Enlibio,Your biological research partner
Xinqidi Biotech Co.,Ltd,Wuhan,China 2008-2022
R&D 14th year

Persister state-directed transitioning and vulnerability in melanoma

Issuing time:2022-06-02 11:16


Melanoma is a highly plastic tumor characterized by dynamic interconversion of different cell identities depending on the biological context. Melanoma cells with high expression of the H3K4 demethylase KDM5B (JARID1B) rest in a slow-cycling, yet reversible persister state. Over time, KDM5Bhigh cells can promote rapid tumor repopulation with equilibrated KDM5B expression heterogeneity. The cellular identity of KDM5Bhigh persister cells has not been studied so far, missing an important cell state-directed treatment opportunity in melanoma. Here, we have established a doxycycline-titratable system for genetic induction of permanent intratumor expression of KDM5B and screened for chemical agents that phenocopy this effect. Transcriptional profiling and cell functional assays confirmed that the dihydropyridine 2-phenoxyethyl 4-(2-fluorophenyl)-2,7,7-trimethyl-5-oxo-1,4,5,6,7,8-hexa-hydro-quinoline-3-carboxylate (termed Cpd1) supports high KDM5B expression and directs melanoma cells towards differentiation along the melanocytic lineage and to cell cycle-arrest. The high KDM5B state additionally prevents cell proliferation through negative regulation of cytokinetic abscission. Moreover, treatment with Cpd1 promoted the expression of the melanocyte-specific tyrosinase gene specifically sensitizing melanoma cells for the tyrosinase-processed antifolate prodrug 3-O-(3,4,5-trimethoxybenzoyl)-(–)-epicatechin (TMECG). In summary, our study provides proof-of-concept for a dual hit strategy in melanoma, in which persister state-directed transitioning limits tumor plasticity and primes melanoma cells towards lineage-specific elimination.


Cellular plasticity describes the capacity of cells to switch from one phenotype to another and is considered a major driver of non-genetic tumor evolution. Recent time-resolved transcriptomic profiling points to an understanding particularly of melanoma as a heterogeneous ecosystem that dynamically transitions through variably differentiated persister states to escape from exogenous pressure such as debulking therapies1,2,3.

One hallmark of persister cells across different cancers is that they transiently rest in a chromatin-mediated slow-cycling state characterized by a high, but reversible nuclear expression of histone H3 lysine demethylases (KDMs)4,5,6. Other features of persister cells are maintenance of embryonic survival or metabolic programs balancing their dependency on oxidative mitochondrial respiration and detoxification of oxygen and lipid radicals5,7,8. To date, the majority of studies have focused on the selective elimination of persister cells as small subsets within the total tumor population by exploiting single vulnerabilities like fatty acid oxidation, autophagy, or ferroptosis9,10,11. Previous approaches to selectively eliminate H3K4 demethylase-expressing melanoma cells (KDM5Bhigh), e.g., based on their biological dependence on oxidative ATP production, showed some effect, but suffered from the high metabolic flexibility of this tumor entity12,13. Surprisingly, the cellular identity of KDM5Bhigh melanoma persisters has not been unraveled so far, missing an important treatment opportunity. Moreover, strategies to manipulate the dynamics in phenotype switching of persister cells and to combine those with cell identity-specific elimination are lacking in the field.

Melanomas are usually characterized by a continuous expression spectrum of KDM5Bhigh, KDM5Bintermediate and KDM5Blow cells. In comparison, benign melanocytic nevi express KDM5B at high levels across the majority of cells14. Single-sorted melanoma cells can rapidly re-establish KDM5B heterogeneity irrespective of their initial KDM5B expression level15. Even the pronounced enrichment for KDM5Bhigh states, which is typically seen under cytotoxic stress in vitro and in vivo5,16,17, rapidly reverts to normal distribution in surviving melanoma cells. This suggests that slow-cycling KDM5Bhighpersister cells represent a transient source for tumor repopulation, but longevity of melanoma requires a dynamic KDM5B tumor composition.

Altogether, this indicates that the slow-cycling KDM5Bhigh persister state could have different, maybe even opposing effects on tumor fate depending on the biological context and might be exploited as a therapeutic target for tumor elimination strategies in melanoma. In this study, we have systematically investigated (i) the actual differentiation phenotype of the KDM5Bhigh melanoma persister state, (ii) the consequences of forcing melanoma cells to ectopically express KDM5B at high levels without the chance to spontaneously revert to normal expression heterogeneity and (iii) whether KDM5B-directed cell state transitioning can be used as a therapeutic vulnerability.


Models for enforced and persistent intratumor expression of KDM5Bhigh states

KDM5B expression in melanoma is usually heterogeneously distributed, whereas benign melanocytic nevi express KDM5B at high levels across the majority of cells (Fig. 1a15). High KDM5B expression in melanoma is significantly associated with poor patient survival (Fig. 1b), while other KDM5 family members lack such an association (Supplementary Fig. 1a). In other cancer entities, KDM5B is inconsistently associated with survival (Supplementary Fig. 1b). To create an experimental model that allows the exogenous induction of KDM5B protein expression, we cloned a Tet-On 3G-system, in which KDM5B expression is driven by a doxycycline-inducible PTRE3G promoter and established stable melanoma cell clones (WM3734Tet3G-KDM5B and WM3734Tet3G-EGFP control, Supplementary Fig. 2a–c). After clonal expansion and doxycycline induction, we observed a dose-dependent upregulation of KDM5B mRNA and protein level (Fig. 1c and Supplementary Fig. 2c), concomitant with an inverse decrease in H3K4me3 protein levels confirming the expression of functional KDM5B (Supplementary Fig. 2d). The KDM5B protein level was stably upregulated under continuous doxycycline treatment for at least 3 weeks (Supplementary Fig. 2d). KDM5B immunostaining showed a shift towards increased signals across the majority of WM3734Tet3G-KDM5B cells (Fig. 1d). From here on, we denote this scenario as enforced KDM5Bhigh cell state.

