MYC pathway activation in triple-negative breast cancer is synthetic lethal with CDK inhibition

We sought to understand whether MYC signaling is an important oncogenic event in triple-negative breast tumors. The I-SPY TRIAL (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis) is an NCI-sponsored multi-institutional study of Stage II and III breast cancer patients to identify diagnostic markers, validate hypothesis, and develop new treatment strategies against breast cancer (Barker et al., 2009; Jones, 2010; Esserman et al., 2012). This clinical trial collects extensive clinical annotation and global mRNA expression from primary tumor samples. Using mRNA expression data from 146 primary patient tumor samples for which hormone receptor and HER2 expression were available, we found that MYC mRNA levels are significantly elevated in triple-negative tumors (Fig. 1 A). We next examined MYC protein and phospho-MYC (T58/S62) expression in 208 independent primary breast tumor samples from a separate cohort using reverse-phase quantitative protein arrays. The 66 triple-negative tumors expressed significantly elevated MYC protein and phospho-MYC (Fig. 1 B) compared with the 142 tumors that were receptor positive. MYC phosphorylation at these sites has been shown to enhance MYC transcriptional activity as well as alter protein stability (Chang et al., 2000; Sears et al., 2000; Seo et al., 2008). Finally, using previously published mRNA expression profiling data (Neve et al., 2006), we examined whether MYC expression in 50 established human breast cell lines correlated with the receptor status. These 50 human breast cell lines have been shown to accurately model aspects of breast cancer biology (Neve et al., 2006) and have been extensively used for mechanistic studies. Compared with cell lines that are not triple negative, triple-negative cell lines had substantially elevated MYC mRNA (Fig. 1 C). Collectively, these three independent datasets confirm the association between elevated MYC expression and triple-negative tumors.

To evaluate MYC activity in primary tumors, we applied a previously validated MYC-regulated transcriptional signature (Chandriani et al., 2009), which is comprised of 352 genes, to the I-SPY dataset. We ordered the patient tumor samples by the Pearson correlation to the MYC gene expression signature (Fig. 2 A), and tumors were divided into high, intermediate, and low correlation groups (see Materials and methods). Triple-negative breast tumors were significantly enriched in the high MYC gene expression group (P < 0.005; Fig. 2 A). These results indicate that a disproportionate number of primary triple-negative breast tumors exhibit elevated MYC function. A MYC gene expression signature has previously been correlated with a basal molecular subtype of breast cancer (Alles et al., 2009; Chandriani et al., 2009; Gatza et al., 2010), which encompasses ∼70% of triple-negative cancers. Whether MYC signaling is also increased in other molecular subtypes of human breast cancer remains unclear. We therefore examined the MYC gene expression signature across different molecular subtypes of breast cancer in the I-SPY dataset. Consistent with prior studies, we found that a high MYC gene expression signature correlated most strongly with the basal subtype; however, we also observed that the luminal B molecular subtype had significantly higher MYC signature than luminal A or HER2 subtypes (Fig. 2 B). Although most luminal B tumors express hormone receptors, this subtype is often associated with increased tumor proliferation and worse outcome than luminal A (Voduc et al., 2010). Our results indicate that along with basal, the luminal B molecular subtype is also associated with increased MYC signaling, which may, at least in part, explain the worse outcome for this tumor type (Cheang et al., 2009).

We next sought to understand the molecular basis for the increased MYC activity in triple-negative breast cancers. Activation or repression of MYC target genes involves multiple signaling complexes comprised of MAX bound to an activator, such as MYC or MYCN, or bound to a repressor such as MXD1-4 or MAX (Grandori et al., 2000; Cowling and Cole, 2006). Thus, MYC/MAX heteromeric complexes can activate transcription, whereas MXD/MAX or MAX/MAX complexes can suppress activation of MYC target genes (Eilers and Eisenman, 2008). Changes in MYC pathway activity in breast cancer are therefore caused by alterations in the architecture of the MYC regulatory network through changing the relative abundance of MAX interacting partners. To our knowledge, the topology of the MYC regulatory network in primary breast tumors has not been previously described. Therefore, we sought to determine if, besides MYC, the mRNA expression of other MAX-interacting genes is also significantly altered in triple-negative primary tumors. Analyzing the mRNA array data available from the I-SPY dataset, we found that multiple MAX-interacting genes are altered in triple-negative tumors (Fig. 2 C). In addition to MYC, we found that the activator MYCN is also up-regulated in triple-negative tumors, whereas the repressor molecules MXD4 and MAX were significantly down-regulated (Fig. 2 C). Thus, the enhanced MYC activity in triple-negative tumors may be a result not only of MYC up-regulation but also of alterations in the expression of other MYC regulatory genes.

We next examined the clinical outcome of patients based on their MYC gene signature in the neoadjuvant I-SPY TRIAL. We found that an increased MYC gene signature was associated with significantly shortened disease-free survival with a median follow-up of 3.9 yr (P = 0.005; Fig. 3 A). Surprisingly, we found that the response of primary tumors at the time of surgery, immediately after the completion of conventional neoadjuvant chemotherapy, did not differ significantly based on MYC signature scores (Fig. 3 B). To further examine the association between MYC signaling and response to neoadjuvant chemotherapy, we divided the total patient population into two categories. One group was composed of those patients who exhibited either complete response or minimal residual cancer burden (RCB 0/I; Symmans et al., 2007) after conventional chemotherapy. The other group was composed of those with substantial residual disease (RCB II/III). For patients who had a dramatic response to neoadjuvant chemotherapy (RCB 0/I), increased MYC signaling did not significantly alter prognosis (Fig. 3 C). In contrast, for those patients whose tumors had only minimal response to neoadjuvant chemotherapy (RCB II/III), an increased MYC signature was associated with early disease recurrence (Fig. 3 D; P = 0.001).

