A prognostic signature of G₂ checkpoint function in melanoma cell lines

Abstract

As DNA damage checkpoints are barriers to carcinogenesis, G₂ checkpoint function was quantified to test for override of this checkpoint during melanomagenesis. Primary melanocytes displayed an effective G₂ checkpoint response to ionizing radiation (IR)-induced DNA damage. Thirty-seven percent of melanoma cell lines displayed a significant defect in G₂ checkpoint function. Checkpoint function was melanoma subtype-specific with “epithelial-like” melanoma lines, with wild type NRAS and BRAF displaying an effective checkpoint, while lines with mutant NRAS and BRAF displayed defective checkpoint function. Expression of oncogenic B-Raf in a checkpoint-effective melanoma attenuated G₂ checkpoint function significantly but modestly. Other alterations must be needed to produce the severe attenuation of G₂ checkpoint function seen in some BRAF-mutant melanoma lines. Quantitative trait analysis tools identified mRNA species whose expression was correlated with G₂ checkpoint function in the melanoma lines. A 165 gene signature was identified with a high correlation with checkpoint function (p < 0.004) and low false discovery rate (≤ 0.077). The G₂ checkpoint gene signature predicted G₂ checkpoint function with 77–94% accuracy. The signature was enriched in lysosomal genes and contained numerous genes that are associated with regulation of chromatin structure and cell cycle progression. The core machinery of the cell cycle was not altered in checkpoint-defective lines but rather numerous mediators of core machinery function were. When applied to an independent series of primary melanomas, the predictive G₂ checkpoint signature was prognostic of distant metastasis-free survival. These results emphasize the value of expression profiling of primary melanomas for understanding melanoma biology and disease prognosis.

Keywords: :

• G₂ checkpoint
• melanoma
• microarray
• ionizing radiation
• oncogene

Introduction

Cancer is a genetic disease, and carcinogenesis is the process of acquisition of the numerous genetic and epigenetic alterations that encode malignant phenotypes.^1,2 Melanoma is much like other solid organ malignancies, with many mutations and chromosomal aberrations that activate proto-oncogenes and inactivate tumor suppressors.^3-5 Common mutations in melanoma include an activating mutation in BRAF codon 600 (V600E) and deletion of the CDKN2A locus that encodes two tumor suppressors, p16INK4A and p19ARF.^4,6,7 Although mutation of TP53 is uncommon in melanoma, being seen in 10–25% of primary lesions,⁴ inactivation of the p53 signaling pathway is common in melanoma cell lines, with 66% displaying a significant defect in p53-dependent G₁ checkpoint response to ionizing radiation (IR)-induced DNA damage.⁸ The common mutation of BRAF, which activates the mitogen-activated protein kinase (MAPK) signaling pathway, appears to be an early event in melanomagenesis, as it is seen as frequently in benign moles (nevi) as in primary melanomas.⁹ Malignant progression of BRAF-mutant lesions appears to require mutation or inactivation of PTEN.¹⁰ Thus, melanoma cell lines can be recovered with mutation in BRAF, inactivation of p53 signaling and severely reduced expression of CDKN2A and PTEN mRNAs. Melanomagenesis involves many genetic and/or epigenetic alterations.

DNA damage checkpoints represent positions of control in the cell division cycle that delay or arrest cell division when DNA damage is detected.¹¹ DNA damage checkpoints are one component of the DNA damage response (DDR) that also includes activation of DNA repair at sites of damage, transcriptional induction and repression and senescence or apoptotic cell death.¹² Many studies have demonstrated that activated oncogenes including mutant BRAF induce growth arrest in normal human cells through a process known as oncogene-induced senescence (OIS).^13-19 OIS as induced by HRAS in normal human fibroblasts produced full activation of the DDR with expression of DDR markers such as γH2AX, phospho-ATM and phospho-p53.¹⁴ Thus has arisen a model that activated oncogenes may induce precocious and aberrant DNA replication, leading to formation of DNA strand breaks and DNA damage checkpoint-induced growth arrest.^13,16 Attenuation of DNA damage checkpoint signaling overrides OIS, permitting outgrowth of oncogene-transformed clones.^14,17,20 Consequently, it is important to determine the functional capacity of DNA damage checkpoints in oncogene-driven malignancies and the biological consequences of checkpoint defects. This study has quantified the functional capacity of the ATM- and ATR-dependent DNA damage G₂ checkpoint²¹ in normal human melanocytes (NHMs) and 35 melanoma cell lines. Defective G₂ checkpoint function was detected in 37% of the melanoma cell lines, associated with significant alterations in gene expression. Some changes in gene expression reflected melanoma subtype-dependent variation, and others were correlated with G₂ checkpoint function across the melanoma subtypes. A G₂ checkpoint-predictive gene signature was found to have prognostic significance for development of distant metastasis in patients with primary melanoma, suggesting that the checkpoint may retard malignant progression.

