Lineage-specific control of TFIIH by MITF determines transcriptional homeostasis and DNA repair

The melanocytic lineage, which is prominently exposed to ultraviolet radiation (UVR) and radiation-independent oxidative damage, requires specific DNA-damage response mechanisms to maintain genomic and transcriptional homeostasis. The coordinate lineage-specific regulation of intricately intertwined DNA repair and transcription is incompletely understood. Here we demonstrate that the Microphthalmia-associated transcription factor (MITF) directly controls general transcription and UVR-induced nucleotide excision repair by transactivation of GTF2H1 as a core element of TFIIH. Thus, MITF ensures the rapid resumption of transcription after completion of strand repair and maintains transcriptional output, which is indispensable for survival of the melanocytic lineage including melanoma in vitro and in vivo. Moreover, MITF controls c-MYC implicated in general transcription by transactivation of far upstream binding protein 2 (FUBP2/KSHRP), which induces c-MYC pulse regulation through TFIIH, and experimental depletion of MITF results in consecutive loss of CDK7 in the TFIIH-CAK subcomplex. Targeted for proteasomal degradation, CDK7 is dependent on transactivation by MITF or c-MYC to maintain a steady state. The dependence of TFIIH-CAK on sequence-specific MITF and c-MYC constitutes a previously unrecognized mechanism feeding into super-enhancer-driven or other oncogenic transcriptional circuitries, which supports the concept of a transcription-directed therapeutic intervention in melanoma.


Introduction
Survival, proliferation, and differentiation of the melanocytic lineage are determined by the Microphthalmia-associated transcription factor (MITF) [1][2][3]. MITF's pivotal role in lineage survival is underscored by the observations that its mutation entails a substantial loss of melanocyte viability and its expression is largely maintained upon malignant transformation [4,5]. MITF's impact on lineage survival is partially explained by direct transcriptional control of the anti-apoptotic BCL-2 [1]. In the cell cycle, MITF exerts opposing context-dependent functions by direct regulation of CDK2 vs. INK4 and p21Cip1 [2,6,7]. In regard to differentiation, MITF is essential to UV-protective pigmentation which is induced by the p53dependent Melanocyte Stimulating Hormone [3,8,9].
Mechanistically, it is still incompletely understood how MITF is linked to melanomagenesis. In 5-20% of human melanomas MITF undergoes genomic amplification and as such it acquires features of a lineage-survival oncogene [10]. In addition, a SUMOylation-defective MITF germline mutation MITF-E318K with increased transcriptional activity has been identified, which predisposes to familial and sporadic melanoma and renal cell carcinoma [11,12]. MITF's oncogenic role is further supported by its EWS-ATF1 dependent upregulation in clear cell sarcoma, which is indispensable for survival and growth of the sarcoma [13]. By contrast, a subset of bulk melanomas (<20%) reveal a low abundance of MITF, which has been linked to an invasive, treatment-resistant phenotype [14]. In addition, single-cell expression analyses identified melanoma cells with a low MITF/AXL ratio in MITF-high bulk melanomas, which may be able to evade senescence and confer treatment resistance [15,16].
Opposing data on MITF's role in UV-dependent DNA damage response pathways and genomic stability have been published and the mechanistic link between MITF and nucleotide excision repair (NER) has not been clearly defined [17,18]. Here we show that MITF impinges upon the functional interface of transcription and nucleotide excision repair (NER) embodied by the general transcription factor IIH [19,20]. TFIIH is a multi-protein complex that is composed of the helicases XPB and XPD, subunits GTF2H1 (p62), p52, p44, p34, p8 (TTDA) which form the core complex as well as the CDK-activating kinase (CAK) sub-complex that contains CDK7, CCNH, and the assembly factor MAT1 [20,21]. XPD bridges the core complex and the CAK sub-complex [22]. TFIIH is not only involved in basal transcription but also in nucleotide excision repair, transactivation of nuclear receptors and in the cell cycle through CDK7 activity of the CAK complex [23,24]. At mitosis CDK1/Cyclin B phosphorylates CDK7 at serine 164 resulting in transcription inhibition and CAK dissociates from TFIIH prompted by degradation of the bridging element XPD [25]. As a TFIIH core element GTF2H1 physically interacts with TFIIE of the transcriptional preinitiation complex [26]. In NER, GTF2H1 contacts DNA damage recognition factors XPC-HR23B and the 3ʹ-endonuclease XPG [27,28].
Our studies identify a previously unrecognized mechanism how a lineage-restricted sequence-specific transcription factor controls genomic and transcriptional homeostasis through transactivation of GTF2H1 as a key component of the TFIIH transcription/repair apparatus. Importantly, the MITF-GTF2H1 axis is preserved in MITFabundant melanomas with potential implications for primary tumor progression and macro-metastatic disease. Moreover, MITF regulates TFIIH kinase (CDK7), which is lost upon depletion of MITF but rescued by its structural homolog c-MYC thus adopting MITF's role in transcriptional homeostasis at the expense of a melanocyte-specific program. The dependence of TFIIH-CAK on the sequencespecific transcriptional master regulators MITF or MYC constitutes a vulnerability in melanoma.

