Loss of CITED1, an MITF regulator, drives a phenotype switch in vitro and can predict clinical outcome in primary melanoma tumours

CITED1 is a non-DNA binding transcriptional co-regulator whose expression can distinguish the ‘proliferative’ from ‘invasive’ signature in the phenotype-switching model of melanoma. We have found that, in addition to other ‘proliferative’ signature genes, CITED1 expression is repressed by TGFβ while the ‘invasive’ signature genes are upregulated. In agreement, CITED1 positively correlates with MITF expression and can discriminate the MITF-high/pigmentation tumour molecular subtype in a large cohort (120) of melanoma cell lines. Interestingly, CITED1 overexpression significantly suppressed MITF promoter activation, mRNA and protein expression levels while MITF was transiently upregulated following siRNA mediated CITED1 silencing. Conversely, MITF siRNA silencing resulted in CITED1 downregulation indicating a reciprocal relationship. Whole genome expression analysis identified a phenotype shift induced by CITED1 silencing and driven mainly by expression of MITF and a cohort of MITF target genes that were significantly altered. Concomitantly, we found changes in the cell-cycle profile that manifest as transient G1 accumulation, increased expression of CDKN1A and a reduction in cell viability. Additionally, we could predict survival outcome by classifying primary melanoma tumours using our in vitro derived ‘CITED1-silenced’ gene expression signature. We hypothesize that CITED1 acts a regulator of MITF, functioning to maintain MITF levels in a range compatible with tumourigenesis.


INTRODUCTION
CITED1 is the founding member of the CITED (CBP/p300-interacting transactivator with glutamic acid [E]/aspartic acid [D]-rich C-terminal domain) family of transcriptional co-regulators and was originally cloned from a differential display screen between pigmented mouse B16 melanoma cells and their dedifferentiated weakly-pigmented derivative, B16F10s. This led to speculation that CITED1 or msg1 (melanocyte specific gene 1) as it was known at that time, was involved in the process of pigmentation (Shioda, Fenner and Isselbacher, 1996). Subsequently, Nair et al. reported that stable overexpression of CITED1 increased the levels of tyrosinase, dopachrome tautomerase (Dct) and melanin in B16 cells, reinforcing the idea that it had a role in melanogenesis (Nair et al., 2001). By 2005, as gene expression profiling became relatively commonplace, CITED1 was identified in several new screens of tumours and cell lines: two studies identified CITED1 as a gene whose expression distinguished nevi from primary melanoma, another found CITED1 to be upregulated in advanced stage melanomas in comparison to benign nevi or melanoma in situ, while expression profiling of an in vitro progression model identified CITED1 among a signature of genes lost in aggressive melanoma lines relative to primary melanocytes in culture (Ryu et al., 2007;Haqq et al., 2005;Talantov, 2005;Smith, Hoek and Becker, 2005). sought to explain the observation that melanoma cells altered between two states: those with high proliferative potential that are less invasive and those with high metastatic potential that are less proliferative. These separate but alternating states are controlled by different transcriptional programs and can be defined by specific gene signatures . MITF expression and many of its known targets (TYR, MLANA) define the 'proliferative' group, while the 'invasive' signature group is characterized by expression of negative regulators of the Wnt signalling pathway (WNT5A, DKK1, CTGF). CITED1 expression was associated with the proliferative pathway signature and subsequently confirmed in an updated and expanded data set to be significantly correlated with the proliferative phenotype (P<1.00E-05, http://www.jurmo.ch/hopp, accessed 19 March 2013) ; (Widmer et al., 2012).
Studies on CITED1 suggest that it is a non-DNA binding nuclear transcriptional co-regulator capable of influencing TGF induced transcription mediated by ligand-induced SMAD hetero-oligomerization; estrogendependent transcription mediated by ER, and Wnt/-Catenin-dependent transcription. These effects are dependent on CITED1-CBP/P300 binding via the conserved CITED family CR2 domain and while CITED1 is thought to act by stabilizing the CBP/P300-ER interaction, in the case of -Catenin it acts to repress transcription by competing for binding with CBP/P300 transcriptional co-activators (Shioda et al., 1998;Yahata et al., 2001;2000;Plisov, 2005).
Microphthalmia-associated transcription factor, MITF, acts as a masterregulator of melanocyte differentiation and as a result has been intensely studied in the field of melanoma research (Widlund and Fisher, 2003;Levy, Khaled and Fisher, 2006).
It is a basic helix-loop-helix leucine zipper transcription factor that recognizes E-box and M-box sequences in the promoter regions of its target genes. Highlighting its importance in the disease, amplification of MITF locus has been found in >15% of metastatic melanomas and germline mutations in MITF that predispose carriers to melanoma development have also been found (Garraway et al., 2005;Bertolotto et al., 2011;. In melanoma cells the target genes of MITF include most notably TYR, MCIR, DCT, MLANA involved in the process of pigmentation; cell cycle regulators such as CDK2 and CDKN1A and the more recently identified BRCA1 gene that has, with other target DNA repair genes, defined a role for MITF in the DNA damage response (DDR) Beuret et al., 2011;Giuliano et al., 2010).
The regulation of MITF is complex and tightly controlled, exhibiting both transcriptional and post-translational regulation. There are several transcript isoforms, of which MITF-M is the dominant form expressed in melanocytes.
Multiple signalling pathways converge on the MITF-M specific promoter that harbours binding sites for PAX3, SOX10, CREB, FOXD3, LEF-1 and BRN2 among other transcription factors Levy, Khaled and Fisher, 2006 inhibitor function, suggesting the existence of at least one positive feedback loop (Sestáková, Ondrusová and Vachtenheim, 2010).
MITF post-translational activity can be affected by phosphorylation, sumoylation, ubiquitination and by binding with proteins that block access to the DNA binding domain such as PIAS3 Levy, 2001). Oncogenic BRAF (but not wildtype BRAF), which is mutated in up to 50% of melanomas, also regulates MITF via simultaneously stimulating MITF activation through ERK phosphorylation, which leads to its subsequent degradation, and by inducing transcription of MITF via BRN2 upregulation (Davies et al., 2002;Wellbrock et al., 2008).
The consensus regarding why the cell invests such effort in maintaining control of MITF levels and why there are so many regulatory mechanisms, is that melanocytes and melanoma are exquisitely sensitive to even small variations in MITF expression. Ultimately its activity must be sustained within the narrow window permissive for continued survival and proliferation. In this study, we characterise the role of CITED1 as a novel regulator of MITF in melanoma.

