SREBP1-dependent de novo fatty acid synthesis gene expression is elevated in malignant melanoma and represents a cellular survival trait

de novo fatty acid biosynthesis (DNFA) is a hallmark adaptation of many cancers that supports survival, proliferation, and metastasis. Here we elucidate previously unexplored aspects of transcription regulation and clinical relevance of DNFA in cancers. We show that elevated expression of DNFA genes is characteristic of many tumor types and correlates with poor prognosis, especially in melanomas. Elevated DNFA gene expression depends on the SREBP1 transcription factor in multiple melanoma cell lines. SREBP1 predominantly binds to the transcription start sites of DNFA genes, regulating their expression by recruiting RNA polymerase II to promoters for productive transcription elongation. We find that SREBP1-regulated DNFA represents a survival trait in melanoma cells, regardless of proliferative state and oncogenic mutation status. Indeed, malignant melanoma cells exhibit elevated DNFA gene expression after the BRAF/MEK signaling pathway is blocked (e.g. by BRAF inhibitors), and DNFA expression remains higher in melanoma cells resistant to vemurafenib treatment than in untreated cells. Accordingly, DNFA pathway inhibition, whether by direct targeting of SREBP1 with antisense oligonucleotides, or through combinatorial effects of multiple DNFA enzyme inhibitors, exerts potent cytotoxic effects on both BRAFi-sensitive and -resistant melanoma cells. Altogether, these results implicate SREBP1 and DNFA enzymes as enticing therapeutic targets in melanomas.

data from 30 diverse cancer types in The Cancer Genome Atlas (TCGA). We found that DNFA enzyme expression varies widely among cancers. Four DNFA enzymes -SCD, FASN, ACLY and ACSS2 -exhibit the highest levels of mRNA expression in skin cutaneous melanoma (SKCM) compared to other tumor types, whereas expression of ACACA is less elevated in melanomas ( Supplementary Fig. 2a). We observed relatively low expression of mRNAs encoding HMGCS1 and HMGCR, two rate-limiting enzymes in the de novo cholesterol synthesis (DNCS) pathway 46 in melanomas. These results indicate that elevated DNFA expression is prevalent among tumors, significantly more so in melanomas than in most others.
To relate elevated DNFA expression with disease progression, we examined its prognostic value using Kaplan-Meier analysis. We divided patients into two groups based on the DNFA gene expression from their tumor biopsies and compared overall survival rates. We found that high expression of SCD ( Fig. 1c and Supplementary Fig. 2d,g), FASN ( Supplementary Figs 1c and 2e,h) and ACACA ( Supplementary Fig. 2b,f) significantly correlates with poor prognosis in all cancers considered collectively. The only exceptions we observed were ACLY (Supplementary Fig. 1e) and ACACA specifically with a top 10% versus bottom 10% expression cutoff ( Supplementary Fig. 2i). Elevated mRNA expression of SCD (Fig. 1d), FASN, ACLY ( Supplementary Fig. 1d,f) and ACACA (Supplementary Fig. 2c) each correlates with poor prognosis in SKCM. We observed for HMGCS1 but not HMGCR that higher expression significantly correlates with poor survival in all tumor types combined ( Supplementary Fig. 3c,e), using 20% as cutoff, however, this was not observed in SKCM ( Supplementary Fig. 3d,f).