Although there was a strong association between elevated MYC expression and triple-negative breast cancer (Figs. 1 and 2), there also appeared to be considerable heterogeneity with regard to receptor status and MYC pathway activation (Fig. 2 A), leading us to explore whether the poor outcome associated with MYC pathway activation (univariate Cox proportional hazards ratio = 19.0 for MYC pathway activation as a continuous measure; 95% confidence interval [CI]: 4.4–82.3; P < 0.001; n = 146) reflects the aggressiveness of triple-negative disease (univariate Cox proportional hazards ratio = 2.2 for triple-negative tumors relative to receptor-positive tumors; 95% CI: 1.2–4.0; P = 0.015; n = 146) or whether MYC pathway activation might be an independent predictor of poor prognosis. In multivariate models that consider both receptor status and MYC pathway activation, the association between receptor status and outcome is considerably diminished (hazard ratio [HR] = 1.6; 95% CI: 0.8–3.0; P = 0.176; n = 146), whereas MYC pathway activation and outcome remain significantly associated (HR = 14.2; 95% CI: 3.1–64.4; P < 0.001; n = 146; Fig. 3 E). These results suggest that MYC pathway activation contains additional information about risk of recurrence that is not reflected in receptor status alone.

Given the poor outcome of patients that have tumors with elevated MYC activity, we sought to identify a therapeutic strategy that could target these tumors. Synthetic lethality between MYC overexpression and the inhibition of the mitotic kinase CDK1 has previously been observed in engineered cells and transgenic mouse models (Goga et al., 2007). However, this synthetic lethality has not been examined in breast cancer. Purvalanol A, an experimental small molecule CDK inhibitor that has higher specificity toward CDK1 (Gray et al., 1998), induced apoptosis in lymphoma cells and hepatocytes engineered to overexpress MYC (Goga et al., 2007). Importantly, CDK1 inhibition has been found to have little effect on the viability of control cells or normal mouse tissues (Goga et al., 2007).

We reasoned that triple-negative breast cancer cells that have arisen to overexpress MYC might also be sensitive to a synthetic-lethal interaction with CDK inhibition. To model our observations from the I-SPY clinical samples, we tested a panel of triple-negative cell lines with elevated MYC expression and a panel of receptor-positive lines that were expected to have lower MYC protein expression based on previously published mRNA data (Neve et al., 2006) for their sensitivity to CDK inhibitors (Fig. 4 A). In these experiments, we tested two CDK inhibitors: purvalanol A, and another CDK inhibitor dinaciclib (Parry et al., 2010) that is currently in phase II clinical trials against various tumor types (Dickson and Schwartz, 2009). Dinaciclib inhibits CDK1, 2, 5, and 9 at concentrations of 1–5 nM, which can be readily achieved in vivo, and exhibits improved pharmacokinetic (PK) and pharmacodynamic (PD) properties, compared with other CDK inhibitors previously evaluated in clinical trials.

Breast cell lines were treated with 10 µM purvalanol A or 10 nM dinaciclib, respectively, and were subjected to a cell viability assay. These chosen concentrations induce cell cycle arrest without causing cell death in normal human epithelial cells (Fig. 4 A; Goga et al., 2007). Human epithelial cells engineered to overexpress MYC (RPE-MYC) undergo cell death when treated with purvalanol A, whereas those with endogenous levels of MYC expression (RPE-NEO) do not, serving as positive and negative controls, respectively (Fig. 4 A; Goga et al., 2007). Dinaciclib exhibited ∼1,000-fold higher potency compared with purvalanol A in inducing apoptosis in epithelial cells engineered to overexpress MYC but not in control RPE-NEO cells (Fig. 4 A). In breast cell lines, CDK inhibitor treatment induced substantial cell death in each of the five triple-negative cell lines tested (range 25–65%), whereas most of the receptor-positive lines showed resistance to such treatment (Fig. 4 A). One receptor-positive line, SKBR3, was sensitive to these inhibitors (eliciting >25% cell death); however, this may be the result of increased MYC expression found in these cells (Fig. 4 A), which was not predicted from previously published mRNA profiling data (Neve et al., 2006). Cell cycle analysis by flow cytometry showed that treatment of both triple-negative and receptor-positive breast cancer cell lines with purvalanol A induced an increased accumulation of G2-M arrested cells, whereas treatment with dinaciclib resulted in cells arrested in both G1-S and G2-M (Fig. 4 B). This is consistent with results showing that purvalanol A is highly selective for CDK1 (Gray et al., 1998), whereas dinaciclib targets both CDK1 and CDK2 with similar potency (Parry et al., 2010).