MITF determines transcriptional activity and is linked to GTF2H1
To test the hypothesis whether MITF acts at the interface of DNA repair and transcription we first assessed the transcription rate of melanocytes upon UVB irradiation (UVR)mediated genotoxic attack by incorporation of 5-ethynyluridine (EU) in the presence or (near) absence of MITF [29]. UVR treatment of MITF-positive neonatal human epidermal melanocytes (NHEM) led to a substantial but transient reduction in transcriptional activity generally due to stalled RNA polymerases at DNA lesions and subsequent transcription-coupled repair (TCR) [30]. By contrast, loss of constitutive expression of MITF in telomerase (TERT)immortalized human melanocytes (IHM) was associated with low transcription rates notwithstanding a faster replication of IHM compared to NHEM (Fig. 1a). In addition, experimental depletion of MITF by RNA interference (RNAi) in 501 mel cells using multiple independent siRNA was associated with a strong reduction in basal transcription (Fig. 1b). On-target specificity of the most efficient MITFdirected siRNA (siMITF 1 ) was validated by retroviral expression of an RNA-interference resistant MITF carrying triple silent mutations, which abolished the senescence-like phenotype of MITF depletion ( Fig. 1c and Supplementary  Figures 1a-c) [17]. In analogy to MITF-negative IHM, UVR treatment of MITF-depleted 501 mel cells did not elicit an oscillatory transcriptional activity in contrast to control siRNA treated cells (Fig. 1d). In a reciprocal experiment, forced expression of MITF in MITF-negative A375 melanoma cells augmented basal transcriptional activity and transcriptional recovery after UVR (Fig. 1e). To screen for potential MITF target genes coding for general transcription and nucleotide excision repair (NER) factors, we analyzed transcriptome data on a panel of primary and metastatic melanomas, melanoma in situ and melanocytes (GSE7553). Masked by a dominating strictly MITF-dependent gene expression signature composed of MLANA, CDK2, PMEL, and other established MITF target genes in neighborhood analyses, we identified a subset of differentially regulated general transcription/NER genes using a less stringent comparative marker selection analysis (Supplementary Figure 2a). Among these we selected the general transcription factor 2H1 (GTF2H1) for further analysis, since it reflects the dual action of TFIIH in transcription initiation and NER [20,31]. The concordant expression of MITF and GTF2H1 transcripts was confirmed in several genetically heterogeneous melanoma lines by RT-PCR (Supplementary  Figure 2b and Supplementary Table 1). Moreover, immunohistochemical analyses of a large panel of primary human cutaneous melanomas (n = 136) spotted onto tissue microarrays (TMA) revealed a moderate but significant correlation between MITF and GTF2H1 at the protein level  Table 2). RNAi-mediated depletion of MITF in melanocytes from various donors and melanoma cells using several independent siRNAs and expression of an adenovirusdriven dominant-negative MITF mutant resulted in a substantial reduction in constitutively expressed GTF2H1 at mRNA and protein levels, in particular under conditions of serum starvation ( Fig. 1h and Supplementary Figures 2d-g). Upon UVR MITF-depletion was associated with markedly decreased GTF2H1 mRNA expression levels over time (Fig. 1i). At the protein level both MITF and GTF2H1 activities showed a clear induction after UVR treatment of primary melanocytes and 501 mel cells (Fig. 1j, k). Stem cell factor (SCF) has previously been shown to be secreted by UV-irradiated keratinocytes and to activate MITF via c-KIT/MAP kinase signaling in melanocytes [32,33]. To examine GTF2H1 regulation in this c-KIT dependent context, we stimulated 501 mel cells with SCF and observed a phosphorylation-dependent MITF mobility shift to the slower migrating, transcriptionally active form and an increase in GTF2H1 expression, respectively. Conversely, the observed SCF-triggered induction of GTF2H1 was completely abrogated upon depletion of MITF (Fig. 1l).

GTF2H1 is a direct transcriptional target of MITF
To assess the directness of the underlying regulatory mechanism, we scrutinized the human GTF2H1 promoter for transcription factor binding sites. Beside SP1 and cAMP response element (CRE) sites, we identified three conserved elements that fit the E box consensus sequence (t)CAC/ TGTG −34, −67, and −340 base pairs upstream of the transcriptional start site (TSS), which were occupied by MITF in ChIP assays (Fig. 2a, b). In cellular models devoid of MITF, the intact GTF2H1 promoter was clearly transactivated by exogenous MITF, exceeding CRE-binding protein (CREB)-mediated effects by treatment with the cAMP agonist forskolin (Fig. 2c). Individual site-specific mutation at the E boxes reduced GTF2H1 promoter activity and combined site-specific mutations at all three E boxes increased the repressive effect on reporter activity as demonstrated in murine and human melanoma cells (Fig.  2d, e). Notably, SUMOylation-defective MITF-E318K, previously implicated in predisposition to familial and sporadic melanoma, significantly enhanced GTF2H1 promoter activity compared to wild-type MITF (Fig. 2f) [11,12].