Results

Defective G₂ checkpoint function in melanoma cell lines

NHM cultures displayed a significant inhibition of mitosis when analyzed 2 h after treatment with IR (1.5 Gy), indicative of G₂ arrest (Fig. 1A). The ratios of mitotic cells in IR-treated cultures in comparison to their sham-treated controls (IR/sham ratio) ranged from 2–20% among eight NHM cultures. The three new NHM cultures that were assayed in the current analysis responded to IR with the same degree of inhibition of mitosis as was seen previously in five other melanocyte cultures.²² NHM strains displayed an average 90% inhibition of mitosis 2 h after 1.5 Gy of IR, indicating that only 10% of melanocytes evaded the checkpoint-induced G₂ arrest.

Figure 1. G₂ checkpoint function in NHM and melanoma cell lines. Flow cytometry was used to quantify mitotic cells with expression of phospho-H3 (ser10).²² G₂ checkpoint function was scored as the ratio of the fractions of cells in mitosis 2 h after treatment with 1.5 Gy IR or a sham treatment (IR/sham ratio). The values shown represent individual determinations in NHM11, NHM16 and NHM18 cultures and mean values in melanoma cell lines (+ sd, n = 2–6). Black bars depict newly determined values while gray bars are from our previous report.²² All melanoma lines with values > 0.3 were determined to have defective G₂ checkpoint function in comparison to NHMs (p < 0.05 after correction for multiple comparisons). The SK-Mel-173 line (*) was also defective. (B) G₂ checkpoint function in melanoma cell lines belonging to three subtypes, “epithelial-like” with wild-type NRAS and BRAF (WT), NRAS- mutant and BRAF-mutant. Results represent mean + sd (WT, n = 5; NRAS, n = 8; BRAF, n = 22). (*) Denotes significantly different from WT (p = 0.002 for NRAS and p = 0.001 for BRAF). NRAS and BRAF were not significantly different (p = 0.1).

Melanoma cell lines displayed significant attenuation of DNA damage G₂ checkpoint function in comparison to NHM, as evidenced by increased fractions of cells present in mitosis 2 h after irradiation (Fig. 1A). When analyzed as a class and compared against the normal melanocytes, the melanoma lines displayed a significant defect in checkpoint function (p = 0.02). When compared on an individual basis and corrected for multiple comparisons, 13 of 35 or 37% of the melanoma lines displayed a significant defect in G₂ checkpoint function (Fig. 1A). The apparent cutoff for effective vs. defective G₂ checkpoint function was 0.28–0.3. The WM2664 melanoma line displayed an average IR/sham ratio of 0.30, which was not significantly different from the NHM control (FDR-adjusted p value = 0.08), while the SK-mel173 line displayed an average ratio of 0.28, which was significantly different from the NHM control (FDR-adjusted P value = 0.04). All melanoma lines with ratios > 0.3 were significantly different from the NHM controls and thus defective for G₂ checkpoint function. The melanoma lines displayed a broad range of responses to IR, with several lines responding to IR with severe G₂ arrest (ratios < 0.05) and three lines with a highly attenuated response (ratios > 0.5).

As previously noted²² and now extended to this larger set of melanoma cell lines, there was significant melanoma subtype-specific variation in G₂ checkpoint function (Fig. 1B). Epithelial-like melanoma lines with wild-type NRAS and BRAF alleles (here designated WT) displayed effective G₂ arrest after IR, while lines with mutant NRAS or BRAF displayed significant attenuation of G₂ checkpoint function (Fig. 1B).

Prior studies indicated that overexpression of RAS family oncogenes could attenuate G₂ checkpoint function.^23-25 To test whether mutation of BRAF was responsible for defective G₂ checkpoint function in melanoma cell lines, V5 epitope-tagged oncogenic B-Raf (V5V600E) under the control of a tetracycline-responsive promoter was induced in a G₂ checkpoint-effective melanoma line with wild-type BRAF and NRAS alleles. Addition of 1 μg/ml doxycycline (dox) to culture medium induced the expression of oncogenic B-Raf in RPMI8332 cells and accordingly enhanced expression of P-MEK1/2 (Fig. 2A). Induction of oncogenic B-Raf was also associated with attenuation of the G₂ arrest in response to treatment with IR (Fig. 2B). The attenuation of G₂ checkpoint function upon induction of oncogenic B-Raf was statistically significant (p = 0.02) but modest, with a 3-fold increase in the fractions of cells evading G₂ arrest (Fig. 2C). A control experiment demonstrated that the addition of 1 μg/ml dox did not attenuate G₂ checkpoint function in the parental line carrying the tetracycline transactivator but without oncogenic B-Raf (not shown). Thus, activation of MAPK signaling likely contributed to the defective G₂ checkpoint function seen in the BRAF-mutant melanoma cell lines, but other changes were probably needed to produce the severe attenuation seen in some of the lines.

Figure 2. Stimulation of MAPK signaling by oncogenic B-Raf attenuates G₂ checkpoint function in melanoma cells. (A) Western immunoblot showing induction of V5-V600E B-Raf and stimulation of MAPK signaling (P-MEK1/2) in the melanoma line RPMI8332. Cells were incubated with 1 μg/ml doxycycline (dox) for 48 h before cell harvest. (B) IR-induced G₂ arrest. Flow cytometry was used to quantify mitotic cells 2 h after treatment with 1.5 Gy IR given at 48 h after adding dox. (C) Average IR/sham ratios with and without induction of oncogenic B-Raf. Induction of oncogenic B-Raf for 24–48 h produced a modest but significant (p = 0.02, n = 5, Student’s t-test) attenuation of IR-induced G₂ arrest.

