Abstract
Poor prognosis in triple-negative breast cancer (TNBC) is due to an aggressive phenotype and lack of biomarker-driven targeted therapies. Overexpression of cyclin E and phosphorylated-CDK2 are correlated with poor survival in patients with TNBC, and the absence of CDK2 desensitizes cells to inhibition of Wee1 kinase, a key cell-cycle regulator. We hypothesize that cyclin E expression can predict response to therapies, which include the Wee1 kinase inhibitor, AZD1775.
Mono- and combination therapies with AZD1775 were evaluated in TNBC cell lines and multiple patient-derived xenograft (PDX) models with different cyclin E expression profiles. The mechanism(s) of cyclin E–mediated replicative stress were investigated following cyclin E induction or CRISPR/Cas9 knockout by a number of assays in multiple cell lines.
Cyclin E overexpression (i) is enriched in TNBCs with high recurrence rates, (ii) sensitizes TNBC cell lines and PDX models to AZD1775, (iii) leads to CDK2-dependent activation of DNA replication stress pathways, and (iv) increases Wee1 kinase activity. Moreover, treatment of cells with either CDK2 inhibitors or carboplatin leads to transient transcriptional induction of cyclin E (in cyclin E–low tumors) and result in DNA replicative stress. Such drug-mediated cyclin E induction in TNBC cells and PDX models sensitizes them to AZD1775 in a sequential treatment combination strategy.
Conclusions:Cyclin E is a potential biomarker of response (i) for AZD1775 as monotherapy in cyclin E–high TNBC tumors and (ii) for sequential combination therapy with CDK2 inhibitor or carboplatin followed by AZD1775 in cyclin E–low TNBC tumors.
TNBC is a subtype of invasive breast cancer with an aggressive phenotype that has decreased survival compared with other types of breast cancers due, in part, to the lack of biomarker-driven targeted therapies. Here we show that TNBCs can be separated into cyclin E–high or -low tumors and those with high cyclin E have a significantly worse prognosis. We show that cyclin E–high tumors are very sensitive to Wee1 kinase inhibition by AZD1775 as monotherapy. To this end, we report on the relationship between cyclin E levels and the sensitivity to Wee1 kinase inhibition providing the mechanistic evidence in support of a new clinical trial (NCT03253679). We also identified cyclin E as a potential predictor of response for the sequential combination therapy with a CDK2 inhibitor or carboplatin followed by AZD1775 in cyclin E–low tumors, providing the scientific rationale for future biomarker-driven clinical trials in TNBC.
Introduction
Triple-negative breast cancer (TNBC) is a subtype of invasive breast cancer that lacks estrogen receptor, progesterone receptor, and HER2 expression (1). TNBC accounts for up to 17% of breast cancers and represents an aggressive phenotype with higher histologic grade, larger tumor size, and more advanced stage at diagnosis, and more likely to affect younger women and African-American women. Patients with TNBC have decreased overall survival compared with patients with other breast cancer subtypes. This decreased survival is due in part to the lack of targeted therapies for TNBCs. While about 40%–50% of patients with TNBC will achieve complete pathologic response (pCR) to standard therapy, there are no biomarkers to select those from the patients who need other targeted approaches. As a result, patients with TNBC have a higher likelihood of developing recurrence within the first 5 years of follow-up mainly due to lack of specific targets (2, 3).
Over 60% of TNBCs harbor TP53 mutations (4–6) and the cell cycle of these tumors is deregulated at the G1–S and G2–M checkpoints. This deregulation often leads to defects in DNA repair pathways, whose proper function is required to respond to DNA replicative stress (7, 8). Exploiting the machinery that regulates cell-cycle checkpoints and DNA repair for cancer therapy has been the focus of numerous investigations; however, translation to patient care has not been very successful. A key challenge has been the lack of consistent and validated predictive biomarkers that can readily identify tumors likely to respond to inhibitors of the DNA repair pathways, specifically those that target DNA replicative stress. One such inhibitor, AZD1775, targets Wee1 kinase, a key regulator of the G2–M checkpoint that negatively regulates entry into mitosis through inactivation of CDK1, arresting cells in G2–M and allowing for DNA repair (9). Wee1 kinase inhibition abrogates the G2–M arrest leading to premature mitosis and also deregulates DNA replication at S-phase leading to DNA damage. (10–13). Cells with DNA replicative stress are therefore vulnerable to Wee1 kinase inhibition (14). Although AZD1775 has been proposed for treatment of TP53-defective cancers (15, 16), studies have shown that the therapeutic efficacy of Wee1 inhibition is independent of TP53 status (11, 17, 18), suggesting alternative pathways that can deregulate G1–S transition (19). In a recent phase I study with AZD1775, none of the patients with a documented TP53 mutation exhibited a response to this agent (20). Given the complex regulation of cell-cycle checkpoints, these seemingly contradictory results are not unexpected, indicating alternative pathways that can deregulate the cell cycle and biomarkers are needed to help predict response to Wee1 kinase inhibition.
Cyclin E, the key regulator of the G1–S transition, activates CDK2 to progress through the S-phase and promote DNA replication (21, 22). Cyclin E levels are tightly cell-cycle regulated, with peak CDK2-associated kinase activity occurring at the G1–S boundary in normally proliferating cells and tissues (23). Cyclin E overexpression has been observed in a broad spectrum of human malignancies, including breast cancer, and is associated with poor outcomes (24–31). Overexpression of cyclin E deregulates the cell cycle at both G1–S and G2–M transitions by accelerating S-phase entry (32), affecting centrosome amplification (33) and stimulating premature mitosis (34, 35). An overactive cyclin E/CDK2 complex also increases DNA origin firing and induces re-replication, causing DNA replication stress in cancer cells (36, 37), potentially sensitizing these cells to Wee1 kinase inhibition. Consistently, CDK2 deletion desensitizes cancer cells to inhibition of Wee1 kinase (38). In this study, we hypothesized that overexpression of cyclin E induces DNA replicative stress and stimulates DNA repair responses, thereby sensitizing cells to AZD1775. We also interrogated whether cyclin E expression levels can predict the response to AZD1775 as a single agent or in combination with agents that can transiently increase cyclin E levels in preclinical studies.