Melanoma is the most aggressive skin cancer type. To control for potential tissue-specific upregulation, we first compared SCD expression in healthy tissues using RNA-Seq data from the Genotype-Tissue Expression (GTEx) database 47 . There, skin SCD expression is low ( Supplementary Fig. 4a); yet among tumor tissues, skin SCD expression is elevated (Fig. 1a). This contrasts with, for example, the liver, which exhibits relatively high SCD expression among healthy ( Supplementary Fig. 4a) and tumor tissues (Fig. 1a). We then compared SCD, ACLY and HMGCS1 expression using normalized RNA-Seq data from the SKCM group in TCGA and the normal skin tissue group in GTEx. SCD and ACLY mRNA expression is markedly elevated in SKCM versus healthy skin tissues ( Supplementary Fig. 4b,c), while HMGCS1 is higher in the healthy tissues ( Supplementary Fig. 4d). We performed principal component analysis (PCA) on RNA-Seq data of DNFA genes from skin tumors and healthy skin tissues ( Supplementary Fig. 4e,f), and found that skin tumors are distinct from healthy skin samples. SCD expression strongly correlates with principal component 1 (PC1) towards skin tumors. Long-chain acyl-CoA synthetase (ACSL1) is a dual function enzyme in DNFA and fatty acid oxidation pathways with low abundance in tumors 4 . When its expression is combined with other DNFA genes in PCA, ACSL1 expression strongly correlates with PC1 towards healthy tissues and results in better separation of skin tumors from healthy skin tissues ( Supplementary Fig. 4e,f). In a separate PCA, DNCS expression, particularly HMGCS1, strongly correlates with PC1 ( Supplementary Fig. 4g). Supplementary Fig. 4e-g suggests difference in lipogenic activities between skin tumors and healthy skin tissues.
Melanomas are transformed from benign, melanocytic nevi, with proliferation of melanocytes commonly triggered by BRAF V600E -activating mutation 48 . To confirm cancer-selective elevation in expression, we compared the DNFA gene expression between malignant melanoma biopsies and benign melanocytic nevus samples ( Supplementary Fig. 5a-d). We observed significantly elevated DNFA gene expression in malignant melanomas. We further analyzed single cell RNA-Seq data from melanoma samples 49. High expression of SCD, FASN, and ACACA was confined to malignant cells, with low expression in healthy adjacent tissue ( Fig. 1e and Supplementary Fig. 5e-h). Somatic mutations of BRAF and NRAS are associated with about 50% and 15% of melanomas, respectively 48 . BRAF and NRAS mutations, while being well-known risk factors and drivers of cancer onset, have limited prognostic significance for overall survival of melanoma patients 50 . There is no significant correlation between DNFA enzyme expression and common oncogenic driver mutations in SKCM (Supplementary Fig. 6a-l), suggesting that elevated DNFA expression in malignant melanomas is mechanistically unrelated to BRAF and NRAS mutation status.

SREBP1 contributes to elevated DNFA gene expression in melanoma cells. To test whether SREBP1
drives elevated DNFA enzyme expression in melanomas, we depleted the SREBP1 mRNA (SREBF1) with antisense oligonucleotides (ASOs) and siRNAs in HT-144 cells. SREBF1 depletion with siRNA and ASO agents was accompanied by decreasing protein levels of SREBP1 and DNFA enzymes (Fig. 2a). The pooled siRNA and ASO agents effectively depleted both the cytoplasmic precursor and the mature nuclear SREBP1 (Fig. 2b,c). Among six tested ASOs targeting SREBF1, ASO-1 and ASO-4 are more potent than single or pooled siRNAs for SREBF1, as 5 nM of ASOs inhibited DNFA mRNA production as effectively as 50 nM of siRNAs ( Supplementary Fig. 7b-f), even when siRNAs decreased the level of SREBF1 mRNA more than ASOs ( Supplementary Fig. 7a). ASOs may also engage in steric translation inhibition of SREBF1 mRNA 51 , which could potentially contribute to their greater potency. We found that ASO-4 inhibits DNFA gene expression commensurately with dosage ( Supplementary Fig. 7g).
