Abstract
Chemotherapy remains the mainstay of treatment for advanced breast cancer; however, resistance is an inevitable event for the majority of patients with metastatic disease. Moreover, there is little information available to guide stratification of first-line chemotherapy, crucial given the common development of multidrug resistance. Here, we describe an in vivo screen to interrogate the response to anthracycline-based chemotherapy in a syngeneic metastatic breast cancer model and identify JNK signaling as a key modulator of chemotherapy response. Combining in vitro and in vivo functional analyses, we demonstrate that JNK inhibition both promotes tumor cell cytostasis and blocks activation of the proapoptotic protein Bax, thereby antagonizing chemotherapy-mediated cytotoxicity. To investigate the clinical relevance of this dual role of JNK signaling, we developed a proliferation-independent JNK activity signature and demonstrate high JNK activity to be enriched in triple-negative and basal-like breast cancer subtypes. Consistent with the dual role of JNK signaling in vitro, high-level JNK pathway activation in triple-negative breast cancers is associated both with poor patient outcome in the absence of chemotherapy treatment and, in neoadjuvant clinical studies, is predictive of enhanced chemotherapy response. These data highlight the potential of monitoring JNK activity as early biomarker of response to chemotherapy and emphasize the importance of rational treatment regimes, particularly when combining cytostatic and chemotherapeutic agents. Mol Cancer Ther; 16(9); 1967–78. ©2017 AACR.
Introduction
Despite significant advances in the systemic treatment of breast cancer, metastatic disease is still considered incurable with the 5-year survival rate at 25% (www.seer.cancer.gov). Thus, the acquisition of resistance is often an inevitable event in advanced disease. As a consequence, there is an urgent need to not only develop new therapeutics but also identify clinically relevant mechanisms of resistance and predictive biomarkers of response for the mainstay treatment for advanced disease, namely chemotherapeutic reagents (1). Chemotherapy selection has traditionally been guided by large randomized trials based only on the origin of disease (2), as opposed to molecularly defined disease subtypes. In addition, with the plethora of clinically approved chemotherapeutics, there is a lack of information to guide personalization of chemotherapeutic approaches and stratify patients on the basis of predicted response (3).
Loss-of-function genetic screening, particularly the exploitation of RNA interference (RNAi), has emerged as a high-throughput and unbiased approach to identify genes associated with a phenotype of interest (4). In vitro RNAi screens have been fundamental to the understanding of mechanism of action of molecularly targeted therapeutics and their resistance pathways (5). However, such screens are unable to differentiate the mechanisms of greatest clinical relevance and may miss additional in vivo–specific drivers of response and resistance (6, 7). In vivo RNAi screening has emerged as a physiologically relevant approach to explore gene function in tumor biology (8–10) and therapeutic response (11). In this study, we have adapted an in vivo short-hairpin (sh)RNAi approach combined with massively parallel sequencing to reveal novel determinants of response to anthracycline chemotherapy in a model of breast cancer metastasis.
Using this approach, we identified c-Jun N-terminal kinase 1 (JNK1) as a clinically relevant determinant of sensitivity to anthracycline chemotherapy. The JNK family of mitogen-activated protein kinases (MAPK) has been implicated in the pathogenesis of a number of tumor types, plus in other pathologies such as autoimmune disease (12, 13). However, multiple roles have been proposed for these kinases, on one hand preventing malignant transformation via induction of apoptosis and on the other promoting cell survival in established tumors. Our study supports a dual role for the JNK family in advanced breast cancer, identifies the potential of monitoring JNK activity as an early biomarker of response, and has implications when considering therapeutic strategies that combine cytotoxic chemotherapy with agents that promote a cytostatic response.
Materials and Methods
Cells and reagents
4T1-Luc cells (Sibtech) were obtained in 2010 and used within 10 passages after resuscitation. Human cells were obtained between 2005 and 2009 from ATCC and short tandem repeat (STR) tested every 4 months using StemElite ID System (Promega). All cells were routinely subject to mycoplasma testing. Details of cell culture, shRNA transduction (Supplementary Table S1), apoptosis, and qPCR (Supplementary Table S2) assays are provided in the Supplementary Methods. Chemotherapy drugs and inhibitors were as follows: 5-flurouracil (Sigma), cyclophosphamide monohydrate (Sigma), doxorubicin hydrochloride (Apollo Scientific), SP600125 (Tocris Bioscience), JNK inhibitor VIII (Calbiochem), and JNK-IN-8 (Calbiochem). See Supplementary Table S3 for details of antibodies and their use.