Although these small molecule CDK inhibitors are selective for either CDK1 (purvalanol A) or for a few CDKs (dinaciclib), it is possible that other non-CDK kinases may also be inhibited and thus may contribute to the overall cell death phenotypes. To address this issue of specificity and further investigate synthetic lethality between MYC overexpression and CDK inhibition, we performed siRNA experiments to knock down specific CDKs in a panel of triple-negative as well as receptor-positive cell lines (Fig. 4 C). We found that the treatment of all three triple-negative cell lines with CDK1 siRNA for 72 h resulted in a significant amount of cell death, whereas the viability of receptor-positive lines was only modestly affected (Fig. 4 C). Interestingly, we found that CDK2 siRNA could also induce cell death, but to a lesser extent in two of the triple-negative lines and in one of the receptor-positive lines, which is in agreement with recent observations that CDK2 function may be essential for the viability of certain cancer types (Molenaar et al., 2009). The higher potency of cell killing induced by dinaciclib may be, in part, a result of its ability to inhibit both CDK1 and CDK2 (Parry et al., 2010). Unexpectedly, we found that CDK1 siRNA treatment dramatically increased CDK2 protein expression in all of the three triple-negative lines and one of the receptor-positive cell lines, whereas CDK2 siRNA did not alter CDK1 protein expression (Fig. 4 C). It remains to be determined whether this CDK2 up-regulation indicates a compensatory mechanism within the cell cycle encountering the loss of the only mitotic CDK.

One possible explanation for the sensitivity of breast tumor cells with high MYC to CDK inhibition is that they are more proliferative. One would expect these cells to therefore also be more sensitive to other chemotherapeutics that target cell proliferation mechanisms. To investigate this possibility, we tested the sensitivity of the breast lines used in this study to the chemotherapeutic agents paclitaxel and doxorubicin (Table 1) that are commonly used in clinical practice to treat breast cancer, including in the I-SPY TRIAL. The cell lines used in this study did not show any MYC expression-specific sensitivity to the chemotherapeutic agents. These cell-based studies are consistent with the observations from the I-SPY TRIAL that a MYC signature was not associated with improved response of primary tumors to conventional chemotherapeutic agents. It is therefore likely that other functions of MYC, in addition to its role in cell proliferation, contribute to the mechanism of synthetic lethality observed.

We next asked whether the observed CDK inhibition-induced cell death in cells with elevated MYC expression is dependent on their MYC status. To address this question, we first used an RNAi approach to knock down MYC protein expression. We found that pretreatment of RPE-MYC cells with MYC-specific siRNA significantly reduced the extent of CDK inhibition-dependent cell death (Fig. 5 A). Similarly, the sensitivities of three triple-negative cell lines treated with MYC siRNA before the addition of purvalanol A were greatly reduced (Fig. 5 B). We next used a lentiviral transduction method to overexpress MYC in receptor-positive cells to examine whether MYC overexpression alone is sufficient to render these otherwise resistant cells sensitive to CDK inhibition. We found that purvalanol A treatment of T47D and HCC1428 cells engineered to overexpress MYC resulted in significantly increased cell death (Fig. 5 C). Thus, these results suggest that elevated MYC expression is necessary and it increases the sensitivity of breast cancer cells to CDK inhibitors. Interestingly, whereas MYC RNAi demonstrated a dramatic decrease in MYC protein expression (Fig. 5 B), it did not appreciably alter cell viability (not depicted).

Although we saw sensitivity of triple-negative cells to CDK inhibition in monolayer cultures (Fig. 4), there is evidence that tumor cells can differ in their response to anti-cancer agents between in vitro and in vivo conditions (Weaver et al., 2002). As a step toward addressing this question, we initially examined the effect of CDK inhibition on cell viability in a 3D Matrigel culture system. 3D epithelial cultures have been used extensively as an in vitro system to more closely approximate the physiological microenvironment of the epithelial cells (Lee et al., 2007). Indeed, it has been shown that a 3D gene signature can more accurately predict outcome in several independent datasets (Fournier et al., 2006; Martin et al., 2008). Two receptor-positive cell lines (T47D and HCC1428) and two triple-negative cell lines (SUM149 and BT549) were seeded onto a solid Matrigel matrix, allowed to form colonies, and subsequently treated with 10 µM purvalanol or 10 nM dinaciclib and allowed to grow for 6 d. In this assay, if the cells undergo cell death, the mean colony size decreases between the initial day and the last day of the treatment incubation time (day 0 and day 6). However, if the cells are insensitive to CDK inhibition-induced cell death, the size of the colonies remains the same in the case of cell cycle arrest or may increase. We found that the mean colony size in the two triple-negative cell lines significantly decreased over a 6-d period after purvalanol A or dinaciclib treatment (Fig. 6, A and B), indicating that the cells in these cultures underwent extensive cell death. In contrast, the colonies formed by the two receptor-positive cell lines did not significantly change (Fig. 6, A and B), suggesting the induction of cell cycle arrest. As positive controls, DMSO-treated cultures for all the four cell lines showed robust growth at day 6 (Fig. 6, A and B). Thus, consistent with what we observed in 2D cultures, receptor-positive cells expressing low MYC appear to be resistant to cell death, whereas triple-negative cells with elevated MYC expression are particularly sensitive to CDK inhibition in 3D cultures.

A stringent test of any therapeutic strategy is the ability to inhibit or regress in vivo tumor growth. Using xenograft transplant models of breast cancer, we examined the in vivo efficacy of inhibiting CDKs in human triple-negative tumors with elevated MYC expression. We focused on dinaciclib because of its higher potency and improved PK properties compared with purvalanol A. We attempted to generate tumor xenografts of three triple-negative and three receptor-positive human breast cancer cell lines. Tumor cells were transplanted into BALB/c Nu/Nu mice and were allowed to form measurable tumors (200–250 mm³ in volume). Two triple-negative cell lines with elevated MYC expression, MDA-MB-231 and HCC3153, formed tumors with high penetrance and were used for subsequent studies. Our attempts to grow tumors of receptor-positive cells were not successful either because of the inability to form tumors or because only small sporadic tumors formed, despite our efforts to increase tumor engraftment by using Matrigel.

