MITF controls the TCA cycle to modulate the melanoma hypoxia response

Abstract In response to the dynamic intra‐tumor microenvironment, melanoma cells adopt distinct phenotypic states associated with differential expression of the microphthalmia‐associated transcription factor (MITF). The response to hypoxia is driven by hypoxia‐inducible transcription factors (HIFs) that reprogram metabolism and promote angiogenesis. HIF1α indirectly represses MITF that can activate HIF1α expression. Although HIF and MITF share a highly related DNA‐binding specificity, it is unclear whether they co‐regulate subset of target genes. Moreover, the genomewide impact of hypoxia on melanoma and whether melanoma cell lines representing different phenotypic states exhibit distinct hypoxic responses is unknown. Here we show that three different melanoma cell lines exhibit widely different hypoxia responses with only a core 23 genes regulated in common after 12 hr in hypoxia. Surprisingly, under hypoxia MITF is transiently up‐regulated by HIF1α and co‐regulates a subset of HIF targets including VEGFA. Significantly, we also show that MITF represses itself and also regulates SDHB to control the TCA cycle and suppress pseudo‐hypoxia. Our results reveal a previously unsuspected role for MITF in metabolism and the network of factors underpinning the hypoxic response in melanoma.


RNA-seq experiments
Understanding the complex interplay between the microenvironment and cancer cells, and more specifically the molecular events underpinning the transition between phenotypic states, termed phenotype switching, can provide therapeutic opportunities. Unlike genetic lesions that are fixed, phenotype switching is dynamic and reversible and is therefore potentially amenable to therapies directed toward inducing cells to convert from drug-resistant to drugsensitive states (Gupta et al., 2009;Saez-Ayala et al., 2013). As such, understanding how microenvironmental cues drive cells toward specific phenotypes, and whether different phenotypic states exhibit specific therapeutic vulnerabilities is an important issue.
In addition to nutrient availability and signals from infiltrating immune cells and the stroma, one of the major intra-tumor microenvironment signals is hypoxia. Tumor growth is associated with poorly organized vasculature leading to reduced oxygen delivery that can fail to meet the demands of tumor cells, and hypoxia is associated with worse prognosis (Bertout, Patel, & Simon, 2008). Hypoxia impacts metabolism (Marchiq & Pouyssegur, 2016), causing a switch away from oxidative phosphorylation toward glycolysis (Semenza, 2013), and can also promote metastasis (Semenza, 2012). In response to low oxygen, cells mount an adaptive response leading to the activation of a set of hypoxia-inducible factors (HIFs) (Ivan et al., 2001;Jaakkola et al., 2001;Mahon, Hirota, & Semenza, 2001). The HIF transcription factors, HIF1α, HIF1β and HIF2α, then drive a program of gene expression directed toward mitigating the effects of hypoxia, including regulation of pH of the extracellular environment and promotion of de novo blood vessel growth (Marchiq & Pouyssegur, 2016;Semenza, 2013). However, increasing evidence appears to suggest that different cell types may exhibit specific hypoxia responses, since there appears to be little overlap (2%-28%) between genes differentially expressed under hypoxia in cells of different origins (Benita et al., 2009;Chi et al., 2006;Denko et al., 2003;Loftus et al., 2017;Widmer et al., 2013).
Indeed, in recognition of its key role in promoting melanomagenesis, MITF has been termed a lineage survival oncogene (Garraway et al., 2005). Given the importance of MITF in determining the phenotypic state of melanoma cells, there is considerable interest in understanding how it might be regulated by the intra-tumor microenvironment.
Consequently, several studies have examined the role of hypoxia in melanoma. Hypoxia leads to transcriptional silencing of MITF via an indirect mechanism involving HIF-mediated up-regulation of the transcription factor bHLHE40/DEC1 that then represses MITF expression (Cheli, Giuliano, et al., 2011a;Feige et al., 2011). Consistent with this, hypoxia can promote increased invasion and de-differentiation in proliferative phenotype melanomas, but not those with a pre-existing invasive phenotype (Widmer et al., 2013). Stabilization of HIF reportedly promotes invasion, with increased HIF-dependent metastasis formation requiring platelet-derived growth factor receptor alpha (PDGFRA) and focal adhesion kinase (FAK) (Hanna et al., 2013). Moreover, hypoxia has been suggested to drive a switch from proliferation-associated receptor tyrosine kinase-like orphan receptor 1 (ROR1) expression to invasion-associated ROR2 expression (O'Connell et al., 2013).
Yet despite these advances, several key questions regarding the impact of hypoxia in melanoma remain. While previous studies have focused on a few selected genes (Widmer et al., 2013), or the hypoxia response in mouse melanocytes (Loftus et al., 2017), the impact of hypoxia on genomewide gene expression or which genes are direct targets of the HIF family in melanoma remains unknown. Nor is it clear whether melanomas exhibit a melanoma-specific hypoxia signature compared to hypoxia in other cell types, whether some melanoma cells with distinct phenotypes exhibit a differential response to hypoxia, or whether MITF may contribute to the adaptive response to hypoxia. Neither is it known if a constitutive pseudo-hypoxia gene expression signature found in some other cancer types under normoxic conditions exists in some melanomas.