In vivo shRNA screen
All animal work was carried out with UK Home Office approval. See text and Supplementary Methods for full screen details, Supplementary Table S4 for list of primers employed in the screen, and Supplementary Table S5 for in vivo chemotherapy regimens. In brief, we used an shRNA library targeting the Cancer 1000 mouse gene set (14) divided into 24 subpools each containing 96 shRNAs. For each subpool, 5 × 105 shRNA expressing 4T1-Luc cells were inoculated intravenously into BALB/c mice (n = 12) on day 0 (Supplementary Fig. S1A), and mice were treated with AC chemotherapy (2.5 mg/kg doxorubicin, 40 mg/kg of cyclophosphamide) or vehicle on days 2, 7, 11, and 16. On day 21, tumor burden was assessed by in vivo IVIS imaging and ex vivo lung weight, and gDNA was extracted from preinoculation cell pellets and tumor-bearing lungs (10). Supplementary Figure S1B outlines the strategy for combining samples prior to PCR amplification and high-throughput sequencing (15). Raw image data were analyzed using GA pipeline v1.8 and short reads aligned to the shRNA library sequences using shALIGN (15) and subject to quality control testing (Supplementary Fig. S1C).
Human datasets
The procedure to generate JNK activity signature (JAS) and proliferation-independent JAS (piJAS) scores based on the H-JNK versus L-JNK gene set of Chang and colleagues (16) is described fully in Supplementary Fig. S2.
piJAS and JAS scores were assessed for breast cancer samples in TCGA (17), METABRIC (18), and the de Ronde (19) (GSE34138) and Hess (20) neoadjuvant trials. In both trials, responders with a pathological complete response (pCR) were defined by the complete eradication of tumor from the breast and axillary lymph nodes, whereas non-responders with residual disease (RD) were those with no or a partial response. Molecular subtypes were defined using the PAM50 classifier (21).
Statistical analysis
Statistical analysis was performed using GraphPad Prism6 and R version 3.2.4 (https://www.bioconductor.org/). Unless otherwise stated, data represent mean values ± SEM, and individual comparisons were made using the two-tailed Student t test.
For the analysis of the human datasets, in the Tukey boxplots, box indicates the 2nd and 3rd quartiles, bar indicates median, whiskers indicate 1.5 IQR (interquartile range), and dots indicate outliers. To assess the assumption of normality in each group, D'Agostino test was used. To assess the equality variances, F test (two groups) or Levene's test (three or more groups) was used. Where the groups had equal variances the significance of the differences of the groups was tested using either unpaired Student t test (two groups) or one-way ANOVA (multiple groups) followed by Dunnett post-hoc testing for correcting multiple two-group comparisons. Where the groups did not have equal variances, unpaired Student t test with Welch correction (two groups) or nonparametric Kruskal–Wallis H test (multiple groups) with Dunn post-hoc testing for correcting multiple two-group comparisons was used. All P values reported were 2-tailed. Full statistical details are provided in Supplementary Table S6.
In all cases; *, P < 0.05; **, P < 0.01; ***, P < 0.001; n.s., not significant.
Results
An in vivo shRNA screen for modulators of chemotherapy response
We set out to develop a clinically relevant in vivo model of cytotoxic chemotherapy treatment in patients with breast cancer with disseminated disease. Anthracyclines, a mainstay in the first-line adjuvant setting for patients with advanced breast cancer, are used in combination regimens; for example, FAC [5-fluorouacil, the anthracycline doxorubicin (Adriamycin), cyclophosphamide] or AC (doxorubicin, cyclophosphamide). In preliminary studies, the presence of 5-fluorouracil within the triple combination, although effective at reducing tumor burden, resulted in significant toxicity in mice (Supplementary Fig. S3), consistent with the greater toxicity of 5-fluorouracil in mouse compared with human cells (22). Consequently, we assessed an AC chemotherapy regime of either high-dose (2.5 mg/kg) or low-dose (1.25 mg/kg) doxorubicin in combination with cyclophosphamide (40 mg/kg) in mice inoculated with 4T1 cells expressing a luciferase transgene (4T1-Luc) transduced with a pool of five nontargeting shRNAs (4T1-Luc NTC; Fig. 1A). Following 4 rounds of chemotherapy, although a loss of body weight was observed with the high-dose doxorubicin combination, this was less severe than the triple FAC combination (Supplementary Fig. S3) and tolerable over the 21-day assay (Fig. 1A). Moreover, the high-dose AC combination, compared with the low-dose combination, led to a significant reduction in tumor burden (Fig. 1B). The high-dose AC combination was used in all further studies.
