The results of our integrative examination of a panel of eight melanoma cell strains, three from short-term cultures, although in need of validation on a larger cohort, revealed underlying processes important for responsiveness to decitabine. The data implicated three major components in Aza responsiveness: a) activation of Wnt signaling; b) re-expression of p21Cip1 in a p53-independent manner and c) activation of two TGFβ pathway genes.
Comparing the gene expression profile of un-treated and treated melanoma cells implicated Wnt signaling based on high expression of the Wnt antagonist SFRP1 only in sensitive cells, which led us to further explore downstream members of this pathway and to identify activated β-catenin as a feature contributing to drug resistance. Although mutations in CTNNB1
are rare in melanomas, activation might be through upstream modulators because a survey of large collection of melanoma tumors in tissue microarrays demonstrated that activated β-catenin in the nucleus is an independent predictor of poor survival 
. The oncogenic potential of β-catenin was validated in a mouse model where stabilized β-catenin repressed p16Ink4a
expression and together with an activated NRas, lead to melanoma development with high penetrance and short latency 
. We showed that the likely effect of activated β-catenin is upregulation of MITF, a potent melanocyte-specific transcription factor by itself considered an oncogene 
. Interestingly only two of the three resistant cell strains, 501 mel and YURIF, harbored activated CTNNB1
mutation. The third one, WW165 expressed constitutively high levels of endogenous MITF (in the absence of any gene amplification). Wnt/β-catenin pathway also interferes with responsiveness of CML to the tyrosine kinase inhibitor Imatinib 
, suggesting a common effect on other cancer cells as well. Therefore, various components of the activated Wnt/β pathway, in particular an activating mutation in β-catenin and high levels of MITF could be considered when selecting patients for this type of therapy, and devising combination therapy.
Protein analysis showed that p21Cip1 was upregulated in a p53 independent manner in two of the sensitive cell strains, but not in a resistant one. However, p21Cip1 was relatively stable and abundant in the other five melanoma cell strains, suggesting the emergence of resistance downstream of this cell cycle suppressor.
Whole genome expression analysis uncovered two reactivated TGFβ-responsive genes Clusterin and TGFBI that were more prominent in Aza-sensitive compared to resistant melanoma cells, and their activation enhanced apoptosis as observed by siRNA mediated gene knockdown. These two proteins can be used as markers, because they are secreted and have the potential to be released into the circulation. Furthermore, TGFBI promoter methylation might be useful as a marker for malignant transformation because it was unmethylated in normal melanocytes and hypo- or fully methylated in freshly isolated primary and metastatic melanoma cells, as well as melanoma tumors.
Our global gene expression analysis uncovered a total of 292 differentially expressed genes (mostly re-expression) across all melanoma strains after Aza treatment, with some products known to be associated with growth arrest. In addition to those described here, we validated the expression of UCHL1, PTPN6, TNFR1, SELENBP1, TNFR1, TNFRSF10D, S100A4, and several MAGE genes by semi-quantitative or real-time RT-PCR, or Western blots. Some of them, such as PTPN6 (protein tyrosine phosphatase, non-receptor type 6), that is expressed primarily in hematopoietic cells, were significantly induced at the protein level in the Aza sensitive YUMAC and YUSAC cells, but very little in the other cell types without any correlation to growth arrest or apoptotic response (Supplementary Figure S3
). We surmise that other activated pathways, such as genes associated with acute inflammatory and immune responses or with activity on neighboring stroma cells, such as IGFBP5 
, are likely to influence drug resistance in vivo
and should be further explored.
We showed that gene reactivation by low-dose Aza in melanoma cells is through two known epigenetic activities of this drug, DNA promoter hypomethylation and histone modification. Other decitabine-responsive genes in our dataset, such as FN1, UCHL1, FUCA1, ICAM1, IL8, SERPINE2, TMEM45A and SFRP2 are also reactivated by HDAC inhibitors 
, and might be modulated through histone modification by Aza as well. Aza can directly and indirectly modify histones as a function of DNMT status. DNMT1 interacts with HDAC1 
and elimination of DNMT1 displaces HDAC1 from target promoters 
. In addition, Aza can affect histone methylation because DNMT1 binds also SUV39H1, a H3K9 methyltransferase, and EZH2 that catalyses the methylation of histone H3 at lysine 27 (H3K27), conferring a suppressive state 
. Decitabine can also reduce the suppressive activity H3 K9 di-methylation by inducing changes in the transcription of enzymes responsible for this covalent modification 
. Taken altogether, the observations reinforce the concept that impact on histone modification should be considered when dissecting the function of decitabine and devising combination therapy that is based on gene reactivation.
The protein validation data highlighted the importance of proteasomal degradation processes in responsiveness to Aza. At least two of the critical growth suppressor proteins, p21Cip1 and Clusterin, undergo proteasomal degradation. This observation led us to infer that a proteasomal inhibitor such as Bortezomib, currently in clinical trials, can synergize with low-dose Aza to alleviate resistance. This prediction was fulfilled in the case of 501 mel resistant cells. The synergistic response to this drug combination was unique because the Hsp90 inhibitor 17-AAG and the IGF1R inhibitor NVP-AEW541 (Novartis), employed at log-range of concentrations, did not show any synergistic growth arrest with Aza (data not shown).
Altogether, our results from this limited panel of melanoma cells suggest that treatment of melanoma patients could be improved by knowledge of the genetic and epigenetic background of individual tumors. In addition, they implicate that proteasomal and HDAC inhibitors might act in synergy with epigenetic modifiers for some patients.