Materials and Methods
Cell lines
76NE6 cyclin E–inducible cell lines were generated from the immortalized human mammary epithelial cell line 76NE6 and cultured in Tet-free medium as described previously (39). HEK-293T cells for lentiviral packaging and the TNBC cell lines HCC1806, MDA436, MDA468, and MDA231 were obtained from the ATCC and cultured as described previously (30, 40). The TNBC cell lines SUM149 and SUM159 were obtained from Asterand Bioscience and cultured in Ham F12 medium with 5% FBS, 5 μg/mL insulin, and 1 μg/mL hydrocortisone. All cells were free of Mycoplasma contamination. Cell lines were regularly (every 6 months) authenticated by karyotype and short tandem repeat analysis at the MD Anderson Cancer Center Characterized Cell Line Core facility.
Mouse and in vivo studies
All animal studies were approved by the MD Anderson Institutional Animal Care and Use Committee and strictly followed the recommendations in the Guide for the Care and Use of Laboratory Animals from the NIH (Bethesda, MD). The generation and preparation of patient-derived xenograft (PDX) models were described in the Supplementary Methods as reported previously (41). A total of 4 × 106 SUM149 or MDA231 cells were injected into the mammary fat pad to generate xenograft models. The mice were given 50 mg/kg AZD1775 (prepared in 0.5 w/v% methyl cellulose 400 Solution) orally or 25 mg/kg dinaciclib (prepared in 20% (2-hydroxypropyl)-β-cyclodextrin) or 30 mg/kg carboplatin (prepared in sterile water) by intraperitoneal injection. The length and width of tumor xenografts were measured by caliper twice per week and the volume of tumor was calculated by the formula volume = length × (width)2/2. The specific treatment conditions for each experiment are provided in Supplementary Methods.
High-throughput survival assay
Cells were treated and their survival examined in 96-well plates over an 11-day period, a method that allows analysis of cytotoxicity of one or more drugs in a wide range of adherent cell lines and provides results that are highly consistent with classic clonogenic assays as described previously (40, 42). At the end of the 11-day assay, the plates were subjected to an MTT (3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide; RPI Corp.) assay as described previously (40, 42). MTT was solubilized, and the absorbance of each well was read at 590 nm using an Epoch microplate spectrophotometer (BioTek). The combination index (CI) for each combination treatment was calculated using the CalcuSyn program (Biosoft). CI < 0.9 indicates the synergy of a combination treatment; 0.9 < CI < 1.1 indicates additivity; and CI > 1.1 indicates antagonism (40, 42). The agents subjected to this assay are AZD1775 (obtained from the Institute of Applied Cancer Science, MD Anderson Cancer Center, Houston, TX), dinaciclib (Merck & Co., Inc.), meriolin5 (ManRos Therapeutics), SNS032 (Selleck Chemicals), roscovitine (ManRos Therapeutics), palbociclib (Pfizer), MLN8237 (Selleck Chemicals), carboplatin (Sigma-Aldrich), cisplatin (Sigma-Aldrich), paclitaxel (Sigma-Aldrich), and epirubicin and doxorubicin (obtained from the pharmacy at The University of Texas MD Anderson Cancer Center, Houston, TX). Additional details regarding the individual treatment conditions and doses of each drug used per cell line are provided in Supplementary Methods.
CRISPR/Cas9 KO
The sgRNA targeting human cyclin E and CDK2 were cloned into pX330GFP, which was transfected into MDA231 cells or MDA231 cyclin E–inducible cells using PEI (polyethylenimine). Green fluorescent protein–positive cells were sorted into 96-well plates at one cell per well. Successful cyclin E knockout (KO) or CDK2 KO in clones was confirming using Western blot analysis.
The cyclin E sgRNA sequence was 5′GGACGGCGGCGCGGAGTTCT3′. The CDK2 sgRNA sequences were 5′CACCGCATGGGTGTAAGTACGAACA3′ and 5′CACCGGATTTCGGCCAAGGCAGCCC3′.
Statistical analysis
For patient cohort studies, the relationship between cyclin E and p-CDK2 scores on IHC was evaluated by using Stata SE statistical software version 12.0 (StataCorp LP). All tests were two-tailed, and statistical significance was set at P < 0.05. Endpoints were recurrence-free survival and overall survival and were calculated from the time of surgery to breast cancer recurrence. Breast cancer cases from the TCGA and METABRIC database were analyzed using the cBioportal website (www.cbioportal.org).
Each cell culture experiment had at least three technical and three biological repeats. Continuous outcomes were summarized with means and SE. Comparisons between two groups were analyzed by two-sided t test using MS Excel or Prism 6. Comparisons among more than two groups were analyzed by one-way ANOVA using Prism 6. Tumor growth in the mouse models was compared by two-sided t test with MS Excel. Kaplan–Meier survival curves in the mouse studies were generated by Prism 6 software (GraphPad), and the significance was determined by a log-rank (Mantel–Cox) test. All graphs were generated either in MS Excel or Prism 6.
Results
High cyclin E levels predict poor prognosis in TNBC
To determine whether overexpression of cyclin E is enriched in TNBC and its relationship with prognosis, we compared the percentages of patient samples with cyclin E alteration (CCNE1 gene amplification and/or RNA upregulation) among all breast cancer subtypes in The Cancer Genome Atlas (TCGA) and METABRIC databases (Fig. 1A; Supplementary Fig. S1A and S1B; refs. 43, 44). We found a higher prevalence of cyclin E alteration in patients with TNBC (52% in TCGA and 40% in METABRIC) than in patients with the estrogen receptor–positive (ER+) subtype (3% in TCGA and 2.3% in METABRIC). Cyclin E DNA/RNA alterations were also correlated with worse outcomes in the TCGA TNBC subset (Fig. 1B). Next, we examined the expression of cyclin E protein levels using IHC staining in 248 TNBC samples in tissue microarrays from the National Cancer Institute and the United Kingdom (Fig. 1C; Supplementary Fig. S1C; Supplementary Table S1; ref. 30). Of these patients, 77% had high expression of cyclin E and significantly shorter time to recurrence than those with low cyclin E expression (Fig. 1D). Together, these findings suggest that TNBCs with elevated cyclin E levels have a higher recurrence rate than those with low cyclin E levels.