An in vivo functional screen to identify modulators of chemotherapy response. A and B, A total of 5 × 105 4T1-Luc NTC cells were injected via the tail vein of BALB/c mice. Groups were treated with 4 doses (red arrowheads) of vehicle, low, or high doxorubicin AC combination (n = 8–10 mice per group). A, Toxicity was monitored by changes in total body weight. B, Lung tumor burden assessed by in vivo IVIS imaging (left) and ex vivo weight (right). Dashed line indicates mean lung weight of aged-matched non–tumor-bearing mice. One-way ANOVA, Tukey honest significant difference (HSD) post-testing. C, Outline of the in vivo screen. D, Day 21, in vivo IVIS imaging (left) and ex vivo lung weight (right) of all mice in the screen (n = 144 mice per treatment arm). E, Short-list of prioritized putative chemotherapy modulators identified in the screen. Hits, number of independent shRNAs significantly enriched/depleted; Gene expression, ranked gene expression (in %) from 4T1 cells freshly isolated from in vivo tumors.
For the in vivo screen, we employed the Cancer 1000 shRNA library consisting of 2,204 shRNAs, targeting a panel of about 1,000 cancer-associated mouse genes (8, 23), with a subpool complexity of 96 shRNAs (24 subpools). The screen strategy is outlined in Fig. 1C. For each of the subpools, 12 BALB/c mice were inoculated with 5 × 105 GFP+ retrovirally transduced 4T1-Luc cells, IVIS imaged after 1 hour and placed into balanced treatment groups (Supplementary Fig. S1A). Mice were treated with four rounds of vehicle or AC chemotherapy and tumor burden in the lungs assessed on day 21 (Fig. 1D). gDNAs extracted from each set of lungs were combined (see Supplementary Fig. S1B) to generate six vehicle and six AC chemotherapy–treated biologic replicates. Following PCR amplification of the shRNAs from the library plasmids, preinoculation cells, and lung samples, products were subject to next-generation sequencing, short reads aligned to the shRNA library reference sequences using shALIGN, and total reads per shRNA were log2-transformed and median-centered per pool using shRNASeq (15). A comparison of the vehicle and AC chemotherapy–treated samples identified 192 shRNAs significantly enriched or depleted (Supplementary Fig. S4A). When genes targeted by these 192 shRNAs were analyzed, the top enriched gene ontology (GO) terms were cell death and signaling pathways (Supplementary Fig. S4B), providing confidence in the screen given the well-defined role of cell death/apoptosis pathways in chemotherapeutic response.
A short list of putative chemotherapy modulators (Fig. 1E) was prioritized on the basis of the identification of multiple significantly enriched/depleted shRNAs targeting the same gene and removing shRNAs that (i) from profiling 4T1 cells freshly isolated from in vivo tumors (24) were in the lower 50th percentile expression range, (ii) from comparing shRNA abundance in the plasmid pool and the preinoculation 4T1-Luc cells negatively affected cell viability, and (iii) from comparing shRNA abundance in the preinoculation 4T1-Luc cells and the vehicle-treated metastasis samples negatively affected metastasis.
From this shortlist, the potential interaction between Mapk8 (Jnk1) and the transcription factor Elk3 (ETS domain–containing protein 3) was of interest, given reports that activated JNK phosphorylates the ETS family member ELK1 (25, 26). Furthermore, the observation that independent shRNAs targeting Mapk8 were enriched in tumors isolated from AC-treated, compared with vehicle-treated, mice implicated a contribution of JNK signaling to tumor cell chemotherapeutic killing in vivo.