A signature of G₂ checkpoint function in melanoma cell lines

The melanoma cell lines were analyzed by DNA microarray to determine global gene expression during basal growth. Expression of mRNA species in individual melanoma cell lines (measured by Cy5) was determined relative to a universal reference standard (measured by Cy3) that was mixed with each sample. Cy5/Cy3 ratios were determined as a measure of gene expression using 44,000 probes on an Agilent microarray. A quantitative trait analysis (QTA) tool was implemented that determined the Spearman correlation coefficient for the relationship between G₂ checkpoint function and gene expression. Table 1 shows that as the p value for this correlation was reduced incrementally from 0.007 to 0.002, the number of selected genes was reduced. At each increment the least absolute shrinkage and selection operator (LASSO) was applied in a leave-one-out cross validation (LOOCV) analysis to return a predicted value for G₂ checkpoint function. The R² correlation between observed and predicted G₂ checkpoint function was maximal (0.62) when the p value of the QTA was ≤ 0.004. At this p value cutoff, the predictive accuracy of LASSO was 85% (i.e., 85% of melanoma lines were correctly predicted to have effective or defective G₂ checkpoint function (results not shown) with 370 probes being selected.

Table 1. QTA and LASSO tools identify G₂ checkpoint-correlated probes

Sig. level^a	0.002	0.003	0.004	0.005	0.006
Probe list^b	205	290	370	434	515
LOOCV- R-sq^c	0.48	0.60	0.65	0.51	0.53

^a Significance threshold of the Spearman correlation test using QTA; ^bNumber of significant probes. The number of probes passing primary filtration was 23,354; ^cPrediction of G₂ checkpoint function using leave-one-out cross-validation (LOOCV). R-sq is the proportion of variation in G₂ checkpoint function that is explained by the model built using the LASSO algorithm.

A second analysis used a Bayesian statistical tool to identify genes whose expression was correlated with G₂ checkpoint function in the melanoma cell lines. The Bayesian tool returned a list of genes with variable false discovery rate (FDR), the lower the FDR the greater the correlation. The 370 probes with the lowest FDR (≤ 0.077) were collected and subjected to LASSO with LOOCV. The R² correlation of the Bayes 370 probe list was 0.64 with a predictive accuracy of 88% (not shown). A total of 165 probes were present on both the QTA and Bayes lists (the QTA370, Bayes370 and overlapping 165 probe lists are given in the Supplemental Materials packet). These probes had a low p value (≤ 0.004) and low FDR (≤ 0.077). For this list of overlapping probes, the R² correlation by LASSO was 0.64, and its predictive accuracy by LOOCV was 94% as shown in Figure 3A. Because the predictive accuracy of the 165 probe intersection list was greater than the QTA370 and Bayes370 lists, it was used for subsequent analyses.

Figure 3. G₂ checkpoint probe list predicts G₂ checkpoint function. (A) The 165 probe list that was identified as the overlap between the QTA 377 probe list with p ≤ 0.004 and the 377 probe Bayes list with FDR ≤ 0.077 was subjected to LASSO with LOOCV. LASSO returned an equation for prediction of G₂ checkpoint function as: ∑_i c_ix_i + 0.399, where c_i and x_i are the coefficient and expression for the i-th transcript, respectively (Supplemental Materials packet). Values of G₂ checkpoint function that were predicted by LASSO were plotted against the observed values. The cutoff value for effective vs. defective G₂ checkpoint function was 0.31. LASSO correctly classified G₂ checkpoint function for 94% of the melanoma cell lines. The two misclassified lines were the checkpoint-effective line SK-Mel-27 with an observed IR/Sham ratio of 0.18 and a predicted ratio of 0.41 and the checkpoint defective line SK-Mel-173 with an observed ratio of 0.28 and a predicted ratio of 0.22. (B) Unsupervised hierarchical cluster of 35 melanoma cell lines using the 165 probes that were correlated with G₂ checkpoint function (p ≤ 0.004 and FDR ≤ 0.077). Gene expression in melanoma cell lines was visualized using the method of Eisen et al.,⁶⁰ which organizes lines and transcripts according to similarity. Transcripts that were expressed at greater levels than the group median are shown in red, and transcripts that were expressed at lesser levels than the group median are shown in green. The intensity of color is proportional to the separation from the median. The two major clades (branches) in the melanoma cell line dendogram largely represented checkpoint-defective (left clade) and checkpoint-effective (right clade) classes. There were eight lines that were misclassified for a correct classification rate of 77%. Melanoma lines with mutations in BRAF and NRAS were represented in both classes. Melanoma lines with wild-type BRAF and NRAS were restricted to the checkpoint-effective class. Chi-square analysis indicated significant separation of checkpoint-effective and -defective lines (χ² = 10.8, p = 0.02).