| Cell lines
All cell lines were grown at 37°C with 10% CO 2 in RPMI-1640 (Gibco BRL, Invitrogen) supplemented with penicillin and streptomycin, 10% fetal bovine serum (FBS, Biosera). For starvation experiments dialyzed serum was used. The 501mel cells expressing the shRNA against HIF1α were constructed using HIF pGIPZ constructs (Open Biosystems/Thermo). The 501mel iMITF cell line was constructed as described previously (Falletta et al., 2017

| Western blotting
Hot SDS-PAGE loading buffer (78.0 mM Tris [pH 6.8], 4% SDS, 20% glycerol, 0.2% bromophenol blue, supplemented with 100 mM DTT) was used to lyse cells before being subjected to SDS-PAGE using 12% acrylamide. Proteins transferred to nitrocellulose membranes (Amersham Biosciences) that were blocked with 5% non-fat milk, in PBS containing 0.1% Tween-20 before probing with primary antibodies (see below) overnight at 4°C. Proteins were detected using anti-mouse, anti-rabbit, or anti-goat immunoglobulin coupled to horseradish peroxidase (Bio-Rad, Santa Cruz) and visualized using an ECL detection kit (Amersham Biosciences) and X-ray film (Fuji).

| Invasiveness assays
Matrigel invasion assays were performed using an invasion chamber from BD Biocoat. Cells were seeded at 2 × 10 5 per insert and cultured overnight in triplicate before treatment with DMOG. After 48-hr incubation, cells remaining above the insert membrane were removed by gentle scraping with a sterile cotton swab. Cells that invaded through the Matrigel to the bottom of the insert were fixed in ethanol for 10 min, washed in PBS, and stained with methylene blue. The insert was then washed in PBS, air-dried, and invading cells counted.

| Antibodies
The primary antibodies used were as follows: mouse anti-HIF1α

| Succinate calorimetric assay
Intracellular succinate level was measured using Succinate Assay Kit (Colorimetric) (Abcam #ab204718) as per manufacturer instruction and normalized to cell counts as determined using TC20 automated cell counter (Bio-Rad).
The sonicated chromatin was cleared by centrifugation at 13,000 ×g, 10 min and the supernatant diluted in 8 ml of ChIP dilution buffer (10 ml total, 1.67 mM Tris (pH 8.0), 167 mM NaCl, 1.2 mM EDTA, 1% Triton X-100, 0.01% SDS) before 80 µg of respective antibodies was added and chromatin rotated in a 50-ml falcon tube overnight. In parallel, 400 µl Dynabeads Protein A or G were washed, resuspended in ChIP dilution buffer, and blocked in 0.5 mg/ml BSA overnight. Immunoprecipitation was carried out using blocked Dynabeads, rotated for 1 hr, and centrifuged (1,500 ×g, 10 min). The beads were resuspended in 1 ml ChIP low salt wash buffer (20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS) and transferred to a fresh microcentrifuge tube, washed, and resuspended in a further 1 ml ChIP low salt wash buffer before transferring to a fresh microcentrifuge tube. Further washing, 2× ChIP low salt wash buffer, 2× ChIP high salt wash buffer (20 mM Tris-HCl (pH 8.0), 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS) and 2× LiCl wash buffer (10 mM Tris-HCl (pH 8.0), 250 mM LiCl, 1 mM EDTA, 1% sodium deoxycholate, 1% NP-40) was done in the same tube. The beads were eluted in 1.2 ml elution buffer (100 mM NaHCO 3 , 1% SDS). Reverse cross-linking of ChIPed-DNA was done at 65°C overnight with addition of 0.3 M NaCl (final concentration), 20 µg RNase A, and 20 µg Proteinase K. Recovery of ChIPed-DNA was done using QIAquick PCR Purification Kit. 6 ml of PB buffer was added to the reversed cross-linked DNA before passing through four columns as described per supplier's instruction. Final elution was done by passing 30 µl water sequentially through all the columns. This fraction was kept for ChIP-seq library preparation, 1 µl of which was assessed on a Bioanalyzer. Samples which passed the QC on the Bioanalyzer (fragment length distribution primarily around 200-400bp) and showed enrichment at expected targets on qPCR were subjected to sequencing on HiSeq 2500 (Illumina) carried out using the Wellcome Trust genomic service, Oxford.