Regulation of GTF2H1 by MITF maintains transcriptional homeostasis
Beyond its essential role in TFIIH complex assembly and stability, the function of GTF2H1 with regard to Fig. 1 MITF regulates GTF2H1 and determines transcriptional output. a Left panel: time course of transcriptional activity detected by 5ethynyl-uridine (EU) incorporation in neonatal human epidermal melanocytes (NHEM) vs. TERT-immortalized human melanocytes (IHM) after UVB irradiation. Middle panel: fluorescence microscopy depicting EU incorporation of single cells. Images were taken at indicated time points after UVR. Nuclear staining by Hoechst 33342 (blue), EU (red), scale: 10 µm. Right panel: immunoblot shows differential expression levels of MITF protein in NHEM and IHM. Actin used as loading control. b MITF mRNA expression (left) and EU incorporation (right) of 501 mel cells after transfection with four different MITF-directed siRNA vs. control siSCR. c Proliferation of 501 mel cells upon siMITF 1 transfection and retroviral expression of wildtype MITF (MITF WT ) and siRNA-resistant mutant MITF (MITF RES ) vs. empty vector (EV). Graphs represent mean absorbance of crystal violet ± SD from technical triplicates (two-tailed unpaired t-test; **, p < 0.01; ***, p < 0.001). d EU incorporation of UVB irradiated 501 mel cells after siMITF 1 vs. siSCR transfection. Immunoblot shows MITF protein levels after siRNA transfection. e EU incorporation of UVB irradiated A375 cells after retroviral expression of wild-type MITF or EV. Immunoblots reflect retrovirus-mediated MITF protein expression compared to EV in A375 cells. Graphs in a, d, and e indicate mean ±SEM of fluorescence intensity in ≥100 nuclei (****, p < 10 -4 , twotailed unpaired t-test). f MITF and GTF2H1 protein expression in representative primary cutaneous melanomas (TMA, tissue microarray comprising n = 136 samples; scale: 50 µm). g Relationship between MITF and GTF2H1 protein expression in primary cutaneous melanomas (p = 0.0109 Pearson correlation coefficient). h Immunoblot analysis of MITF and GTF2H1 expression under siMITF 1 vs. siSCR control in NHEM. i GTF2H1 mRNA expression under siMITF 1 vs. control siRNA after UVB irradiation of 501 mel cells. Mean ± SD normalized to GAPDH is presented (**, p < 0.01; ***, p < 0.001; twotailed unpaired t-test). j Time course of MITF and GTF2H1 protein expression after UVB irradiation of 501 mel cells showing an increase in phosphorylated MITF (marked by upper arrowhead; lower arrowhead, unphosphorylated MITF) and in GTF2H1. k Immunoblot of MITF and GTF2H1 expression after UVR of NHEM including early time points (5ʹ, 15') displaying a rapid increase in phosphorylated MITF accompanied by an up-regulation of GTF2H1. l MITF and GTF2H1 protein expression after SCF stimulation in 501 mel cells transfected with siMITF 1 vs. control siSCR. Note rapid and transient shift of MITF to slower migrating phosphorylated form. j-l: lower and upper arrowheads indicate non-phosphorylated MITF at 54 kd and phosphorylated MITF at 60 kd, respectively. Experiments depicted in a-e, h-l were performed at least twice with very similar results transcriptional activity is poorly defined and it has not been linked to specific repair or transcription deficiency phenotypes in contrast to other TFIIH components such as XPB, XPD, and TTDA (GTF2H5) [34]. To independently confirm that differential regulation of GTF2H1 by MITF feeds back on global transcription, we performed RNA sequencing (RNA-seq) under siRNA-mediated knockdown of MITF or GTF2H1 vs. scrambled siRNA with spike-in normalization [35]. Transcriptomes were similarly composed and showed differential gene regulation upon normalization of relative expression levels, but revealed an overlapping almost universal repression upon depletion of MITF or GTF2H1 compared to controls as demonstrated by correction to cell equivalents (Supplementary Figure 3a). RNA-seq data were validated by qRT-PCR of randomly selected transcripts after normalizing mRNA input to cell number (Supplementary Figure 3b). In accordance with the established role of TFIIH in POLR2-and POLR1dependent transcription, experimental disruption of the MITF-GTF2H1 axis resulted in a decrease in total, messenger and ribosomal RNA (Supplementary Figure 3c) [36,37]. To minimize convolving cell cycle effects on global transcriptional output, we performed thymidine block experiments combined with RNA interference, which caused a significant expansion of non-cycling cell populations with low RNA content under MITF or GTF2H1 depletion compared to control siRNA . This finding indicates that MITF and its transcriptional target GTF2H1 have a profound impact on global transcriptional activity, which clearly exceeds the transcriptional demand of cell cycle activity (Supplementary Figure 3d).

MITF enhances nucleotide excision repair through regulation of GTF2H1
Given MITF's DNA damage responsiveness and control of GTF2H1 we sought to evaluate whether MITF activity impacts global NER capacity. To this end, we quantitatively assessed NER as a function of MITF or GTF2H1 by UV dosage-dependent unscheduled DNA synthesis (UDS) in primary melanocytes and melanoma cells utilizing a 5ethylene-2ʹ-deoxyuridine (EdU) incorporation assay [38], which revealed compromised repair rates under MITF as well as GTF2H1 knockdown. In addition, MITF-high primary melanocytes showed a significantly greater unscheduled DNA synthesis than immortalized MITF-low melanocytes (Fig. 3a, b and Supplementary Figure 4a). Accordingly, repair of cyclobutane pyrimidine dimers (CPD) as a predominant type of UV-induced DNA lesion as well as cisplatin-induced primary DNA mono-adducts and subsequently formed intrastrand crosslinks, both of which constitute NER substrates, was substantially delayed upon depletion of MITF or GTF2H1 in primary melanocytes and melanoma cells, respectively (Fig. 3c, and Supplementary  Figures 4b-d) [39]. The essential 3ʹ-endonuclease XPG, which physically interacts with GTF2H1 via a pleckstrin homology domain [28], was not recruited to DNA lesions after focal UVC-irradiation of MITF-depleted nuclei indicating a disassembly of NER complexes (Fig. 3d). Although NER was shown to be transcription-independent by simultaneous measurement of EU and EdU incorporation following UVR (50 J/m 2 ) and 1-h pulse treatment with the CDK9 inhibitor 5,6-Dichloro-1-β-D-ribofuranosylbenzimidazole (DRB) in the presence of MITF, secondary transcriptional effects of MITF or GTF2H1 depletion on NER factors cannot be ruled out with certainty ( Fig. 3e, f). Hence, we sought to assess NER in transcriptionally homeostatic melanoma cells with different expression levels of constitutive MITF. In a complementary gain-of-function experiment, we measured unscheduled DNA-synthesis after UVR upon retroviral expression of low-copy number MITF vs. empty vector control in MITFnegative A375 cells. In accordance with the observed reduction in UDS upon MITF and GTF2H1-directed RNA interference, we observed a strong positive association between the level of MITF expression and NER activity, which concurred with increased survival and colony formation of UV-irradiated MITF-abundant melanoma cells

Modulation of GTF2H1 affects growth of the melanocytic lineage
To assess whether depletion of GTF2H1 recapitulates the senescence-like phenotype induced by knockdown of MITF, we utilized several independent siRNAs directed against GTF2H1, which caused a significant growth inhibition in vitro in a variety of melanoma cell lines ( Fig. 4a and Supplementary Figures 5a-c). In long-term colony formation assays, lentiviral shRNA-mediated GTF2H1 or MITF depletion almost completely abrogated colony growth of 501 mel cells (Fig. 4b). In vivo, GTF2H1directed RNA interference repressed growth and dissemination of 501 mel cells in subcutaneous and distantly spread xenograft mouse models resulting in significantly superior tumor-free and overall survival of the GTF2H1-knockdown cohorts compared to the control cohorts in Kaplan-Meier analyses (Fig. 4c, d). In