Hierarchical clustering of the melanoma cell lines using the 165 probe list produced two major clades with significant separation of checkpoint-effective and -defective lines (χ² = 10.8, p = 0.02). The predictive accuracy of this binary hierarchical cluster analysis was 77% (Fig. 3B).

Genes on the 165 probe list

Gene ontology analysis using Expression Analysis Systematic Explorer identified one biological category that was significantly over-represented in the 165 probe list, the lysosome. Eight lysosomal genes were on the list, STX3, USE1, FAM176A, ARSA, CTSA, DPP7, M6PR and ACP5. The commercial Ingenuity Pathways analytic tool did not identify a biological process or pathway as significantly associated with the G₂ checkpoint signature list when correcting for multiple comparisons using the Benjamini-Hochberg method. However, manual curation of the 165 probe list revealed numerous genes whose expression could be expected to influence cell proliferation and/or G₂ checkpoint function (Fig. 4 and Table 2). Many probes on the list act at steps in a KIT receptor tyrosine kinase-initiated, MAPK-dependent signaling cascade leading to DNA replication and culminating in mitosis/cytokinesis. A total of 31 probes on the 165 probe list encoded proteins that perform biological functions that are connected to this signaling cascade. Numerous biochemical activities were represented on the list, including ubiquitin ligation, de-ubiquitination, protein kinase, protein phosphatase, transcription factor, proteosome subunits and histone methylation and acetylation (Table 2).

Figure 4. G₂ checkpoint-correlated genes encode mediators of a receptor tyrosine kinase signaling cascade initiating DNA replication and culminating in mitotic cell division. KIT activates the MAPK signaling cascade to initiate DNA replication. DNA replication stress produces DNA double-strand breaks, which activate the sensor checkpoint kinases ATM and ATR and the transducer checkpoint kinases Chk1 and Chk2. Inhibition of Cdc25B blocks the activation of mitosis-promoting factor (cyclin B1/Cdk1) to enforce the G₂ checkpoint. Once activated in the nucleus, mitosis-promoting factor activates the anaphase-promoting complex to complete chromatid segregation (karyokinesis) and cell division (cytokinesis). Additional signaling from MITF and TGFβ impinges on the core pathway. Checkpoint-correlated gene products are enclosed in ovals near the proteins or processes that they regulate (see Table 2 for annotations). Arrows denote activation and blunt ends denote inhibition.

Table 2. G₂ checkpoint-correlated genes encode products that mediate regulation of cell division and the G₂-M transition

Gene name	Biochemical activity	Biological function	References
DLG3	Guanylate kinase	Polarity, proliferation	61
PTPRM	protein tyrosine phosphatase	Migration, proliferation	62
MAP3K5	MAPK kinase kinase	P38 kinase activation	40
RHOQ	GTP binding	Motility, attachment	63
YPEL4	Interacts with Vault proteins	Regulate MAPK pathway	64
DOCK10	Rho guanine exchange factor	migration	65
DAAM2	Stabilize Dvl3/Axin2 binding	Wnt signaling pathway	66
β-catenin	Transcription factor	Wnt signaling pathway	38
ICAM1	Surface glycoprotein	Adhesion, metastasis	67
TWSG1	BMP antagonist	SMAD repression	68
MKL2	Transcription factor	Response to serum	69
HSF4	Transcription factor	Regulate senescence	70
WDHD1	Initiation of DNA replication	Load DNA pol α	26
KAT2B	Histone acetyltransferase	Transcriptional coactivation; regulated by WDHD1	71 and 72
MMS22L	Ubiquitin ligase	Restart collapsed forks	73
ARNTL2	Transcription factor	Circadian rhythm	74
HERC2	Ubiquitin ligase	DNA repair, replication and checkpoint function	75
ANKRD28	Phosphatase subunit	Inactivate AuroraA	76
CTDSPL2	phosphatase	Dephosphorylate RNA polII	77
MNAT1	Subunit of CAK	CDK activation, transcription, DNA repair	50
PP4R1	Phosphatase subunit	Checkpoint recovery	78
SETDB2	Histone methylase	Enhance prophase to anaphase transition	79
SPG20	Microtubule interacting	Enhance cytokinesis	80
CKS1B	CDK1 regulator	Direct cyclinB1/Cdk1 to APC/C, p27 degradation	81 and 82
FOXA1	Transcription factor	Stimulate ubiquitylation of cyclin B1	83
FBXO5	Ubiquitin ligase	Inhibit APC	84
WAPAL	Cohesin interacting protein	Sister chromatid resolution	85
H2AFY2	Chromatin condensation	Senescence-associated heterochromatin, melanoma suppression	52
PSMB6	Y Subunit of proteosome	Ubiqutin-mediated Proteolysis	86
PSMD7	Subunit of proteosome	G2/M transition	87

The G₂ checkpoint signature gene list included proteins that regulate cell proliferation and migration at numerous steps in the proliferation signaling cascade (Fig. 4 and Table 2). Several proteins regulate mitogen-activated protein kinase signaling (RHOQ, YPEL4, DOCK10, MAP3K5), others regulate melanogenesis through the master transcription factor MITF (β-catenin, DAAM2, ICAM1), and a third cluster appears to influence signaling through the TGF-β receptor (TWSG1, MKL2). A substantial number of proteins on the list participate in DNA metabolism and mitotic cell division. WDHD1, KAT2B, MMS22L, ARTNL2 and HERC2 are implicated in various forms of DNA metabolism or damage response and MNAT1, SETDB2, CKS1B, PP4R1, FBXO5, FOXA1, WAPAL and SPG20 are associated with steps of mitosis (prophase, metaphase, anaphase, telophase) culminating in cytokinesis. Thus, the core machinery of the cell division cycle did not appear to be altered in melanoma cell lines in a way that was correlated with G₂ checkpoint function, while many regulators of the core machinery were.