To evaluate potential activators of DNFA expression, we depleted SREBF1, SREBF2, and co-activators MED15 and CREBBP 28 with siRNAs, and examined DNFA expression across three melanoma cell lines: HT-144 (Fig. 2d,e), A375 ( Supplementary Fig. 7i,j) and MEL-JUSO ( Supplementary Fig. 8a,b). We observed a similar range of mRNA reductions (50-70%) for most DNFA enzymes after SREBF1 depletion (Fig. 2d, Supplementary  Figs 7i and 8a). Depletion of MED15 and CREBBP individually or together impacts DNFA gene expression, but to a lesser extent than SREBF1 depletion ( Supplementary Figs 7i and 8a). SREBF2 depletion exerts potent effects on the expression of the DNCS genes HMGCS1 and HMGCR in HT-144 and A375 cells ( Supplementary Fig. 7h,k), in line with previous studies in hepatocytes 52 . SREBF2 also affects DNFA enzyme expression, especially in concert with SREBF1 depletion (Supplementary Fig. 8a-d). The role of SREBP2 in regulation of DNFA genes may be transitive via SREBP1 53 , since SREBF1 expression decreases after SREBF2 depletion (Fig. 2e); or it may be acting in a partially redundant manner 54 . Thus, we confirmed the transcription regulatory role of SREBP1 in promoting DNFA gene expression in multiple melanoma cell lines. ( 55 . FASN and SCD displayed increased expression at consecutive time points ( Supplementary Fig. 8c-f). SREBF1 depletion by siRNA blocked DNFA gene activation on days 3 and 5, whereas DNFA enzyme protein (Fig. 2f) and mRNA levels (Fig. 2g,h) were elevated in response to overexpression of nSREBP1a, the abundant isoform in proliferating embryonic cells and in cancers, as well as nSREBP1c, which is the predominant isoform in adult liver and adipose tissues 56 . With a longer transactivation domain, nSREBP1a interacts more avidly with co-activators and thus exhibits stronger transcription activity than nSREBP1c 27,57 . Consistently, we observed higher expression of DNFA genes after overexpressing nSREBP1a than nSREBP1c. These results indicate that SREBP1 expression controls DNFA gene expression in melanoma cells. SREBP1 directly regulates DNFA pathway genes through RNAP II recruitment and productive elongation. To characterize the transcriptome changes after SREBF1 depletion, we carried out RNA-Seq analysis after SREBF1 depletion with pooled siRNAs and individual ASOs in HT-144 cells, followed by PCA on RNA-Seq data. To identify the possible off-target effect of the ASOs, we performed RNA-Seq analyses on   (g,h) mRNAs were analyzed with RT-qPCR assay. The bar graphs show the relative expression of DNFA enzymes to that of pcDNA3 (control) transfection group. Data are expressed as mean ± SD and quantified from triplicates. One-way ANOVA tests were performed. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.
www.nature.com/scientificreports www.nature.com/scientificreports/ two constructs that target different sequence regions of SREBF1 mRNA: ASO-1 and ASO-4. The two principal components (PCs) in the PCA biplot represent over 70% of the overall gene expression changes ( Fig. 3a and Supplementary Fig. 9a). SREBF1 and SCD are among the top six contributors to data separation and they align with PC2 ( Fig. 3a, Supplementary Fig. 9b). SREBF1 siRNA, ASO-1 and ASO-4 produced similar vertical separations to all negative controls in PC2 (Fig. 3a), and they all reduced expression of FASN and SCD compared to negative controls (Supplementary Fig. 9e-g). These results indicate that the most profound effects after SREBF1 depletion by both ASOs and siRNA are on DNFA genes. ASO-1 has significant lateral separation from ASO-4 and SREBF1 siRNA on PC1 (Fig. 3a). The changes in SPIRE1 and USP9X expression are the major contributors to PC1 ( Supplementary Fig. 9c) and are only affected by ASO-1 ( Supplementary Fig. 9h,i). This result indicates that ASO-1 has specific off-target effects on SPIRE1 and USP9X. Multiple negative controls are grouped together, and ASO-4 is close to SREBF1 siRNA on the PCA biplot, consistent with the results from hierarchical clustering analysis of the same RNA-Seq dataset ( Supplementary Fig. 9d). Hence, ASO-4 has similar specificity as pooled SREBF1 siRNAs for SREBF1 depletion.
To explore the overall biological pathways affected by SREBF1 depletion, we examined differentially expressed genes (DEGs) in the RNA-Seq dataset (siSREBF1 group vs siNegative group) ( Supplementary Fig. 9j). Using gene-set enrichment analyses (GSEA) with Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and gene ontology (GO) terms on DEGs, we determined that the downregulated genes after SREBF1 depletion are primarily involved in fatty acid biosynthesis and lipid metabolism (Fig. 3b, and Supplementary Fig. 9k), as expected. Cellular inflammatory response pathways are significantly enriched in upregulated genes after SREBF1 depletion (Fig. 3c, and Supplementary Fig. 9l), including Toll like receptor and tumor necrosis factor (TNF) signaling pathways that mediate cytotoxicity 58 .