Gene expression analysis
RNA was isolated (4 replicates for each treatment) using a Qiagen RNeasy Plus mini-kit and the quality determined using a Bioanalyser (Agilent).
Replicates were cell samples from separate wells, but plated on the same day and derived from the same passage number. Gene expression experiments were performed using the Illumina HT12 array covering more than 47,000 transcripts and known splice variants across the human transcriptome. The raw data was quantile normalized and Illumina control probes were removed from subsequent analysis using BASE (Vallon-Christersson et al., 2009). The data were exported to MeV, log2 transformed and gene and sample centred (Saeed et al., 2006). SAM (significance of microarray analysis) was performed using a two-group comparison; for the siRNA experiment the groups cases there was a median false discovery risk of 10 false-positive transcripts.
Hierarchical clustering was performed to visualize the data. 1009 probes were significantly altered by TGF1 treatment while 312 probes were found to be significantly altered in the siRNA experiment (208 upregulated and 104 downregulated). These data can be found in Supplemental files S7 and S8, respectively. DAVID was used to assist in functional annotation of the final gene lists (Huang et al., 2007) For the publically available data cited, 120 melanoma cell lines from three cohorts ( (Johansson, Pavey and Hayward, 2007); ; (Greshock et al., 2010)) analysed by Affymetrix gene expression microarrays were collected, individually MAS5 normalized, and merged into a single cohort. Probe sets were collapsed into single genes and mean-centred across the entire cohort. These 120 cell lines and their associated normalized expression data can be found in supplemental data S10.  (Wellbrock et al., 2008). A pRL-Renilla Luciferase reporter vector was used as a control for each transfection.
CITED1 was overexpressed using a pRc/CMV containing a N-terminal HAtagged human CITED1 (transcript isoform 1) referred to as pCITED1 in the text (Shioda, Fenner and Isselbacher, 1996). An empty CMV-promoter expression plasmid, pcDNA3.1 (+) was used a negative control. Recombinant human transforming growth factor-1 (TGF1), #PHG9203 was purchased from Invitrogen. For the A2058 gene expression experiment, cells were exposed to either 5 or 10ng/ml TGF1 in serum-free media for 24 hours. In the case of the Luciferase reporter assay, cells were serum starved the day after transfection for 3 hours and exposed to 5ng/ml TGF1 in serum free media for 24 hours prior to harvesting.