Inhibition of JNK signaling impairs proapoptotic signaling and promotes resistance to anthracyclines in mouse and human models of breast cancer
To validate the effects of inhibiting JNK activity in vivo, 4T1-Luc cells were transduced with two independent shRNAs targeting Mapk8 (Jnk1; shMapk8-1; shMapk8-2) that were distinct from the shRNAs in the Cancer 1000 library (Fig. 2A). As the JNK family contains two other members that are targets for pan-JNK inhibitors, cells were also transduced with two independent shRNAs targeting Mapk9 (Jnk2; shMapk9-1; shMapk9-2; Fig. 2A). Mapk10 (Jnk3) was not targeted, as Jnk3 expression is primarily restricted to neural tissue (27) and was not expressed in 4T1 cells. Effective downregulation of either Mapk8 or Mapk9 expression had no impact on 4T1-Luc cell proliferation in vitro (Fig. 2B). However, when inoculated into the tail vein of vehicle-treated mice, Mapk8-knockdown cells showed a reduced tumor burden compared with the control (Empty, shNTC) cells (Fig. 2C, left; Fig. 2D). Importantly, relative to their respective vehicle-treated arms (Fig. 2C, right), there was a statistically significant decrease in tumor burden in the AC chemotherapy–treated mice inoculated with shEmpty and shNTC 4T1-Luc cells, whereas the difference between the vehicle and AC arms for the shMapk8-1 and shMapk8-2 cells did not reach significance. For Mapk9-knockdown cells, no significant difference for shMapk9-1 cells was observed; however, shMapk9-2 cells showed a statistically significant reduction in tumor burden in AC-treated mice, consistent with Jnk2 not being identified as a significant hit in the in vivo shRNA screen.
Downregulation of Mapk8 (Jnk1) or Mapk9 (Jnk2) expression in vivo. 4T1-Luc cells were transduced with two independent shRNA lentiviruses targeting Mapk8 or Mapk9. A, qPCR analysis. Mean of three independent experiments; one-way ANOVA, Tukey honest significant difference (HSD) post-hoc testing. B, Cell viability (n = 6 wells per time point). Mean of two independent experiments; two-way ANOVA. C, A total of 5 × 105 shRNA 4T1-Luc cells were injected via the tail vein of BALB/c mice on day 0 and treated with vehicle or AC chemotherapy as in Fig. 1A. Mice were culled on day 19 and (left) ex vivo lung weight recorded. P < 0.05 (one-way ANOVA). Individual comparisons were n.s. [Tukey honest significant difference (HSD) post-hoc testing]. To quantify chemotherapy response (right), AC relative to vehicle mean lung weight; one-way ANOVA, Tukey honest significant difference (HSD) post-hoc testing. D, Representative hematoxylin and eosin (H&E)-stained lung sections from C. Scale bar, 1 mm.
To confirm that the effect of JNK downregulation was due to loss of JNK catalytic activity, 4T1-Luc cells were treated with increasing concentrations of three chemically distinct pan-JNK small-molecule inhibitors, SP600125 (28), JNK inhibitor VIII (29, 30), and JNK-IN-8 (13). JNK inhibition was monitored by a decrease in Ser63 phosphorylation of the JNK downstream target c-Jun (Fig. 3A). In combination with doxorubicin treatment, SP600125 resulted in a dose-dependent reduction in doxorubicin-mediated cell killing (Fig. 3B). However, SP600125 alone at higher concentrations impaired the proliferation of 4T1-Luc cells (Fig. 3C). Therefore, to quantify the level of SP600125 antagonism, the observed inhibition of the combined agents was compared with the expected inhibition on the basis of the single-agent response curves. Using a Bliss additive model (31), which is appropriate where the two agents are not directed against the same target, antagonism was observed across the SP600125 and doxorubicin drug concentrations (Fig. 3D). More strikingly, in a colony formation assay (Fig. 3E), SP600125 treatment alone had little effect but was able to effectively rescue the doxorubicin-mediated severe reduction in colony number. Of note, although SP600125 alone did not impair colony number, colony size was reduced consistent with a cytostatic effect of JNK inhibition.
Inhibition of JNK activity antagonizes response to doxorubicin in vitro. A, Immunoblotting of 4T1-Luc cells treated with the JNK inhibitors SP600125, JNK inhibitor VIII, or JNK-IN-8 for 24 hours. B and C, 4T1-Luc cells treated with doxorubicin and SP600125 (B) or SP600125 alone (C). Cell viability was assessed at 72 hours. Data in B are normalized relative to the appropriate SP600125 concentration in the absence of doxorubicin. D, Using raw data from B, the expected combined inhibition was calculated according to the Bliss additive single agent model. Heatmap indicates deviation between observed and expected combined inhibition. E, A total of 1 × 102 4T1-Luc cells per well were seeded into 6-well plates, cultured for 7 days, and then treated with doxorubicin and/or SP600125 for 4 days (n = 3 per condition). Left, Representative crystal violet stained plates. Right panel, normalized colony number relative to vehicle treatment. Mean values from three independent experiments. F, 4T1-Luc cells were plated in 96-well plates (n = 6 per condition) to achieve confluency at 24 hours when JNK inhibitors (1 μmol/L), doxorubicin (1 μg/mL), or the combination was added. Cell viability was assessed 48 hours later. Equivalent results were obtained in three independent experiments.