To assess the direct regulatory roles of SREBP1 on target genes in cancers, we analyzed public ChIP-Seq data for SREBP1 from lung cancer, breast cancer and chronic myeloid leukemia (CML) cell lines. ChIP-Seq peaks primarily localize at the proximal promoter regions around TSS ( Supplementary Fig. 10a,b). De novo motif sequences identified from SREBP1 ChIP-Seq peaks match the known SREBP1 binding motif ( Supplementary  Fig. 10c). High DNFA gene expression was observed in lung and breast cancers (Fig. 1a). We identified genes that are present in SREBP1 ChIP-Seq data from both lung cancer line A549 and breast cancer line MCF7 ( Supplementary Fig. 10d), and that are also DEGs in our RNA-Seq data. We reasoned that the "overlapping" genes common to all three data sets are likely regulated directly by SREBP1. We therefore performed functional network analysis on this subset. This analysis revealed genes in many pathways including the PI3K/AKT pathway   ChIP-qPCR signals were compared between cells cultured in 10% FBS and 1% ITS medium conditions. Data were presented as mean ± SD and quantified from triplicates. Two-way ANOVA tests were performed. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.
www.nature.com/scientificreports www.nature.com/scientificreports/ and the RNAP II elongation complex, and the expected DNFA pathways (Supplementary Fig. 10e). We then used our RNA-Seq data to divide the overlapping genes into upregulated and downregulated sets, and performed GSEA on each. The downregulated genes participate primarily in lipid metabolism pathways ( Supplementary  Fig. 10f), which confirms SREBP1 as a direct activator of DNFA genes. Inflammatory response pathways are significantly upregulated in DEGs from RNA-Seq analysis (Fig. 3c), but seem not to be direct targets of SREBP1 ( Supplementary Fig. 10g). We suspect that downregulation of DNFA pathways may change the homeostasis of cellular fatty acids and exert further impact on inflammatory response pathways as well as cell death 59 .
To elucidate the molecular mechanism of SREBP1-governed DNFA gene activation in melanoma cells, we used a chromatin immunoprecipitation (ChIP)-qPCR assay to detect occupancy of SREBP1, RNA polymerase II (RNAP II), and H3K36me3 -a histone marker associated with transcription elongation 60 on DNFA genes. SREBP1 depletion by ASO-4 diminished SREBP1, RNAP II and H3K36me3 signals at the SCD promoter ( Fig. 3d-f). We observed similar (albeit lower-magnitude) results for FASN ( Supplementary Fig. 11a-c). These ChIP-qPCR results together with the RNA-Seq data ( Supplementary Fig. 9e-g) suggest that removal of SREBP1 at DNFA promoters inhibits transcription activity and mRNA production. We performed ChIP-qPCR analyses in SREBP1-activating (1% ITS medium, no lipids) and SREBP1-repressing (10% FBS medium, with lipids) conditions. We found that 1% ITS medium dramatically increases SREBP1 occupancy at the transcription start sites (TSS) of SCD (Fig. 3g) and FASN (Supplementary Fig. 11f) in HT-144 cells. The strong RNAP II binding peaks at TSS of SCD and FASN in both 10% FBS and 1% ITS culture conditions ( Fig. 3h and Supplementary Fig. 11g) indicate promoter-proximal pausing of RNAP II 61 . Furthermore, culturing cells in 1% ITS medium increased the occupancy of actively elongating RNAP II (RNAP II S2P) ( Supplementary Fig. 11d,i), but not poised RNAP II (RNAP II S5P) 62 (Supplementary Fig. 11e,j), at the TSS as well as the gene bodies of both SCD and FASN. Accordingly, the histone marker H3K36me3 was elevated along both gene bodies in cells cultured with 1% ITS ( Fig. 3i and Supplementary Fig. 11h). These results suggest that SREBP1 binding near the TSS involves both RNAP II recruitment and stimulation of productive elongation on DNFA genes.

Essential role of DNFA in proliferation and survival of untreated and drug-resistant melanoma cells.