The triple-negative tumor-bearing mice were then treated twice weekly for 2 wk with dinaciclib (50 mg/kg/dose i.p.) or vehicle control. The mice treated with diluent experienced a ∼120–150% increase in growth of tumors during the experimental period (Fig. 7, A and B). In contrast, dinaciclib-treated mice had a dramatic response with ∼50% tumor regression (Fig. 7, A and B). In independent experiments, we also examined the effects of dinaciclib on the established MDA-MB-231 and HCC3153 xenografts 24 h after treatment. We found a dramatic decrease in the overall phosphorylation levels of presumptive CDK substrates that have the consensus CDK motif, as well as in the phosphorylation of a validated CDK1 substrate PP1-α (protein phosphatase 1–α; T320; Dohadwala et al., 1994; Blethrow et al., 2008) in primary tumors (Fig. 7 C). This was accompanied by the induction of apoptosis as demonstrated by PARP cleavage in tumor tissues (Fig. 7 C). Collectively, these results demonstrate that small molecule inhibition of CDKs represents a novel and feasible treatment strategy against human triple-negative breast cancers.

We next sought to understand the mechanism that underlies the synthetic-lethal interaction between elevated MYC expression and CDK inhibition in epithelial cells. We previously reported that CDK inhibition–induced apoptosis was independent of p53 but required activation of the mitochondrial intrinsic apoptotic pathway (Goga et al., 2007). We therefore reasoned that inhibition of CDKs might either increase the activity of proapoptotic BCL-2 family members or decrease the activity of prosurvival factors. To test this hypothesis, we examined the protein expression of components of the mitochondrial intrinsic pathway in matched RPE cells engineered to overexpress MYC (RPE-MYC) or control cells (RPE-NEO). RPE cells have modest levels of endogenous MYC expression (Fig. 4 A), which makes them a suitable model to study their response to MYC overexpression. We found that a proapoptotic BH3-only member BIM was substantially up-regulated in RPE-MYC cells (Fig. 8 A) as has been previously observed in other cell types engineered to overexpress MYC (Egle et al., 2004; Hemann et al., 2005). Surprisingly, BIM was dramatically further up-regulated after RPE-MYC cells were treated with purvalanol A, whereas BIM protein levels remained undetectable in RPE-NEO cells throughout the time course (Fig. 8 A). BIM up-regulation in RPE-MYC cells coincided with the cleavage of PARP, an event which indicated induction of apoptosis (Fig. 8 A). Protein quantification revealed that the extent of BIM protein up-regulation in this particular cell line was 2.2-fold (Fig. 8 B). This level of BIM up-regulation is likely to be significant because BIM up-regulation induces apoptosis unless its activity is concomitantly suppressed by overexpression of antiapoptotic BCL-2 family members (O’Connor et al., 1998). Quantitative PCR (TaqMan) analysis showed up-regulation of BIM mRNA (∼2.5-fold) after treatment of RPE-MYC cells with purvalanol A (Fig. 8 C). These results suggest that BIM up-regulation after CDK inhibition is, in part, a result of increased BIM mRNA levels. BIM mRNA also increased in RPE-NEO cells after purvalanol A treatment but remained substantially lower than the levels observed in RPE-MYC cells (Fig. 8 C). Interestingly, we found that dinaciclib treatment of RPE-MYC cells resulted in the up-regulation of not only the BIM-EL isoform found with purvalanol A treatment but also the shorter isoforms BIM-L and BIM-S (Fig. 8 D), which have been shown to be significantly more potent in inducing apoptosis (O’Connor et al., 1998). This may, at least in part, explain the higher potency associated with dinaciclib compared with that of purvalanol A (Fig. 4 A).

We postulated that the cell death observed in MYC-overexpressing cells after CDK inhibition is dependent on a threshold level of BIM, which is not reached in RPE-NEO cells. To test this hypothesis, we asked whether BIM is necessary for cell death induced by CDK inhibition in the context of MYC overexpression. Pretreatment of RPE-MYC cells with BIM-specific siRNAs protected the cells from purvalanol A–induced cell death (Fig. 8 E). These findings establish BIM as a major contributor to CDK inhibitor-induced apoptosis. We next tested triple-negative breast cell lines with endogenously elevated MYC expression and observed similar BIM up-regulation upon CDK inhibitor treatment (Figs. 8, F–G). To determine if BIM was required for the induction of cell death in these cells after CDK inhibitor treatment, we generated stable cell lines expressing a control or BIM shRNA (Fig. 8 H). We found that BIM depletion attenuates cell death after purvalanol A treatment (Fig. 8 I). These results indicate that the mechanism of CDK inhibitor–induced apoptosis in triple-negative cells with elevated MYC expression involves BIM up-regulation.

