MITF and c-Jun antagonism interconnects melanoma dedifferentiation with pro-inflammatory cytokine responsiveness and myeloid cell recruitment

Abstract

Inflammation promotes phenotypic plasticity in melanoma, a source of non-genetic heterogeneity, but the molecular framework is poorly understood. Here we use functional genomic approaches and identify a reciprocal antagonism between the melanocyte lineage transcription factor MITF and c-Jun, which interconnects inflammation-induced dedifferentiation with pro-inflammatory cytokine responsiveness of melanoma cells favouring myeloid cell recruitment. We show that pro-inflammatory cytokines such as TNF-α instigate gradual suppression of MITF expression through c-Jun. MITF itself binds to the c-Jun regulatory genomic region and its reduction increases c-Jun expression that in turn amplifies TNF-stimulated cytokine expression with further MITF suppression. This feed-forward mechanism turns poor peak-like transcriptional responses to TNF-α into progressive and persistent cytokine and chemokine induction. Consistently, inflammatory MITF(low)/c-Jun(high) syngeneic mouse melanomas recruit myeloid immune cells into the tumour microenvironment as recapitulated by their human counterparts. Our study suggests myeloid cell-directed therapies may be useful for MITF(low)/c-Jun(high) melanomas to counteract their growth-promoting and immunosuppressive functions.

Figure 1. Dedifferentiated melanoma cells have a higher inflammatory responsiveness and pathway activity.

(a) Outline of bioinformatic analysis. (b) Hierarchical clustering and heatmap analysis of melanocytic gene expression to define differentiation status of the melanoma cell lines. Gene expression data were log₂ transformed and mean centred. (c) Expression of the TNF response gene set in TNF-treated (72 h) or -untreated cells grouped by their differentiation status (MITF^high, differentiated; MITF^low, dedifferentiated). Significance was determined by analysis of variance and Tukey's honest significant difference test correction for multiple comparisons. **P<0.01. (d) Analysis of relative mRNA expression levels on a logarithmic scale by qRT–PCR normalized to ubiquitin (UBC). Cells were treated with different concentrations of TNF for 24 h or left untreated. Error bars indicate s.d. from biological triplicates. (e) Correlation of the expression of individual genes with the expression of the TNF response signature in the BROAD melanoma cell line panel (n=88). Gene probes were ordered by an increasing Pearson's correlation value. Top negatively or positively correlated genes are highlighted. (f) Hierarchical clustering and heatmap analysis of the BROAD melanoma cell line panel using the TNF response gene set and a manually curated pigmentation gene set (abbreviated as Pigm.). Gene expression data were log₂ transformed and mean centred.

Figure 2. MITF suppresses global inflammatory responsiveness.

(a) Outline of experimental setup. (b) Validation of MITF siRNA knockdown efficiencies by qRT–PCR. siNT represents a pool of non-targeting siRNA controls. Error bars indicate s.d. from technical replicates of a representative experiment (repeated at least three times). (c) Quantification of the number of gene probes that were deregulated by at least twofold in response to TNF (24 h) in siMITF- or siNT-transfected MZ7 melanoma cells using gene expression microarrays. (d) Gene-probe clustering and heatmap analysis of TNF-induced genes (>twofold). The two main clusters are indicated at the right side. Gene expression data were log₂ transformed and mean centred. (e) Quantification and dissection of the TNF response patterns of the two clusters shown in d. Each box represents the distribution of the normalized expression of all genes within the respective cluster. The indicated significance levels were determined by analysis of variance and Tukey's honest significant difference test for correction of multiple comparisons. ***P<0.001. (f) Analysis of relative mRNA expression levels by qRT–PCR normalized to ubiquitin (UBC) on a logarithmic scale. Cells were transfected with the indicated siRNAs and treated with different concentrations of TNF for 24 h or left untreated. Error bars indicate s.d. from biological triplicates. (g) Left and middle panel: analysis of relative mRNA expression levels on a logarithmic scale by qRT–PCR normalized to UBC in cells with doxycycline (Dox)-inducible MITF expression treated with TNF (24 h). Error bars indicate s.d. from biological triplicates. Right panel: western blotting of Dox-induced GFP–MITF expression detected with anti-MITF. (h) ELISA assay for detection of secreted IL-1β from supernatants of TNF-treated MZ7 cells transfected with control (siNT) or siMITF siRNAs. Error bars indicate s.d. from biological triplicates.

Figure 3. Dedifferentiated melanoma cell state is characterized by reciprocal c-Jun/AP-1 upregulation.

(a) Left panel: outline of bioinformatic GSEA analysis. Right panel: top 20 enriched gene sets from the GSEA using the MSigDB C3 transcription factor targets gene set collection. (b) GSEA plot for the top ranking NF-κB target gene set. ES, enrichment score; NES, normalized enrichment score; FDR, false discovery rate. (c) Immunoblot analysis of p65/RelA protein levels in nuclear lysates of indicated cell lines treated with TNF (30 min) or left untreated. The nuclear protein histone H1 was used as loading control. (d) GSEA plot for the top ranking AP-1 target gene set. (e) Anti-correlation of JUN or FOSL1 mRNA expression with MITF in three independent melanoma cell line panels based on microarray gene expression data (Affymetrix gene probes are indicated). Significance of negative Pearson's correlation values was determined by a two-sided correlation test. (f) Immunoblottings for Mitf, c-Jun, Fosl1 and actin in human MITF^high and MITF^low melanoma cell lines.

Figure 4. MITF directly suppresses c-Jun that accounts for inflammatory hyperresponsiveness caused by MITF loss.