Cyclin E overexpression sensitizes TNBC to Wee1 kinase inhibition in vitro and in vivo
Overexpression of cyclin E has been shown to promote genomic instability by causing DNA replication stress and deregulation of the G1–S transition in various in vitro and in vivo model systems (33, 34, 45–47). Wee1 kinase, which prevents premature mitotic entry by inhibiting CDK1 and CDK2, may be essential for preventing massive DNA damage and cell death when cyclin E is overexpressed (48–50). To test whether overexpression of cyclin E sensitizes TNBC cells to Wee1 kinase inhibition, we evaluated the sensitivity of six TNBC cell lines to the clinically available Wee1 inhibitor AZD1775 (51). Expression of cyclin E was highly predictive (r = 0.90) of response to AZD1775 in all six TNBC cell lines (Fig. 1E; Supplementary Fig. S2A). Among other key cell-cycle modulators examined, only p-CDK1 (Y15), which is a known downstream effector of Wee1 kinase, and p-CDK2 (T160) were moderately predictive of AZD1775 sensitivity in only two (HCC1806 and MDA231) of six cell lines (Supplementary Fig. S2B). Knockdown of cyclin E in HCC1806 and MDA231 desensitized them to AZD1775 (Fig. 1F; Supplementary Fig. S2C and S2D; Supplementary Table S2), while induction of cyclin E in 76NE6 and MDA231 cells significantly increased sensitivity to AZD1775 (Fig. 1G; Supplementary Fig. S2E and S2F; Supplementary Table S2). Induction of tumor-specific cytoplasmic low molecular weight cyclin E (39, 52) in 76NE6 cells also increased sensitivity to AZD1775 (Supplementary Fig. S2G).
To examine the duration of cyclin E overexpression required to sensitize cells to Wee1 kinase inhibition, we evaluated the response of MDA231 and 76NE6 cyclin E–inducible cells to AZD1775 by inducing the expression of cyclin E from 24 hours to 10 days (Supplementary Fig. S3A). Induction of cyclin E for 48 hours was sufficient to maximally sensitize cells to Wee1 kinase inhibition (Supplementary Fig. S3B–S3D; Supplementary Table S2). These results suggest that as long as cyclin E can be transiently induced during treatment with AZD1775, the cells can be sensitized to Wee1 kinase inhibition.
Next, we examined the sensitivity to AZD1775 in vivo in one cyclin E–low and two cyclin E–high TNBC PDX models, which were selected from 25 TNBC PDXs on the basis of the cyclin E levels (ref. 41; Supplementary Fig. S4). PDX1 has a normal CCNE1 copy number and low cyclin E expression, while PDX2 and PDX3 exhibit CCNE1 amplification and cyclin E overexpression (Fig. 1H; Supplementary Table S3). The histology of these PDX models was similar to the original patient tumors (Supplementary Fig. S5). CDK2 and p-CDK2 are cooverexpressed with cyclin E in PDX2 and PDX3 (Fig. 1H). Mice representing each PDX model were treated with either vehicle or 50 mg/kg AZD1775 for four cycles and euthanized at the end of treatment or when the maximum allowable tumor burden was reached (schema, Supplementary Fig. S6A). Consistent with our in vitro results, AZD1775 treatment significantly decreased tumor burden and improved survival of the cyclin E–overexpressing PDX2 and PDX3 models, but not the PDX1 model (Fig. 1I–K; Supplementary Fig. S6B). Collectively, these results suggest that cyclin E overexpression in cell lines and PDX models predicts response to Wee1 kinase inhibition.
Cyclin E–overexpressing cells rely on Wee1 kinase to suppress DNA replicative stress and stimulate DNA damage repair
Because both cyclin E and Wee1 kinase are involved in maintaining normal DNA replication fork progression (12, 36, 47, 53), we next examined whether the progression of DNA replication in TNBC cells with high cyclin E is altered upon treatment with AZD1775 using DNA fiber assays. MDA231 cyclin E–inducible cells were pulse labeled with iododeoxyuridine for 60 minutes followed by chlorodeoxyuridine labeling for 30 minutes, and the speed of DNA replication was quantitated. Compared with control cells (0.59 kb/minute), DNA replicative fork progression was slower in cells with either cyclin E overexpression (0.43 kb/minute) or Wee1 kinase inhibition (0.29 kb/minute; Fig. 2A and B; Supplementary Fig. S7A and S7B). Moreover, the replication forks in cyclin E–induced cells treated with AZD1775 not only progressed at only 25% the speed of those in control cells (Fig. 2A and B), but also showed severe stress (i.e., 75% of these replication forks progressed slower than 0.15 kb/minute; Supplementary Fig. S7B).
We also quantitated replication protein A (RPA) foci, an indicator of replicative stress (54, 55), in the MDA231 cyclin E–inducible cells. Consistent with the results of the DNA fiber assay, the RPA foci–positive cells were significantly more frequent (66%) in the cyclin E–induced cells treated with AZD1775 than in the control cells (4%), the untreated cyclin E–induced cells (28%) or the AZD1775–treated uninduced (27%) cells (Fig. 2C and D; Supplementary Fig. S7C). Furthermore, the levels of key regulators of DNA replicative stress: ATR, p-CHK1(S345), Wee1, and p-CDK1(Y15), also increased upon cyclin E induction and decreased upon cyclin E downregulation in 76NE6 and MDA231 cyclin E–inducible and HCC1806 cyclin E knockdown cell lines (Fig. 2E; Supplementary Fig. S7D). Collectively, these results suggest that replication fork progression and replicative regulation pathways in cells with high cyclin E is protected by Wee1 kinase activity.