| RNA-seq
RNA was extracted using RNeasy kit (QIAGEN #74106), and QC on the Bioanalyzer (for RIN ≥9.5). ERCC ExFold RNA Spike-In Mixes (Ambion) was added prior to Library prep using QuantSeq Forward kit (LEXOGEN #0.15.96), using 500 ng starting material to minimize the PCR amplification step. Samples were sequenced on HiSeq 4000 (Illumina) carried out using the Wellcome Trust genomic service, Oxford.

| Bioinformatics for ChIP-seq
Each replicate contained two technical replicates (same library sequence in two separate flow cells) which were stitched together using UNIX to generate a single fastq. Raw fastq files were fastQCed to check the read quality and PCR duplication, processed, and mapped to human genome build hg19 (GRCh37, February 2009) using Bowtie (Langmead & Salzberg, 2012;Langmead, Trapnell, Pop, & Salzberg, 2009) allowing for 2 mismatches. Mapped SAM-files were used for peak calling using the Homer package (Heinz et al., 2010). The background files used were generated by performing a parallel ChIP-seq experiment using HA antibody against 501mel parental cell lines. Peak annotation, genome ontology analysis, de novo motif identification, and bedgraph generation were carried out using the Homer package with peaks being assigned to the nearest gene.
Peaks were further filtered for those with peak score <10 to increase stringency of the analysis. Peaks were identified as co-occupying a genomic location if the peak summit or the start coordinate of the peak lay within 200 bp.

| Bioinformatics for RNA-seq
Fastq files were treated as for ChIP-seq. Raw fastq files were then trimmed of poly-A using cutadapt (Martin, 2011) and mapped using STAR (Dobin et al., 2013) against hg38 (GRCh38, 2015. Counts per gene from STAR were used as input for differential gene expression analysis using EdgeR (Robinson, McCarthy, & Smyth, 2010). Reads for each sample set were first filtered for genes whose expression is <1 count per million prior to glmQLFTest. Genes with a p ≤ 0.05 and fold-changes above 2 were taken for further analysis. In the case of shHIF1α, genes whose differential gene expression was dampened by ≥8-fold were taken as significant. Heatmaps of RNA-seq samples were generated from the edgeR-library normalized reads of genes whose differential gene expression has p ≤ 0.05 and fold change ≥2 before center normalized and cluster using ComplexHeatmaps (Gu, Eils, & Schlesner, 2016).

| GSEA and GSVA analyses
GSEA analyses were carried out using javaGSEA2-3.0 (Subramanian et al., 2005). 10,000 permutations were carried out for each probed gene set. Maximum gene set size was set to 800 to accommodate the Verfaillie invasive gene set. GSVA analyses were performed using the Bioconductor package GSVA (Hänzelman, Castelo, & Guinney, 2013). The gene sets used were obtained from the Molecular Signatures Database (Subramanian et al., 2005). The GSVA matrix was then clustered and displayed as heatmap using Pheatmap (https ://cran.r-proje ct.org/web/packa ges/pheat map/index.html).

