, phalloidin in white. Scale bar = 50um. (E) Immunofluorescence imaging of coverslip cell infiltration assay with HS294T-tdT (orange) melanoma cells in co-culture with either HaCaT-CTRL or HaCaT-TWIST (cyan). Insets highlight areas of melanoma cell infiltration into keratinocyte monolayers. Scale bar = 500 μm. (F) Quantification of E. Infiltrating HS294T melanoma cells from each image were counted and averaged across four images per well. Resulting cell counts were normalized to average cell counts of HaCaT-CTRL from each set. N = 9, 3 sets, 3 replicates/wells per set. * = p<0.05 by t-test. (G) Immunofluorescence imaging of coverslip cell infiltration assay with SKMEL2-tdT (orange) melanoma cells in co-culture with either HaCaT-CTRL or HaCaT-TWIST (cyan). Insets highlight areas of melanoma cell infiltration into keratinocyte monolayers. Scale bar = 500 μm. (H) Quantification of G. Infiltrating SKMEL2 melanoma cells from each image were counted and averaged across four images per well. Resulting cell counts were normalized to average cell counts of HaCaT-CTRL from each set. N = 9, 3 sets, 3 replicates/wells per set. * = p<0.05 by t-test.

scRNA-sequencing shows unique keratinocyte-melanoma communication with twist1a/b overexpression in keratinocytes.

(A) Schematic of scRNA-sequencing protocol. Melanoma and surrounding tissue were dissected from 26-weeks old zebrafish from either CTRL or TWIST conditions as shown in Figure 3. Samples were dissociated to single cell suspensions for FACS isolation of keratinocytes (GFP) and melanoma (tdTomato). Keratinocytes and melanoma were recombined per condition at a ratio of 7:3 for enrichment of keratinocytes for scRNA-sequencing. Schematic in panel A created with BioRender.com. (B) UMAP dimensional reduction and feature plots of scRNA-sequencing dataset. CTRL and TWIST samples were sequenced, and cell types were identified using eGFP+ for keratinocytes and tdTomato+ for melanoma, after which the datasets were integrated. (C) UMAP of keratinocyte clusters from CTRL and TWIST conditions combined, showing Normal Keratinocyte Cluster (NKC), Tumor-Associated Keratinocyte (TAK), and Twist-High cluster (TWIST-specific). CTRL: 32.9% TAK. TWIST: 14.3% TAK, 56.3% Twist-High. (D) UMAP highlighting keratinocyte clusters, with a Tumor-Associated Keratinocyte (TAK) cluster, a Normal Keratinocyte Cluster (NKC), and a Twist-High cluster unique to the TWIST condition. (E) Melanoma cell state analysis of the melanoma cluster unique to the Twist condition indicated by arrow in Figure 5C. (F) Schematic overview of CellChat analysis. In CTRL condition, we analyzed Ligand-Receptor pairs with NKC as sender and CTRL-Melanoma as receiver. In TWIST condition, we analyzed L-R pairs with both NKC and Twist-High as sender and TWIST-Melanoma as receiver. Schematic in panel F created with BioRender.com (G) CellChat analysis results. L-R pairs shown at p<0.01, with color scale indicating communication probability of L-R pair.

Enrichment of keratinocyte signatures in microregions with melanocytic atypia.

(A-D) Visualization of a subset of reanalyzed GeoMx microregions (MRs); GeoMx data (n=8 MRs) is reanalyzed from Vallius, Shi, Novikov, et al. biorxv 2025. Immunofluorescence staining is shown for morphology markers DNA (grey), panCK (green) and SOX10/MART1 (red; middle and bottom rows). The selected MRs are illustrated with white polygons (middle), and the shaded area within the MRs represent the area for oligonucleotide barcode collection (referred as a “segmentation mask” per NanoString; bottom). Serial sections were H&E-stained and the regions mapping to the sequenced MRs are shown (top row). The MRs were annotated as epidermis containing morphologically normal melanocytes (panel A), and regions of melanocytic atypia (precursor; panels B-D). Scale bars, 50 μm. (E–G) Melanocyte-adjusted TAK and TWIST gene-signature activity across reanalyzed MRs (n=8 GeoMx MRs). (E) Heatmap showing gene-signature scores derived from the top 100 upregulated genes defining the TAK and TWIST programs across MRs. Each column represents an MR, and the MRs are annotated for sample ID and histopathological stage, with melanocyte fraction (melanocyte count divided by total nuclei count within the MR) per GeoMx MR displayed as a top annotation.

Model of melanoma restraint by keratinocytes

Schematic model of melanoma-keratinocyte interactions in the skin. During early melanoma development, melanocytes reside within an epidermis composed of normal keratinocytes. In the context of low TWIST expression (TWISTlo keratinocytes), melanoma cells adopt an invasive, undifferentiated phenotype and are able to breach the epidermal compartment. In the presence of EMT/TWISThi keratinocytes, melanoma cells are restrained within the epidermis and adopt a more differentiated, less invasive cell state. These findings suggest that EMT-like changes in tumor-associated keratinocytes, rather than promoting tumor spread, act as a cell-extrinsic barrier to melanoma invasion through altered cell-cell interactions in the local skin microenvironment. Created with BioRender.com.

Melanoma in TWIST conditions were non-invasive independent of anatomical location

(A) Brightfield and fluorescence images of two sample fish from the TWIST condition. Melanoma are pigmented in brightfield images. Keratinocytes are labeled by eGFP and melanoma are labeled by tdTomato in fluorescence images. Scale bar = 5mm. (B) IHC of cross-sections through the zebrafish body and melanoma. Red staining indicates positivity for hBRAFV600E. Scale bar = 1mm.

Melanoma migration from coverslip

(A) Immunofluorescence imaging of coverslip cell infiltration assay with HS294T-tdT (orange) melanoma cells in co-culture with parental HS294T (uncolored). Scale bar = 500um. (B) Quantification of the migration of HS294T-tdTomato melanoma cells in co-culture with parental HS294T melanoma cells. Each data point is the sum of 4 images per well. Mean ± SE, n = 9 wells from 3 independent experiments. (C) Immunofluorescence imaging of coverslip cell infiltration assay with SKMEL2-tdT (orange) melanoma cells in co-culture with parental SKMEL2(uncolored). Scale bar = 500um. (D) Quantification of the migration of SKMEL2-tdTomato melanoma cells in co-culture with parental SKMEL2 melanoma cells. Each data point is the sum of 4 images per well. Mean ± SE, n = 8 wells from 3 independent experiments.

scRNA-sequencing of zebrafish melanoma

(A) Violin plots of scRNA-sequencing data comparing CTRL and TWIST conditions after quality control. Samples were filtered by number of features between 500 and 5000, with less than 10% mitochondria genes per cell. (B) Top 6 GSEA Hallmark pathways of differentially expressed genes between TAK and NKC clusters. (C) Twist1a, twist1b, and eGFP expression in keratinocyte clusters. (D) UMAP of melanoma samples from either the CTRL or TWIST conditions, identified by seurat clusters. 7 distinct clusters were identified in the CTRL melanoma sample vs. 8 distinct clusters in the TWIST melanoma sample. (E-F) GSEA analysis showing pathways upregulated in TWIST vs. CTRL melanoma: (E) GO:Biological Process pathways and (F) Hallmark pathways.