To evaluate the roles DNFA in cell proliferation and survival, we cultured melanoma cells in the lipid-free 1% ITS medium, using insulin as a growth factor to stimulate proliferation. The metastatic melanoma-derived cell lines HT-144 and A375 can proliferate in both 10% FBS and 1% ITS media but remain quiescent in 0% FBS medium (Fig. 4a,b). By contrast, the primary melanoma-derived cell line, MEL-JUSO, remains quiescent in both lipid-free (1% ITS and 0% FBS) media (Fig. 4c). We depleted SREBF1 with ASO-4 in several melanoma cell lines cultured under the three medium conditions, to obtain dose-response curves and IC50 values. We found that ASO-4 decreased viability of both proliferative and quiescent cells in all conditions (Fig. 4d-i). We reason that, although cancer cells may employ lipid uptake 63 , DNFA is required for cell survival regardless of external lipid availability under conditions similar to those we tested. We observed diminished viability at higher ASO concentrations (greater than 5 nM), which correlates with more efficient depletion of SREBP1 and more marked decrease of DNFA gene expression ( Supplementary Fig. 7g). Comparing the two growth factor-containing culture conditions, we find that ASO-4 is a more potent inhibitor of cell growth/survival in 1% ITS medium than in 10% FBS medium with HT-144, A375 and LOXIMVI cells. In contrast, WM1552C, MeWo and MEL-JUSO cells are relatively insensitive to ASO-4 treatment. The implication is that, although DNFA seems to be vital for cell survival, medium lipid availability may mitigate the impact of SREBP1 inhibition in cell lines that are sensitive to ASO-4.
BRAF inhibitors (BRAFi, e.g. vemurafenib) improve survival for patients with metastatic melanomas harboring the BRAF V600E mutation 64 . However, the rapid emergence of treatment-surviving tumors limits clinical benefits 65,66 . We derived two BRAFi-resistant cell lines with prolonged vemurafenib treatment: HT-144BR from a vemurafenib-sensitive cell line HT-144, and LOXIMVIBR from a vemurafenib-insensitive cell line LOXIMVI (Fig. 5a,c). Both HT-144BR and LOXIMVIBR cell lines continuously proliferate in the presence of vemurafenib (2 μM). Using cell viability assay, we found that HT-144 and HT-144BR displayed similar sensitivities to ASO-4 treatment, as did LOXIMVI and LOXIMVIBR (Fig. 5b,d). We reason that these cell lines all depend on SREBP1 for survival regardless of whether they are sensitive or resistant to vemurafenib.
To confirm that DNFA enzyme inhibition is the crucial driver for reduced cell viability, we then used small molecule inhibitors of FASN and SCD enzymes, which reportedly decreased tumor growth in preclinical studies [67][68][69] . In 10% FBS (lipid-containing) medium, FASN and SCD inhibitors decreased viability in both HT-144 and HT-144BR cells (Fig. 5e,g), although the effect was less potent than SREBF1 depletion by ASO-4. However, we observed much stronger effects on cell survival when combining the two inhibitors, confirmed by Bliss independence analysis 70 as robust positive synergy (Fig. 5f,h). DNFA thus appears vital to melanoma cell survival even in vemurafenib-resistant cells.

Continued and elevated DNFA gene expression in melanoma cells after BRAF inhibitor treatment.
To explore the role of DNFA in BRAFi resistance, we compared DNFA gene expression in untreated and vemurafenib-resistant cell lines. We observed modestly higher DNFA gene expression in the vemurafenib-resistant HT-144BR and LOXIMVIBR cells as compared with the untreated HT-144 and LOXIMVI cells (Fig. 6a,b). We then investigated the short-term impact of vemurafenib on DNFA gene expression in HT-144 cells treated with various doses of vemurafenib. After one-day treatment, vemurafenib exerted low-dose stimulation and high-dose inhibition of DNFA gene expression, exhibiting hormesis (i.e. bell-shaped dose-response curves) 71 (Fig. 6c,d, Supplementary Fig. 12a-e). We observed dose-dependent induction of PPARGC1A expression with vemurafenib treatment (Supplementary Fig. 12f) as expected, given that the BRAF/MEK pathway directly suppresses PPARGC1A expression and fatty acid oxidation in melanomas 72,73 . DNFA inhibition in response to high-dose vemurafenib treatment may correlate with onset of cell death, whereas DNFA stimulation in response to low-dose vemurafenib treatment suggests rapid cellular resistance.