In addition to the proapoptotic BCL-2 family member BIM, we found that all of the three prosurvival BCL-2 family members, BCL-2, BCL-xL, and MCL-1, were also up-regulated in RPE-MYC cells compared with RPE-NEO cells in the absence of purvalanol A (Fig. 8 A). This is consistent with the hypothesis that cancers that have evolved to sustain high MYC expression have reached a balance between pro- and antiapoptotic factors to limit spontaneous apoptosis (Lowe et al., 2004; Hemann et al., 2005). This is also consistent with previous observation that co-overexpression of MYC and BCL-2 occurs in high-grade human breast tumors (Sierra et al., 1999). In our study, however, the expression levels of these prosurvival members did not change appreciably upon CDK inhibition (Fig. 8 A). Thus, treatment of tumors with elevated MYC expression, such as triple-negative breast cancers, with CDK inhibitors may tip the balance in favor of apoptosis by increasing BIM expression.

There is an evident and urgent need to develop targeted therapeutic strategies against triple-negative breast cancer. To identify these therapeutic targets, a better understanding of the biology of triple-negative breast cancer is therefore needed. In this study, we investigated the biology of triple-negative cancers and identified that MYC signaling is elevated in these tumors. Prior studies have examined MYC expression in breast cancers; however, we find, for the first time, that the expression of multiple MAX binding partners, which can regulate MYC activity, is altered in triple-negative tumors and may therefore contribute to MYC pathway activity.

Several studies have shown that the basal breast cancer subtype exhibits enrichment for a MYC transcriptional gene signature (Alles et al., 2009; Chandriani et al., 2009; Gatza et al., 2010). However, basal breast tumors account for only ∼70% of triple-negative tumors, and the importance of MYC signaling in the remaining 30% has previously been unknown. In clinical practice, tumors are routinely evaluated for estrogen receptor, progesterone receptor, and HER2 receptor status and, thus, the pathological determination of the triple-negative subtype is more clinically relevant for deciding on a course of treatment. We found that 33 of 36 triple-negative tumors (92%) in the I-SPY TRIAL had a high or intermediate MYC gene signature score (Fig. 2 A) that correlated with worse outcome (Fig. 3, A and D). The present study also demonstrates, for the first time, that MYC signaling is associated with diminished disease-free survival in patients whose tumors exhibited poor response to neoadjuvant chemotherapy (Fig. 3 D).

To determine if elevated MYC signaling can be exploited to treat triple-negative breast cancer, we assessed the utility of a synthetic-lethal approach between MYC up-regulation and CDK inhibition. Small molecule inhibition of CDK activity was effective at inducing significant cell death in triple-negative cell lines with elevated MYC expression as well as in mouse xenograft models. We found that the mechanism of such cell death included the up-regulation of the proapoptotic BCL-2 family member BIM. Thus, this study represents a significant step forward in identifying apoptotic mechanisms for treating triple-negative breast cancers.

Although prior studies have focused on the many biological functions of MYC, how this information could be translated to develop novel ways to treat breast cancer remains unclear. In an emerging era of personalized medicine, it is crucial to identify what specific patient populations would most benefit from a given treatment strategy. We found that, for patients who experienced a limited tumor response to conventional neoadjuvant chemotherapy, disease-free survival was strongly influenced by MYC expression. Insight into components of MYC signaling, therefore, will provide potential targets that could be exploited as new therapies for this subset of breast cancer patients. However, despite up-regulation of MYC signaling in a variety of human cancers, the potential utility of direct MYC inhibition remains unclear. For instance, MYC knockdown via RNAi has not been found to diminish the viability of cultured tumor-derived cells (Guan et al., 2007), consistent with our observations in this study with breast cancer cells. In contrast, inhibition of MYC-MAX dimerization using an experimental compound, 10058-F4, led to the induction of apoptosis in cultured human leukemia cells (Huang et al., 2006). Furthermore, in vivo inhibition of MYC transcriptional activity using a conditional dominant-negative mutant induced cell death and caused tumor regression in a KRAS-initiated mouse lung tumor model (Soucek et al., 2008). Whether the differences among these observations are a result of the methods used (i.e., knocking down MYC protein expression versus inhibition of its transcriptional activity) or to the extent that each cancer type depends on deregulated MYC activity requires further investigation.

An alternative approach to directly inhibiting MYC is to use a MYC-dependent synthetic-lethal strategy. We previously identified a form of synthetic-lethal interaction between MYC overexpression and CDK1 inhibition using engineered cell lines as well as in vivo model systems (Goga et al., 2007). More recent studies have shown that RNAi-mediated or small molecule inhibition of two additional cell cycle kinases, CDK2 and aurora kinase B, respectively, has synthetic-lethal interactions with MYC overexpression in certain cancer cell types (Molenaar et al., 2009; Yang et al., 2010). Another study has also reported that agonist-mediated activation of a TRAIL (tumor necrosis factor–related apoptosis-inducing ligand) receptor (DR5) induces MYC-dependent synthetic lethality (Wang et al., 2004). These distinct forms of synthetic lethality appear to require fundamentally different cellular mechanisms of cell death, including different requirements for an intact p53 tumor suppressor pathway (Wang et al., 2004; Goga et al., 2007; Molenaar et al., 2009; Yang et al., 2010). For example, 44–82% of primary basal breast tumors either lack p53 or harbor p53 mutant alleles (Sørlie et al., 2001; Carey et al., 2006). Therefore, CDK1-MYC synthetic lethality, which is p53 independent (Goga et al., 2007), may be particularly useful in treating these tumors.