| Hypoxia in melanoma correlates with invasiveness
Previous work has established an inverse correlation between hypoxia and expression of differentiation markers in melanoma in a restricted set of melanoma tissue sections and has suggested that hypoxia may induce de-differentiation and invasiveness in melanoma or in melanocytes (Cheli, Giuliano, et al., 2011a;Loftus et al., 2017;Widmer et al., 2013). However, whether hypoxia correlates with invasion in melanoma in general has not been examined in detail. We therefore analyzed the TCGA melanoma cohort for correlations with the previously published Elvidge hypoxia gene expression signature comprising 171 genes that are up-regulated in response to hypoxia (Elvidge et al., 2006). Each melanoma was ranked by a score corresponding to the average expression of the genes in the Elvidge hypoxia gene set. We then examined each individual melanoma for the expression of the Verfaillie melanoma invasive gene expression signature (Verfaillie et al., 2015). While there was some variation in the invasive gene expression signature score between individual melanomas, a moving average of the Verfaillie signature showed a high degree of correlation with the hypoxic signature ( Figure S1a). The noise in the moving average was largely abolished when we compared the Verfaillie invasive signature to the Elvidge hypoxia signature in a single-cell RNA-seq analysis (Tirosh et al., 2016) of melanoma cells from dissociated tumors ( Figure S1b), suggesting the noise may arise from non-melanoma cells within the TCGA melanoma samples. Note that only 31 genes are found in both gene sets, and these do not account for the correlation between the hypoxia and invasive signatures. The hypoxia signature was also strongly inversely correlated to MITF expression (Figure S1c), consistent with hypoxia repressing MITF to drive a de-differentiated phenotype (Cheli, Giuliano, et al., 2011a), and correlated strongly with the expression of AXL ( Figure S1d) encoding a receptor tyrosine kinase linked to an MITF-low, AXL-high drug resistance phenotype (Dugo et al., 2015;Konieczkowski et al., 2014;Muller et al., 2014). Gene set enrichment analysis (GSEA) of the top and bottom 75 TCGA melanomas ranked by the Elvidge hypoxia gene expression signature also confirmed a strong enrichment in the top 75 hypoxic melanomas for the Verfaillie invasive gene set ( Figure S1e) and epithelial-mesenchyme transition (EMT)-associated genes (HALLMARK EMT) ( Figure S1f). As expected, given the inverse correlation in melanoma between proliferation and invasion (Carreira et al., 2006), the 75 TCGA melanomas exhibiting the highest hypoxic gene expression exhibited a reduced proliferative gene expression signature (Verfaillie et al., 2015) compared to the bottom 75 ( Figure S1g). That hypoxia could induce invasion was confirmed using DMOG, a cell-permeable prolyl-4-hydroxylase inhibitor, to impose a hypoxia gene expression program. As anticipated, DMOG transiently induced HIF1α expression and increased invasiveness in both IGR37 and 501mel human BRAF V600E -mutated melanoma cell lines ( Figure S1h).
Hypoxia should reduce oxidative phosphorylation that occurs in mitochondria (Semenza, 2013), and hypoxia-mediated suppression of MITF that controls expression of PPAR gamma cofactor 1 alpha Recent advances in melanoma therapy have seen a shift away from BRAF targeted therapies toward those aimed at reactivating the immune system. However, as resistance to immune checkpoint therapies is frequently encountered, we asked whether the Elvidge hypoxia signature would also correlate with a recently characterized gene expression signature that correlates with innate anti-PD-1 resistance (IPRES) (Hugo et al., 2016). Strikingly, GSVA of the top 75 TCGA melanomas ranked by the Elvidge hypoxia signature showed they were very strongly enriched for the IPRES signature ( Figure   S1k), as were a subset of the CCLE melanoma cell lines ranked by the Elvidge hypoxia signature ( Figure S1l). Collectively these analyses indicate that in melanomas, hypoxia correlates with invasion, drug, and immune checkpoint inhibitor resistance and negatively correlates with mitobiogenesis, differentiation, and proliferation. They also indicate that a subset of melanoma cell lines exhibit a constitutive pseudo-hypoxia gene expression signature even when grown under normoxic conditions.