MITF depletion causes loss of c-MYC and TFIIH kinase
By contrast, under experimental depletion of MITF in 501 mel cells recombinant GTF2H1 rescued only minimal transcriptional activity and cellular growth as shown by loss of phosphoactivities of RNA pol II CTD and CDK7, diminished EU incorporation, and minimal increase in proliferation (Fig. 5a-c and Supplementary Figure 6a). Strikingly, repression of MITF caused a substantial loss of MYC and CDK7 the latter of which was barely rescued by GTF2H1 (Fig. 5a). Several independent siMITF 1-3 almost completely abrogated MYC mRNA expression, whereas CDK7 transcripts were reduced by ∼2-fold (Supplementary Figure 6b). MITF-directed RNAi did not affect c-MYC mRNA expression in MITF-negative melanoma (A375) and non-melanocytic tumor cells (MCF7 breast cancer and ChIP primer annealing sites (colored arrows). TSS transcriptional start site, ATG, start codon. b ChIP, in vivo occupancy of MITF at GTF2H1 promoter in 501 mel cells. E box enrichment compared to intron 1. Mean ± SD from technical triplicates is presented. IgG (AB) was used as control antibody. c Normalized luciferase activity of GTF2H1 promoter after transfection with wild-type MITF vs. empty vector (EV) ± forskolin in A375 melanoma cells. Graphs represent mean ± SD from technical triplicates. d, e Normalized luciferase activity of GTF2H1 promoter containing E box site mutations in B16V or 501 mel cells ± forskolin. Mean ± SD from technical triplicates. b-e Experiments were performed twice with comparable results. f Transactivation of GTF2H1 promoter after wild-type MITF vs. SUMOylation-defective MITF-E318K transfection of HEK 293T cells as mean of fold-induction ± SD from 10 independent experiments (biological replicates). Right panel: immunoblot showing MITF protein levels in transfected HEK 293T cells. b-f *, p < 0.05; **, p < 0.01; ***, p < 0.001; two-tailed unpaired t-test) Analysis of transcriptional activity of 501 mel cells after 1-h DRB pulse treatment vs. DMSO control by measurement of EU incorporation. Graphs indicate mean ± SEM of fluorescence intensity in ≥100 nuclei. Depicted pictures were taken 6 h after UVR (scale: 25 µm). f UV-induced UDS assay in 501 mel cells in analogy to e. Graphs indicate mean ± SEM of fluorescence intensity in ≥100 nuclei (scale: 25 µm). g UV-induced UDS assay in melanoma cells with low vs. high abundance of MITF. Graphs indicate mean ± SEM of fluorescence intensity in ≥100 non-replicating nuclei. h Clonogenic cell survival assay of UV-irradiated MITF-abundant 501 mel cells vs. MITFnegative A375 cells. Graph represents mean ± SD from technical triplicates. a-h Experiments were done at least twice with similar results. All graphs were analyzed with a two-tailed unpaired t-test. **, p < 0.01, ****, p < 0.0001, n.s. not significant  Figures 6d and e).
To further explore the loss of CDK7 mechanistically, we first focused on posttranslational modifications leading to destabilization of CDK7. To this end, we depleted the direct MITF target CDK2, which has been shown to phosphorylate CDK7 on Thr170 and Ser164 in vitro as a potentially activating feedback loop, without discernible effect on CDK7 expression (Fig. 5d) [41]. Under the hypothesis that CDK7 is targeted for proteasomal degradation we utilized the proteasome inhibitors MG132 or bortezomib, both of which caused a marked increase in FLAG-tagged CDK7ubiquitin conjugates immunoprecipitated from 501 mel cells (Fig. 5e). Accordingly, MG132 treatment of melanoma cells caused an accumulation of CDK7 and rescued its expression under experimental depletion of MITF or c-MYC, the latter of which is an established target of several ubiquitin ligases (Fig. 5f) [42]. As anticipated, proteasome inhibition resulted in a substantial accumulation of c-MYC in the presence and upon depletion of MITF as well as MYC-directed RNAi (Fig. 5f).
As a next step, we asked whether MITF or c-MYC could stabilize CDK7 through a posttranscriptional mechanism mediated by protein-protein interaction. To this end, we engineered a series of wild type and deletion mutant HAtagged MITF and FLAG-tagged CDK7 fusion proteins for eukaryotic expression. In addition, we generated GST-CDK7 and wild type and deletion mutant FLAG-tagged MITF or c-MYC fusion proteins for prokaryotic expression. Immunoprecipitation experiments showed binding of recombinant CDK7-FLAG to wild type and DNA-binding-domain (DBD)-mutated MITF (MITF-HA R215Δ) (Supplementary Figure 7a). The interaction of MITF and CDK7 was independently confirmed utilizing GST-CDK7 (Supplementary Figure 7b). Further mutagenesis of MITF revealed binding of N-and C-terminal domains of MITF to GST-CDK7 (Supplementary Figure 7c). In similar experiments, we employed wild type and FLAG-tagged c-MYC deletion mutants which demonstrated binding of MYC box II transactivation domain to CDK7-GST ( Supplementary  Figures 7d and e). However, in rescue experiments utilizing a MYC wild type expression construct in comparison to transcription-and dimerization-deficient mutants of MITF or MYC with intact CDK7-binding domains, CDK7 was only preserved upon expression of transcription-competent endogenous MITF or recombinant wild type c-MYC (Fig.  5g). This finding clearly indicates a transcription-dependent regulation of CDK7 by MITF or MYC without discernible impact of the observed protein-protein interaction on CDK7 stability.