The idea of targeting cell cycle kinases to selectively kill tumor cells, which often exhibit higher proliferation rates than nontumorigenic cells, is appealing and has indeed led to the clinical development of several small-molecule CDK inhibitors (Shapiro, 2006; Malumbres et al., 2008). However, there have been several issues associated with their clinical development. Previous generations of CDK inhibitors generally suffered from low in vivo potency and poor PK/PD properties. More recent third-generation CDK inhibitors, including dinaciclib, exhibit significantly improved PK/PD properties and offer greater promise for in vivo use. Indeed, dinaciclib was well tolerated in phase I trials and is currently being evaluated in phase II trials against various tumor types (Dickson and Schwartz, 2009; Parry et al., 2010). Prior clinical studies have also suffered from a lack of understanding of which tumor types are the most likely to be responsive to CDK inhibitors (Malumbres et al., 2008). Therefore, selection of patient cohorts based on molecular targets, such as MYC overexpression, may improve the therapeutic potential of CDK inhibitors in clinical trials.

In the present study, we found that CDK inhibition increases BIM protein levels not only in the model cells engineered to overexpress MYC but also in a panel of patient-derived triple-negative breast cancer cell lines with elevated MYC expression. Elevated BIM plays a direct role in CDK inhibition–induced cell death. Prior studies have found that BIM isoforms can be regulated at both the transcriptional and posttranslational levels. Protein expression of the shortest isoform Bim-S is sufficient to potently induce apoptosis (O’Connor et al., 1998). In contrast, the activity of the longest isoform BIM-EL can also be modulated by distinct phosphorylation mechanisms that can either stabilize the protein or induce its degradation via the proteasome pathway (Hübner et al., 2008). In our studies, we found that both isoforms can be up-regulated in epithelial cells after CDK inhibitor treatment.

Previous studies have shown that BIM expression is elevated upon MYC overexpression (Egle et al., 2004; Hemann et al., 2005). Having uncovered that CDK inhibition triggers induction of BIM expression, we became interested in studying the relationship between the protein expression levels of MYC and BIM in a panel of untreated breast cancer cell lines. We did not find a correlation between MYC and BIM expression in these cells regardless of their molecular subtypes or receptor status (unpublished data). Therefore, in a given cellular context, it is likely that the relative increase in BIM activity in response to CDK inhibition, not the absolute basal level of BIM expression, determines whether or not apoptosis can be initiated. Thus, the protein stoichiometry among BIM and antiapoptotic BCL-2 family members such as BCL-2, BCL-xL, and MCL-1 is likely to dictate if apoptosis is triggered. In this respect, a combinatorial approach of CDK inhibition with inhibition of antiapoptotic BCL-2 family members would be predicted to have a synergistic effect in inducing cell death in MYC overexpressing triple-negative tumors. Indeed, several BH3 mimetics (Chonghaile and Letai, 2008), namely ABT-737/263 (Oltersdorf et al., 2005) and obatoclax (Nguyen et al., 2007), are currently under development for clinical use.

In conclusion, we have shown utility for small molecule CDK inhibitors in the treatment of triple-negative breast tumors with elevated MYC expression. It is likely that CDK inhibitors will be effective not only against MYC-overexpressing triple-negative tumors but also for aggressive receptor-positive tumors with elevated MYC expression (i.e., luminal B). Methods for immunohistochemical analysis on paraffin-embedded primary tumor biopsies for MYC protein expression have been previously described (Gurel et al., 2008; Ruzinova et al., 2010). In addition, gene expression profiling methods on primary breast tumors are becoming available for use in the clinic (van ’t Veer et al., 2002). Such technologies could be applied to assess whether MYC signaling is elevated in patient tumor samples (Alles et al., 2009; Chandriani et al., 2009; Gatza et al., 2010). Thus, these detection methods for MYC activity have the potential to be translated for routine use in the clinic. Considering the lack of established targeted therapeutics against triple-negative tumors, we propose that MYC status will prove useful as a predictive biomarker of response to CDK inhibitors for the treatment of triple-negative breast cancers.

Expression microarray results were available for 149 tumors before treatment. Three tumors could not be designated as triple negative or receptor positive, owing to a lack of staining for one or more receptor types, leaving 146 available for analysis. Details of the I-SPY Trial protocol, patient array data, and characteristics of enrolled patients are described elsewhere (Esserman et al., 2012) and are also available at the NCI I-SPY Data Portal (http://ispy.nci.nih.gov/). All patients included in our analysis had pretreatment core biopsies for which gene expression measurements had been determined using 44K microarrays (Agilent Technologies). Microarray hybridization was performed as per the manufacturer, the background was subtracted, and Lowess-normalized log2 ratio of Cy3 and Cy5 intensity values were calculated. For MYC mRNA levels, a MYC-specific probe (A_32_P60687) was compared across the 146 samples for which receptor status was known. Patients enrolled in the trial received conventional neoadjuvant chemotherapy that included doxorubicin and cyclophosphamide, and/or paclitaxel per the I-SPY protocol before definitive surgical resection. The extent of residual disease was quantified using RCB and reported by RCB class (RCB 0-III; Symmans et al., 2007).