Notable probes on the G₂ checkpoint signature list included CKS1B, WDHD1 and H2AFY2 (Fig. 5). CKS1B is a subunit of the cyclin B1/Cdk1 mitosis-promoting factor (MPF) kinase and mediates MPF interaction with the anaphase-promoting complex/cyclosome (APC/C) that ubiquitinates cyclin B1 to trigger the metaphase-to-anaphase transition in mitotic cells. CKS1B mRNA was comparatively low in primary and hTERT-transduced melanocytes and high in melanoma cell lines, consistent with the enhanced proliferation of the melanoma cell lines (Fig. 5). Expression of CKS1B was also high in G₂ checkpoint-effective melanoma cell lines and low in checkpoint-defective lines. All three melanoma subtypes displayed reduced expression with decreased G₂ checkpoint function, but only in the BRAF-mutant subtype lines was the correlation significant (p = 0.02).

Figure 5. Genes may regulate G₂ checkpoint function in a melanoma subtype-dependent and -independent manner. NHMs and NHM lines that were transduced with hTERT are depicted in black. WT melanoma lines are depicted in blue, NRAS-mutant lines in red and BRAF-mutant lines in purple. Gene expression is depicted as the ratio of mRNA expression in experimental samples (Cy5) divided by expression in a pool of cancer cell lines (Cy3, Stratagene universal reference mRNA). CKS1B and WHDH1 generally were expressed at higher levels in melanoma cell lines than in NHMs. Among melanoma lines, all three subtypes displayed a negative correlation between CKS1B and WDHD1 expression and G₂ checkpoint function. H2AFY2 was expressed at low levels in NHMs expressing telomerase and in some NRAS- and BRAF-mutant melanoma lines. Expression of H2AFY2 was uniformly high in “epithelial-like” WT lines. Among BRAF-mutant melanoma lines H2AFY2 was not correlated with G₂ checkpoint function.

WDHD1, also known as AND1 and CTF4, has been shown to contribute to loading of DNA pol α at replicon origins through a direct interaction with a checkpoint mediator protein Tipin.²⁶ Tipin is a component of the replication fork protection complex that mediates both the intra-S checkpoint response to UV-induced DNA damage^27,28 and the G₂ checkpoint response to IR-induced DNA damage.²⁹ Low WDHD1 expression was associated with defective G₂ checkpoint function (Fig. 5). The correlation of WDHD1 expression with G₂ checkpoint function was significant for both the NRAS-mutant (p = 0.05) and BRAF-mutant (p = 0.01) melanoma lines. WDHD1 expression was also higher in the melanoma cell lines than in the primary and telomerized melanocytes.

H2AFY2, also known as macro-H2A2, is a minor histone variant that is concentrated in senescence-associated heterochromatin.³⁰ Expression of H2AFY2 was uniformly high in checkpoint-effective WT melanoma cell lines, in which decreased G₂ checkpoint function was correlated with increased expression of H2AFY2 (Fig. 5, p = 0.02). The reduced expression of H2AFY2 in NRAS- and BRAF-mutant melanoma cell lines was not correlated with reduced G₂ checkpoint function (p > 0.05). Further inspection of the data revealed substantial variation in expression of H2AFY2 in NHM cultures and melanoma cell lines. In comparison to primary NHM cultures, H2AFY2 expression was suppressed in NHM lines that were transduced with hTERT to induce telomerase activity. Expression of H2AFY2 in NHM primary cultures and NHM-hTERT lines did not differ with supplementation of melanocyte growth medium with phorbol myristylate (PMA) or endothelin-1 (ET1). The NHM-hTERT lines displayed a very low level of expression of H2AFY2 similar to that seen in three of nine melanoma lines with NRAS mutations and 13 of 23 melanoma lines with BRAF mutations. In contrast, the WT melanoma lines uniformly displayed high expression of H2AFY2.

NHM and melanocytic nevi express low levels of telomerase and nodular melanoma, metastatic melanoma and melanoma cell lines express high levels of telomerase^31,32 indicating that induction of telomerase is associated with malignant transformation of melanocytes. The repression of H2AFY2 expression in melanocytes that were transduced with hTERT to induce telomerase activity and immortality suggested that expression of H2AFY2 might be connected to the senescence pathway that recognizes deprotected telomeres that develop during cellular aging. Telomerase enzyme activity was quantified to test whether the elevated expression of H2AFY2 in WT melanoma cell lines affected expression of telomerase. Melanoma lines with high expression of H2AFY2 (PMWK, RPMI8322) displayed high levels of telomerase similar to those seen in melanoma lines with intermediate or low expression of H2AFY2 (Fig. S1). Thus, it appeared that expression of H2AFY2 was not correlated with the induction of telomerase activity in melanoma.