| Identification of a core hypoxic response signature between melanoma cell lines
Although these data provide an indication of the how tumors respond to hypoxia, the microenvironment within tumors is highly complex and it is possible that additional signals within the hypoxic microenvironment could contribute to the correlations observed.
Moreover, it is unclear whether all melanoma cells will exhibit a common hypoxia response, or whether different phenotypic subpopulations of cells within a tumor will mount a different hypoxia response.
To address these issues, we examined the gene expression signature of three different BRAF V600E mutant melanoma cell lines in response to hypoxia over time in biological triplicate using a 3'RNA-seq approach. The cell lines used were IGR39, that is MITF-low, highly de-differentiated, invasive, and drug-resistant (Konieczkowski et al., 2014); IGR37 that is MITF-positive, non-invasive, and isolated from the same patient as IGR39 (Luis et al., 1989); and 501mel (Shamamian et al., 1994), that expresses high levels of MITF, is non-invasive, and was isolated from a different patient than the IGR37 and IGR39 cell lines. Importantly, rather than examining gene expression at a single 24 hr time point, as has been done previously for a mouse melanocyte cell line (Loftus et al., 2017), we chose to assess the effects of hypoxia over time. This is because the stabilization of HIF1α in response to low oxygen tends to be short-lived while HIF2α mediates a longer term response (Koh & Powis, 2012) and we wished to capture both short-and long-term effects on gene expression as well as any dynamic changes. Hypoxia, and specifically HIF1α, imposes a metabolic shift away from oxidative phosphorylation and toward glycolysis (Semenza, 2013). Consistent with this, all three lines exhibited enrichment in expression of gene sets associated with glycolysis ( Figure 1d). We were especially interested in any difference in the response between the differentiated IGR37 and undifferentiated IGR39 cell lines. In normoxic conditions, these two cell lines possess substantially different gene expression programs ( Figure S2a). GSVA of several EMT-asso- Interestingly, no genes implicated in invasion or EMT were commonly up-regulated, and although DMOG could trigger invasion ( Figure S1h), we were unable to increase invasiveness in the melanoma cell lines grown in 1% oxygen (not shown). This result was unexpected because hypoxia is known to induce metastatic spread and EMT signatures are enriched in hypoxic melanoma tumors ( Figure   S1a,b,e,f). This may be because we assayed for gene expression changes at 1% oxygen, which is sufficient to induce HIF1α and reprogramming of metabolic gene expression, but lower oxygen levels found within tumors and mimicked by DMOG may be necessary to promote invasion.

| HIF target genes in melanoma
To identify which genes might be direct HIF targets, we exam-  Table   S3. The recognition motif for each factor derived from the ChIPseq analysis (Figure 2b) reflected the known consensus, ACGTG, though for both HIF2α and HIF1β, the consensus motif was extended to CACGTG, an E-box of which around 50% possessed a 5′ T flanking residue, a hallmark of MITF binding sites (Aksan & Goding, 1998). Of the HIF binding sites detected, 1,473 were bound by both HIF1β and HIF2α, while 357 were co-occupied by HIF1α and HIF1β (Figure 2c). Only 48 exhibited co-occupancy by HIF1α and HIF2α alone, but 917 were bound by all three factors.
Since the HIF family binds DNA as heterodimers, the detection of all three would suggest exchange of dimers at the same location.  Table S4.

| MITF is transiently up-regulated by HIF1α
The results so far indicate that different melanomas cell lines exhibit a distinct hypoxia response, but share the key elements of metabolic reprogramming known to occur in other cell types in low oxygen conditions. However, we noted that in some experiments after 48 hr in hypoxia the IGR37 cell line, though not the MITF-negative cell line IGR39, or 501mel, exhibited increased pigmentation (Figure 3a).
Pigmentation is a differentiation function of melanocytes in which the production of the melanin is a consequence of the activity of a set of pigmentation enzymes, including tyrosinase (TYR), tyrosinase-related protein 1 (TYRP1), and dopachrome tautomerase (DCT) within specialized organelles termed melanosomes (Park & Gilchrest, 1999). MITF coordinates melanin production by directly regulating most, if not all, genes implicated in melanin synthesis, and melanosome genesis (e.g., PMEL and MLANA) and transport (e.g.,

