Mechanical confinement governs phenotypic plasticity in melanoma

Abstract

Phenotype switching is a form of cellular plasticity in which cancer cells reversibly move between two opposite extremes: proliferative versus invasive states^1,2. Although it has long been hypothesized that such switching is triggered by external cues, the identity of these cues remains unclear. Here we demonstrate that mechanical confinement mediates phenotype switching through chromatin remodelling. Using a zebrafish model of melanoma coupled with human samples, we profiled tumour cells at the interface between the tumour and surrounding microenvironment. Morphological analysis of interface cells showed elliptical nuclei, suggestive of mechanical confinement by the adjacent tissue. Spatial and single-cell transcriptomics demonstrated that interface cells adopted a gene program of neuronal invasion, including the acquisition of an acetylated tubulin cage that protects the nucleus during migration. We identified the DNA-bending protein HMGB2 as a confinement-induced mediator of the neuronal state. HMGB2 is upregulated in confined cells, and quantitative modelling revealed that confinement prolongs the contact time between HMGB2 and chromatin, leading to changes in chromatin configuration that favour the neuronal phenotype. Genetic disruption of HMGB2 showed that it regulates the trade-off between proliferative and invasive states, in which confined HMGB2^high tumour cells are less proliferative but more drug-resistant. Our results implicate the mechanical microenvironment as a mechanism that drives phenotype switching in melanoma.

Fig. 1. Confinement induces an undifferentiated neuronal gene program.

a, Schematic detailing the workflow of spatial transcriptomics and scRNA-seq experiments performed on zebrafish melanomas. b, Uniform manifold approximation and projection (UMAP) of human melanoma scRNA-seq dataset from Jerby-Arnon et al.. Cluster annotations from the original paper are labelled. Tumour cell clusters are outlined. c, Gene module scoring for interface genes extracted from zebrafish spatial transcriptomics and scRNA-seq data, projected onto tumour cells outlined in b. The red arrow denotes the subpopulation with the highest expression of interface genes. d, Cell state classification for melanoma differentiation states identified by Tsoi et al.. Cells were classified on the basis of the highest expression of the gene modules indicated. e, Module scores for melanoma cell state genes from Tsoi et al. in interface cells. f, Normalized expression per cell in UMAP space for the indicated genes. The red arrow indicates the interface cluster identified in b. g, Top 20 most highly upregulated genes in the human interface cluster. Neuronal genes are labelled in purple. h,i, GSEA barcode plot for the Gene Ontology (GO) pathways ‘cell fate specification’ (h) and ‘regulation of neuron differentiation’ (i). Normalized enrichment score (NES) and false discovery rate (FDR) are labelled. j,k, Immunofluorescence of adult zebrafish tissue sections highlighting the centre of the tumour (j) and tumour–TME interface (k). Individual nuclei are pseudocoloured and displayed without image overlay at right. l, Schematic of in vitro confinement workflow using a polydimethylsiloxane (PDMS) piston. m, Principal component analysis plot for each RNA sequencing (RNA-seq) replicate. Percentage variance for each principal component (PC) is labelled. n = 3 biological replicates for each condition. n, Top 10 most highly upregulated pathways from GSEA of confined cells relative to unconfined cells. NES and FDR are indicated. Scale bars, 100 μm (a), 10 µm (j,k).

Fig. 2. Perinuclear acetylated tubulin cage assembles in response to confinement.

a, Representative stills from confocal imaging of A375 cells stained with SiR-tubulin. MTOC, microtubule-organizing centre. b, Line intensity profile of perinuclear tubulin intensity over time from the images shown in a. MTOC is highlighted (a,b). a.u., arbitrary units. c, Schematic detailing immunofluorescence staining of sections from adult zebrafish melanomas. d,e, Immunofluorescence images of acetylated tubulin staining at the invasive front (d) compared with the centre of the tumour (e). f, A375 cells stained with antibodies labelling acetylated tubulin (green) and total tubulin (purple). g, Inset of regions labelled in f. h, Quantification of whole-cell acetylated tubulin intensity. Each point represents one cell. Unconfined, n = 27 cells from three images; confined, n = 80 cells from nine images. Horizontal lines, mean; box, s.e.m.; vertical lines, s.d. P value is indicated (two-sample t-test; two-sided). i,j, A375 cells treated with 1-µm nocodazole (NZ) for approximately 18 h and stained for acetylated tubulin (yellow, top and bottom), total tubulin (purple) and Hoechst (blue) at ×63 (i) and ×20 (j) magnification. Scale bars, 50 µm (d,e,j), 20 µm (a,f), 10 µm (g,i). Illustrations in c were created using BioRender (https://biorender.com). Source Data