Prognostic significance of G₂ checkpoint signature list

To test the biological significance of the G₂ checkpoint signature list, an independent melanoma microarray database was analyzed. Winnepenninckx et al. quantified gene expression in 58 primary melanomas that did or did not develop distant metastases within 4 y of primary diagnosis.³³ The primary Agilent microarray data were downloaded from an open source and loaded into BRB array tools to generate Cy5/Cy3 ratios as indices of gene expression; a total of 6307 probes passed filtration and were available for analysis.³⁴ Of the 165 probes on the G₂ checkpoint signature list, 32 were present in the filtered melanoma data set. These 32 probes were analyzed with a survival risk prediction (SRP) tool to generate two groups, one with low risk of development of distant metastasis, the other with high risk (Fig. 6). The 32 probes that were in the G₂ checkpoint gene signature returned a significant separation of outcome in the melanoma patient samples (p = 0.018). Thus, in an independent melanoma data set, genes whose expression was correlated with DNA damage G₂ checkpoint function were prognostic of melanoma outcome.

Figure 6. The G₂ checkpoint signature list is prognostic for melanoma progression. Primary melanomas were analyzed for global gene expression using the Agilent microarray platform. A 6307 probe set was downloaded from an open source and analyzed for probes that were prognostic for development of distant metastases within four years of resection of the primary tumor.^8,33 The 165 probe set that was correlated with G₂ checkpoint function was used with survival risk prediction to generate two patient groups, one with low probability of distant metastasis, one with high probability. The separation of the two groups in the Kaplan-Meier survival plot was significant (χ² = 10.8, p = 0.002).

A similar analysis of DNA damage G₁ checkpoint function in melanoma cell lines generated a different signature of gene expression that also was prognostic of melanoma progression.³⁴ The G₁ and G₂ checkpoint signatures contained no overlapping probes. To test whether a combination of the two prognostic tests performed better than either individual test, SRP was done to identify patients who were classified as high risk in both signatures or low risk in both signatures. The separation of metastasis incidence for the high/high and low/low groups was greater that than obtained using the G₂ signature alone but less than that obtained using the G₁ signature alone. The χ² values for the individual tests were 5.6 (p = 0.018) and 10.9 (p = 0.001) for G₂ and G₁, respectively, and the χ² value for the combined test was 9.7 (p = 0.002). Combining the G₁ and G₂ checkpoint signatures did not improve prognostic identification of patients with high risk to develop a distant metastasis, in comparison to the G₁ signature used alone.

Discussion

Melanoma cell lines displayed substantial variation in DNA damage G₂ checkpoint function with 37% being classified as defective in comparison to NHM cultures. Expression of oncogenic B-Raf appeared to attenuate G₂ checkpoint function only modestly, implying that other genetic or epigenetic alterations are required to produce the severely defective G₂ checkpoint function seen in some melanoma cell lines. Bioinformatic analyses of global gene expression identified 165 probes that were correlated with G₂ checkpoint function with low p-values and low FDR. Gene products that are translated from these mRNAs were not core components of the cell cycle but rather mediators. Probes on the G₂ checkpoint signature list were prognostic for melanoma progression in an independent data set.

Defects in G₂ checkpoint function were restricted to melanoma cell lines with mutant BRAF and NRAS oncogenes. Many reports have associated defects in G₂ checkpoint function with activation of the MAPK signaling pathway,^23-25,35 but few studies have focused on the effects of oncogenic B-Raf. Expression of oncogenic B-Raf at levels sufficient to activate MEK1/2 produced a modest but significant attenuation of G₂ checkpoint function in the RPMI8322 melanoma line. As some melanoma cell lines displayed a severe attenuation of the G₂ checkpoint, the results suggest that other melanoma-associated phenotypic alterations also influence the G₂ checkpoint response to IR-induced DNA damage.

Two proteins in the melanomagenesis pathway, PTEN and AKT, have been reported to interact with or influence the checkpoint transducer kinase CHK1.^36,37 Inactivation of PTEN appeared to reduce CHK1 kinase activity within the nucleus by cytoplasmic sequestration.³⁷ AKT was shown to phosphorylate CHK1 on ser180 to promote its cytoplasmic sequestration.³⁷ Reduced activity of PTEN or activation of AKT therefore might attenuate the G₂ checkpoint response to IR-induced DNA damage by deregulating CHK1. BRAF mutations appear to cooperate with inactivation of PTEN in melanomagenesis.^4,10,38 In the current series, many of the melanoma lines with mutations in BRAF were identified with severe reductions in PTEN mRNA levels. However, PTEN mRNA expression was not correlated with G₂ checkpoint function (not shown). The WM35 melanoma line with mutant BRAF and a severe defect in G₂ checkpoint function expressed PTEN mRNA at levels seen in NHM and telomerase-expressing NHM. Moreover, the checkpoint-effective PMWK line displayed very low expression of PTEN mRNA associated with homozygous deletion of the PTEN gene locus (results not shown). Reduced expression of PTEN mRNA did not account for the reduced G₂ checkpoint function in BRAF-mutant melanoma cell lines. Activation of AKT was recognized in many of the melanoma cell lines studied here and the “epithelial-like” melanoma lines with wild-type NRAS and BRAF expressed lower levels of phospho-AKT than the lines with oncogene mutations.³⁹ Mutational inactivation of PTEN and/or mutational activation of AKT (without alteration in mRNA levels) might contribute to reduced G₂ checkpoint function in BRAF-mutant melanoma cell lines. A parallel branch of the MAPK pathway leading to activation of ERK1/2, p38SAPK and JNK includes MAP3K5 and MAP3K9. Inactivating base-substitution mutations in MAP3K5, a member of the 165 probe G₂ checkpoint signature list, were found in 9% of melanoma cell lines.⁴⁰ MAP3K5 mRNA was negatively correlated with G₂ checkpoint function; thus, reduced expression of MAP3K5 mRNA was associated with increased fractions of cells evading IR-induced G₂ arrest.