Expression of the MYC signature (Chandriani et al., 2009) in the I-SPY dataset was examined by first converting the platform-specific probe identifiers in both studies to UNIGENE cluster identifiers (Unigene Build 219 Homo sapiens). Data from multiple probes mapping to the same UNIGENE cluster IDs were averaged. The MYC pathway activity refers to the MYC gene signature score centroid that was calculated as previously described (Chandriani et al., 2009). The Pearson correlation of each tumor’s expression profile of these genes to the MYC signature centroid was determined. The tumors were then ordered, from high to low, by the Pearson correlation value. Tumors that lacked conclusive receptor status information were excluded from analysis. The remaining tumors were divided into tertiles representing three groups with high, intermediate, or low correlation to the MYC signature centroid. Recurrence-free survival was analyzed using Cox Proportional Hazards models that were evaluated using Wald’s Test. This analysis was performed using the survival package in the R Environment for Statistical Computing (http://www.r-project.org/). Associations between categorical variables were evaluated using Fisher’s Exact Test. This analysis was performed using the stats package in the R Environment for Statistical Computing (http://www.r-project.org/).

Protein was extracted from the human tumors, and reverse-phase protein lysate microarray was done as described previously (Tibes et al., 2006; Hennessy et al., 2007; Hu et al., 2007; Liang et al., 2007). Briefly, lysis buffer was used to lyse frozen tumors by homogenization. Tumor lysates were normalized to 1 µg/µl concentration with the use of bicinchoninic acid assay and were boiled with 1% SDS, and the supernatants were manually diluted in six or eight twofold serial dilutions with lysis buffer. An arrayer (2470; Aushon Biosystems) created 1,056 sample arrays on nitrocellulose-coated FAST slides (Schleicher & Schuell BioScience) from the serial dilutions. A slide was then probed with a validated primary MYC and phospho-MYC T58/S62 antibody (Cell Signaling Technology), and the signals were amplified with a catalyzed system (Dako). A secondary antibody was used as a starting point for amplification. The slides were scanned, analyzed, and quantitated with the use of MicroVigene software (VigeneTech Inc.) to generate serial dilution signal intensity curves for each sample with the logistic fit model: ln(y) = a + (b − a)/(1 + exp^{c*[d−ln(x)]}). A representative natural logarithmic value of each sample curve on the slide (curve mean) was then used as a relative quantification of the amount of each protein in each sample. The levels of unphosphorylated and phosphorylated-MYC (T58/S62) in each sample was expressed as a log-mean centered value after correction for protein loading with the use of the mean expression levels of >150 proteins as previously described (Tibes et al., 2006; Hennessy et al., 2007; Hu et al., 2007; Liang et al., 2007; Stemke-Hale et al., 2008; Gonzalez-Angulo et al., 2009).

The propagation of human breast cell lines used in this study and their global mRNA expression profiling has been previously described (Neve et al., 2006). Engineered human epithelial cell lines RPE-NEO and RPE-MYC cells were previously described (Goga et al., 2007). The MYC-overexpressing versions of the receptor-positive cell lines T47D and HCC1428 were established by infecting the cells with the lentivirus prepared using pLVX-AcGFP-N1 plasmid with a full-length human MYC cDNA cloned into the EcoRI–XhoI sites or with recombinant retrovirus expressing MYC to generate the MCF10A-MYC cells..

CellTiter (Promega) cell viability assay shown in Fig. 4 A was performed in 96-well plates, using a Safire2 plate reader (TECAN) that runs Magellan software, according to the manufacturer’s instruction. Each cell line was plated onto 10 wells per experiment and the assay was repeated at least five times. For the rest of the cell viability experiments described throughout this manuscript, a flow cytometry–based Guava ViaCount viability assay (Millipore) was performed according to the manufacturer’s instruction.

Cell lines were treated with DMSO, 10 µM purvalanol A, or 10 nM dinaciclib (provided by the Drug Synthesis and Chemistry Branch, Developmental Therapeutics Program, Division of Cancer Treatment and Diagnosis, National Cancer Institute [Bethesda, MD] and MERCK) for 72 h. After treatment, cells were fixed in 70% ethanol and stained with propidium iodide to measure DNA content. Samples were analyzed on a LSRII flow cytometer. Cell populations were gated to exclude cell debris and doublets and cell cycle distribution was determined using FlowJo analysis software (Tree Star).

Cultured cells were washed with ice-cold PBS and harvested directly into radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris, 150 mM NaCl, 0.5% sodium-deoxycholate, 1% NP-40, 0.1% SDS, 2 mM EDTA, pH 7.5) containing COMPLETE protease inhibitor cocktail (Roche) and phosphatase inhibitors (Santa Cruz Biotechnology, Inc.). Isolated tumor tissues were first washed in ice-cold PBS and homogenized on ice using Tissue Tearor (Biospec Products, Inc.) in RIPA buffer containing protease inhibitors and phosphatase inhibitors. Protein concentrations were determined by performing DC Protein Assay (Bio-Rad Laboratories) using BSA as standard. The following antibodies were used for western analyses: MYC (Epitomics), β-actin (Sigma-Aldrich), PARP (Cell Signaling Technology), BIM (Assay Designs), Puma (Cell Signaling Technology), Bid (R&D Systems), Bax (Cell Signaling Technology), Bak (Santa Cruz Biotechnology, Inc.), BCL-2 (Cell Signaling Technology), BCL-xl (Santa Cruz Biotechnology, Inc.), MCL-1 (Abcam), Phospho-(Ser) CDK substrates (Cell Signaling Technology), PP1-α (Epitomics), PP1-α (pT320; Epitomics), CDK1 (Santa Cruz Biotechnology, Inc.), and CDK2 (Santa Cruz Biotechnology, Inc.).