Fig. 1: Expression modulation of the histone H3K4 demethylase KDM5B/JARID1B.
figure 1

a Anti-KDM5B immunostaining of a melanoma patient sample (left) compared to a benign human nevus (right). Highly positive nuclei are dark red, medium positive nuclei are light red, low expressing nuclei are blue. Isotype controls are shown in the upper right corners. Depicted are representative images of different melanoma or nevi samples (n = 5 each). b Kaplan–Meier survival curves of cutaneous melanoma patients were calculated from the TCGA data set based on cut-point optimization for KDM5B expression (high expression, red, vs. low, green, TCGA browser tool UCSC Xena and GraphPad Prism). Sample sizes are indicated in the patient at risk table (# of risk). Significance was tested by Long-rank (Mantel–Cox) test. c Quantitation of KDM5B mRNA induction after 24 h of doxycycline (Dox) treatment as assessed by qPCR. Shown is one representative example (mean, n = 2). d Anti-KDM5B nuclear immunostaining of WM3734Tet3G-KDM5Bcells after Dox-titration at the indicated concentrations for 24 h (left, representative images; right, quantitation shown as normalized frequency distribution of nuclear staining intensity). Shown is one representative out of three clones. e Flow cytometric detection of endogenous KDM5B protein levels after treatment with Cpd1 for 72 h. Mean ± SD (n = 4); two-sided t-test. f Anti-KDM5B nuclear immunostaining of three different melanoma cell lines (WM3734, WM88, MelJuso) after 72 h of 10 µM Cpd1 treatment (left, representative pictures; right, quantitation shown as normalized frequency distribution of nuclear staining intensity). g Time course of KDM5B protein levels after treatment of WM3734 cells with Cpd1 (10 µM) plus cycloheximide (CHX, 50 µg/ml, n= 2). h Ubiquitin protein conjugates were immunoprecipitated in WM3734 cells after Cpd1 treatment for 72 h. The total ubiquitinated cellular protein content was detected by a pan-ubiquitin-HRP antibody (upper row). The fraction of ubiquitinated KDM5B protein was detected by a KDM5B-specific antibody (middle row). Neg4 was used as structure homolog compound control (n = 2). To improve visualization of the KDM5B bands resulting from immunoprecipitation, the contrast was enhanced for both treatments equally. Source data are provided as a Source Data file.

To obtain a second, independent method for the modulation of KDM5B expression, which additionally allows simple application to several melanoma cell lines or in vivo, we developed a cell-based compound screening assay applying our previously published KDM5B-promoter-EGFP-reporter construct stably expressed in WM3734 melanoma cells (WM3734KDM5Bprom-EGFP cells15, Supplementary Fig. 3a). This fluorescence-based model facilitated monitoring the dynamic nature of the KDM5B transcription state. After screening a 7500-compound library, primary hits were counter-screened in a CMV-promoter-EGFP-reporter assay and confirmed hits were further validated in dose–response curves and independent assays (Supplementary Fig. 3b–d). The main criteria for hit compound selection were the absence of immediate overall cell toxicity at 72 h of treatment and modulation of the KDM5B-promoter-driven expression of EGFP (abbreviated K/EGFP, for more details see “Methods”). Compounds that directly increased K/EGFP signals were not considered because of possibly masking effects by passive enrichment for KDM5Bhigh cells as previously seen for various cytotoxic drugs5. Instead, K/EGFP signal decrease was of particular interest because it could indirectly indicate a persistent increase of the endogenous KDM5B protein level as a result of chemical compound treatment (for example, through negative feedback of high KDM5B protein levels on mRNA transcription).

As confirmed by flow cytometric analysis and by immunocytology, the top hit compound 2-phenoxyethyl-4-(2-fluorophenyl)-2,7,7-trimethyl-5-oxo-1,4,5,6,7,8-hexahydroquinoline-3-carboxylate (PubChem name BAS00915510, here abbreviated Cpd1) reproducibly increased KDM5B protein levels irrespective of the genotype of the melanoma cell lines tested (Fig. 1e, f and Supplementary Table 1). A structurally homologous compound that failed to change K/EGFP transcription and endogenous KDM5B protein levels was selected as negative control for subsequent experiments (abbreviated Neg4, Supplementary Table 1 and Supplementary Fig. 3c, d). KDMs are histone demethylases and transcriptional regulators and as such mainly function in the nucleus18,19,20,21,22. Accordingly, quantitation of the KDM5B expression localization in nuclear vs. cytoplasmic compartments confirmed a Cpd1 time-dependent increase of nuclear KDM5B (Supplementary Fig. 3e) phenocopying the effect of Tet-On 3G-induced expression of KDM5B. Again, this was independent of the melanoma cell type tested (Supplementary Fig. 3f).