The cyclin E–induced DNA replication stress may also increase DNA damage and activate DNA repair pathways (36, 47, 48). Consistent with this hypothesis, immunofluorescence staining of γH2AX foci (marker of DNA double-strand breaks) and RAD51 or 53BP1 foci (markers of DNA repair) revealed that cyclin E induction significantly increased frequencies of foci-positive cells for all three markers compared with uninduced MDA231 cells (Fig. 2C and D; Supplementary Fig. S7C). Treatment with AZD1775 increased γH2AX-positive cells, but not RAD51 or 53BP1-posivitve cells. Specifically, the majority (84%) of cyclin E–induced MDA231 plus AZD1775-treated cells are γH2AX positive. However, only 15% and 16% of these cells were RAD51 and 53BP1 positive, respectively (Fig. 2C and D; Supplementary Fig. S7C). These results suggest that the majority of the cyclin E–induced, AZD1775-treated MDA231 cells harbor unrepairable DNA damage.
Treatment of cyclin E–induced MDA231 and 76NE6 cells with AZD1775 resulted in significant accumulation of sub-G1 (Fig. 2F; Supplementary Fig. S8A) and cleaved PARP (Fig. 2E) compared with control cells. Furthermore, in these cyclin E–induced AZD1775-treated MDA231 and 76NE6 cells, the percentage of cells in G1 phase dropped to 10.5%–13.5% while the polyploid population increased up to fourfold compared with controls (Fig. 2F; Supplementary Fig. S8A). Measurement of multinucleation and micronucleation (indicators of mitotic catastrophe) revealed that treatment of high cyclin E MDA231 cells with AZD1775 resulted in accumulation of cells with abnormal nuclei in 48% of these cells but only 4.5% of uninduced and/or untreated controls (Fig. 2G; Supplementary Fig. S8B). Collectively, these results suggest that cyclin E overexpression induces DNA replicative stress and increases DNA double-strand breaks, which can be repaired via Wee1 kinase activation and the ATR–CHK1 pathway. However, when Wee1 kinase was inhibited by AZD1775 in the cyclin E–high cells, DNA repair pathways were unable to repair DNA damage sufficiently, leading to accumulation of abnormal nuclei and cell death.
CDK2 is required for cyclin E–mediated sensitization to Wee1 kinase inhibition in TNBC cells
Kinase activities of CDK2 and cyclin E in both 76NE6 cyclin E–inducible cells and HCC1806 cyclin E–downregulated cells (Supplementary Fig. S9A) were positively associated with levels of p-CDK2 (T-160; Supplementary Fig. S7D), suggesting that the kinase associated function of cyclin E is CDK2 dependent in these cells. Moreover, in patient samples (from our two TNBC patient cohorts; Fig. 1C and D), 85% (164/192) of all cyclin E–overexpressing TNBC tumors also overexpressed p-CDK2 (P = 0.0001; Supplementary Table S4), and that cooverexpression of cyclin E and p-CDK2 (but not p-CDK2 alone) was significantly (P = 0.0004) associated with recurrence (Fig. 3A; Supplementary Fig. S9B). On the basis of these findings, we next interrogated whether CDK2 is essential in sensitizing cyclin E–overexpressing cells to Wee1 kinase inhibition.
To this end, we compared the DNA damage and response abilities of the mutated cyclin ER130A, which prevents its interaction with CDK2 (39), to wild-type cyclin E in 76NE6-inducible cells. While the induction of wild-type cyclin E resulted in significant (P < 0.0003) accumulation of γH2AX and RAD51 foci-positive cells (Fig. 3B) and sensitized cells to AZD1775 (Fig. 1G), the induction of the mutant cyclin ER130A did not sensitize cells to AZD1775 nor increase the γH2AX and RAD51 foci–positive cells compared with the vector alone and the uninduced control cells (Fig. 3B and C; Supplementary Fig. S9C). Furthermore, knockdown of CDK2 in HCC1806 cells significantly (P = 0.0001) reduced their sensitivity to AZD1775 (Supplementary Fig. S9D) and decreased the levels of DNA replication stress regulators such as p-CHK1, ATR, and Wee1 (Supplementary Fig. S9E).
Next, we explored whether CDK2 depletion could rescue cyclin E–induced DNA replicative stress and AZD1775 sensitivity. CDK2 was knocked out (KO via CRISPR/CAS9) in MDA231 cyclin E–inducible cells, and the indicators of DNA replicative stress (RPA foci), DNA damage (γH2AX foci), and DNA repair (RAD51 foci) were measured. The percentages of cells with foci for all three indicators approached that of parental cells with basal levels of CDK2 and cyclin E expression (Fig. 3D and E; Supplementary Fig. S10A). The progression of DNA replication forks in CDK2 KO cells was faster than that in the parental cells under both cyclin E–uninduced (0.98 kb/minute in CDK2 KO vs. 0.59 kb/minute in parental) and cyclin E–induced (0.69 kb/minute in CDK2 KO vs. 0.46 kb/minute in parental) conditions (Fig. 3F and G; Supplementary Fig. S10B and S10C). Although the replicative forks of CDK2 KO cells with cyclin E overexpression (mean 0.69 kb/minute) progressed slower than those of the uninduced CDK2 KO cells, these forks still progressed faster than those of the parental cyclin E–uninduced cells (mean 0.56 kb/minute). These results suggest that cyclin E-induced replicative stress, DNA damage, and DNA repair are all dependent on a functional CDK2.