As proliferation rate is known to be associated with chemotherapy response (32), it was critical to confirm that the resistance mediated by JNK knockdown or inhibition was independent of its cytostatic effect. Compared with the effects observed in subconfluent cells (Fig. 3B) and on the colony size in the colony formation assay (Fig. 3E), SP600125 treatment alone, at two different concentrations, had no significant impact on cell viability in confluent cultures (Fig. 3F and Supplementary Fig. S5A), supporting the concept that the JNK inhibition results in cytostasis rather than cytotoxicity. In contrast, in these confluent cultures, doxorubicin retained its ability to significantly impair cell viability and this effect was significantly abrogated by SP600125 in the combination treatment. Equivalent results were obtained using the higher specificity inhibitors, JNK inhibitor VIII and JNK-IN-8. Similarly, in confluent cultures, treatment with the three JNK inhibitors alone did not promote apoptosis, whereas the number of apoptotic cells was substantially increased following doxorubicin treatment (Fig. 4A). Again, all three JNK inhibitors significantly abrogated doxorubicin-induced cell apoptosis.
JNK inhibition impairs doxorubicin-mediated apoptosis. A, 4T1-Luc cells were plated in 96-well plates (n = 6 per condition) to achieve confluency at 24 hours when JNK inhibitors (1 μmol/L), doxorubicin (1 μg/mL), or the combination was added for 24 hours. Quantification of apoptotic cells in three independent assays. B, Immunoblotting of confluent 4T1-Luc cells treated for 24 hours as in A. C, 4T1-Luc treated for 24 hours as in A before staining for active mitochondrial Bax (arrowheads). Scale bar, 50 μm. Active Bax+ cells quantified in 8 fields per treatment. Mean values from 2 independent experiments.
As further support for JNK modulating doxorubicin resistance independent of its cytostatic effect, we examined downstream apoptotic effectors. JNK inhibition had no effect on the total levels of Bax or cleavage of caspase-3. Conversely, doxorubicin treatment enhanced c-Jun phosphorylation and substantially increased the level of cleaved caspase-3, effects that were effectively rescued by all three JNK inhibitors (Fig. 4B). Phosphorylation of the proapoptotic Bcl-2 family member Bax results in activation and translocation to the mitochondrial membrane, and an antibody specific for the active confirmation of Bax (33) only stains caspase-3–positive cells (Supplementary Fig. S5B). JNK inhibition alone did not promote Bax activation but, at two different concentrations, all three JNK inhibitors significantly abrogated doxorubicin-induced Bax activation (Fig. 4C and Supplementary Fig. S5C). Equivalent results were obtained with human MDA-MB-231 cells (Fig. 5A). In conclusion, active JNK promotes sensitivity to doxorubicin by stimulating translocation of active Bax to the mitochondria and enhancing apoptosis.
JNK inhibition impairs doxorubicin response in human breast cancer cells. A, MDA-MB-231 cells treated with SP600125 (5 μmol/L), doxorubicin (1 μg/mL), or both and the number of activated Bax+ cells quantified as described in Fig. 4C. Scale bar, 50 μm. B, qPCR analysis of MAPK8 (JNK1; left) and MAPK9 (JNK2; right) in ER+ (red) and ER− (blue) breast cancer cell lines. Mean values (n = 3). C, Immunoblotting of cells treated for 22 to 24 hours with SP600125 (5 μmol/L), doxorubicin (1 μg/mL), or both. D, A total of 3 × 103 cells per well were seeded into 96-well plates. Twenty-four hours later, doxorubicin was added at the indicated concentrations with either vehicle or 5 μmol/L SP600125, n ≥ 3 wells per data point. Cell viability was assessed 72 hours later. Equivalent results were obtained in 4 independent experiments.