Due to the pronounced nuclear enrichment of KDM5B seen across the majority of melanoma cells upon Cpd1 treatment (Fig. 1f), we hypothesized that Cpd1 might affect the natural homeostasis of KDM5B protein as a potential mode of action. Longitudinal treatment of melanoma cells with the protein synthesis inhibitor cycloheximide revealed a steady decrease of KDM5B protein yielding a 79 and 98% decrease after 24 and 72 h, respectively. Co-treatment with Cpd1 resulted in an elevated KDM5B level (382 and 147% at 24 and 72 h), suggesting that Cpd1 might attenuate KDM5B protein degradation and by this maintain its nuclear expression level (Fig. 1g). To study the cellular degradation pathway of KDM5B, we established a proteolysis targeting chimera (PROTAC) model23,24(Supplementary Fig. 4a). This model is based on the binding of SLF′-thalidomide to FKBP12-tagged KDM5B. SLF′ is a synthetic binding ligand of FKBP12, while thalidomide recruits the E3 ubiquitin ligase cereblon for ubiquitination and subsequent proteosomal degradation of targeted KDM5B. Applying SLF′-thalidomide to FKBP2-tagged KDM5B in WM3734 melanoma cells, we could demonstrate that KDM5B protein is quickly proteosomally degraded in an E3 ubiquitin ligase-dependent manner (Supplementary Fig. 4b). Since proteasomal degradation is a calcium-dependent process25,26 and the dihydropyridine structure of Cpd1 indicates a function as calcium channel blocker, we tested if Cpd1 affects the intracellular calcium content and the degradation level of ubiquitinated protein in melanoma cells. Indeed, both acute and long-term Cpd1 treatments led to a significant reduction in the store-operated calcium entry (Supplementary Fig. 4c) paralleled by an accumulation of ubiquitinated KDM5B (Fig. 1h).

Enforced expression of KDM5B directs melanoma cells towards a slow-growing persister state

Standard proliferation (MTT) and apoptosis (caspase 3 plus annexin V measurement or 7AAD flow cytometry) assays indicated a rather cytostatic than apoptotic or toxic effect in short-term Cpd1-treated melanoma cell lines or WM3734Tet3G-KDM5B cells (Fig. 2a, left panel and Supplementary Fig. 5a, b). For Cpd1, a maximum feasible dose of 10 µM was assumed based on cross-cell line comparisons in MTT assays (Supplementary Fig. 5c) and was used for all following experiments. In congruence with reports on KDM5B-associated therapy persistence5,16,17, we observed decreased drug susceptibility to MAPK signaling pathway inhibitor (MAPKi) treatment upon KDM5B upregulation irrespective of the melanoma cell line tested (Fig. 2a, right panel).

Fig. 2: In vitro and in vivo effects of enforced KDM5B expression.
figure 2

a MTT assays with Cpd1 started at day −7, and then MAPKi was added for 72 h. Readout after 72 h of MAPKi (5 µM PLX4720 + 0.5 µM GDC-0973) and in total 10 days of Cpd1. Cell numbers after 10 days of DMSO or Cpd1 without MAPKi (left panel) and after 72 h of MAPKi compared to “no MAPKi control” (right). Mean ± SD (n = 3), two-sided t-test (****p ≤ 0.0001). b Clonogenic growth after KDM5B induction (9, 16, and 20 days). Mean ± SD, two-sided t-test. Shown is one out of six clones. c Soft agar colony formation after 52 days of KDM5B induction. Shown is one representative experiment (n = 3). Mean ± SD; two-sided t-test. d Growth of xenografted WM3734Tet3G-KDM5B cells. Doxycycline treatment started day 34. Mean ± SEM with n = 6 mice in the control and n = 10 mice in the treatment group, linear mixed-effect spline model. e Clonogenic growth assay of WM983B and WM1366 cells continuously treated over 7 and 9 days with 10 µM of Cpd1 or Neg4 (n = 3). f Clonogenic growth assay of 451Lu and resistant 451Lu BR cells treated over 9, 16, and 20 days with 10 µM of Cpd1 vs. DMSO or Neg4 controls (n = 2). g Two-dimensional colony formation after continuous treatment with 10 µM Cpd1 for 19 days. Mean, n = 3. h Soft agar colony formation under constant Cpd1 treatment at the indicated doses over 30 days (one representative experiment out of n = 2) and i compared to short-term treatment for 72 h before seeding without treatment continuation (mean, n = 2). j Collagen invasion of WM3734 spheroids under 10 µM of Neg4 or Cpd1 at day 10. Shown is one representative experiment (n = 3). Mean ± SD; two-sided t-test. k WM3734 melanoma spheroids at day 10 after collagen embedding. l, m Growth of xenografted WM3734 cells (left, n = 9 mice per group) and syngeneic CM cells (right, n = 3 per control and n = 5 mice per treatment group). Cpd1 treatment was started when tumors were palpable. Mean ± SEM, linear mixed-effect spline model. Source data are provided as a Source Data file.