Fig. 3. HMGB2 is a confinement-induced marker of invasion.

a, HMG family expression in interface cells from zebrafish melanoma scRNA-seq. b, Normalized hmgb2a/hmgb2b expression. P values are noted (Wilcoxon rank-sum test; two-sided). c, HMGB2 expression per cell in human melanoma scRNA-seq data from Jerby-Arnon et al.. The arrow indicates interface cluster. d, Mean HMGB2 expression per cluster. P value calculated using Wilcoxon rank-sum test with Bonferroni’s correction; two-sided. e, Zebrafish melanoma stained for HMGB2 and Hoechst. f, Inset of region indicated in d. Elongated HMGB2-high cells are labelled. g, Correlation between nuclear circularity and HMGB2 intensity. Red dashed line, line of best fit by linear regression. h, Immunofluorescence targeting HMGB2 in confined A375 cells. i–k, HMGB2 intensity (i), Hoechst intensity (j) and HMGB2 intensity normalized to Hoechst (k) per cell. Unconfined, n = 49 cells from three images; confined, n = 97 cells from nine images. Horizontal lines, mean; box, s.e.m.; vertical lines, s.d. P value is indicated (two-sample t-test; two-sided). l, TurboID workflow. m, Nesprin 2 protein abundance; n = 3 replicates per condition. Horizontal line, median; hinges, first and third quartiles; whiskers, range. NLS, nuclear localization signal. n, HMGB2 expression in confined A375 cells. siNT, non-targeting siRNA; siSYNE2, SYNE2-targeting siRNA. o, Quantification of HMGB2 intensity. siNT unconfined, n = 72 cells from eight images; siNT confined, n = 94 cells from eight images; siSYNE2 unconfined, n = 48 cells from eight images; siSYNE2 confined, n = 64 cells from eight images. P value is indicated (analysis of variance with Tukey post hoc test; two-sided). p, Images showing acetylated tubulin (yellow) and Hoechst (magenta) in confined A375 cells. q, Quantification of acetylated tubulin intensity in confined cells. siNT, n = 66 cells from eight images; siSYNE2, n = 104 cells from eight images. P value is indicated (two-sample t-test; two-sided). Horizontal line, median; edges, upper and lower quartiles; whiskers, non-outlier minima and maxima (o,q). Scale bars, 50 µm (e), 10 µm (f), 25 µm (h,n,p). Illustrations in l were created using BioRender (https://biorender.com). Source Data

Fig. 4. Confinement-mediated stabilization of HMGB2 increases chromatin accessibility at neuronal loci.

a, Representative stills from time-lapse imaging of A375 cells expressing HMGB2–GFP and subjected to FRAP. The yellow dashed region indicates the photobleached area. Time is relative to photobleaching. Scale bars, 10 µm. b,c, FRAP recovery curves for HMGB2–GFP in unconfined (b) and confined (c) cells. Each curve represents fluorescence recovery within the area photobleached on a single cell. d,e, Representative plots showing a two-component exponential equation fit to HMGB2–GFP fluorescence recovery curves in unconfined (d) and confined (e) cells. f, Relative proportion of slow-diffusing HMGB2–GFP. Horizontal lines, mean; box, s.e.m.; vertical lines, s.d. P value is indicated (two-sample t-test; two-sided). Unconfined, n = 45 cells; confined, n = 54 cells (b–f). g, Volcano plot of differentially expressed peaks upon HMGB2^OE. P-value cut-off, 0.05; fold change cut-off, log₂(1.25) and log₂(−1.25). h, Top 10 enriched GO: Biological Process (BP) pathways from genes mapped to open chromatin loci upon HMGB2^OE. P values are indicated (two-sided hypergeometric test with Benjamini–Hochberg correction) (g,h). i,j, Tornado plots showing chromatin accessibility at loci linked to genes from the ‘Lee neural crest stem cell up’ (i) and ‘regulation of nervous system development’ (j) pathways. k, Homer de novo motif analysis of transcription factor motifs in open chromatin regions in promoter regions of genes upon HMGB2^OE. P values are indicated (binomial test). Source Data