Similarly, the KO or knockdown (shRNA) of CDK2 inhibited the cyclin E–mediated increase of DNA replication stress regulators (such as p-CHK1 (S345), p-CDK1 (Y15), ATR, and Wee1 kinase) in both MDA231 and 76NE6 cyclin E–inducible cells (Fig. 3H; Supplementary Fig. S10D and S10E). To examine whether the ablation or knockdown of CDK2 can also rescue cells from cyclin E–induced sensitivity to AZD1775, 76NE6 CDK2 shRNA and MDA231 CDK2 KO cells were treated with increasing concentrations of AZD1775 in the presence (induced) or absence (uninduced) of cyclin E. The results revealed that KO or knockdown of CDK2 (in the presence or absence of cyclin E) made the cells more resistant to AZD1775 compared with their respective controls (Fig. 3I and J; Supplementary Fig. S10F and S10G). To explore the effect of pharmacologic inhibition of CDK2 on sensitivity to AZD1775 induced by cyclin E overexpression, MDA231 cyclin E–inducible cells were treated with dinaciclib (a CDK1/2/5/9 inhibitor) and AZD1775 (Supplementary Fig. S10H). The addition of dinaciclib increased the cell survival of the cyclin E–induced (but not the control) cells to AZD1775 across all concentrations we examined (Fig. 3K; Supplementary Fig. S10I). Collectively, these results (Fig. 3; Supplementary Figs. S9 and S10) underscore the requirement of CDK2 for cyclin E–induced sensitivity to Wee1 kinase inhibition.
Sequential combination treatment of dinaciclib followed by AZD1775 in TNBC with low cyclin E levels is synergistic in vitro and in vivo
Because in the uninduced (cyclin E–low) MDA231 cells, the combination treatment of dinaciclib and AZD1775 results in more cell killing than the AZD1775 treatment alone (Fig. 3K; Supplementary Fig. S10I), we next hypothesized that the combination of these two agents is likely to be synergistic in cyclin E–low TNBC cells. To test this hypothesis, we compared the efficacy of the dinaciclib and AZD1775 combination in MDA231 cyclin E–inducible cells, using a high-throughput survival assay that allows comprehensive evaluation of two or more drugs given either in sequence or concomitantly (schematically depicted in Supplementary Figs. S10H and S11A; refs. 40, 42).
The results revealed that only when cyclin E–low (uninduced) cells were treated sequentially with dinaciclib followed by AZD1775 was the combination therapy synergistic, with a CI of 0.8. On the other hand, the concomitant treatment with both agents, in the presence or absence of cyclin E, resulted in an antagonistic response (CI 1.14–1.5). Furthermore, the sequential combination of dinaciclib followed by AZD1775 was only additive (CI of 1.0) in cyclin E–induced cells (Fig. 4A, Supplementary Table S5). To validate these findings, we treated the panel of six TNBC cell lines (from Fig. 1E) either with sequential treatment of dinaciclib followed by AZD1775 (D→AZD) or with concomitant treatment of both agents (D+AZD; Supplementary Fig. S11B). The only condition resulting in a synergistic interaction of the two agents, with CIs of 0.3–0.8, was the sequential administration of the drugs (D→AZD) in the cyclin E–low TNBC cell lines (SUM149, SUM159, and MDA231). All other conditions, sequential or concomitant in cyclin E–high TNBC cells (HCC1806, MDA436, MDA468) or concomitant in cyclin E–low TNBC cells, resulted in either antagonistic or additive interactions, with CIs of 1.0–1.7 (Fig. 4B; Supplementary Table S5), which were also confirmed by clonogenic assays (Fig. 4C). The D→AZD treatment resulted in apoptosis in the cyclin E–low TNBC cells, as shown by (i) higher percentages of sub-G1, (ii) an increased annexin V–positive population, and (iii) accumulation of cleaved PARP in the combination-treated cells compared with the single agent–treated cells (Fig. 4D; Supplementary Fig. S11C; Supplementary Table S6). Cell-cycle analyses revealed that the changes in G1, S, or G2–M phases between the high and low cyclin E cells did not correlate with the efficacy of the D→AZD treatment (Supplementary Fig. S11D).
To confirm the synergism of CDK2 inhibition followed by Wee1 kinase inhibition in the cyclin E–low TNBC cells, we evaluated the efficacy of the sequential combination of other CDK2 inhibitors (meriolin5, SNS032, and roscovitine) with AZD1775. Results revealed that only the sequential combination of these CDK2 inhibitors followed by AZD1775, but not the concomitant treatment combination, caused synergistic interaction in the cyclin E–low TNBC cells (Fig. 4E; Supplementary Table S7).
To evaluate the efficacy of D→AZD treatment in vivo, we administered these drugs sequentially in orthotopic SUM149 xenograft and TNBC PDX models with either low or high cyclin E expression. For each model, we treated the mice first with 25 mg/kg dinaciclib once per day for 2 days, followed by 50 mg/kg AZD1775 twice per day for 2 days, followed by 3 days off; this 7-day cycle was repeated for 4–10 cycles (Supplementary Fig. S12A). Consistent with the in vitro results, D→AZD sequential treatment prolonged the mouse survival in PDX1 model (low cyclin E) synergistically (P = 0.0006) and in SUM149 xenograft model with a trend toward synergism (P = 0.097; Fig. 4F and G; Supplementary Table S8). In both models, the D→AZD sequential treatment also significantly decreased tumor growth compared with those treated with either of the single agents (Fig. 4F and G; Supplementary Fig. S12B and S12C; Supplementary Table S9). In the cyclin E–amplified PDX2 model, which was sensitive to AZD1775 alone (Fig. 1J), the D→AZD combination showed no significant improvements compared with AZD1775 alone (Fig. 4H; Supplementary Fig. S12B and S12C; Supplementary Tables S8 and S9).
We next interrogated whether the concomitant treatment (D+AZD) of the cyclin E–low PDX1 model could also result in a survival benefit compared with single-drug treatment. To this end, mice were treated concomitantly with 25 mg/kg dinaciclib once per day and 50 mg/kg AZD1775 twice per day on days 1 and 2 of each cycle, followed by 5 days off (Supplementary Fig. S12A). This 7-day treatment cycle was repeated for up to six cycles. The D+AZD arm did not show a significant difference in tumor growth compared with the single-drug arm and showed much less effectiveness compared with the D→AZD arm (Fig. 4G; Supplementary Fig. S12B and S12C). Furthermore, the D+AZD treatment arm had no additional survival benefit compared with no treatment or single-drug treatment arm (Fig. 4G). Finally, the D→AZD treatment was well tolerated in mice, as indicated by the lack of weight loss in any of the treatment arms during the treatment (Supplementary Fig. S12D) as well as normal whole-blood counts measured at the end of 6–10 cycles (Supplementary Fig. S13). Collectively, these results suggest that the sequential combination treatment using dinaciclib followed by AZD1775 is much more effective than the concomitant combination in vivo in the cyclin E–low PDX model.