JNK activity modulates AC response in human breast cell lines
In human breast cancer cells, MAPK8 (JNK1) expression was notably higher in the estrogen receptor (ER)-negative (MDA-MB-231, MDA-MB-468) compared with the ER+ (MCF7, ZR75.1) cell lines, whereas levels of MAPK9 (JNK2) expression were not associated with hormone receptor status (Fig. 5A). To examine levels of JNK activity, phosphorylation of c-Jun was monitored by immunoblotting, with protein loading and exposure times equivalent to allow comparison between cell lines (Fig. 5B). Basal JNK activity did not correlate with either MAPK8 or MAPK9 levels but, as observed in the 4T1-Luc mouse cells (Fig. 4B), doxorubicin treatment of the human breast cancer lines enhanced c-Jun phosphorylation and this enhanced JNK activity was blocked by SP600125 (Fig. 5B). Of note, JNK inhibition resulted in reduced levels of total c-Jun consistent with the autoregulation of c-Jun by its own activity (34).
In the 4T1-Luc cells, doxorubicin-mediated cell killing has an IC50 of 40 ng/mL. The human breast cancer cells lines showed variability in their response to doxorubicin (Fig. 5C) with the ER− lines have IC50 values of about 7 ng/mL (MDA-MB-231) and about 4 ng/mL (MDA-MB-468), whereas the ER+ lines were more resistant with IC50 values of about 60 ng/mL (ZR75.1) and about 13 ng/mL (MCF7). Interestingly, although SP600125 impaired the growth of the cell lines to varying degrees, it only antagonized the doxorubicin-mediated cytotoxicity of the human ER− cell lines (Fig. 5C), consistent with the results obtained with the ER− 4T1 cell line.
Activity of the JNK signaling cascade is associated with triple-negative breast cancer and is associated with poor prognosis
The data presented support a critical role for JNK in the activation of apoptosis in response to anthracycline chemotherapy. While it has been reported that JNK signaling is required for induction of the mitochondrial apoptotic pathway, the transcriptional response downstream of JNK, including activation of the proto-oncogene c-Jun, is primarily believed to be oncogenic (12, 35). This indicates a dual role for JNK, on one hand promoting the survival of tumor cells whereas on the other inducing apoptosis in response to genotoxic insult. Given that in cell lines, JNK activity does not correlate with MAPK8 or MAPK9 expression, we further explored the association of JNK signaling activity in different human breast cancer subtypes and in their response to chemotherapy.
In hepatocellular carcinoma, a gene expression signature of JNK activity has been associated with poor overall survival (16). However, this signature contains a number of proliferation-associated genes that could confound potential differences in patient prognosis, an issue of particular importance in breast cancer where a high proliferation index is associated with poor prognosis (36, 37). Consequently, proliferation associated genes were removed from the JNK activity signature (JAS) as described in Supplementary Fig. S2 to generate a piJAS. Using this piJAS, proliferation-independent JNK activity scores were generated for samples classified by receptor status (Fig. 6A and B and Supplementary Table S6) or molecular subtype (Supplementary Fig. S6A and S6B and Supplementary Table S6) in both the METABRIC and TCGA gene expression datasets containing 1,992 and 522 primary breast cancers, respectively. In both datasets, significantly higher piJAS scores were observed in triple-negative (TN; i.e., ER−, PR−, HER2−) and basal-like tumors. To confirm this association was indeed proliferation-independent, we applied the original JAS to the same datasets (Supplementary Fig. S6C–S6F and Supplementary Table S6). The results were broadly similar; however, it was notable that a higher JAS score was observed in the luminal B versus the luminal A tumors (Supplementary Fig. S6E and S6F), consistent with the higher proliferation index associated with luminal B tumors (37). The TCGA dataset also includes reverse-phase protein array (RPPA) data for 402 breast cancer samples probed with a phospho-JNK antibody. Levels of phospho-JNK again correlated with molecular subtype, with highest phospho-JNK detected in HER2+ and basal-like tumors (Supplementary Fig. S6G).
piJAS is associated with poor prognosis and response to neoadjuvant chemotherapy in breast cancer. See Supplementary Table S6 for full statistical details. Numbers of samples in each category are indicated. A and B, Tukey boxplots of piJAS scores in the METABRIC and TCGA breast cancer datasets based on the tumor receptor status. C and D, Kaplan–Meier analysis of breast cancer–specific survival [disease-specific survival (DSS)] of patients with triple-negative breast cancers in the METABRIC study who did not receive chemotherapy (C) or were chemotherapy-treated (D) using weighted averaged expression of piJAS. Clinical information is shown in Supplementary Table S7. Patients surviving more than 180 months were censored. E and F, Analysis of neoadjuvant study of HER2-negative patients in de Ronde and colleagues' dataset. G and H, Analysis of neoadjuvant FAC/T study of Hess and colleagues. piJAS scores in all (E and G) and triple-negative (F and H) breast cancer cases divided into those with pCR and residual disease (RD).