Fig. 5. Confinement promotes drug tolerance by downregulating proliferation.

a, Schematic detailing the influence of confinement on melanoma phenotype switching. b, Enrichment scores for GO:BP pathways related to cell division from RNA-seq of confined cells relative to unconfined cells. FDR is indicated. c, Stills from confocal imaging of A375 cells stably expressing the FastFUCCI reporter and stained with SiR-tubulin. d, Quantification of FUCCI signal over time in confined cells; n = 37 cells from four videos. Error bars, s.e.m. e, Schematic of in vitro invasion assay workflow. f,h, Western blot for HMGB2 (top) and tubulin (f; loading control; bottom) or β-actin (h; loading control; bottom) for A375 cells transfected with the indicated siRNAs (f) or plasmids (h). siHMGB2, HMGB2-targeting siRNA. g,i, Quantification of in vitro invasion assay results for A375 transfected with the indicated siRNAs (g) or plasmids (i); n = 3 biological replicates. P value is indicated (two-sample t-test; two-sided). EV, empty vector; OE, overexpression (of HMGB2). j, Representative images of adult zebrafish with melanomas generated by means of Transgene Electroporation in Adult Zebrafish (TEAZ), 12 weeks after electroporation. NT, non-targeting; sgRNA, single-guide RNA. k, Tumour surface area over time for melanomas induced in zebrafish using TEAZ. Error bars, s.e.m. l, Tumour surface area at 12 weeks after TEAZ. P value is indicated (two-sample t-test; two-sided). Horizontal line, median; box edges, 25th and 75th percentiles; whiskers, data range. NT sgRNA, n = 10 fish; sgRNA targeting HMGB2 (sgHMGB2), n = 8 fish (j–l). m, Representative haematoxylin and eosin (H&E) images of tumour invasion in the indicated conditions at 12 weeks after TEAZ. n, Schematic of the drug treatment experiment workflow. o, Tumour volume over time for the indicated conditions; n = 11 mice per condition from two biological replicates. Error bars, s.e.m. *P < 0.05. P value calculated using likelihood ratio tests with a fitted exponential; two-sided (P = 0.043). Scale bars, 25 µm (c), 2 mm (j), 250 µm (m). Illustrations in e were created using BioRender (https://biorender.com). Mouse cartoons in n were adapted from Wikimedia, under a Creative Commons licence CC BY-SA 3.0. Source Data

Extended Data Fig. 1. Interface cells are found in human patient samples.

a. tSNE of human melanoma scRNA-seq dataset from Jerby-Arnon et al. Cluster annotations from the original manuscript are labeled. Tumor cell clusters are outlined. b. Gene module scoring for interface genes extracted from zebrafish spatial transcriptomics and scRNA-seq data, projected onto tumor cells outlined in a. Red arrow denotes the subpopulation with highest expression of interface genes. c. Cell state classification for melanoma differentiation states identified by Tsoi et al. Cells were classified based on highest expression of the gene modules indicated. d-f. Tumor and interface cells classified by treatment status. g-h. Normalized expression per cell for the indicated genes. Red arrow indicates the interface cluster identified in b.

Extended Data Fig. 2. Confinement does not cause apoptosis and induces a neuronal gene program.

a,c,e. IF of confined A375 cells labelled with antibodies against the apoptosis markers Annexin V (a), cleaved caspase-3 (c), and cleaved PARP (e). b,d,f. Intensity per cell for the indicated markers. b. Unconfined: n = 250 cells from 6 images. Confined: n = 225 cells from 6 images. d. Unconfined: n = 261 cells from 6 images. Confined: n = 185 cells from 6 images. f. Unconfined: n = 245 cells from 6 images. Confined: n = 174 cells from 6 images. g. IF of A375 cells confined for ~18 h and then left to recover for 24 h (right) or unconfined cells (right). a,c,e,g. Scale bars, 25 µm. h. Euler diagram depicting co-expression of differentially expressed (DE) genes with an adjusted P-value < 0.05. i. Results from pathway analysis on the co-upregulated genes between both datasets using the GO:BP pathway set. Pathway analysis performed with Enrichr. Neuronal pathways are highlighted. Source Data

Extended Data Fig. 3. ATAT1 inactivation abolishes acetylated tubulin.