Transient CDK2 inhibition–induced cyclin E sensitizes cyclin E–low TNBC cells to Wee1 kinase inhibition
To examine the mechanism of the sequential treatment with dinaciclib followed by AZD1775, we interrogated whether dinaciclib could induce the transient upregulation of cyclin E levels in the TNBC cells because cyclin E induction sensitized these cells to Wee1 kinase inhibitor (Fig. 1G; Supplementary Fig. S3B–S3D). Therefore, we compared the cyclin E levels among all six TNBC cell lines with or without 24-hour dinaciclib treatment and after the treatment withdrawal. Cyclin E protein levels in TNBC cells that were synergistic to the sequential D→AZD treatment were increased and remained high, even after dinaciclib was withdrawn for 24 hours (Fig. 5A). On the other hand, in TNBC cells that did not have synergistic response to the sequential D→AZD treatment, cyclin E levels did not increase.
The synergistic TNBC cells (i.e., cyclin E–low cells) had significantly less cyclin E mRNA (P < 0.04) than the cyclin E–high TNBC cells (Fig. 5B). Furthermore, the cyclin E mRNA levels increased upon dinaciclib treatment in the cyclin E–low cells, but not in the cyclin E–high cells (Fig. 5C). The binding of H3K27ac (which distinguishes active from inactive/poised enhancers; ref. 56) on the cyclin E enhancer increased significantly in cyclin E–low cells (MDA231 and SUM159) following treatment with dinaciclib, but not in cyclin E–high cells (MDA468; Fig. 5D; Supplementary Fig. S14A), indicating increased transcriptional activity on the cyclin E gene in the cyclin E–low cells upon dinaciclib treatment. These results suggest that dinaciclib treatment could induce cyclin E expression transcriptionally in the cyclin E–low TNBC cells.
The increased cyclin E levels, following dinaciclib-mediated transient induction, were associated with increased CDK2 kinase activity and p-CDK2 (T160) levels only upon removal of dinaciclib for 24 hours (D+24 lanes of Fig. 5A; Supplementary Fig. S14B). Moreover, the dinaciclib-induced cyclin E also significantly increased the percentage of cells with RPA foci and cells with both RPA and RAD51 foci at 24 hours after dinaciclib treatment (Fig. 5E; Supplementary Fig. S14C). These results suggest that while dinaciclib is necessary to induce cyclin E levels, the drug has to be removed for cyclin E to be functional.
KO of cyclin E (CRISPR/CAS9) in MDA231 cells resulted in antagonistic interaction of the sequential D→AZD treatment (Fig. 5F), suggesting the requirement of cyclin E for the observed synergism. To examine the contribution of CDK2 to the sequential treatment, cyclin E and/or CDK2 were downregulated (shRNA) in the cyclin E–low TNBC cells (Supplementary Fig. S14D). Downregulation of CDK2 increased the CI values of the D→AZD treatment, and the double knockdown with cyclin E changed the interaction of the sequential combination from synergistic to additive (CI = 1; Fig. 5G). Moreover, other CDK2 kinase inhibitors, including roscovitine, SNS032, and meriolin5, increased cyclin E protein levels in MDA231 and SUM159 cells (Fig. 5H). Finally, the sequential combination of roscovitine and meriolin5 with AZD1775 became antagonistic in the cyclin E KO MDA231 cells (Fig. 5I). These results suggest that only when the CDK2 inhibitor is removed does the transient induction of cyclin E by CDK2 kinase inhibition lead to a functional cyclin E/CDK2 complex with increased DNA stress, sensitizing TNBC cells to Wee1 kinase inhibition.
Sequential combination treatment of carboplatin followed by Wee1 kinase inhibitor in TNBC is synergistic in vitro and in vivo
We next examined the pharmacologic generalizability of transient cyclin E upregulation in mediating sensitivity to AZD1775 by conducting a small-scale screening of clinically available targeted and systemic therapeutic agents in two of the cyclin E–low cell lines, MDA231 and SUM159. These agents included a CDK4/6 inhibitor (palbociclib), an aurora kinase A inhibitor (MLN8237), and several chemotherapeutic agents, including carboplatin, cisplatin, paclitaxel, epirubicin, and doxorubicin. The synergistic efficacies of sequential treatment using these agents followed by AZD1775 were significantly correlated with the induction status of cyclin E (P = 0.0014; Fig. 6A and B; Supplementary Tables S6 and S7; Supplementary Table S10), suggesting that cyclin E expression could be a potential pharmacodynamic marker for predicting response to these sequential combinations of drugs with AZD1775.
RPA and RAD51 foci staining revealed that 24 hours of carboplatin treatment was sufficient to stimulate DNA replicative stress (RPA foci) in 50% and DNA repair (RAD51 foci) in 38% of the MDA231 cells (Fig. 6C). The percentage of cells that were positive for both RPA and RAD51 foci significantly increased upon carboplatin treatment and represented the main population of foci-positive cells (63% of RPA-positive and 82% of RAD51-positive), suggesting that most of cells with DNA replication stress are undergoing DNA repair (Fig. 6C; Supplementary Fig. S15A and S15B). These indicators of DNA replicative stress and DNA repair were further increased following the removal of carboplatin for 24 hours (Fig. 6C; Supplementary Fig. S15A and S15B). Cell-cycle analysis revealed that cells were blocked at both the S-phase and the G2–M phase with carboplatin and at the G2–M phase following 24 hours of removal from carboplatin (Fig. 6D). Because Wee1 kinase regulates both the DNA replicative checkpoint and the G2–M checkpoint (10, 11, 14, 57), these results suggest that the TNBC cells treated with carboplatin could respond to Wee1 kinase inhibition both before and after the removal of carboplatin. As predicted, the concomitant combination of carboplatin and AZD1775 in cyclin E–low TNBC cells were also synergistic in vitro (Supplementary Fig. S15C).