MYC rescues general transcription in the absence of MITF
In accordance with the aforementioned, retroviral expression of c-MYC preserved CDK7 at mRNA and protein levels, restored transcriptional activity, and prevented a senescence-like phenotype as shown by reduced β-galactosidase staining and p27 downregulation (Fig. 6a-e). As predicted, c-MYC was unable to maintain transcription of melanocyte-restricted MITF target genes resulting in dedifferentiation as demonstrated for PMEL, MLANA, and TYRP1 pigmentation genes (Supplementary Figure 8a). Since the relationship between MITF and c-MYC had not been clearly defined yet, we analyzed accessible single-cell RNA-seq data from a panel of metastatic melanomas, which showed a strong positive correlation between the expression of MITF and c-MYC transcripts in individual cells of a given heterogeneous tumor cell population and across various tumor specimens (Fig. 6f) (GSE72056) [16]. The concordant expression of MITF and c-MYC transcripts was confirmed in a panel of genetically heterogeneous melanoma lines by RT-PCR (Supplementary Figure 8b). Notably, at the protein level MYC appeared to be decoupled from MITF in ∼15-20% of metastatic human melanoma samples (n = 18) and melanoma cell lines (n = 7) reflected by a low MITF/MYC expression ratio in whole cell lysates as demonstrated in immunoblot analyses ( Supplementary  Figures 8c and d).
Previously, we identified fuse binding protein 2 (FUBP2) as a potential MITF-target gene, which belongs to the family of far upstream element binding factors implicated in pulse-regulation of c-MYC through TFIIH [1,43]. Hence, Fig. 4 Disruption of MITF-GTF2H1 axis represses melanoma growth. a Proliferation of 501 mel cells after transfection with siMITF 1 or three different siGTF2H1 1-3 vs. siSCR. Graphs represent mean ± SD of crystal violet (CV) absorbance from technical triplicates (two-tailed unpaired t-test). Right panels, mRNA expression of MITF and GTF2H1 under RNAi as indicated. b Colony formation assay of transduced 501 mel cells expressing shRNA directed against MITF or GTF2H1 compared to shSCR control. Right panel, bars represent mean ± SD mean of the number of colonies from technical triplicates (two-tailed unpaired t-test). a, b Biological replicates revealed similar results. c End point analysis of tumor growth in shGTF2H1 vs. shSCR expressing melanoma xenografts after subcutaneous transplantation into SCID mice. Tumor growth in shGTF2H1 cohort (n = 3 of 9 animals) was associated with insufficient repression of GTF2H1 determined by expression analysis in FACS-sorted GFP-positive tumor cells (one-tailed unpaired t-test). Right panel, Kaplan-Meier analysis of tumor-free survival. shGTF2H1 cohort, n = 9 mice; shSCR cohort, n = 10 mice. d Immunohistochemical analysis exhibiting pulmonary spread of genetically modified shGTF2H1 vs. shSCR expressing melanoma cells after tail vein injection. H&E, hematoxylin & eosin staining; MITF, C5 monoclonal antibody (red signal). Arrowheads, melanoma cell infiltration. Right panel, Kaplan-Meier analysis of overall survival. shGTF2H1 cohort, n = 5 mice; shSCR cohort, n = 6 mice. P, log-rank test we hypothesized that MITF-dependent regulation of FUBP2 might at least partially affect MYC expression. In agreement with our hypothesis, in silico ChIP-seq analyses identified occupancy of MITF at proximal E box containing promoter sequences of FUBP2 (KHSRP) in primary melanocytes and melanoma cell lines (Fig. 6g). MITF depletion

TFIIH kinase is transactivated by MITF and MYC in the melanocytic lineage
To explore the regulatory mechanism of CDK7 transcription, we first treated several MITF-dependent melanoma lines under MITF-directed RNAi with 12-O-tetradecanoylphorbol-13-acetate (TPA), a potent activator of protein kinase C. TPA leads to activation of MAP kinase and subsequent phosphorylation of MITF S73 which mediates upregulation of transcriptional activity of MITF [33]. Across all melanoma lines tested, depletion of MITF was associated with a significant decrease in CDK7 transcription over time. Successful activation of MITF by TPA treatment was confirmed by immunoblot analysis which demonstrated a rapid shift of MITF protein to the slower migrating form (∼60kd) followed by consecutive upregulation of MYC and CDK7 (Fig. 7a, b). Next, in silico ChIP-seq analyses revealed binding of MITF to an E box (tCACGTG) consensus sequence −86 base pairs upstream of the transcriptional start site of CDK7 in primary melanocytes, primary melanoma and melanoma cell lines. In analogy, we identified MYC enrichment at the identical site in the CDK7 promoter in various ChIP-seq data sets from breast cancer and B-cell malignancies suggesting that both MITF and MYC are able to directly transactivate CDK7 (Fig. 7c). To provide evidence that MITF and MYC occupy the CDK7 promoter in the melanocytic lineage, we performed ChIPexperiments in MYC-overexpressing vs. empty vector control expressing 501 mel cells confirming enrichment of both MITF and MYC at the E box sequence in the CDK7 promoter (Fig. 7d, e). Interestingly, endogenous MITF also bound the CDK7 promoter under conditions of c-MYC overexpression (Fig. 7d). Likewise, c-MYC bound the CDK7 promoter in the presence of MITF, but also under conditions of MITF depletion (Fig. 7e). In regard to simultaneous presence of MITF and MYC at a given CDK7 promoter, these findings could implicate higher affinity binding of MITF but MYC might be able to replace gradually decreasing MITF in the regulation of CDK7.