a. Schematic detailing generation of ATAT1 knockdown cell lines. b. Expression of ATAT1 mRNA in the noted cell lines. Expression is normalized to β-actin and the non-targeting control condition. For each cell line, n = 3 biological replicates for a total of n = 12 technical replicates. c-d. IF images showing expression of acetylated tubulin (yellow, top; bottom), HMGB2 (cyan, top) and actin (magenta, top) in unconfined (c) and confined (d) cells. Scale bars, 20 µm. e-f. Quantification of acetylated tubulin (e) and HMGB2 (f) intensity in the noted conditions. sgNT unconfined: n = 55 cells. sgNT confined: n = 60 cells. sgATAT1_1 unconfined: n = 75 cells. sgATAT1_1 confined: n = 73 cells. sgATAT1_2 unconfined: n = 41 cells. sgATAT1_2 confined: n = 94 cells. n = 8 images were used for quantification for each condition. ***, P < 0.001; *, P < 0.05; n.s., P > 0.05. Illustrations in a were created using BioRender (https://biorender.com). Source Data

Extended Data Fig. 4. Localization of tubulin post-translational modifications in confined cells.

a-h. IF of A375 cells labeled with antibodies targeting acetylated tubulin, in addition to tyrosinated tubulin (a-b), detyrosinated tubulin (c-d), polyglutamylated tubulin (e-f), and polyglycylated tubulin (g-h). a,c,e,g. Scale bars, 25 µm. b,d,f,h. Scale bars, 10 µm.

Extended Data Fig. 5. Confinement specifically upregulates HMGB2.

a. Stills from confocal imaging of A375 cells expressing HMGB2-GFP. Cells are pseudocolored by HMGB2 intensity. b. HMGB2-GFP intensity per cell over time, normalized to intensity at the first time point acquired. n = 37 cells from 6 movies. c,e. Immunofluorescence images of A375 cells stained with antibodies targeting HMGB1 (c) and HMGA1 (e). Scale bars, 25 µm. d,f. Quantification of intensity in confined/unconfined cells for the indicated markers. Each point represents 1 cell. Horizontal lines, mean; box, SEM; vertical lines, SD. d. Unconfined: n = 105 cells from 9 images. Confined: n = 76 cells from 9 images. f. Unconfined: n = 125 cells from 9 images. Confined: n = 80 cells from 9 images. Source Data

Extended Data Fig. 6. Interface cells are present in human samples and other cancers.

a. Immunofluorescence performed on a human melanoma tissue sample showing enrichment of HMGB2 and acetylated in elongated nuclei at the tumor border. b. Visualization of nuclear shape relative to distance from the tumor border. c. Inset of a. d. Schematic of a human melanoma tissue microarray colored by the presence of HMGB2+ and/or acetylated tubulin+ cells in each sample. e. Quantification of d. f,i,l,o. Immunofluorescence images of confined MIA-PaCa-1 (f), Panc-1 (i), HTB-4 (l), and HTB-9 (o) cells showing HMGB2 (red/orange) and acetylated tubulin (white). Scale bars, 25 µm. g-h,j-k,m-n,p-q. Quantification of HMGB2 (g,j,m,p) and acetylated tubulin (h,k,n,q) intensity. U = unconfined, C = confined. P-values are indicated. For quantification of nuclear HMGB2 intensity: MIA-PaCa-2 unconfined: n = 74 cells from 8 images. MIA-PaCa-2 confined: n = 128 cells from 8 images. Panc-1 unconfined: n = 159 cells from 14 images. Panc-1 confined: n = 119 cells from 9 images. HTB-4 unconfined: n = 63 cells from 8 images. HTB-4 confined: n = 78 cells from 9 images. HTB-9 unconfined: n = 121 cells from 8 images. HTB-9 confined: n = 103 cells from 8 images. For quantification of whole-cell acetylated tubulin intensity: MIA-PaCa-2 unconfined: n = 41 cells from 8 images. MIA-PaCa-2 confined: n = 84 cells from 8 images. Panc-1 unconfined: n = 82 cells from 14 images. Panc-1 confined: n = 77 cells from 9 images. HTB-4 unconfined: n = 26 cells from 8 images. HTB-4 confined: n = 33 cells from 9 images. HTB-9 unconfined: n = 75 cells from 8 images. HTB-9 confined: n = 88 cells from 8 images. Source Data

Extended Data Fig. 7. Expression of commonly mechanosensitive transcription factors in confined melanoma cells.