To evaluate the in vivo efficacy of the combination treatment using carboplatin and AZD1775, we first interrogated whether carboplatin treatment could increase cyclin E protein levels in the orthotopic MDA231T (MDA231 cells passaged through mice) xenograft models by treating the mice with various doses (0, 15, or 30 mg/kg) of carboplatin (1 dose/week) for 3 weeks. Consistent with the in vitro studies, cyclin E and p-CDK2 (T160) protein levels increased following carboplatin treatment (Fig. 6E). Next, we randomized MDA231T tumor–bearing mice into five arms: (i) vehicle (V), (ii) carboplatin treatment (carbo), (iii) AZD1775 treatment (AZD), (iv) concomitant treatment using carboplatin and AZD1775 (carbo+AZD), and (v) sequential treatment using carboplatin followed by AZD1775 (carbo→AZD). The mice were treated with 30 mg/kg carboplatin and 50 mg/kg AZD1775 in a 7-day cycle and repeated for 3–7 cycles (Supplementary Fig. S15D). Consistent with the in vitro studies, the carbo→AZD sequential combination significantly reduced tumor growth compared with the vehicle and single-drug arms and extended the median survival time from 15 days in the vehicle arm to 47 days, which was more effective than the carbo+AZD concomitant combination (Fig. 6F and G). Furthermore, the sequential carbo→AZD combination treatment was statistically synergistic using our two study endpoints of survival (P = 0.026) and tumor growth rate (P = 0.011; Supplementary Tables S8 and S9). However, the concomitant carbo+AZD combination was not synergistic with either study endpoints (Supplementary Tables S8 and S9). All single and combination treatment arms were well tolerated, as indicated by the lack of significant changes in body weight or in complete blood counts of non–tumor-bearing mice at the end of four cycles (Supplementary Figs. S15E and S16).
Discussion
Here we report that TNBCs with cyclin E overexpression have a worse prognosis and a higher rate of DNA damage than other TNBCs. We also found that PDX models derived from TNBCs that overexpress cyclin E are very sensitive to AZD1775 owing to CDK2-dependent activation of the DNA replication stress pathway. Moreover, coexpression of cyclin E and p-CDK2 in TNBC tumors is correlated with poor prognosis in >70% of all TNBC cases examined, suggesting AZD1775 could be a promising therapy for these patients. Finally, we found that the sequential treatment of PDX models with low cyclin E levels using agents that induce cyclin E expression followed by AZD1775 causes significant and sustained growth inhibition. Collectively, our results suggest that cyclin E alteration could be used as a potential biomarker to select patients whose tumors are likely to respond to AZD1775 either as a single agent or in combination with agents that can transiently induce cyclin E.
Patients with TNBC have limited treatment options beyond standard chemotherapy, which has substantial toxicities. Some therapeutics, including PARP inhibitors given in combination with chemotherapy have shown promise; however, owing to the lack of predictive biomarkers of response, many patients do not experience a response to such therapies. Genomic sequencing for genes involved in DNA repair pathways (in addition to BRCA1/2) across tumor types suggests that many patients with advanced malignancies may be candidates for DNA repair–targeted therapeutics such as AZD1775 (58). We propose that cyclin E expression could be used to identify patients who would either benefit from AZD1775 as monotherapy (in cyclin E–high TNBC tumors) or in sequential combination therapies with carboplatin or dinaciclib (in cyclin E–low tumors).
Our laboratory has previously reported on the prognostic role of cyclin E in breast cancer, across all subtypes and patient cohorts (30, 31). Specifically, we examined the expression of cyclin E, by IHC, in 2,494 breast cancer patients of all subtypes from four different patient cohorts. In multivariable analysis, compared with other prognostic factors (e.g., estrogen receptor status, progesterone receptor status, HER2 status, Ki67 status), cyclin E staining was associated with the greatest risk of recurrence. These studies suggest that cyclin E is likely to identify patients with the highest likelihood of recurrence consistently across different patient cohorts and breast cancer subtypes. In this study, we propose that cyclin E may also be used as a potential predictive biomarker for AZD1775-based therapies. Using the guidelines set forth by the Reporting Recommendation for Tumor Marker Prognostic Studies (REMARK; ref. 59), we evaluated the utility of cyclin E both as a prognostic and a potential predictive marker (Supplementary Table S11). While the role of cyclin E as a prognostic marker meets the REMARK guidelines, the criteria for the use of cyclin E as a predictive biomarker of response to AZD1775 have only been partially met, because the studies presented here are the preclinical evaluation of cyclin E expression as a biomarker of response to AZD1775 in cell lines and PDX models. However, these preclinical studies provided the needed rationale for a biomarker-driven clinical trial (NCT03253679) entitled “Wee1 inhibitor AZD1775 in treating patients with advanced refractory solid tumors with CCNE1 amplification.” This new phase II trial will enroll 35 patients with the histologically advanced solid tumors harboring cyclin E amplification. The patients enrolled in this trial will receive AZD1775 on days 1–5 and 8–12, which will repeat every 21 days in the absence of disease progression or unacceptable toxicity. Hence, linking the functionality of cyclin E amplification or overexpression to Wee1 kinase in individual tumors is critical for patient selection for AZD1775 therapy.
This study also provides mechanistic evidence in preclinical models to support the use of cyclin E to predict the response to AZD1775 in TNBC. We show that the replication stress pathway is activated upon cyclin E overexpression, which is biochemically functional and CDK2 dependent and results in sensitivity to Wee1 inhibition. The replicative stress pathway (Fig. 6H) as measured by the changes in the ATR-CHK1 axis, RPA foci–positive cells, and DNA replication progression are all modulated by either knock out or overexpression of cyclin E or CDK2 levels in TNBC cells.