RAB27A
) (Goding & Arnheiter, 2019). The increase in pigmentation in the IGR37 cells under hypoxia was therefore surprising since it is well established that the regulator of melanoma/melanocyte differentiation, MITF, is down-regulated in hypoxia (Cheli, Giuliano, et al., 2011a;Feige et al., 2011). However, we had already observed that MITF is up-regulated in both the IGR37 and 501mel cell lines 24 hr post-hypoxia (Figure 1f). We therefore re-examined the expression of MITF and several of its pigmentation-associated target genes over time. Strikingly, in 501mel cell mRNA for MITF and many of its downstream differentiation targets were transiently up-regulated at 12 hr following exposure to hypoxia, but expression was returned to baseline or below by 24 hr (Figure 3b). Using a 501mel cell line expressing an shRNA targeting HIF1α, the up-regulation of MITF and its target genes also occurred in low oxygen conditions, but was severely delayed ( Figure 3c). IGR37 cells, which undergo a less robust hypoxic response than 501mel cells (Figure 1c), also up-regulated MITF and its target genes (Figure 3d Together with the RNA-seq analysis (Figure 3b,c), this result is consistent with hypoxia transiently up-regulating MITF and its downstream target genes via HIF1α binding.
We next asked why shHIF1α led to greater reduction in MITF levels at later times than was observed in control cells. Previous studies have identified bHLHE40/DEC1, a transcription factor directly upregulated by HIF1α in hypoxia as a direct repressor of MITF (Cheli, Giuliano, et al., 2011a;Feige et al., 2011). However, our observation that shRNA knockdown of HIF1α led to enhanced silencing of MITF suggested a mechanism for MITF repression that was bHLHE40/ DEC1-independent. Consistent with this, while bHLHE40/DEC1 Since hypoxia promotes increased expression of genes implicated in glycolysis and shHIF1α prevented the increased expression of glycolysis gene sets (Figure 3i), we hypothesized that under hypoxia MITF expression might be sustained by increased glucose uptake, consistent with low glucose suppressing MITF expression (Ferguson et al., 2017). In 501mel cells, glucose deprivation led to a progressive decrease in MITF expression (Figure 3j) accompanied by up-regulation of ATF4, a transcriptional repressor of MITF (Falletta et al., 2017;Ferguson et al., 2017) that is induced under conditions of translation stress (Harding et al., 2000). This is reminiscent of the decrease in MITF mediated by translational reprogramming mediated by eIF2α phosphorylation observed on glutamine limitation (Falletta et al., 2017). Since eIF2α phosphorylation is a key determinant of melanoma phenotype (Falletta et al., 2017;Maida et al., 2019), our observations are consistent with elevated glucose uptake under hypoxia contributing to the maintenance of MITF expression.

| MITF regulates a cohort of hypoxic response genes
The fact that MITF and its target genes are transiently up-regulated by hypoxia led us to ask whether MITF could co-regulate a set of HIF targets. Given the related consensus motifs for DNA binding by MITF (CACGTG) and by HIF (ACGTG) (Figure 2b), it seemed likely that at least some genes would be co-bound and co-regulated. Using a threshold for the ChIP-seq analysis of a peak score above 10, a total of 882 MITF target sites were also recognized by a HIF1α/ HIF2α-HIF1β combination (Figure 4a). Of these, 317 sites were co-occupied by all four transcription factors, while 500 sites are unique to MITF-HIF2α-HIF1β and 65 are specific to MITF-HIF1α-1β.
Examining the read densities from the ChIP-seq analysis of all four factors ranked by peak score of HIF1α (Figure 4b) revealed extensive co-occupancy at bound sites, though it was not likely that MITF would bind to sites simultaneously with the HIF factors. As might be expected, given the extended sequence requirement for MITF DNA binding (Aksan & Goding, 1998) compared to HIF, the consensus at the co-bound sites match that of classical MITF TCACGTG targets rather than the shorter ACGTG hypoxia response element ( Figure 4c). Some examples of well-characterized hypoxia response genes co-regulated by MITF are highlighted in Figure 4d, with their induction in response to inducible MITF shown in Table 1. For example, MITF and HIF binding to distinct sites in the VEGFA gene is shown in Figure 4e, and other examples of co-bound and regulated genes are shown in Figure S4a. Notable among the co-regulated genes are VEGFA that stimulates angiogenesis and the glucose transporter SLC5A9, consistent with MITF promoting glucose uptake like HIF1α (Figure 4d; Table 1). Some genes robustly up-regulated by hypoxia and sharing a common binding site for MITF and HIF were also up-regulated by MITF. Using a previously described (Falletta et al., 2017) (Figure 4h), raising the possibility that, in addition to MITF, these HIFs may directly contribute to any transient increase in expression of these genes under hypoxia (Figure 3b-d). However, we noted that the peak score of the HIFs on the pigmentation genes ( Figure 4h) was considerably lower than known HIF targets such as BHLHE40 or SLC8A2 (Figure 4g) and consequently defining whether the HIFs contribute directly to the regulation of these genes will require further experimentation.