a-b, d-e, g-h, j-k. IF images of A375 cells stained with antibodies against YAP (a-b), Twist (d-e), Snail (g-h) or SMAD3 (j-k). Scale bars, 20 µm. c,f,i,l. Quantification of intensity in confined/unconfined cells for the indicated markers. Each point represents one cell. Horizontal line, median; box edges, lower and upper quartiles; whiskers, upper and lower limits of data without outliers. U = unconfined; C = confined. c. Unconfined: n = 27 cells from 6 images. Confined: n = 111 cells from 6 images. f. Unconfined: n = 33 cells from 6 images. Confined: n = 52 cells from 6 images. i. Unconfined: n = 56 cells from 6 images. Confined: n = 54 cells from 6 images. l. Unconfined: n = 19 cells from 4 images. Confined: n = 25 cells from 6 images. Source Data

Extended Data Fig. 8. Loss of HMGB2 does not affect perinuclear acetylated tubulin assembly.

a. Schematic showing generation of HMGB2^KO A375 cell lines using the LentiCRISPR approach. b. Western blot showing expression of HMGB2 in A375 cells stably expressing each of the 6 HMGB2 gRNAs. Red boxes indicate KO cell lines selected for further analyses. c. Quantification of HMGB2 expression in b relative to tubulin (loading control). d. Immunofluorescence of acetylated tubulin (white) and HMGB2 (red/orange/yellow) intensity in cells expressing the indicated sgRNAs. Scale bars, 25 µm. e. Quantification of acetylated tubulin intensity. sgNT unconfined: n = 58 cells from 8 images. sgNT confined: n = 69 cells from 8 images. sgHMGB2_1 unconfined: n = 71 cells from 8 images. sgHMGB2_1 confined: n = 104 cells from 8 images. sgHMGB2_2 unconfined: n = 82 cells from 8 images. sgHMGB2_2 confined: n = 77 cells from 8 images. Source Data

Extended Data Fig. 9. Tubulin stabilization upregulates nuclear HMGB2.

a-d. A375 cells treated with the indicated tubacin concentrations (b-d) or DMSO as a vehicle control (a) and stained with antibodies labelling acetylated histone H3 and acetylated tubulin. Scale bars, 25 µm. e-f. Quantification of whole-cell acetylated tubulin intensity (e) and nuclear histone H3 acetylation (g). ***, P < 0.001. n.s., not significant. g. HMGB2-GFP accumulation in A375 cells treated with DMSO, tubacin or Taxol. Scale bars, 25 µm. Time after applying confinement is indicated. h. HMGB2-GFP intensity in confined cells over time. i. HMGB2-GFP intensity per cell at the final time point imaged (~16 h). h-i. DMSO: n = 38 cells from 7 movies. Tubacin: n = 31 cells from 7 movies. Taxol: n = 10 cells from 3 movies. j. HMGB2-GFP and SiR-tubulin intensity over time for confined cells treated with DMSO or 1 µM nocodazole. Scale bars, 25 µm. k. HMGB2-GFP intensity per cell at the final time point imaged (~ 16 h). NZ = nocodazole. Horizontal lines, mean; box, SEM; vertical lines, SD. l. HMGB2-GFP intensity over time. Error bars, SEM. k-l. DMSO: n = 19 cells from 3 movies. Nocodazole: n = 30 cells from 3 movies. Source Data

Extended Data Fig. 10. The LINC complex is required for HMGB2 enrichment upon confinement.

a. Immunofluorescence images showing expression of nesprin-2 in A375 cells. Scale bars, 25 µm. b. Quantification of nesprin-2 intensity in A375 cells. Unconfined: n = 93 cells from 8 images. Confined: n = 85 cells from 8 images. c. Expression of SYNE2 mRNA in the noted conditions. Expression is normalized to β-actin and the non-targeting control condition. For each cell line, n = 3 biological replicates for a total of n = 12 technical replicates. d. Immunofluorescence images showing expression of lamin A/C (blue/green) and phalloidin in A375 cells. Scale bars, 25 µm. e. Quantification of lamin A/C intensity. Unconfined: n = 180 cells from 8 images. Confined: n = 135 cells from 8 images. f. Representative AFM force maps of the nuclear region of A375 cells. Scale bars, 10 µm. g. Quantification of nuclear stiffness. Unconfined: n = 70 cells from 3 biological replicates. Confined: n = 71 cells from 3 biological replicates. Source Data