Bioinformatic analysis of the G₂ checkpoint signature list identified the lysosome as a potential modifier of checkpoint function. The lysosome is a key organelle in the process of autophagy that degrades intracellular proteins by a non-proteosomal mechanism.⁴¹ Autophagy has been linked to activation of cytoplasmic ATM,⁴² a key transducer kinase in the G₂ checkpoint immediately upstream of CHK1.^12,43,44 Autophagy is also regulated by protein acetylation and deacetylation,^45,46 and recent studies demonstrate the requirement of protein acetylation and deacetylation in the response to DNA double-strand breaks.⁴⁷ The presence of the histone acetyltransferase KAT2B on the G₂ checkpoint gene list provides a possible connection to lysosomes and autophagy.^47,48

The list of checkpoint-correlated probes contained many genes that are connected to regulation of mitosis, including CKS1B, WHDH1 and MNAT1. A recent study has associated alterations in CKS family proteins with override of the replication checkpoint that blocks the initiation of mitosis in cells with incomplete DNA replication.⁴⁹ The protein product of WDHD1, AND1, is known to interact with Tipin, an established mediator of G₂ checkpoint function.²⁹ MNAT1 is a subunit of CAK, the CDK-activating kinase that phosphorylates CDK1 on thr161.⁵⁰ It is conceivable that a variety of combinations of alterations in gene expression are manifested as a functional defect in the ability to respond to IR-induced DNA damage and arrest in G₂.

H2AFY2 appeared to be on the G₂ checkpoint signature list because of its high expression in the “epithelial-like” melanoma lines and reduced expression in NRAS- and BRAF-mutant lines. Among the BRAF-mutant lines, expression of H2AFY2 was not correlated with G₂ checkpoint function. Macro-H2A.2, the protein product of the gene locus, has been linked to cellular aging being induced in senescent cells.⁵¹ Macro-H2A proteins appear to concentrate in senescence-associated heterochromatin, suggesting that macro-H2A may help to organize a repressive chromatin structure that enforces senescence. Macro-H2A was shown to repress CDK8 in a melanoma cell line as a potential mechanism of tumor suppression.⁵² In the current data series, there was no correlation between H2AFY2 and CDK8 mRNA expression among the 35 melanoma cell lines (data not shown). The repression of H2AFY2 expression in melanocytes that were transduced with hTERT to induce telomerase activity and immortality suggested that expression of H2AFY2 might be connected to the mechanisms of regulation of telomerase and immortality in melanoma. However, no correlation between H2AFY2 and telomerase enzyme activity was observed in the melanoma cell lines, and lines with high expression of H2AFY2 expressed telomerase activity at the same levels as lines with low expression of H2AFY2. The biological significance of high expression of H2AFY2 in WT melanoma cell lines remains to be determined.

Given that DNA damage checkpoints slow cell cycling and stabilize the genome,^53-55 it is axiomatic that defects in checkpoint function enhance neoplastic growth and genomic instability, thereby fueling malignant progression. This facet of cancer biology may be reflected in the result that signatures of gene expression that were predictive of G₁ and G₂ checkpoint function were also prognostic for a measure of melanoma disease progression, i.e., development of a distant metastasis. The G₁ and G₂ checkpoint signatures did not contain any overlapping transcripts, and there was no correlation between G₁ and G₂ checkpoint function in the melanoma lines (R² < 0.001). The result that combination of the G₁ and G₂ signatures did not improve prognosis suggests that the DNA damage G₁ and G₂ checkpoints use different pathways to enforce a common barrier to melanomagenesis.

Materials and Methods

Molecular biology and viral transduction

A V5-tagged oncogenic BRAF (V600E) insert was synthesized and cloned into the pRetro-X-Tight-Hyg (Clonetech) vector by Blue Heron Biotechnology. The V5-BRAF insert was confirmed by sequenceing (data not shown). Virus was packaged and cells infected using standard methods as described previously.⁵⁶ The RPMI8322 cell line that is wild-type for NRAS and BRAF was first infected with the Retro-X-TetON-Advanced (Clonetech) virus to express the TetON transactivator. Following 1 wk of selection in 500 µg/ml G418 (Life Technologies), the cells were infected with the V5-BRAF-expressing virus. These cells were subsequently selected in 100 µg/ml Hygromycin B (Roche) for 1 wk prior to use.