Discussion
TFIIH is a major effector of global genome repair as well as transcription-coupled repair (TCR) [44]. By direct regulation of GTF2H1 MITF controls both NER pathways as demonstrated here by an augmented unscheduled DNA synthesis and recovery of RNA synthesis after UVR utilizing 5-ethynyl-2'-uridine incorporation as a surrogate of TCR under conditions involving highly abundant MITF. In addition, the observed loss of CDK7 under experimental depletion of MITF abrogates the restart of transcription after completion of repair [22]. MITF's impact on global transcription superimposes over TCR and prevents a formal discrimination between direct and indirect MITF-dependent effects on strand-specific repair, but RNAi-mediated repression of the TCR-specific helicase Cockayne syndrome protein B (CSB) in melanoma cells recapitulated the NER phenotype under MITF depletion associated with a marked decrease in unscheduled DNA synthesis <12 h after UVR (Supplementary Figure 9) [45]. The experimental evidence presented here primarily pinpoints GTF2H1 as a transcriptional target of MITF activity accounting for the observed NER as well as general transcription phenotypes modulated by MITF, however additional targets operating at the interface of repair and transcription could be involved in particular with regard to TCR.
Within a given tumor cell population the dichotomous behavior of MITF shuttling between low and high expression levels is largely enigmatic, but it may reflect invasive or dormant vs. proliferative states with varying transcriptional demands [14]. Invasive behavior has been associated with high expression of AXL and therapy resistance [15,16]. However, a proliferative state is essential to primary tumor progression and macro-metastatic manifestation. In this regard, it has been unclear how melanoma cells with low MITF survive and proliferate in vivo, the majority of which apparently do not exhibit elevated AXL protein Fig. 6 c-MYC rescues general transcription and prevents senescence in the absence of MITF. a Immunoblot analysis of 501 mel cells under retrovirus-driven c-MYC expression compared to EV and subsequent siMITF 1 vs. siSCR transfection. b MITF, c-MYC, and CDK7 mRNA expression in 501 mel cells under conditions analogous to a. Relative expression was measured by qRT-PCR, normalized to GAPDH and given as mean ± SD from technical triplicates (two-tailed unpaired ttest). c Transcriptional activity measured as EU incorporation in 501 mel cells in analogy to a and b. Graphs indicate mean ±SEM of fluorescence intensity in ≥250 nuclei (two-tailed unpaired t-test). d Proliferation of 501 mel cells in analogy to a-c. Graphs represent mean ± SD of crystal violet absorbance from technical triplicates (twotailed unpaired t-test). e SA-β-gal positivity in 501 mel cells at day 3 in analogy to d. Data represent mean ± SD from technical triplicates (two-tailed unpaired t-test). Micrographs display SA-β-gal signals and Hoechst 33342 nuclear staining under corresponding conditions (scale: 50 µm). f Regression analysis of MITF and MYC single-cell mRNA expression from a panel of metastatic melanomas (GSE72056). g ChIP-seq tracks of MITF-binding signals at FUBP2/KHSRP gene locus in primary melanocytes, and 501 mel and COLO829 BT168F melanoma cell lines. Green arrowheads indicate MITF-binding consensus sequences (E box) at the promoter and first intronic region of the FUBP2/KHSRP gene. GEO accession numbers are listed under Materials and Methods. h MITF, c-MYC, and FUBP2 mRNA expression in 501 mel cells after siSCR, siMITF 1 , or siFUBP2 transfection. Relative expression was measured by qRT-PCR, normalized to GAPDH and given as mean ± SD from technical triplicates (twotailed unpaired t-test). i Immunofluorescence and immunoblot analyses of MITF, MYC, and FUBP2 protein expression in analogy to h. DRAQ5 used for nuclear staining and actin used as loading control (two-tailed unpaired t-test of c-MYC fluorescence intensity in 100 nuclei is indicated). a-e, h, i Experiments were repeated twice with comparable results expression [16]. In contrast to the correlative association between MITF and c-MYC in single-cell RNA-seq data sets on metastatic melanoma, preliminary data indicate that expression of c-MYC appears to be decoupled from MITF at the protein level in bulk MITF-low tumor cells and TERT-immortalized melanocytes ( Supplementary Figures  8c and d). We propose that a gradual loss of MITF as sometimes observed during melanoma progression might be supplanted by a variable decrease in MYC turnover rather than transcriptional upregulation allowing for accumulation of MYC protein, which preserves CDK7 and acts as a compensatory switch mechanism to maintain transcriptional homeostasis and to modulate proliferative activity at the expense of melanocyte-specific gene activation (Fig. 6a-e and Supplementary Figure 8a). In accordance, c-MYC has previously been implicated in suppression of BRAF V600Einduced senescence in melanocytes and melanoma progression, while loss of CDK7 results in senescence [46,47]. Transcriptional downregulation of MYC might account for the resistance of melanoma cells toward BET bromodomain inhibition by the small molecule JQ1, which has been shown to downregulate MYC transcription and to disrupt the MYC-dependent transcriptional program in multiple myeloma, resulting in potent antiproliferative effects as opposed to the melanoma models examined in this study (Supplementary Figure 8f) [48]. Nevertheless, transcriptional re-wiring of c-MYC has recently been described in BRAF-inhibitor resistant melanoma sensitizing to bromodomain inhibition [49]. However the MITF/MYC ratio may be, the transcription dependency of melanoma cells is not significantly altered by a gradual decrease in MITF expression and concomitant de-differentiation as reflected by similar sensitivities of melanoma cells with low or high abundance of MITF toward the covalent CDK7 inhibitor THZ1. Our studies identify c-MYC as a surrogate of MITF which directly activates CDK7 likely accounting for the non-selective efficacy of THZ1 as observed here and previously by others in the context of super-enhancer-driven oncogenesis ( Supplementary Figures 8g and h) [50,51].
Together, our findings place MITF as a safeguard into the UVR-induced DNA-damage response pathway activating NER and transcription, which represent a significant conceptual advance in our understanding of global transcriptional regulation in the melanocytic lineage under genotoxic stress (Fig. 8). Moreover, the dependency of TFIIH-CAK on the melanocyte master regulator MITF or its structural homolog c-MYC extends the conceptional idea of a CDK7-targeted treatment approach to melanoma far beyond super-enhancers as determinants of oncogenesis. Primary human melanocytes were cultured in MCDB153 media (Sigma) with 4% FCS, 1% penicillin/streptomycin, 5 mg/ml insulin, 1 mg/ml transferrin, 0.6 ng/ml human basic FGF, 10 ng/ml TPA and 13 mg/ml BPE or TICVA media containing Ham's F10, 7% FBS, 1% penicillin/streptomycin, 2 mM glutamine, 50 ng/ml TPA, 0.1 mM IBMX, Na 3 VO 4 , and 1 mM dbcAMP. IHM were cultured in Pri-Grow II medium (ABM) with 10% FBS. Cells were grown at 37°C and 5% CO 2 . Authenticity of human cell lines has been confirmed by STR analysis. Species of animal lines was confirmed by Cytochrome-C Oxidase I DNA barcoding (DSMZ, Braunschweig). Cell lines were free of mycoplasma contamination.

Real-time RT-PCR analysis
Total RNA was isolated using Trizol (Invitrogen) or RNeasy Plus Mini Kit (Qiagen) and converted into cDNA by reverse transcriptase (Invitrogen or Promega). cDNA expression was quantified using FastStart DNA Master PLUS SYBR Green I Kit (Roche) on the Light Cycler 480 machine (Roche). Data were normalized to GAPDH or B2M expression for reproduction. Gene specific primer sets are available on request.