www.nature.com/scientificreports www.nature.com/scientificreports/ To rule out vemurafenib-specific (perhaps off-target) effects, we evaluated dabrafenib, another BRAFi employed in clinical practice 74 . We monitored DNFA gene expression in HT-144 cells after one day treatment of dabrafenib. Cell death was less evident after high-dose dabrafenib treatment, perhaps due to lower off-target toxicity 75 . We found that expression of DNFA genes decreased after low-dose dabrafenib treatment but increased at effective high-dose treatment ( Supplementary Fig. 13a-g). Dabrafenib at 2.5 µM concentration significantly stimulated gene expression of ACLY, SCD, ACSL1 and PPARGC1a (Supplementary Fig. 13b,e,f and h).
The effectiveness of BRAF inhibition is widely attributed to downstream inhibition of MEK, and then ERK. Resistance due to reactivation of MEK/ERK bypasses that effect. In A375 cells ( Supplementary Fig. 14a-h), a cell line with reported MEK/ERK reactivation associated with vemurafenib treatment 76 , vemurafenib exerts little induction of DNFA gene expression nor dose-dependent induction of PPARGC1A. An adjuvant ERK inhibitor (ERKi, SCH772984) has been employed clinically to overcome vemurafenib resistance 77 , by directly contributing to overall BRAF/MEK/ERK pathway inhibition 78 . When combining 0.5 μM vemurafenib and 0.5 μM SCH772984  Fig. 14i). We saw evidence of a dose-response relationship between ERKi and induction of DNFA gene expression, where treatment of SCH772984 at 1 μM achieved stronger induction of DNFA gene expression ( Supplementary  Fig. 14i). Our overall interpretation is that BRAF/MEK/ERK pathway inhibition upregulates DNFA gene expression, possibly through AKT activation 79 , which then contributes to vemurafenib tolerance in melanoma cells.
To further characterize the role of SREBP1 and DNFA upregulation in response to BRAF inhibition, we combined ASO-4 and vemurafenib treatment in HT-144 cells. We observed that SREBF1 depletion abolished the SCD and ASCL1 induction accompanying vemurafenib treatment by RT-qPCR assays ( Fig. 6e-g). Finally, we investigated cell viability after co-treatment with ASO-4 and vemurafenib. ASO-4, alone or in combination with vemurafenib, effectively killed HT-144 (Fig. 6h) and A375 cells ( Supplementary Fig. 14j). However, there is a mild antagonistic effect between ASO-4 and vemurafenib at low doses according to Bliss analysis (Fig. 6i, Supplementary Fig. 14k). We regard this as corollary to DNFA stimulation by vemurafenib ( Fig. 6e-g,  Supplementary Fig. 14a-g), while noting that, consistent with our findings above for A375, vemurafenib treatment alone yielded little induction of DNFA gene expression in A375 compared to HT-144, and the antagonistic effect was correspondingly lower. In both cell lines, high-dose ASO-4 exerts dominant cell killing effect over vemurafenib in combination treatment.

Discussion
Cancers frequently exhibit reprogrammed metabolic traits such as elevated DNFA 3 that act to sustain active proliferation and cell survival under adverse conditions, and support the process of tumorigenesis and metastasis, as well as resistance to targeted therapies. The diverse genetic paths cancers take to achieve such traits have frustrated efforts to exploit them clinically [80][81][82] .