Next, we tested how melanoma cell populations behaved, when the KDM5B expression spectrum was constantly directed to a higher level without the chance to dynamically revert to normal heterogeneity. Enforced expression of KDM5B in the WM3734Tet3G-KDM5B model reduced cell numbers in long-term two-dimensional (2D) growth assays and decreased anchorage-independent 3D-colony formation in a dose-dependent manner (Fig. 2b, c and Supplementary Fig. 5d). Growth of naïve WM3734 and WM3734Tet3G-EGFP control cells was not affected by doxycycline (Supplementary Fig. 5e). The KDM5B-mediated effect on tumor proliferation was then studied in vivo. We allowed xenografted WM3734Tet3G-KDM5B cells to establish tumors up to approximately 150 mm3 before doxycycline was supplemented to the drinking water. This led to a significant tumor growth delay over nearly 2 weeks (Fig. 2d; p < 0.05, linear mixed-effect model).

Long-term treatment with Cpd1 significantly reduced the growth of different melanoma cell lines irrespective of their genotype (Fig. 2e, Supplementary Fig. 6a–c show examples for treatment up to 16 days). Also, melanoma cells that had previously developed resistance to MAPK inhibition responded to chemical enforcement of the KDM5B phenotype supporting the observation that chronically drug-resistant melanoma cell populations re-establish normal KDM5B heterogeneity after initial KDM5B enrichment (Fig. 2f and Supplementary Fig. 6d–g). 7AAD staining suggested that this effect is more the result of the antiproliferative effect of upregulated KDM5B than of compound toxicity (Supplementary Fig. 6h).

In addition, limited dilution assays and 3D-colony formation indicated a loss of tumor repopulation properties of melanoma cells under constant exposure to Cpd1 (Fig. 2g, h). In contrast, pre-treatment of melanoma cells with Cpd1 for only 3 days before seeding onto soft agar was not sufficient to reduce the number of formed colonies again suggesting spontaneous reversibility of the KDM5B phenotype under normal conditions (Fig. 2i and Supplementary Fig. 6i). Continuous treatment with Cpd1 also reduced the invasive capacity of melanoma cells in a 3D-collagen spheroid model (Fig. 2j). Interestingly, Cpd1 strongly impaired the capacity of melanoma cells to form proper spheroids; especially when treatment commenced before collagen embedding (Fig. 2k).

Finally, we tested Cpd1 in a xenograft and an immunocompetent syngeneic mouse melanoma model5,27. The phenotypic effects of Cpd1 treatment were confirmed beforehand in orthogonal in vitro assays for the murine CM cells, same as for the WM3734 cells (Supplementary Fig. 6j, k, Fig. 1e, f and Supplementary Figs. 3d, 5c, 6b, c, respectively). In both mouse models, Cpd1 treatment resulted in a significant antitumor growth effect in comparison to the vehicle control (Fig. 2l, m). Ex vivo immunostaining confirmed a Cpd1-induced shift towards high KDM5B expression states (Supplementary Fig. 6l).

In sum, our results suggest that KDM5B expression dynamics can be limited by exogenous genetic or chemical modulation providing a possibility to manipulate KDM5B-dependent tumor maintenance.

KDM5B-mediated cell cycle arrest and inhibition of cytokinetic abscission

Although persister cells have been repeatedly described to be slow-cycling in various cancer entities4,6,9, their true cellular identity and proliferation phenotype remains elusive in melanoma. We first demonstrated by DNA content cell cycle analyses that enforced KDM5B expression over 72 h dose-dependently induces cell cycle delay across various genetically different melanoma cell lines (Fig. 3a and Supplementary Fig. 7a). Interestingly, time-lapse cell cycle imaging of single FUCCI-WM164 cells28 revealed that Cpd1 treatment increases the total cell cycle length by 43.8% compared to DMSO- and Neg4 controls due to longer time spent in both G1 and S/G2/M phases (Fig. 3b and Supplementary Fig. 7b, c). Longitudinal DNA content cell cycle analysis under continuous treatment with Cpd1 up to 7 days (Neg4 as negative control) demonstrated that the Cpd1-induced delays of G1 and G2/M can occur in a temporally consecutive manner (Supplementary Fig. 7d). In sum, these experiments underscore the role of heterogeneity and temporal plasticity for effect size measurements across single melanoma cells, especially for the interpretation of phenotypic cell cycle readouts from single time points.

Fig. 3: KDM5B-mediated cell cycle arrest and inhibition of cytokinetic abscission.
figure 3