Extended Data Fig. 11. ChIP-sequencing and RNA-sequencing identify HMGB2 targets in A375 cells.

a. Schematic detailing ChIP-seq experiment workflow. b. Composite plots showing read density at target gene TSS in HMGB2WT and KO cells. c-e. Integrated genome browser tracks representing HMGB2 binding and H3K4me3 signal near the TSS of FOSL1 (c), NOTCH2 (d), and KMT2A (e). One representative replicate per condition is shown (of two replicates per condition performed). A full list of HMGB2 targets can be found in Supplementary Table 6. f. Western blot for HMGB2 (top) and tubulin (loading control, bottom) in stable A375 lines infected with lentivirus encoding HMGB2 or an empty vector control. g. Representative images of cells from cell lines indicated in f. h. Volcano plot of differentially expressed genes upon HMGB2^OE. i. Heatmap of Z-scored expression of selected neuronal and invasive genes across replicates. j. Double waterfall plot of top GO biological processes pathways by normalized enrichment score (NES). Neuronal pathways are labeled in purple. k. Top 10 GO biological processes pathways by NES upregulated upon HMGB2^OE. Schematic in a was created using BioRender (https://biorender.com).

Extended Data Fig. 12. Characterization of HMGB2 targets in SKMEL5 cells.

a. Western blot showing expression of V5-tagged HMGB2 in SKMEL5 cells, and ChIP-seq experimental workflow. b. Composite plots of read density at target gene TSS in HMGB2EV and OE cells. c-d. Integrated genome browser tracks representing HMGB2-V5 (purple) and HMGB2 (blue) binding and H3K4me3 (green) and input (gray) signal around the TSS of MITF (c), and NOTCH2 (d). One representative replicate per condition is shown (of two replicates per condition performed). A full list of HMGB2 targets in SKMEL5 cells can be found in Supplementary Table S8. e. Western blot showing expression of HMGB2 in SKMEL5^KO cells and RNA-seq experiment workflow. f. Principal component analysis of SKMEL5 RNA-seq samples. g. Expression of selected genes in SKMEL5-HMGB2^KO cells relative to SKMEL5-NT cells. h. HOMER de novo motif analysis showing enrichment of a BRN2/POU3F2 motif in the promoter region of HMGB2 targets identified from ChIP-seq.

Extended Data Fig. 13. Confined cells do not display fast amoeboid migration and are drug tolerant.

a. Stills from time-lapse imaging of confined A375 cells expressing HMGB2-GFP. Nuclei are pseudocolored. Scale bars, 20 µm. b. X-Y velocity over time per cell. c. Histogram showing mean velocity per cell. Dotted line indicates mean velocity across all cells. b-c. n = 38 cells from 7 movies. d. Hoechst (DNA) staining of cells treated with Taxol and confined (right) or unconfined control (left). Scale bars, 50 µm. e. Spaghetti plot showing tumor volume over time in A375-HMGB2EV (blue) and A375-HMGB2OE (red) mouse xenografts treated with dabrafenib/trametinib. Raw data can be found in Supplementary Table S10. Source Data

Extended Data Fig. 14. Structure-function analysis of HMGB2.

a. HMGB2 protein sequence (Uniprot). Functional domains are labeled. b. Schematics illustrating each deletion construct. c. Immunofluorescence images showing expression of GFP-tagged constructs outlined in a. Scale bars, 25 µm. d-e. Quantification of in vitro proliferation (d) and invasion (e) assay results for A375 cells expressing the indicated constructs (n = 3 biological replicates). f. Quantification of nuclear HMGB2-GFP intensity for the noted constructs. FL: n = 94 cells from 3 images. ΔAT: n = 105 cells from 3 images. ΔA-box: n = 111 cells from 3 images. ΔB-box: n = 82 cells from 3 images. g. Schematic detailing in vivo tumor growth assay. h-i. Tumor volume over time (h) and at endpoint (i; 22–24 days post transplant). n = 10 mice per condition from 2 biological replicates. j. Immunohistochemistry targeting acetylated tubulin in cells at the tumor-microenvironment interface. Mouse cartoons in g adapted from Wikimedia, under a Creative Commons licence CC BY-SA 3.0. Source Data

Extended Data Fig. 15. Confinement governs phenotypic plasticity in melanoma.

Model for the role of confinement in melanoma phenotype switching: nuclear compression induces HMGB2 contact with chromatin, increasing chromatin accessibility and gene expression at neuronal loci.