Antibodies and Immunoblotting
The following antibodies were used: anti-CITED1, #AB15096 from Abcam anti-CDKN1C/P57, #2557 were purchased from CellSignaling Technology and anti--Actin (AC-15), #A5441 from Sigma-Aldrich. Cell lysates were resolved by SDS-PAGE (pre-cast gels purchased from Life Technologies) and transferred to 0.45 μm PVDF membranes purchased from Millipore by electroblotting. The membranes were blocked in 5% non-fat milk in TBST prior to incubation with primary antibodies diluted 2.5% non-fat milk. The blots were probed with the appropriate secondary antibodies (Pierce Biotechnology) in 5% non-fat milk. The membranes were developed using ECL (GE Healthcare).

Cell cycle analysis
Flow cytometry was performed on a FACSCalibur (BD Biosciences) and subsequently analysed using ModFit (Verity House Software). Briefly, following transfection, confluent cells were detached, washed in 1XPBS and fixed in 70% ethanol. Prior to analysis they were stained with a propidium iodide solution and a 20G syringe was used to obtain a homogenous single cell solution. All events were saved (up to 20,000 events per replicate) ungated, using BD Cell Quest and the data exported to ModFit where following selection of the appropriate ploidy status, a standard auto-analysis fit using autolinerarity was performed. We found that a 2-cycle aneuploiddip/tetraploid was appropriate for HT144 and A2058 while 1-cycle diploid was suitable for A375. The Alamar blue assay reagent was purchased from Invitrogen (Life positive/negative droplets was selected using the Bio-Rad QuantaSoft TM data analysis suite to calculate the relative copies/µl of each transcript.

RESULTS
TGF induces expression of the invasive signature genes while suppressing a cohort of proliferative signature genes including

CITED1
Hoek et al. noted that many of the genes that defined the invasive phenotype were commonly TGF-driven while at the same time only the proliferative signature phenotype cells were sensitive to TGF growth inhibition in vitro . That MITF levels decrease and invasiveness is enhanced in response to TGF stimulation was also confirmed subsequently (Pierrat et al., 2012;Pinner et al., 2009). In agreement, we showed that the melanoma cell line A2058 upregulates WNT5A in response to TGF exposure and that exogenous Wnt-5a in turn, increased their invasive potential (Jenei et al., 2009). For the present study, in an effort to examine what other phenotype specifying genes were directly regulated by TGF, we performed gene expression analysis and found TGF treatment resulted in both upregulation of invasive signature genes and suppression of genes characterizing the proliferative phenotype (Fig. 1a) has a slightly different but overlapping gene profile based on the top ranked differentially expressed genes (Fig. 1b). Both MITF and CITED1 are in the proliferative cohort and their response to TGF treatment was confirmed at protein level in A2058 cells (Fig. 1C). set that had matching gene symbols in our data (Fig. 2a). We also confirmed the correlation in cell lines derived from our own lab (Fig. S1). This was important as inconsistency in interlaboratory phenotype signatures has previously been reported (Widmer et al., 2012). We could additionally confirm expression at the protein level ( Fig. 2b)

Gene expression analysis reveals CITED1 silencing can induce a phenotype-switch
To investigate the function of CITED1 in melanoma, we transiently downregulated its expression using CITED1 targeting siRNA. We choose the  (Fig. 3a,b). A heatmap of the expression profiles clearly illustrates that the shift is due to a general induction of the 'proliferative´ and suppression of the 'invasive' cohort (Fig. 3c). It was apparent that the #3 siCITED1 siRNA was not as effective at switching the cells as the #1 siCITED1, this was observed consistently throughout our experiments and may be due to the fact that #3 siCITED1 was not as successful at silencing CITED1 (Fig. 3b, inset).