Most DNFA enzymes are primarily regulated at the transcriptional level in a coordinated manner 6 , thus the abundance of their mRNAs can be employed as a simple surrogate for DNFA activities. Our results show that elevated DNFA gene expression may serve as a prognostic marker for some cancer types, even though they are not considered onco-drivers. Moreover, increased DNFA pathway expression appears to be intrinsic to malignant cancer cell types independently of onco-drivers, including constitutively active BRAF mutants. This may be consistent with the notion of increased DNFA as an "oncosustenance" pathway, with well-definable mechanistic contributions to cancer survival and proliferation. Once the oncosustenance pathway becomes active, it may not always matter which onco-drivers acted prior to cancer onset. Indeed, it has already been suggested that malignant cancer cells may have ongoing reliance upon oncogene-induced signaling pathways, but not upon initial onco-drivers 83 . It is also possible that SREBP1 transcription auto-regulation 28 might "lock in" elevated DNFA expression in malignant cells. The scope of therapeutic investigations, which frequently focus on mutated oncogenes, could be broadened to include oncosustenance mechanisms. www.nature.com/scientificreports www.nature.com/scientificreports/ The prevalence of elevated DNFA expression in certain tumor types such as melanoma suggests the presence of survival pressure for increased lipogenesis, which may therefore represent a therapeutic linchpin. However, high doses of individual pharmacological inhibitors of SCD and FASN enzymes were reportedly required to impact cell viability in melanomas and other cancers 67,69 . This phenomenon may be attributable to compensatory upregulation of SREBP1 and corresponding elevation of DNFA pathway expression, triggered by individual DNFA enzyme inhibition 84 . However, the combined effect of multiple DNFA enzyme inhibitors is synergistic and potent, at least in vitro (Fig. 5e-h). We demonstrated in melanomas that inhibition of SREBF1 by 5 nM ASO-4 treatment results in 50% reduction in the expression of multiple DNFA genes. The potent cell death effect after ASO-4 treatment is likely due to the strong synergistic effects of 50% reduction across the entire DNFA enzymatic cascade that is coordinately regulated by SREBP1. Therefore, inhibition of multiple DNFA enzymes could be of therapeutic benefit due to a potent cumulative effect on lipogenesis, as observed with SREBP1 mRNA-inhibiting ASOs and DNFA enzyme-targeting cocktails.
Previous studies have shown that SREBP1 binding and RNAP II recruitment to DNFA gene promoters represents the primary mechanism for transcription activation 85,86 . In accord with this, we observed that RNAP II accumulated at the proximal promoter regions of SCD and FASN. Under lipid-depleted (SREBP1-activating) cell culture conditions, we observed two indications of productive transcription elongation: increased RNAP II with serine 2 phosphorylation at its C-terminal domain (RNAPII-S2p) 62 and the histone marker for transcription elongation (H3K36me3) 87 along the gene bodies of FASN and SCD. Based on these findings we suggest a two-step working model for regulation of RNAP II machinery: RNAP II recruitment to proximal promoters at lipogenic genes, followed by RNAP II release for productive elongation. This model appears to explain elevated and synchronous DNFA gene expression in melanomas, and perhaps other cell types. It is presently unclear exactly how RNAP II is released for productive elongation.
BRAF-mutated melanomas are frequently treated with vemurafenib, a targeted therapy to inhibit the oncogenic BRAF pathway. Because vemurafenib sometimes fails to inhibit MEK/ERK (key targets downstream of BRAF) 88 , current clinical regimens combine inhibitors of both BRAF and MEK and/or ERK for treating metastatic melanomas 89,90 . However, even when combined inhibition is achieved, resistance arises via genetic alterations that upregulate the PI3K/AKT pathway 91 . Our determination that elevated DNFA is mechanistically important for resistance to targeted therapies dovetails with a similar recent finding by Talebi et al. 92 . They reported that (c,d) HT-144 cells were treated with different dosages of vemurafenib for 0.5 or 1 day in 1% ITS medium. SCD expression was assayed by RT-qPCR analysis. Expression of SCD from all treatment groups was normalized to expression under DMSO treatment at day 0.5 (normalized as 1). Relative gene expression was compared between 0.5-day and 1-day treatment groups. Data were presented as mean ± SD and quantified from triplicates. Two-way ANOVA tests were performed. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. (e-g) HT-144 cells were transfected with ASO-4 (5 nM) and then treated with vemurafenib (100 nM) in 1% ITS medium for one day. RT-qPCR assay analyzed DNFA gene expression from treatment groups relative to expression under ASO-Neg and DMSO treatment at day 1 (normalized as 1). Data are presented as mean ± SD and quantified from triplicates. One-way ANOVA tests were performed. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. www.nature.com/scientificreports www.nature.com/scientificreports/ SREBP1 processing is downregulated in vemurafenib-responsive cells, while BRAFi-resistant melanoma cells maintain SREBP1 processing to protect them from ROS-induced cell death. We find that (1) BRAFi treatment is associated with DNFA stimulation in a dose-response relationship and (2) elevated DNFA gene expression is a vital survival trait of melanoma cells both before and after they achieve resistance to combined BRAF/MEK inhibition. A speculative explanation is that, compared to cells vulnerable to BRAFi treatment, resistant cells achieve a state that relies more on DNFA activities and less on BRAF activities. We suspect that increased PI3K/ AKT signaling activates SREBP1 and DNFA for survival in BRAFi-resistant melanoma cells 79 . There is ongoing investigation to demonstrate mechanisms of cell state-related drug resistance.