a Propidium iodide cell cycle analysis of Cpd1-treated WM3734 (above) or Dox-treated WM3734Tet3G-KDM5B cells (below) after 72 h. Shown are representative data (n = 4). b Quantitation of the G1 (upper panel) and S/G2/M (lower panel) cell cycle duration by real-time cell cycle imaging of FUCCI-WM164 cells pretreated with Cpd1 (10 µM) vs. DMSO or Neg4 controls (10 µM) for 72 h. Scatter dot plots represent mean ± SD (n = 3); Kruskal–Wallis test. c Time-lapse imaging analysis of WM3734 cell numbers during treatment with Cpd1 (10 µM) vs. DMSO or Neg4 controls (10 µM) up to 72 h (15 different areas, n = 2). d Time-lapse microscopic movies were analyzed for the time to complete cytokinesis (15 different cells, n = 2). e Quantitation and representative immunofluorescence staining of midbodies in Cpd1-treated (72 h) WM164 cells (left) and Dox-treated WM3734Tet3G-KDM5B cells (right) vs. respective controls. Depicted are single fluorescence channels: blue (DAPI, nucleus); yellow (α-tubulin, microtubules); green (Aurora B kinase, midbodies); red (phalloidin, F-actin). Dot plots show the ratio of midbodies/cell per field of view, 12–16 fields, n = 3. Mean ± SD, left, one-way ANOVA: right, two-sided t-test. f Quantitation of midbody decrease in the KDM5B-PROTAC model. Proteasomal degradation of KDM5B was induced by SLF′-thalidomide (SLF′-t) concurrently applied with Cpd1 (“con”) or applied after Cpd1 pre-treatment for 3 days (“pre”). Neg4 was used as control. Dot plots show the ratio of midbodies/cell nuclei as median values with the interquartile range (4 coverslips, n = 2; one-way ANOVA). g Venn diagram and gene ontology analysis of significantly regulated genes upon Cpd1 (10 µM) treatment of WM3734 cells as detected by mass spectrometry and RNA sequencing. h Quantitation of downregulation of selected cytokinesis regulators as assessed by mass spectrometry and RNAseq (left, n = 1) or qPCR (right, mean ± SD, n = 2) in Cpd1-treated WM3734 and Dox-treated WM3734Tet3G-KDM5B cells. Source data are provided as a Source Data file.

To further unravel the molecular mechanisms underlying the antiproliferative effect of KDM5B, we performed single-cell time-lapse microscopy of Cpd1-treated WM3734 cells vs. Neg4-treated control cells (Supplementary Movies 13). The quantitation of cell numbers after 72 h confirmed a reduction in cell proliferation (Fig. 3c). Interestingly, we observed that some cells undergoing cell division showed a prolonged time to complete cell abscission (Fig. 3d). We hypothesized that this phenomenon may be caused by a so far unknown regulatory role of KDM5B in cytokinesis (final process of physical separation of dividing cells). We aimed to quantitate this by immunostaining of intercellular midbodies, i.e. structures that represent microtubule-rich membrane bridges that connect two daughter cells shortly before their membranes fully dissever29. Indeed, we found a significant increase in the number of midbodies in a fraction of Cpd1-treated WM164 cells (p ≤ 0.0001) as well as doxycycline-treated WM3734Tet3G-KDM5B cells (p = 0.0033) (Fig. 3e). Importantly, the Cpd1-mediated increase of midbodies was rescued by SLF′-thalidomide-induced degradation of KDM5B in WM3734Tet3G-KDM5B-FKBP12cells (Fig. 3f). Transcriptional and proteomic profiling of KDM5B-enforced WM3734 cells by RNAseq and label-free quantitative mass spectrometry after 72 h of 10 µM Cpd1 revealed a small overlap of 11 genes/proteins, of which 7 have known functions during cytokinesis (Fig. 3g and Supplementary Data 1 highlighted in bold). For example, AURKB, MKI67, KIF4A, UBE2C, RRM2, KIF1C1, and ANLN are attributed to the GO cytoskeleton-dependent cytokinesis. For the expression of a subset of genes (AURKB, KIF4A, SHCBP1, and UBE2C), we found a downregulation in both Cpd1-treated WM3734 as well as doxycycline-treated WM3734Tet3G-KDM5B cells (Fig. 3h). AURKB, KIF4A, and UBE2C are known to regulate spindle assembly and coordinate abscission30,31. SHCBP1 is involved in midbody organization and cytokinesis completion32. Thus, irrespective of the model used, the above-mentioned genes may be functionally involved in the observed increase in midbodies and the delay in cell doubling.

Our experiments suggest so far that exogenous expression of KDM5B can direct melanoma towards a slow-cycling tumor phenotype across the majority of tumor cells, where different molecular mechanisms might cooperate to prevent cell proliferation. Regarding potential clinical implications, reducing persister state plasticity could provide a therapeutic gain in time by delaying tumor progression. However, this approach might be more effective, if slow-cycling and potentially tumor repopulating persister states are additionally eradicated by exploiting state-specific drug sensitivities.

Enforced KDM5B expression leads to lineage reprograming

To unravel potential changes in cellular identity and underlying transcriptional programs after enforced KDM5B expression, we performed RNA sequencing of our genetic and chemical KDM5B induction models (Tet-On 3G and Cpd1) head-to-head. WM3734Tet3G-KDM5B cells were treated for 24, 48, and 72 h with doxycycline and then harvested for analysis. In parallel, naïve WM3734 and patient-derived short-term cultured CSM152 cells were treated with Cpd1 and analyzed after 72 h (Supplementary Data 2). We first checked if Cpd1 also transcriptionally phenocopies the effect of Tet-On 3G-mediated KDM5B induction. Indeed, we found that both of our models can regulate a previously published KDM5B target gene motif (SCIBETTA_KDM5B_TARGETS_DN (DN = “down”) motif, 79 genes) to a similar degree21 (Supplementary Fig. 8a). Subsequent Gene Set Enrichment Analysis (GSEA) followed by Cytoscape visualization revealed a transcriptional landscape that matched other expected KDM5B-associated pathways like chromosomal remodeling or ATP-ase metabolism. Also, genes involved in cell cycle/mitosis, DNA damage response, RNA processing, and immune response were found to be regulated by KDM5B (Supplementary Fig. 8b). Transcripts that control cell cycle and mitosis revealed a steadily increasing and statistically significant regulation from 24 to 48 and 72 h of KDM5B induction pointing to a fundamental and temporally dynamic influence on the cell proliferation machinery (Supplementary Fig. 8c).