CITED1 is a reciprocal regulator of MITF and impacts MITF target gene expression
A heatmap highlights the identity of the only the significantly differentially induced transcripts between siNEG and both #1 & #3 siCITED1 (Fig. 4a). Of most relevance, we found MITF, a known driver of the proliferative phenotype switch and many of its previously known downstream targets, these also included genes categorized by Gene Ontology annotation (GO) as related to pigmentation and UV/DNA damage response (Fig. 4a) McGill et al., 2006;Sánchez-Martín et al., 2002;Strub et al., 2011). We could confirm that indeed MITF protein levels were affected by siCITED1 in HT144 cells and that conversely, overexpression of CITED1 in A2058 cells, resulted in downregulation of MITF (Fig. 4b). Strub et al. identified a large number of genomic targets of MITF by ChIP-seq analysis .
A comparison of the genes differentially expressed by siCITED1 compared to siNEG, revealed that there was significant enrichment of these potential targets (Fig. S2a). Notably, genes both up and down regulated by siCITED1 are represented among genes defined as having MITF-occupied promoters ( Fig. S2b). We also found that downregulation of MITF using siRNA in HT144 cells (Fig. 4c) and in WM293A, and SKMEL5 cells (Fig. S3a, b) resulted in decreased protein expression of CITED1 suggesting reciprocity between these factors.

Induction of MITF by CITED1 silencing transiently restrains cell cycle progression and impacts cell viability
To investigate the effect of CITED1 silencing on melanoma cells behaviour we analysed the cell cycle distribution following siRNA treatment, by flow cytometry. In siCITED1 treated HT144 cells we saw G1 accumulation as indicated by an increase in the diploid G1 fraction and a concomitant reduction in the total S-phase fraction peaking at 33 hours but also observed at 48 and 72 hours post-transfection in comparison to siRNA control HT144 cells. Again, the effect was apparent but not as pronounced using the #3 siCITED1 ( Fig S4a). Similar effects were seen in #1 and #3 siCITED1 treated A2058 and A375 cells (Fig. S4b, c).
Owing to the previously reported dependency of MITF-induced cell cycle arrest on CDKN1A/P21 we investigated the levels of several cyclin-dependant kinase inhibitors following CITED1 silencing (Carreira et al., 2005). We found that CDKN1A/P21 was transiently increased in siCITED1 treated HT144 cells relative to the siNEG treated HT144 cells (Fig 5b). In contrast, in A2058 cells, which do not have detectable levels of CDKN1A/P21 (Fig. S5), the levels of CDKN1C/P57 were supressed in response to CITED1 overexpression (Fig 5b).
We hypothesised therefore that melanoma cells can utilise either CDKN1A/P21 or CDKN1C/P57 to mediate cell cycle arrest induced by MITF and this is reflected in the expression levels of the alternate CDK inhibitors in different melanoma cell lines (Fig. S5).
In agreement with the cell cycle data, an Alamar Blue assay revealed a significant reduction in cell viability as measured by metabolic activity over 5 days in HT144 cells treated with siCITED1 (Fig 5c). The effect was apparent but not as pronounced in the #3 siCITED1 sample.

The effect of CITED1 silencing on MITF is transient and mediated via promoter activation
We observed that the peak upregulation of MITF and CDKN1A/P21 protein following siCITED1 treatment varied from transfection to transfection, being seen between 24-48 hours post-transfection but appearing as unchanged or even downregulated after this time (Fig. 6a, upper panel). In agreement, later timepoints of the cell cycle analysis (=/>72 hours) exhibited little or no change in G1/S-phase distribution or even a reverse pattern (Fig. 5a HT144, and data not shown: A2058, A375). We therefore sought to examine the transcriptional dynamics more closely, map the changes in MITF following CITED1 silencing and see if they corresponded to cell behaviour and changes at the protein level. We used a quantitative droplet digital PCR based assay The rapid MITF transcriptional response to CITED1 manipulation suggested to us that the effect could be directly mediated at the promoter level. To test this hypothesis, we over expressed an MITF-M promoter-reporter construct and CITED1 in A375 cells. We chose A375 cells, as while they had less endogenous CITED1 and MITF than HT144 or A2058 so as not to cause interference with the assay, we also knew that they could respond adequately as they had an identical G1 accumulation/S-phase decrease to both HT144 and A2058 cells following CITED1 silencing (Fig S4c). treatment alone or combination with CITED1 overexpression (Fig. 6c). There did not appear to be an additive or synergistic effect using both TGF treatment and CITED1 overexpression suggesting TGF may be dependent on CITED1 for MITF suppression.