| MITF represses its own expression
In the course of these experiments, we noted that induction of ectopic expression of HA-tagged MITF led to repression of endogenous MITF mRNA (Table 1). Western blotting using anti-MITF antibody to

| MITF controls succinate dehydrogenase to suppress pseudo-hypoxia
The results so far suggest a complex relationship between MITF and the hypoxic response, with MITF able to affect the regulation of a set of hypoxia-responsive genes in melanoma. Moreover, although low levels of MITF are also associated with a higher hypoxia gene expression signature in the TCGA melanoma cohort ( Figure S1c), consistent with previous work (Cheli, Giuliano, et al., 2011a;Feige et al., 2011) indicating suppression of MITF in vivo in response to low oxygen,  (Selak et al., 2005), a multi-subunit complex that catalyzes the conversion of succinate to fumarate, or by inhibition of SDH by malonate that is generated by carboxylation of oxaloacetate by pyruvate carboxylase in cells undergoing oxidative stress (Reed, Ludwig, Bunce, Khanim, & Gunther, 2016). To ask whether a deregulated TCA cycle could be responsible for the pseudo-hypoxia signature observed in some MITF-low melanoma cell lines, we first confirmed that both malonate and succinate can lead to elevated HIF protein levels in melanoma cells (Figure 6d). We then used a mass spectrometry approach to interrogate the metabolite profile of the MITF-high, non-invasive melanoma cell line IGR37, and the IGR39 cell line derived from the same patient that is MITF-low, invasive, and exhibits a pseudo-hypoxic signature (Figure 1e). Since the increased pseudo-hypoxia signature is predominantly associated with MITF-low cell lines, we also profiled the MITF-high cell line 501mel in which MITF was depleted using a specific siRNA. In focusing on TCA cycle metabolic intermediates, we found that several were elevated in the MITF-low pseudo-hypoxic IGR39 melanoma cells as well as in MITF-depleted 501mel cells (Figure 6e; boxed metabolites in Figure 6b). These included succinate itself as well as the SDH