As KDM5B has been reported to guide developmental processes in a highly conserved fashion33,34,35, we asked if enforced KDM5B expression could affect the differentiation state of melanoma cells. Indeed, we observed for genetic as well as chemical KDM5B enforcement an impressive downregulation of mesenchymal (SARRIO_EPITHELIAL_MESENCHYMAL_TRANSITION_DN, 145 genes)36 and proliferative (GERBER_PROLIFERATION_SPOT A, 405 genes)37 gene motifs (Fig. 4a, FWER p < 0.05 was considered as statistically significant). Additionally, melanocytic differentiation genes were specifically regulated in both models (GO_PIGMENTATION38,39, 86 genes, Fig. 4b). Most strikingly, enforced KDM5B expression was followed by a time-dependent transcriptional shift, in which cells were gradually transitioning from undifferentiated and neural crest-like to transitory-melanocytic and melanocytic gene motifs over time (Fig. 4c and Supplementary Table 2). Accordingly, KDM5B knockdown showed an inverse gene expression towards undifferentiated states (Fig. 4c, control WB in Supplementary Fig. 8d). In an independent control experiment, differentiation reprograming was reproduced for Cpd1 and occurred also in a time-dependent manner (Supplementary Fig. 8e, Supplementary Table 2 and Supplementary Data 3). Importantly, KDM5B-associated differentiation reprograming was similarly induced by Cpd1 in patient-derived CSM152 melanoma cells (Fig. 4a, b and Supplementary Fig. 8a, e) and was also confirmed in silico for endogenous gene expression in single melanoma cells isolated from human tumor tissue (n = 1253 melanoma cells from 19 tumors40). Here, KDM5B expression was significantly higher in cells with melanocytic gene signatures compared to cells with undifferentiated signatures (Fig. 4d).

Fig. 4: KDM5B leads to a differentiation-directed phenotypic shift.
figure 4

a, b Heatmaps of the SARRIO_EPITHELIAL_MESENCHYMAL_TRANSITION_DN36, GERBER_PROLIFERATION_SPOT A signatures37 (a) and GO_PIGMENTATION signature38,39 (b) from WM3734Tet3G-KDM5B cells after 72 h of Dox treatment and WM3734 and CSM152 cells after 72 h of Cpd1 treatment; red, upregulated; green downregulated genes. Significance is indicated by FWER p < 0.05. cHeatmap of the Tsoi differentiation trajectory66 for WM3734Tet3G-KDM5B cells after 24, 48, and 72 h Dox treatment (left), WM3734 KDM5B knockdown (sh) versus control (scr) cells (right); red, upregulated, green downregulated genes. d Violin plot displaying endogenous KDM5B expression levels according to the transcriptional differentiation level in single-cell RNA-sequenced human melanomas. Asterisks represent a p value <0.0001 from one-way ANOVA with Dunnett’s multiple comparison test. Here the comparison was made to the undifferentiated group.

KDM5B is an epigenetic determinant of cytokinesis and differentiation gene transcription

Considering KDM5B’s known role as a histone demethylase with direct involvement in transcriptional regulation, we performed an in silico analysis of KDM5B and H3K4me3 ChIP-Seq data using the ChIP-Atlas41 set to “all cell type class“ as well as H3K4me3 ChIP-Seq data from breast cancer42 and melanoma cells43,44. This suggested a correlation between KDM5B DNA-binding motifs and H3K4me3 chromatin marks across a number of cytokinesis genes (breast cancer and melanoma cell lines) and melanocytic differentiation genes (melanoma cell lines) (Fig. 5a, b). Specifically, this included the cytokinesis genes AURKB, KIF4A, SHCBP1, and UBE2C as well as the differentiation genes MITF, MLANA, NGFR, AXL, TYR, TJP1, CDH2, ZEB1 and SMAD1, which we showed to be regulated in a Cpd1-dependent manner (Figs. 3g, h and 4a–d).

Fig. 5: KDM5B-dependent cytokinesis and differentiation gene transcription.
figure 5

In silico analysis of cytokinesis and mitotic spindle assembly genes (a) or differentiation genes (b) of KDM5B- and H3K4me3-ChIPseq data from breast cancer cell lines SUM185, SUM159, MCF7, HCC2157, T47D, and MDA231 (GSE4607342) and melanoma cell lines MM27, MM13, MM16 (GSE7185444) and A375 (GSE9983543)) and ChIP-Atlas41. c KDM5B knockdown in MaMel63a cells was confirmed by immunoblotting 4 days after siRNA transfection. d KDM5B target gene expression of cytokinesis-related genes (AURKB, KIF4A) and differentiation-related genes (DCT, MelanA) was analyzed by qPCR 4 days after KDM5B knockdown and 3 days of Neg4 or Cpd1 treatment (10 µM). Shown is one representative example (mean ± SD, n = 2). Source data are provided as a Source Data file.