(a) Analysis of transcriptional changes of canonical AP-1 family members caused by loss of MITF using microarrays. The quantification is based on two biological replicates for each cell line. (b) ChIP-seq profile showing significant 3HA–MITF binding peaks in the JUN genomic region (black arrowheads). The input track is shown as a control. Two additional tracks indicate potential enhancer regions (H3K27ac) in human foreskin melanocytes (HFM) and from the Encode project. The scale bar indicates the size of the genomic region in kilobases (kB). (c) Analysis of relative c-Jun mRNA expression levels by qRT–PCR normalized to ubiquitin (UBC). Cells were treated with TNF for 24 h or left untreated. Values are referenced to siNT/-TNF for each cell line. Error bars indicate s.d. from biological triplicates. (d) Immunoblottings for Mitf, c-Jun and actin corresponding to the experiment described in c. (e) Heatmap analysis of genes strongly dependent on c-Jun in the context of MITF loss and TNF stimulation (24 h) in Ma.Mel15 cells using gene expression microarrays. Candidate genes are highlighted on the right side. (f) Independent validation of the heatmap results by qRT–PCR determining relative mRNA expression levels normalized to UBC. Expression values are referenced to the maximum value in each panel. Error bars indicate s.d. from biological duplicates.

Figure 5. c-Jun is critical for inflammation-induced dedifferentiation and the reciprocal gain of inflammatory responsiveness in melanoma cells.

(a) Scatter plot of log₂ fold changes of MITF and c-Jun caused by TNF treatment in MITF^high melanoma cell lines based on re-analysis of gene expression data (GSE51221). The significance of negative Pearson's correlation was determined by a one-side correlation test. (b) Immunoblottings for Mitf, c-Jun and actin from protein lysates of melanoma cells treated with TNF for the indicated times. (c) Experiment as described in b, but quantification of relative mRNA expression by qRT–PCR normalized to ubiquitin (UBC) and referenced to the respective highest value of the kinetics in MZ7 cells. Error bars denote s.d. of technical replicates. This representative experiment was performed in parallel to the immunoblot time course shown in b. The time courses were independently repeated three times. (d) Left panel: immunoblotting of c-Jun-GFP expression in Ma-Mel-102 cells detected by anti-GFP. Right panel: immunoblotting for Mitf and actin in c-Jun-GFP-expressing cells treated with TNF for the indicated times. (e–g) Experiment as described in d, but quantification of relative mRNA expression levels normalized to UBC and referenced to the highest expression value in each panel. Error bars indicate s.d. from biological triplicates.

Figure 6. The inflammatory MITF^low/c-Jun^high cell state correlates with increased myeloid cell infiltration in human melanomas.

(a–c) Expression of c-JUN, the TNF response gene signature and MLANA across TCGA melanoma samples ordered by increasing MITF levels from the left to the right in each panel. The black line (corresponding y axis at the right side of each panel) shows the increasing MITF expression levels. Grey bars represent the respective expression levels of c-JUN, the TNF response signature and MLANA in each individual sample. The coloured lines reflect the expression trends as determined by a moving average algorithm with a sample window size of n=20. (d–f) Same analysis as in a–c, but showing the expression of the melanoma dedifferentiation marker gene AXL, the myeloid immune cell marker gene CD14 and the T-cell gene CD3D. (g–i) Gene expression data of from the Lund melanoma cohort analysed as shown in a–f. The plots indicate the expression c-JUN, MLANA and CD14 in samples ranked by increasing MITF level. (j) Immunohistochemical analysis for gp100, CD45 and CD14 in representative cases from the Lund melanoma cohort. Small panels show zoom-in views, for example, to visualize morphology of CD45+/CD14+ tumour-infiltrating immune cells. Scale bars, 50 μm.

Figure 7. A stable inflammatory MITF^low/c-Jun^high cell state recruits myeloid cells in syngeneic Hgf-Cdk4^R24C mouse melanomas.

(a) Cartoon summarizing the establishment of dedifferentiated mouse melanomas cell lines as described in this and our previous study. (b) Immunoblottings showing the expression of Mitf, c-Jun and Fosl1 in the indicated mouse melanoma cell lines. Expression of actin served as loading control. (c) Chemokine expression measured by ELISA assay in the supernatant of the indicated mouse melanoma cell lines. Error bars indicate s.d. from biological triplicates. (d) Tumour growth kinetics of transplanted HCmel3, HCmel3-R and HCmel10 syngeneic melanomas shown per individual mouse (n=6 per group). Tumour growth was measured as diameter. Numbers (for example, 3/6) indicate tumour take rate (3 out of 6). (e) Immunohistochemical analysis for gp100 (melanocytic differentiation marker) and Gr-1 (myeloid cell marker) in the respective syngeneic melanomas shown in d. Size bars indicate the magnification of the pictures in each panel (50 and 10 μm). The panels at the right show a zoom-in view of the Gr-1 stain, to visualize morphology of individual Gr-1+ myeloid immune cells. (f) Immunoblottings for Mitf, c-Jun and actin from HCmel3 and HCmel10 tumour lysates. (g) Summary of flow cytometry analysis to quantitatively characterize tumour-infiltrating immune cells in Hcmel3 and HCmel10 syngeneic melanomas (n=3 in each group, mean percentage±s.e.m., unpaired two-tailed Student's t-test; *P<0.05, ***P<0.001). (h) ELISA assay for Tnf and Ccl2 from tumour lysates from HCmel3 and HCmel10 melanomas. (n=4 in each group, unpaired two-tailed Student's t-test; *P<0.05, ***P<0.001; horizontal lines and whiskers indicate quartiles).