Increased JNK activity predicts response to neoadjuvant chemotherapy in human breast cancer
To address whether JNK activity had prognostic value in determining response to chemotherapy in human breast cancers, we focused on the triple-negative tumors given that JNK activity was highest in these cancers. Importantly, the triple-negative breast cancers are most likely to be treated with chemotherapy, given a lack of targeted therapeutics. Furthermore, despite having greater chemotherapy response rates, patients with triple-negative breast cancer have the highest relapse and poorest survival rates (21, 38), with the majority of events occurring within the first 60 months after diagnosis (39, 40). In triple-negative breast cancers that were not treated with chemotherapy (Fig. 6C; see Supplementary Table S7 for clinical details), high JNK signaling, as monitored by a high piJAS score, showed a significant correlation with a poorer patient outcome (breast cancer–specific deaths), indicating that a high JNK activity is associated with a more aggressive tumor phenotype. Conversely, in the chemotherapy-treated patients, JNK signaling had no prognostic value (Fig. 6D), indicating that, despite conferring a more aggressive phenotype, high JNK activity is also associated with an increased response to chemotherapy.
On the basis of these findings, we turned to JNK activity as a predictor of chemotherapy response in randomized neoadjuvant trials. The study of de Ronde and colleagues (19) contains gene expression profiling from 178 biopsies of HER2-negative breast cancers taken prior to neoadjuvant anthracycline chemotherapy. Again, highest piJAS scores were found in the triple-negative and basal-like breast cancers (Supplementary Fig. S7A and B and Supplementary Table S6). Critically, the piJAS score was significantly higher in pathologic complete response (pCR) group compared with the non-responders with residual disease (RD) whether considering all patients (pCR in 28 of 165; Fig. 6E) or only the triple-negative patients (pCR in 24 of 52; Fig. 6F). Of note, the observation that 24 of 28 patients with a pCR had triple-negative tumors is consistent with this subtype having a higher response rate to chemotherapy. Finally, this analysis was extended to a second neoadjuvant dataset reported by Hess and colleagues (20), containing gene expression profiling of 133 breast cancers prior to FAC/T (FAC plus taxane) chemotherapy. There was no significant difference in piJAS score between the ER+ and triple-negative tumors but, within the molecular subtypes, basal-like cancers displayed a significantly higher piJAS score than the luminal A and luminal B subtypes (Supplementary Fig. S7C and D and Supplementary Table S6). Again, a significantly higher piJAS score was detected in the pCR group compared with those with residual disease, whether assessing all patients (pCR in 34 of 133; Fig. 6G) or only triple-negative patients (pCR in 13 of 27; Fig. 6H).
Discussion
In vitro RNAi screens have been critical for understanding mechanisms of resistance to molecularly targeted agents in cancer (5, 7); however, such approaches have not been informative in studying resistance to cytotoxic chemotherapy (7), which remains a mainstay of treatment for metastatic disease. The study design models the use of adjuvant chemotherapy, following surgery, to prevent the establishment of metastatic disease. The goal was to integrate RNAi technology with this model to identify modulators of chemotherapy response in metastatic breast cancer in vivo. We demonstrate the feasibility of this approach and identify the JNK pathway as critical for response to doxorubicin, a prototypical member of the anthracycline chemotherapeutic class.
The JNK pathway is associated with a number of disease states including diabetes, cancer, as well as inflammatory and neurodegenerative disorders (12). As a result, the JNK family represents an attractive drug target and a number of inhibitors have been developed. However, paradoxically, studies have demonstrated both a protumorigenic (16, 41, 42) and a tumor-suppressive (43, 44) role for the JNK family, with duration of activation and context both implicated in defining the response (45). A major consideration in addressing these contrasting results is that reports of JNK acting as a tumor suppressor have studied the formation of primary tumors (46–48), where JNK-mediated activation of apoptosis has an inhibitory role. In contrast, the in vivo screen described here represents an advanced tumor setting and our functional analyses confirm the complex interplay of prosurvival and proapoptotic signaling via JNK, which has important implications for the use of JNK inhibitors as potential therapeutics in cancer (12).