Several previous studies have examined the association of genes involved in DNA repair (such as TP53, SIRT1, and RAD18), DNA synthesis (such as SETD2, RRM2, and Polκ), and G1–S phase regulation (such as SKP2, CUL1, and CDK2) with sensitivity to Wee1 kinase inhibition (12, 38, 60, 61), however, these findings have not been readily translated into the clinic. For example, while preclinical studies have shown that cancer cells with TP53 mutation are more sensitive to Wee1 kinase inhibition than those with wild-type TP53 (10, 15, 16), none of the patients with only TP53 mutation responded to AZD1775 treatment in a recent phase I trial (NCT01748825) (20). In the same trial, two patients carrying both BRCA and TP53 mutations achieved only partial response. These data indicate that owing to the complexity of cell-cycle checkpoint regulation, additional biomarkers are required to determine sensitivity to Wee1 kinase inhibitors.
Our study also identified an alternate treatment strategy in cyclin E–low TNBC tumors. We found that certain chemotherapeutic (i.e., platinum) or targeted (i.e., dinaciclib) agents are capable of transiently inducing the expression of cyclin E. Once cyclin E levels are induced, the cells and tumor models are then responsive to AZD1775 in both in vitro and in vivo models. As such we propose that treatment strategies combining a platinum agent first, followed by AZD1775 in sequence may have activity in cancers with low cyclin E levels. In a recent study, combination treatment effect of three agents; roscovitine, AZD1775, and AZD6738 (ATR inhibitor; ref. 62) in MDA231, revealed that inhibition of the three pathways being targeted may be antagonistic; however, the AZD1775 dose–response studies as well as the short-term growth inhibition assay system were not consistent with other published studies (38, 63).
In summary, the strengths of the current study are (i) establishing the prognostic value of cyclin E for TNBC patients; (ii) providing the needed evidence to design and implement a biomarker-driven clinical trials with AZD1775 as a single agent for cyclin E–high tumors (NCT03253679) and in combination with a platinum agent for cyclin E–low tumors. However, our study has some limitations. Because the prediction of response to the AZD1775-based treatments was investigated on the basis of cell-based and mouse models, they do not meet the REMARK criteria. Therefore, further clinical studies are required to establish cyclin E as a predictive biomarker for the mono- and combination AZD1775-based treatment strategies. The expression of cyclin E will need to be examined in all patients to stratify them into cyclin E–low and cyclin E–high cohorts. We propose that the cyclin E–high cohort is likely to respond to AZD1775 monotherapy, while the cyclin E–low cohort is more suited to the sequential carboplatin followed by AZD1775 treatment. Furthermore, the levels of cyclin E should also be measured following carboplatin administration and prior to AZD1775 in a window-trial setting to monitor whether carboplatin leads to the induction of cyclin E protein. Sequential cyclin E–inducing chemotherapy with carboplatin followed by AZD1775 may also prove to be less toxic than the combination administered concomitantly because carboplatin is used as a biological modulator (of cyclin E) rather than as a cytotoxic chemotherapy. As such, carboplatin can be dosed at a nontherapeutic dose at lower doses that would lead just to the induction of cyclin E. Our in vitro and in vivo studies suggest that the dose needed for inducing cyclin E is lower (50 mg/kg) than those used for therapeutic response in other studies (85–120 mg/kg; refs. 64, 65). Such dosing of carboplatin could be translated clinically to help reduce the significant adverse events currently associated with carboplatin therapy (66). These proposed clinical trials could have significant therapeutic implications for cyclin E–dependent cancers for which no targeted therapy is currently available.
Disclosure of Potential Conflicts of Interest
F. Meric-Bernstam reports receiving commercial research grants from Novartis, AstraZeneca, Taiho, Genentech, Calithera, Debio, Bayer, Aileron, PUMA, CytoMx, Zymeworks, Curis, Pfizer, and eFFECTOR, and is a consultant/advisory board member for Genentech, Inflection Biosciences, Pieris, Xencor, Samsung Bioepis, Sumitomo Group, Spectrum, OrigiMed, and Audoro. K. K. Hunt is a consultant/advisory board member for Armada Health. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: X. Chen, K.K. Hunt, K. Keyomarsi
Development of methodology: X. Chen, K.H. Low, K.W. Evans, K. Keyomarsi
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Chen, A. Alexander, Y. Jiang, S. Vijayaraghavan, M. Yi, D.C. Ellis, K.-L. Cheung, I.O. Ellis, F. Meric-Bernstam, K. Keyomarsi
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Chen, A. Alexander, C. Karakas, K.R. Hess, J.P. Carey, M. Yi, K.-L. Cheung, S. Fu, F. Meric-Bernstam, K.K. Hunt, K. Keyomarsi
Writing, review, and/or revision of the manuscript: X. Chen, A. Alexander, K.R. Hess, S. Vijayaraghavan, M. Yi, D.C. Ellis, K.-L. Cheung, I.O. Ellis, S. Fu, K.K. Hunt, K. Keyomarsi
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Chen, T. Bui, I.O. Ellis, K.K. Hunt
Study supervision: X. Chen, K.K. Hunt, K. Keyomarsi
Acknowledgments
Research reported in this publication was supported by the National Cancer Institute of the NIH under award (P30CA016672) to The University of Texas MD Anderson Cancer Center, R01 grants CA1522218, CA223772 and Cancer Prevention and Research Institute of Texas (CPRIT) RP170079 grants (to K. Keyomarsi), the NCI Experimental Therapeutics Clinical Trials Network grant 3UM1CA186688-03S2 (to F. Meric-Bernstam), the Susan G. Komen for the Cure grant KG100521 (to K.K. Hunt), the Susan G. Komen post-doctoral fellowship grant PDF14302675 (to J.P.W. Carey), and the CPRIT RP170067 training grant (to S. Vijayaraghavan).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.