In summary, our work demonstrates that melanomas engage the DNFA pathway for cell survival -more so during drug resistance -and employs SREBP1 to promote transcription activation and elongation of DNFA enzyme genes. Although immunotherapy has emerged as a highly promising treatment for a subset of melanoma patients 93 , it may be hampered by significant complications 94 . SREBP1 and/or DNFA enzyme inhibition may represent potential therapeutic alternatives worthy of further exploration for the treatment of melanoma and possibly other cancer types characterized by elevated DNFA.
The BRAF inhibitor vemurafenib (S1267) and ERK inhibitor SCH772984 (S7101) were purchased from Selleck Chemicals. Dabrafenib was kindly provided by R. Corcoran at MGH. SCD inhibitor MF-438 (569406, Sigma) and FASN inhibitor GSK 2194069 (5303, Tocris) were dissolved in dimethyl sulfoxide (DMSO) to yield 50 mM stock solutions for in vitro studies. To generate vemurafenib-resistant cells, parental cells were exposed to increasing concentrations of vemurafenib (from 1 μM to 2 μM) for three months. The resistance was confirmed by measuring cell viability under vemurafenib treatment.

Plasmids, siRNAs and
Reverse transcription quantitative PCR (RT-qPCR) and RNA-Seq assays. RNAs were isolated from cultured cells using the RNeasy Mini Kit (Qiagen). RNAs were treated with RNase-free DNase (Qiagen). RNA concentrations were quantified with Qubit RNA BR Assay Kit (Thermo Fisher Scientific). One μg RNAs were used for cDNA synthesis with RNA to cDNA EcoDry Premix (TaKaRa) containing both random hexamer and oligo(dT) 18 primers (Double Primed). qPCR was carried out in triplicates on a LightCycler 480 (Roche) using LightCycler 480 SYBR Green I Master (Roche). qPCR primers were designed by MGH primer bank (https:// pga.mgh.harvard.edu/primerbank/) and the primer sequences are listed in Table 1. Relative gene expression was calculated with the 2 −ΔΔCt method 98 , and normalized to the 18S housekeeping gene. The mean of negative control samples was set to 1.

RNA-Seq and ChIP-Seq analyses.
For RNA-Seq, Illumina sequencing reads (FASTQ files) were checked with FASTQC for quality control and then aligned to the human genome (GRCh38.86). Genome index generation and sequence alignment were performed using STAR software 99 , followed by sorting and indexing of BAM files with SAMtools. Raw counts of reads mapped to genes were calculated using HT-Seq 100 . Differential expression analysis was performed using DESeq2 R package 101 . KEGG pathway analysis of differentially expressed gene lists and principal component analysis (PCA) were performed using functions within DESeq2. The complete RNA-Seq data set has been submitted to the National Center for Biotechnology Information (NCBI) in the Gene Expression Omnibus (GEO) database (series entry GSE122707). SREBP1 ChIP-Seq in multiple cancer cell lines were previously published 102 . Bam files of SREBP1 and IgG control ChIP-Seq from the same cell lines were downloaded from ENCODE (https://www.encodeproject.org). SREBP1 binding peaks were called with Model-based Analysis of ChIP-Seq (MACS) and then annotated with ChIPseeker R package 103 . We performed de novo motif analysis of the SREBP1-binding sites with HOMER software (http://homer.ucsd.edu/homer/). Overlapping genes between RNA-Seq and ChIP-Seq datasets were identified by BioVenn 104 . Pathway annotation network analysis was performed on Cytoscape using ClueGo with REACTOME pathway 105 .  Table 2. The primers used for ChIP-qPCR assays.