The CITED1-silenced gene signature predicts outcome in primary melanoma
The 'proliferative' and 'invasive' signature phenotypes have served to define the gene expression classification of melanoma cell lines. However, primary tumours and metastatic lesions have also been molecularly classified into several distinct groups by gene expression profiling (Harbst et al., 2012;Jonsson et al., 2010). The four-class structure found in tumours consists of the 'pigmentation', 'proliferative', 'high-immune' and 'normal-like' subgroups with a subset falling into an unclassifiable cohort (Jonsson et al., 2010). We used the same tumour classification to subtype the 120 cell lines that had publically available expression data and could show that the tumour 'pigmentation' subgroup that highly expresses MITF, corresponds to the cell line 'proliferative' phenotype described by Hoek et al. Accordingly, the tumour 'proliferative' and 'high-immune' subgroups comprise the cell line 'invasive' phenotype (Fig. 7a). It is worth noting that the names of the tumour subgroups were derived from a description of the differentially expressed genes that comprised each molecular classification while the 'invasiveproliferative' switching phenotypes were named to reflect the behaviour to explain the confusing occurrence that both classifications have a group referred to as 'proliferative' although they are not equivalent.
The overlap between the primary tumour classifying and cell line classifying systems allows us to infer that CITED1 expression is most likely restricted to a subset of MITF high 'pigmentation' subtype tumours. As the tumour subtype classification was shown to be prognostically significant in primary melanomas we were interested to know if CITED1 expression itself was independently predictive of outcome. Previously we reported on the analysis of 223 primary lesions using the Illumina WG-DASL protocol (Harbst et al., 2012). As the CITED1 probe in this assay did not produce reliable data we instead derived a CITED1-silenced gene signature score based on the differentially expressed genes from the HT144 siCITED1 experiment (Fig. 3).
We therefore effectively created a multi-gene surrogate expression signature rather than using CITED1 gene expression itself. We subsequently interrogated the gene expression data on the primary melanoma lesions using a nearest centroid approach derived from the CITED1-silenced gene signature. This revealed that primary melanomas with a gene expression signature most similar to the CITED1-silenced signature (CITED1low-class) had a significantly better outcome than those with a signature most disparate from the CITED1-silenced signature (CITED1high-class) (Fig. 7b). Importantly, the CITED1 signature classing had independent prognostic information (HR gene-signature, these data indirectly imply that CITED1 expression is a potential prognostic indicator in primary melanomas and the transcriptional program influenced by CITED1 expression determines tumour behaviour in vivo. counter-intuitively, a lineage-specifying differentiation factor can behave as a potent oncogene (Hoek and Goding, 2010;Carreira et al., 2006;Cheli, Giuliano, et al., 2011). The rheostat model (Fig. S6) (Carreira et al., 2006;Loercher et al., 2005;Carreira et al., 2005). At the extreme high end of MITF expression lies differentiated melanocytic cells, while the lowest levels can lead to senescence and irreversible cell death.

DISCUSSION
Between these two extremes however it is thought that melanoma cells can oscillate from a low-MITF 'invasive' to a high-MITF 'proliferative' state via phenotype-switching.
We hypothesise that the role of CITED1 in melanoma is to maintain levels of MITF compatible with tumour progression and effectively tip the balance in favour of cell cycle progression rather than MITF-induced G1-arrest. This is supported by our findings that downregulation of CITED1 using siRNA results in a phenotype switch to a more pigmented state driven by increased MITF