To functionally support that the observed effects of Cpd1 on gene expression are truly KDM5B-dependent, we have performed siRNA knockdown of KDM5B and subsequently measured transcriptional expression of selected cytokinesis and differentiation genes upon concurrent Cpd1 treatment (Fig. 5c, d). As expected, cytokinesis genes, which we have found downregulated after genetic (and Cpd1-mediated) KDM5B upregulation in Fig. 3h, were upregulated at mRNA level after siRNA knockdown of KDM5B as compared to the scrambled control (e.g., AURKB, KIF4A, Fig. 5d, Neg4 treatment). Accordingly, differentiation genes, which we found to be upregulated upon KDM5B induction (Fig. 6a), were downregulated after KDM5B knockdown. When we treated with scrambled control cells with Cpd1, we again saw the expected downstream effects of KDM5B induction, i.e., mRNA downregulation of cytokinesis and upregulation of differentiation genes (gray bars). When we added Cpd1 to cells with siRNA KDM5B knockdown, Cpd1 could not maintain the KDM5B protein level (Fig. 5c, no KDM5B protein was produced that could be preserved from proteasomal degradation) and, consequently, could not reverse siRNA-dependent regulation of KDM5B downstream genes (white bars).

Fig. 6: Enforced KDM5B expression facilitates melanocytic lineage-directed elimination by TMECG.
figure 6

a Quantitation of mRNA after 24 h, 48 h, 72 h and 7 days of Cpd1 treatment of MaMel63a cells as assessed by qPCR. Mean ± SD. Shown is one representative example. b Regulation of differentiation, cytokinesis, and mitotic spindle assembly genes as detected by cDNA microarray analysis after KDM5B shRNA knockdown in WM3734 cells (n = 1). c, d Immunoblotting of melanocytic lineage and (de-)differentiation markers after 24 h of KDM5B induction in WM3734Tet3G-KDM5B cells (c) and after 72 h of Cpd1 treatment in MaMel63a cells (d). Shown are representative data (n = 2). e Anti-MITF immunostaining (upper panel) and Fontana-Masson staining (lower panels) of CM melanoma tumor grafts from Cpd1-treated vs. control mice. f MTT cell viability assay of WM3734 cells. Representative example is shown left (mean ± SD, n = 2) and corresponding IC50 values on the right. TMECG was either concurrently given together with Cpd1 (“con”) or added 3 days after Cpd1 pre-treatment (“pre”). Readout was performed after 72 h of TMECG treatment. g Persister-state-directed therapy model in vivo. Left: schematic representation of treatment dosing and timing in immunodeficient NMRI-(nu/nu)-nude mice. Right: tumor volumes of WM3734 xenografts (endpoint at day 30). TMECG was either concurrently given together with Cpd1 (“con”) or added one week after Cpd1 pre-treatment (“pre”). Mean ±SEM (6 mice in TMECG and Cpd1 control group, five mice in “con” and seven mice in “pre” group). Significance was determined by two-sided Mann–Whitney test. Source data are provided as a Source Data file.

Enforced KDM5B expression leads to lineage-directed vulnerability

The observed KDM5B-dependent cell state reprograming not only confirms commitment to melanocytic differentiation, it could also sensitize melanoma cells for secondary phenotype-specific drugs. We sought to take advantage of 3-O-(3,4,5-trimethoxybenzoyl)-(−)-epicatechin (TMECG), a tyrosinase (TYR)-processed antimetabolic agent previously described to eliminate melanoma cells in a lineage-specific way45. Thus, we first assessed whether Cpd1 treatment truly activates the melanocytic differentiation/pigmentation machinery in melanoma cells. Indeed, we found a significant time-dependent Cpd1-induced transcription of the melanocytic master regulator MITF and other downstream differentiation genes like DCT, TYR, and MART-1/MLANA in pigmentation-competent melanoma cells (Fig. 6a). Conversely, knockdown of KDM5B was associated with a decrease in differentiation gene expression (Fig. 6b). Immunoblotting of WM3734Tet3G-KDM5B cells confirmed a KDM5B-dose-titratable increase of MITF protein, whereas typical markers of mesenchymal cell phenotypes such as CDH2, AXL, ZEB1, or SMAD1 proteins were decreased (Fig. 6c). A similar tendency of regulation of these markers at protein level was confirmed for Cpd1 (Fig. 6d). Lastly, immunostained sections from syngeneic melanomas after Cpd1 treatment (Fig. 2m) confirmed enriched MITF protein expression in vivo (Fig. 6e, upper row). Fontana-Masson staining additionally indicated a focal increase of intracellular melanin production (Fig. 6e, lower row).

We next tested lineage-directed melanoma cell sensitization by Cpd1 in preclinical models. We tested combinations of Cpd1 with TMECG in three melanoma cell lines in vitro (WM3734, MaMel63a, WM983B, Fig. 6f and Supplementary Fig. 9). MTT assays revealed that TMECG in combination with Cpd1 was more effective in cells, which were pretreated with Cpd1 for 3 days. This complemented our prior observation that KDM5B-directed cell reprograming towards cell differentiation is a time-dependent process (Fig. 4c and Supplementary Fig. 8e). TMECG alone showed a limited effect on cell numbers and, therefore, was defined next to Cpd1 as mono-treatment control for subsequent in vivo combination therapies (Fig. 6g). The effect of Cpd1 alone on in vivo tumor growth compared to vehicle control was assessed beforehand (Fig. 2l). To study the influence of the temporal sequence of TMECG and Cpd1 in vivo, we set up a therapy model, in which WM3734 cells were xenografted to establish tumors on the back of immunodeficient NMRI-(nu/nu)-nude mice over 20 days (Fig. 6g, left). Co-treatment of Cpd1 and TMECG was started either simultaneously or consecutively, i.e., the tumors were pretreated with Cpd1 for one week before TMECG was added. In line with our in vitro observations, melanoma tumor growth was significantly reduced when established tumors were primed to differentiation before state-specific elimination started (Fig. 6g, right; p < 0.05, Mann–Whitney test). Concurrent treatment of Cpd1 and TMECG failed to reduce tumor size within the time span of our animal protocol. A possible long-term effect of Cpd1 also in the concurrent setting could not be assessed in this model.