Cell culture and assay of G₂ checkpoint function

Normal human melanocytes (NHMs) and melanoma cell lines were grown and assayed for G₂ checkpoint function as previously described.²² Cells were routinely assayed for mycoplasma contamination using commercial kits (Geneprobe, Gene-Probe and PlasmoTest^™, Invivogen). Contaminated cultures were cleansed with ciprofloxacin and/or a commercial antibiotic (Plasmocin, Invivogen). Results are reported only for mycoplasma-negative cell cultures. NHM cultures were grown in growth medium supplemented with phorbol myristylate (PMA) as previously described²² or when indicated PMA was replaced with endothelin-1 (ET1) at 100 ng/ml. To induce expression of telomerase and generate immortal lines NHM were transduced with hTERT by retroviral infection and selection with puromycin (50 μg/ml) as previously described⁵⁶ then grown with PMA or ET1.

G₂ checkpoint function was quantified by flow cytometric determination of the fraction of cells that were positive for expression of phospho-ser10 histone H3 as measured 2 h after irradiation with 1.5 Gy IR using a RS2000 Biological Irradiator (Rad-Source). The fraction of cells in mitosis in IR-treated cultures was expressed as a percentage of the same fraction scored in sham-treated controls.

The RPMI8322 cell line and its derivative were grown in high-glucose DMEM (Life Technologies or Cellgro) supplemented with 10% bovine growth serum (Hyclone). Addition of 1 μg/ml doxycycline (dox) to culture medium activated the tetracycline transactivator and induced expression of V5-tagged, oncogenic B-Raf (V5V600E). The parental RPMI8322 melanoma line was shown to display a severely defective G₁ checkpoint response to ionizing radiation, demonstrating functional inactivation of p53 signaling.⁸ Expression of V5V600E was monitored by immunoblot analysis using antibodies against V5 (Sigma), B-Raf (Cell Signaling) and phospho-MEK1/2 (Cell Signaling). Cells were assayed for G₂ checkpoint function 24 or 48 h after addition of 1 μg/ml dox.

G₂ checkpoint function was compared between individual melanoma cell lines and a group of eight NHM strains, based on a linear mixed effects model⁵⁷ to account for the potential correlation of multiple measurements (median of 3 with interquartile range of 2–4) per cell line. A one-sided t-test used the predicted empirical Bayesian estimate of each cell line from the output of the linear mixed effects model to identify melanoma cell lines with significantly reduced IR-induced mitotic inhibition in comparison to the set of NHMs. P values were adjusted for multiple comparisons using the method of Benjamini and Hochberg.⁵⁸

Signature of G₂ checkpoint function

Analysis of gene expression using cDNA microarrays was done as previously described.^8,22 mRNA expression in NHM and melanoma cell lines (Cy5 dye) was expressed relative to a universal human mRNA standard (Cy3 dye). The log₂ ratio of Cy5 to Cy3 intensities was computed for each probe.

Analysis of microarray data was done using BRB array tools software.⁵⁹ Quantitative trait analysis (QTA) used the Spearman correlation coefficient to identify genes whose expression was correlated with G₂ checkpoint function in melanoma cell lines. A Bayesian method for selection of significant genes was applied as previously described²² to identify G₂ checkpoint-correlated genes with low false-discovery rate (FDR < 0.077). The least absolute shrinkage and selection operator (LASSO) was used with leave-one-out cross validation (LOOCV) to predict G₂ checkpoint function from gene expression signatures. Hierarchical clustering was also used to predict G₂ checkpoint-effective and -defective melanoma cell lines.

Gene ontology analysis was done using DAVID 6.7 (http://david.abcc.ncifcrf.gov/). The Ingenuity Pathways tool was used to search for biological functions and pathways that are associated with genes in the G₂ checkpoint signature. The survival risk prediction BRB array tool was used to determine whether a G₂ checkpoint gene signature predicted development of distant metastasis in an independent data set³³ as was done previously using a G₁ checkpoint gene signature.⁸

Abbreviations:
BrdU	=	5′-bromo-deoxyuridine
DDR	=	DNA damage response
dox	=	doxycycline
ET1	=	endothelin-1
FDR	=	false-discovery rate
IR	=	ionizing radiation
LASSO	=	least absolute shrinkage and selection operator
LOOCV	=	leave-one-out cross-validation
MAPK	=	mitogen-activated protein kinase
NHM	=	normal human melanocyte
OIS	=	oncogene-induced senescence
PMA	=	phorbol myristylate
SRP	=	survival risk prediction
TTA	=	tetracycline transactivator
QTA	=	quantitative trait analysis

Acknowledgments

We are grateful to Drs Jayne Boyer and Leena Nylander-French for providing primary melanocyte cultures. Supported in part by PHS grants ES10126, CA16086 and ES014635. This research was supported in part by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Supplemental Materials

Supplemental materials may be found here: 
www.landesbioscience.com/journals/cc/article/24067