Intriguingly, the kinases MAP3K1 and MAP2K4, upstream members of the JNK pathway, are the fourth and seventh most frequently mutated genes in breast cancer, respectively (17), and exhibit a mutually exclusive pattern (49). The majority of these mutations are predicted to result in loss of function (7, 8) and are found almost exclusively in luminal disease (17, 49). In addition, MAP2K4 is frequently associated with LOH events (17, 18). The absence of these mutations in triple-negative/basal-like breast cancers suggest that signaling via the JNK pathway may be an important driver of proliferation and survival in these subtypes, a concept supported by the demonstration of the differential sensitivity of ER+ and ER− breast cancer cell lines to JNK inhibition (Fig. 5C and D). Furthermore, the increased JNK activity in ER− disease (Fig. 6A and B) may create a unique vulnerability and provide an explanation for the higher sensitivity of ER− tumors to chemotherapy (49).
Using a previously developed signature of JNK pathway activity (16), we confirmed the association of JNK activity with the triple-negative and basal-like breast cancer subtypes in two large breast cancer gene expression datasets (17, 18). Furthermore, when proliferation genes were removed from the signature, this correlation was maintained confirming the association is not driven by the higher proliferation rate of these subtypes (17). In the prognostic setting, the proliferation-independent signature was associated with poor outcome only when chemotherapy was absent from the treatment regime, further supporting a dual role of the JNK pathway in promoting both tumor aggressiveness (see Fig. 2C and D) and response to chemotherapy. Critically, in two independent neoadjuvant trials of anthracycline chemotherapy (19, 20), the proliferation-independent signature was predictive of response.
These findings have two important implications when considering therapeutic strategies. First, they predict that the development of JNK inhibitors for treatment of advanced breast cancers would have dual effect in JNK-active tumors acting as cytostatic agents but also inhibiting JNK downstream stress response and apoptosis pathways, thereby antagonizing anthracycline-mediated cytotoxicity. Consequently, treatment with a JNK inhibitor would require careful planning. For example, if combined with cytotoxic therapeutics, it would be important to assess whether treatment was more effective as an interspersed rather than combination regime. Second, in both of the neoadjuvant studies, biopsies were collected prior to treatment, indicating that endogenous JNK activity is associated with response. However, our cell line data indicate that increased JNK activity in response to anthracycline treatment, as opposed to basal activity, would be more accurate in predicting response. Determining whether JNK activity can predict therapeutic response prior to treatment and/or act as a dynamic biomarker to guide early therapeutic switching would require identifying a robust immunohistochemical and/or gene expression assay for monitoring JNK signaling in serial biopsies taken pre- and during neoadjuvant anthracycline-based chemotherapy, and correlating changes in JNK activity with clinical response. Ultimately, the predictive value of JNK signaling would need to be explored in response to other chemotherapeutic and targeted agents, as this may have potential in guiding stratification or sequencing of treatment.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: M. Ashenden, Q. Gao, C.M. Isacke
Development of methodology: M. Ashenden, A. van Weverwijk, N. Murugaesu
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Ashenden, A. van Weverwijk, A. Fearns, M. Iravani
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Ashenden, A. Fearns, J. Campbell, M. Iravani
Writing, review, and/or revision of the manuscript: M. Ashenden, J. Campbell, Q. Gao, M. Iravani, C.M. Isacke
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Ashenden
Study supervision: C.M. Isacke
Grant Support
This work was funded by Breast Cancer Now, including a Breast Cancer Now studentship to M. Ashenden. We acknowledge NHS funding to the NIHR Biomedical Research Centre at The Royal Marsden and the ICR.
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.
Acknowledgments
The results published here are in whole or part based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/). This study makes use of METABRIC data generated by the Molecular Taxonomy of Breast Cancer International Consortium with funding provided by Cancer Research UK and the British Columbia Cancer Agency Branch. We thank Fredrik Wallberg, David Robertson, Frances Daley, Kerry Fenwick, and Iwanka Kozarewa for invaluable assistance in this project; Scott Lowe for providing the Cancer 1000 shRNA library; Anita Grigoriadis and Olivia Fletcher for statistical advice; and Christopher Lord and Nicholas Orr for advice on the planning and analysis of the screen.
Footnotes
Note: Supplementary data for this article are available at Molecular Cancer Therapeutics Online (http://mct.aacrjournals.org/).
- Received October 31, 2016.
- Revision received April 15, 2017.
- Accepted May 18, 2017.
- ©2017 American Association for Cancer Research.