To summarize, our study provides proof-of-concept for a dual hit strategy in melanoma, in which persister state-directed transitioning limits cell state plasticity and primes tumor cells towards differentiation-specific elimination (Fig. 7).

Fig. 7: Phenotype-specific dual hit strategy in melanoma.
figure 7

KDM5B persister state-directed transitioning limits tumor plasticity and heterogeneity and primes melanoma cells towards lineage-specific elimination.


Next to genetic tumor evolution, particularly phenotypic cell plasticity is an emerging problem in cancer therapy46. As melanoma cells are rapidly shifting between different transcriptional, cell proliferation, and differentiation programs, such phenotypes may transiently overlap and rapidly adapt to any strategy that specifically attacks single drug resistance mechanisms1,2,3,47. Thus, one question of this study was if limiting phenotypic cell state dynamics could become a starting point for a dual hit strategy in melanoma. The slow-cycling KDM5Bhigh melanoma cell phenotype was used as a model for proof-of-concept.

So far, the real nature of slow-cycling KDM5Bhigh persister cells in melanoma was unknown. Our data suggest that KDM5B directs the transcriptional identity of melanoma cells from undifferentiated towards melanocytic lineage differentiation. KDM5B-induction upfront to therapy changed the frequency of intrinsically resistant melanoma cells similar to the recently reported DOT1L-inhibited state3. Inhibition of the histone H3K79 methyltransferase DOT1L by gene knockout in a CRISPR-Cas9 screen favored a neural crest cell differentiation program (NGFRhigh/EGFRhigh state) and increased the number of cells primed to resistance upfront to BRAF inhibitor treatment. Future studies on single-cell transcriptional and proteomic level have to decipher, if KDM5B induction and DOT1L inhibition drive cellular identity towards the same end stage and how strong these effects depend on the actually active gene expression level and time required for phenotypic reprogramming.

In accordance with reports on KDM5B-dependent cell fate decisions and lineage commitment in other tissues types (e.g. neural differentiation34 or hematopoiesis48), our study suggests that KDM5B acts as a highly dynamic coordinator of both differentiation and cell division programs in melanoma. We show that KDM5B efficiently suppresses cell proliferation via downregulation of genes, which control cytoskeleton-dependent cytokinesis like AURKB, KIF4A, UBE2C, and SHCBP130,31,32. Our midbody analysis estimates that a considerable fraction of KDM5B-enforced melanoma cells needs prolonged time to complete cell abscission, which might contribute the observed overall cell cycle delay. As KDM5B is not affected by inactivating mutations in most melanomas according to currently available genetic profiling data like TCGA (mutation frequency of 5.9% according to CBioPortal), it may represent an ideal way for melanoma cells, irrespective of their mutational background to secure a slow-cycling state whenever required. Conceptually, this could mean that melanoma cells exploit the slow-cycling differentiated KDM5Bhigh state to immediately survive selective pressure, but must decrease KDM5B expression again and resume cell cycle progression in favor of more proliferative cell phenotypes to ensure long-term tumor repopulation (plasticity addiction).

Although we consider the KDM5Bhigh phenotype as differentiated, we are aware that any terminology that is applied to describe phenotypic cell states in melanoma needs to be used with caution because of their highly volatile nature. Under prolonged KDM5B induction, we observed changes in RNA signatures and protein marker profiles over time. Depending on the time point chosen for experimental read-out, marker constellations, or even functional behavior like cell proliferation can be misleading for assessment of whether a certain cell state is advantageous or disadvantageous for tumor cell survival or drug killing. In this regard, our results unveiled a “Janus-faced” role of the KDM5Bhigh melanoma cell state. On the one hand, melanoma growth and cell invasion are impaired, when the slow-cycling state is sustained. On the other hand, melanoma cells can exploit this state to immediately survive targeted or cytotoxic therapies. However, as demonstrated here, when KDM5Bhigh state-directed transitioning is combined with a state-specific elimination strategy (instead of unspecific debulking drugs), control of tumor growth is possible. In fact, multiple applications of cell state-directed elimination are conceivable now; particularly as a rescue strategy or to deepen response in residual disease after state-of-the-art cancer therapies. Thus, we plan to systematically test Cpd1/TMECG in sequential therapy regimens with other therapies in future studies ideally using a chemically improved version of Cpd1 with optimized solubility.

Article classification: Biological abstract
Share to:
Add:Room A11-329, 1st Floor, No.1, SBI Venture Street, Optics Valley, East Lake
New Technology Development Zone, Wuhan, China.
Certificate NO.:U18Q28010569R0S