VersaDoc Imaging System (4000 MP; Bio-Rad Laboratories) was used to quantify protein expression. BIM up-regulation at the protein level was determined by normalizing presaturated BIM signals to those of β-actin. To determine the relative MYC protein expression levels across a panel of breast cell lines, the MYC signals acquired through anti-MYC Western blotting were normalized against total fluorescence from the Ponceau S–stained bands as previously described (Nijjar et al., 2005). This method was necessary as a result of significantly different protein expression levels of major house keeping proteins across breast cell lines used in this study.

Purvalanol A (Sigma-Aldrich) was reconstituted in 100% DMSO. Unless otherwise indicated, it was used at a concentration (10 µM) previously shown to induce cell cycle arrest in various mammalian cell lines (Gray et al., 1998; Goga et al., 2007). Dinaciclib was reconstituted either in 100% DMSO for cell culture use or in 20% HPBCD (hydroxypropyl β cyclodextrin) for mouse studies.

The establishment of 3D cultures was performed as previously described (Lee et al., 2007). Briefly, cells were seeded on top of the solidified Matrigel (BD) layer at 2,000 cells per well in 96-well plates and incubated for 6 d to allow for multicellular colonies to form. Phase contrast images were obtained at this point and designated as day 0. Cultures were treated with DMSO, 10 µM purvalanol A, or 10 nM dinaciclib and incubated for an additional 6 d before obtaining images of treated and control cultures and designated as day 6. Images were analyzed using ImageJ image analysis software (National Institutes of Health). The background was subtracted to remove differences in intensity as a result of shadow effects within the culture images, and the mean area of the colonies within the field was determined using the particle analysis algorithm within ImageJ. To decrease the chance of debris being included in the analysis, particles with an area of 5,000 pixels (3.1 µm) or less (the mean size of noncellular debris within the images as determined empirically) were excluded. The mean surface area of the day 0 cultures for each cell line was set to 100%, and the mean surface area of day 6–treated cultures were determined relative to day 0. The Student’s t test was used to determine statistical significance.

For all the breast cancer cell lines used in this study, 10⁷ cells in 200 µl PBS were subcutaneously injected into immunodeficient female mice (BALB/c nude/nude) aged ∼6–8 wk. The tumors were allowed to grow for 3–4 wk (with the tumors reaching ∼200–250 mm³ in volume) before the animals were treated with either dinaciclib at 50 mg/kg or vehicle alone (20% HPBCD) via i.p. injection. All animal experiments were approved by the University of California San Francisco institutional animal care and use committee.

Unless otherwise noted, siRNA against human CDK1, CDK2, MYC, and BIM (BCL2L11), respectively, and a pool of nontargeting control siRNA (siGENOME SMART pool siRNA; Dharmacon) were used according to the manufacturer’s protocol. For the MYC knockdown experiments described in Fig. 5 A, the liposomal siRNA preparations against human MYC and luciferase (negative control), respectively, were provided by Alnylam Pharmaceuticals, Inc. The retroviral shRNA constructs used in this study are pMKO shRNA Bim (plasmid 17235; Addgene; Schmelzle et al., 2007) and pMKO shRNA GFP (plasmid 10675; Addgene; Masutomi et al., 2005).

RPE cells, with or without constitutive MYC overexpression, were treated with purvalanol A for 0, 24, 48, and 72 h. Total RNA was isolated from cells using mirVana (Ambion) and digested with DNaseI to remove contaminating DNA (Ambion). cDNA was prepared from 1 µg of total RNA using Superscript II reverse transcription kit (Invitrogen). Real-time PCR was performed using probes specific for human BIM and β-ACTIN (ABI), according to the manufacturer’s instructions. Samples were run in triplicate on a Real-Time Thermal Cycler (Bio-Rad Laboratories). Variation of BIM expression was calculated using the ΔΔCT method (Livak and Schmittgen, 2001) with β-ACTIN mRNA as an internal control.

A list of 352 genes that comprise the MYC gene signature used for the analysis in this study is provided in Table S1.

We thank the patients, investigators, and institutions that participated in the I-SPY TRIAL. We are grateful to Drs. Luika Timmerman, Koei Chin, Wen-Lin Kuo, Adi Gazdar, Stephen Ethier, and Joe Gray for providing cell lines. We thank Drs. Clifford Hudis and J. Michael Bishop, and Ms. Brittany Anderton for comments and critical reading of the manuscript.

We acknowledge the following support: California Breast Cancer Research Program post-doctoral (D. Horiuchi and L. Kusdra) and pre-doctoral (N.E. Huskey) fellowships, NCI 2T32CA108462 post-doctoral training grant (L. Kusdra), the Howard Hughes Medical Institute (S. Chandriani), the UC Cancer Coordinating Committee (N.E. Huskey), AHA Scientist Development Grant SDG3420042 (J.W. Smyth), NCI 1K23CA121994 (A.M. Gonzalez-Angulo), the Susan G. Komen Foundation (A.M.Gonzalez-Angulo, G.B. Mills, and A. Goga), NCI 1K08CA104032, 1R01CA136717 (A. Goga), a UCSF SPORE 5P50CA058207 Developmental Project (A. Goga), and V-Foundation Scholar Award (A. Goga). We also acknowledge the I-SPY program for additional statistical support: NCI SPORE, CA58207; ACRIN, U01 CA079778 and CA080098; and CALGB, CA31964 and CA3360.

The authors do not declare any competing financial interests.