GST pull down
Full length versions and deletion mutants of human MITF, MYC, and CDK7 were expressed as GST-fusion proteins in AVB100 bacteria. For MITF and MYC, the GST-tag was removed by digestion with Prescission protease (GE Healthcare, 27-0843-01). For GST pull down, 12.5 µg of GST-CDK7 were bound to 10 µl of GSH-sepharose (GE Healthcare, 17075601) in the presence of protease inhibitors for 1 h at RT. Subsequently, beads were washed (2×) with PBS, blocked with 1% BSA in PBS for 30 min at RT, incubated for 1.5 h at 4°C with 3-6 µg GST-free MITF and MYC, respectively, washed with PBS and eluted with 20 mM GSH (Sigma, 4251) in 100 mM Tris-HCl (pH 8.0) for Western blot analysis; free GST served as binding control.

UV-induced unscheduled DNA synthesis (UDS)
5-ethynyl-2ʹ-deoxyuridine (EdU) incorporation was used to quantify UDS. Cultured cells on coverslips were irradiated at different doses of UVB (10-100 J/m 2 ) in a CL-1000 Ultraviolet Crosslinker (UVP). Subsequently, cells were incubated with serum-free media supplemented with 10 µM EdU and 1 µM of fluorodeoxyuridine (FdU) for 6 h. Cells were then fixed with 3.7 % paraformaldehyde, permeabilized with 0.5% triton X-100 and after washing with PBS acid extraction was performed using Bouin's fixative to reduce background signals. Incorporated EdU was detected using the Click-iT EdU Imaging Kit (C10337, Invitrogen) following the manufacturer's recommendations. Images were obtained using a fluorescence microscope (Leica DMIL). Fluorescence signal intensity was measured in ≥100 nuclei and expressed in arbitrary units (AU) using ImageJ software.

Measurement of Pt-(GpG) adducts in DNA
Plated cells were treated with cisplatin (20 µg/ml) for 4 h and maintained in drug-free media for up to 48 h. Cell aliquots were harvested, immunostaining and analysis were performed as described before [53]. Mean values ± 95% confidence intervals of ≥100 cells per sample were calculated and expressed as arbitrary fluorescence units (AFU).

Assessment of RNA synthesis
After UVR cells were incubated with 1 mM 5-ethynyluridine (EU) for 1 h and subsequently fixated with 3.7% formaldehyde in PBS for 15 min. EU incorporated into newly synthesized RNA was detected using the Click-iT RNA Alexa Fluor 594 Imaging Kit (Invitrogen) following the manufacturer's recommendations. Images were obtained using a fluorescence microscope (Leica DMIL).
Fluorescence signal intensity was measured in ≥100 nuclei and expressed in arbitrary units using ImageJ software. Total, mRNA and rRNA were measured in triplicates by spectrophotometry (NanoDrop) and fluorescence electropherogram on a Bioanalyzer instrument (Agilent) and normalized to cell numbers. To determine RNA content per cell, cells were stained using the metachromatic dye acridine orange (AO, Sigma Aldrich) at a concentration of 10 μg/ml and subsequently analyzed by flow cytometry on a BD FACS Canto.

Gene expression array analysis
The gene expression profiles on melanoma tissues were downloaded from the Gene Expression Omnibus (GEO) database (GSE7553). Primary and metastatic melanoma samples (supplemented by melanoma in situ and epidermal normal melanocytes) were separated into two classes based on MITF expression. A sample was considered MITF-high, if it exhibited a signal greater than 4000 from the probe set 207233_s_at. Samples were sorted according to MITF expression level and sample type. Comparative Marker Selection analysis was carried out with GenePattern (http:// genepattern.broadinstitute.org/gp) with 10,000 permutations. Probe sets with P-values < 0.05 are considered significant. Heat maps were generated with GenePattern Heatmap Viewer.

GTF2H1 promoter mutagenesis and luciferase reporter assays
A fragment of the human GTF2H1 promoter (bp −882 to +159, relative to the transcription start site) was generated by PCR and cloned into the pGl-3 basic vector (Promega) upstream of the luciferase reporter gene. In addition, three E box sequences in the −882/+159 bp GTF2H1 promoter were mutated from CACGTG to GAGGTG using sitedirected mutagenesis (Stratagene). GTF2H1 promoter constructs were co-transfected with pGL4.73 [hRluc/SV40] plasmids (Promega) upon serum withdrawal. A wild-type MITF pcDNA3.1 expression vector vs. control vector was co-transfected into A375 melanoma cells. Forskolin (Sigma Aldrich) was given as indicated above. To assess the responsiveness of GTF2H1 promoter after overexpression of wild-type MITF vs. SUMOylation-defective MITF, wildtype MITF or SUMOylation-defective MITF-E318K pCMX-PL2 expression vector vs. control vector was cotransfected into HEK 293T cells. Depending on the experimental context readout was performed 48 or 72 h after transfection using the Dual Luciferase Reporter Assay System (Promega) according to the manufacturer´s protocol.