| D ISCUSS I ON
In vivo, melanoma cells transition though distinct phenotypic states in response to a changing microenvironment, and most notably can switch between invasive and proliferative phenotypes characterized by low and high levels of MITF activity respectively (Hoek et al., 2008;Hoek & Goding, 2010). Since melanoma cell lines isolated from human tumors tend also to fall into either proliferative or invasive, slow-growing phenotypes , it seems likely that established lines reflect specific phenotypic states within tumors (Tsoi et al., 2018), including those detected using single-cell RNA-seq (Rambow et al., 2018),  (Benita et al., 2009;Chi et al., 2006;Denko et al., 2003;Loftus et al., 2017;Widmer et al., 2013); our data reveal that even cells from the same tissue of origin or patient will respond differently to low oxygen, with only a relatively small number of core genes regulated by hypoxia in common be- to hypoxia toward enhanced glucose uptake and glycolysis appears to represent a core hypoxia program (Semenza, 2013).
One important and unexpected result presented here is that in contrast to previous work (Cheli, Giuliano, et al., 2011a;Feige et al., 2011), we find that MITF is up-regulated by HIF1α in response to hy-  (Carreira et al., 2006), tumor-initiating (Cheli, Guiliano, et al., 2011b), and drug-and immunotherapy-resistant phenotypes (Dugo et al., 2015;Konieczkowski et al., 2014;Landsberg et al., 2012;Muller et al., 2014;Riesenberg et al., 2015;Tirosh et al., 2016). It was not surprising therefore that hypoxia, a known trigger for invasiveness, was identified as a repressor of MITF expression (Cheli, Giuliano, et al., 2011a;Feige et al., 2011). The mechanism reported is indirect, with activation of HIF1α leading to up-regulation of the transcription factor bHLHE40/DEC1, one of the common 23 genes we identify as induced in all three melanoma cell lines examined after 12 hr in hypoxia, and its consequent binding and repression of the MITF promoter. In contrast to this simple scenario, our results reveal a previously unsuspected complexity in the interplay between the hypoxic response and MITF (Figure 7g). Surprisingly, at early times, hypoxia induces the expression of MITF and its downstream target genes, an effect blunted in the presence of shRNA that prevents accumulation of HIF1α. The activation of MITF is likely to be directly mediated by the HIFs since ChIP-seq analysis revealed that they bind upstream from the melanocyte-specific MITF-M promoter. Moreover, in response to hypoxia, the HIF-mediated activation of genes implicated in glycolysis elevates glucose import and processing (Semenza, 2013 (Falletta et al., 2017), low MITF may be insufficient to trigger invasion. This is most likely because in melanoma invasion reflects in part a response to a low nutrient supply environment (Falletta et al., 2017) and that by promoting proliferation, for example by activating CDK2 (Du et al., 2004), MITF imposes a high nutrient demand state that is incompatible with nutrient limitation. For example, 24 hr following glutamine deprivation MITF is suppressed to facilitate a switch to invasion (Falletta et al., 2017).
However, at 4 hr following nutrient deprivation, MITF is activated to amplify the response to low nutrient levels. Thus, there are strong parallels between the response to low glutamine and low oxygen.
MITF can activate expression of HIF1α (Busca et al., 2005) and PGC1α (Haq et al., 2013;Vazquez et al., 2013). It seems likely therefore that MITF's role at early times following exposure to low oxygen is to amplify the hypoxia response with the aim to buy time for cells to maintain their gene expression and metabolic program in anticipation that oxygen supply may be rapidly restored. Consistent with this, MITF, like HIF1α, can up-regulate VEGFA, though MITF binds at different sites, but can also activate a cohort of direct hypoxiaregulated HIF target genes by binding the same sites as the HIFs.
Only if oxygen levels remain low or are further reduced would MITF be down-regulated by bHLHE40/DEC1, which is also up-regulated by our inducible MITF and is directly bound by MITF (Feige et al., 2011) to enable cells to establish an invasive program. We found no evidence of an invasive gene expression signature in melanoma cell lines at 12 hr post-hypoxia, a time when a range of HIF-bound genes implicated in metabolic reprogramming were already induced. If oxygen levels remain low for an extended period, it seems likely that cells will reverse the initial up-regulation of MITF expression.
We also establish a new metabolic role for MITF in controlling the TCA cycle, directly binding and regulating the gene encoding the key SDH subunit SDHB, with reduced MITF levels correlating to elevated succinate, a known inhibitor of the prolyl hydroxylase that promotes degradation of HIF1α (Selak et al., 2005). The regulation of SDHB by MITF provides a mechanism to promote a prolonged hypoxia response; the repression of MITF by bHLHE40/DEC1 would lead to decreased SDHB levels and consequently increased succinate that is able to inhibit the degradation of HIF1α. Although we have focused here on the impact of succinate on the hypoxia response, succinate is a key metabolic intermediate that plays a role in many biological processes (Tretter et al., 2016). These include mitochondrial ROS production, that may contribute in part to the elevated SDH inhibitor malonate in MITF-depleted cells, the succinylation of proteins (usually on lysines), and epigenetic events including inhibition of histone demethylases and the ten-eleven translocation family of 5-methylcytosine hydroxylases (Tretter et al., 2016). Significantly, succinate has been termed an oncometabolite, with SDHB mutations being pathogenic (Saxena et al., 2016). As such, the low levels of SDHB in MITF-low cells may contribute to disease progression in melanoma.
Finally, we reveal that MITF can repress its own expression. This

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest. DRM provided antibodies, FB and CRG provided resources and supervision, and PL, IL, and CRG wrote the manuscript.

DATA AVA I L A B I L I T Y
The RNA-seq and ChIP-seq datasets reported in this study have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession numbers GSE95280 and GSE132624).