Next-generation RNA sequencing
Next-generation RNA sequencing and spike-in normalization were done as previously described [35]. Library preparation was performed with NEBNext® Ultra™ RNA Library Prep Kit for Illumina® (NEB E7530) following manufacturer's recommendations using 640 ng total RNA each. ERCC RNA spike-In reference sequences (Ambion, 4456739) were incorporated into the human reference assembly (Ensembl GRCh38.77). TopHat (v2.0.13) was used to align reads to resulting reference sequences and read counts were calculated with HTSeq (v0.6.1) [54,55]. Subsequent processing was done with DESeq2 and R [56,57]. For the assessment of total expression, data were normalized based on reads aligned to ERCC sequences. For measuring differential expression between siMITF and siGTF2H1, size factors were estimated by DeSeq2 without taking the ERCC sequences into account. RNA-seq data reported in this article have been deposited in the European Nucleotide Archive and are available under the accession code PRJEB30337.
Proliferation, senescence, colony formation, and clonogenic cell survival assay To measure proliferation, cells were seeded at low confluence in 24-well plates in triplicate and fixed in 3.7% paraformaldehyde on indicated days for subsequent staining with crystal violet (CV) 0.05%. After washing the content of cellular CV was extracted with acetic acid 10% and absorbance was measured at 595 nm. For growth-rescue experiments retroviral vector pBABE-puro-GTF2H1 or pBABE-puro-c-MYC vs. pBABE-puro-empty were employed. For quantitative assessment of senescence, 501 mel cells were transduced with different retroviral vectors (pBABE-puro-MITF WT , pBABE-puro-MITF RES , or pBABE-puro-MYC vs. pBABE-puro-empty) followed by puromycin selection and subsequent transfection with siMITF vs. siSCR RNA. Cells were seeded in 24-well plates in triplicate and kept in culture for three days. SA-βgal staining was carried out using the senescence β-galactosidase staining Kit (Cell Signaling) following the manufacturer's recommendations. For assessment of clonogenic growth in soft agar pLKO.1-puro-CMV-tGFP shRNA lentiviruses expressing scrambled, MITF or GTF2H1 shRNA were generated using HEK 293T cells co-transfected with compatible packaging plasmids. 501 mel cells were infected with lentivirus followed by puromycin selection. Transduced 501 mel cells were grown in 0.35% soft agar media for 30 days. The number of colonies per plate was quantified after staining with 0.05% crystal violet. To determine clonogenic cell survival, 501 mel and A375 cells were seeded in six-well plates (500, 750 or 1000 cells per well), and UVB-irradiated at the indicated doses. After irradiation cells were maintained in complete media at 37°C and 5% CO 2 for 12 days. Subsequently, cells were fixed with 3.7% paraformaldehyde and stained using 0.05% crystal violet for quantification of colonies. The percentage of survival was calculated relative to non-irradiated cells.

TMA staining
Tissue microarrays (TMA) comprised n = 140 primary cutaneous melanomas (MM). Immunostaining of MITF, S-100 and GTF2H1 was done in 136 out of 140 eligible TMA-spotted melanoma samples. For immunohistochemistry (IHC) we used anti-human MITF (Dako, M3621) or anti-GTF2H1 (AbD, MCA4041Z) at a dilution of 1:100 for 40 min at RT or overnight at 4°C. Permanent AP Red (Zytomed, ZUC001-125) was used for detection and slides were counterstained by hematoxylin. AP red signal frequencies and intensities were evaluated semi-quantitatively by assignment of an expression score (0, negative; >0 ≤ 1.5, low; >1.5 ≤ 2.5, intermediate; >2.5 high). IHC analyses were performed in a blinded fashion by two independent examiners. IHC staining of MITF was re-validated by immunofluorescence co-staining of MITF and S-100 for unambiguous detection of melanoma cells (Supplementary Figure 2c). Slides were incubated at 4°C overnight with anti-human MITF (Dako, M3621) and anti-S-100 (Progen, 16100) antibodies at a dilution of 1:100 and 1:200, respectively. Anti-mouse Alexa 488-coupled and anti-rabbit Alexa 555-coupled (Molecular Probes Europe BV) were used as secondary antibodies. Nuclei were stained by DAPI. Isotype-specific antibodies were used as negative controls.

Animal studies
All animal studies were performed according to ARRIVE guidelines and regulations of institutional animal care and use after approval by the Institutional Animal Care and Use Committee of the City of Hamburg (V1/591-00. 33 No. 131/ 13). Severe combined immunodeficiency disorder (SCID) female 8-to 12-week-old mice (strain: Prkdcscid) (Charles River Laboratories) were used as allograft recipients. 501 mel cells were transduced with SCR-shRNA or GTF2H1-shRNA expressing lentiviruses (pLKO.1-puro-CMV-tGFP shRNA; Sigma-Aldrich). Viability of transduced cells (>90%) was verified at the day of transplantation. Genemodified cells were either injected into subcutaneous tissue of the dorsolateral flank (2×10 6 cells) or into the tail vein (1×10 6 cells). Animals were not randomized with regard to treatment allocation. Two animals were excluded from further analyses (one animal each in subcutaneous and IV xenotransplantation experiments from GTF2H1 knockdown cohorts) due to fatal adverse events during general anesthesia. Subcutaneous tumors were measured manually in two diameters. Tumor volume was calculated according to the formula: (width 2 × length)/2. Weight loss of ≥20% or ulceration after tumor engraftment was defined as stopping rule. End point was defined by tumor burden according to clinical assessment score or end of follow-up in the absence of tumor growth (day 117 after subcutaneous injection). Primary tumors and lung tissue were harvested and processed according to standard procedures. An approximate number of pulmonary (micro)metastases was determined by examining ten sections from the right lung of each animal separated by a minimum distance of 40 nm taken from the middle of each paraffin block. Pulmonary metastases were counted by two independent examiners; mean of ten sections was multiplied with the number of sections and correction factor 0.8 in order to estimate the total number of distantly spread cells per lung. Cell clusters consisting of ≤2 cells and intravascular cell clusters were excluded from the analysis.

Statistical analyses
With regard to the tissue microarray Fisher's exact test and chi-square-test were used to analyze frequency tables. Mann-Whitney Rank Sum test was used to compare groups of patients (25%/75% percentile) and Spearman correlation coefficient was calculated to describe the correlation of parameters. In mouse experiments, tumor-free as well as overall survival was calculated from the date of xenotransplantation to death. Distant spread of melanoma cells in IV xenotransplantation studies was assessed in a blinded fashion by two independent investigators. Kaplan-Meier analyses were performed to evaluate the association between survival and gene-modulating intervention. Logrank tests were applied to assess statistical significance using Graph Pad Prism software. Sample size was estimated according to N g,min = 2×[(t 2alpha + t 2beta )/(D rel/s )] 2 with P1 (alpha) = 0.05 and P2 (beta) = 0.2 to achieve a power of 80%. One-tailed unpaired Student's t-test, two-tailed paired Student's t-test or unpaired Student's t-test was used to perform other statistical analyses of parametric data except for Cspl-induced Pt-(GpG) intrastrand adducts, which were tested for 95% CI. In general, distribution of data was evaluated for ≥100 parametric data points if available. Distribution/variance of data was evaluated using F-test. In case of unequal distribution non-parametric tests were applied such as Mann-Whitney U-test (two-tailed). Experimental samples/data were not excluded from the analyses except for two mice inoculated with shGTF2H1expressing 501 mel cells which inadvertently died during general anesthesia.