MAPK Reliance via Acquired CDK4/6 Inhibitor Resistance in Cancer

Dysregulation of the cell cycle is a hallmark of cancer. Although initial attempts to target the cell cycle with nonspecific cyclin-dependent kinase (CDK) inhibitors were clinically unsuccessful, a new generation of selective CDK4/6 inhibitors has emerged that has shown clinical promise across multiple cancer types, including breast, melanoma, and colorectal cancer (1, 2). Currently, three CDK4/6 inhibitors have entered clinical trials: palbociclib (Ibrance), ribociclib (Kisqali), and abemaciclib (Verzenio) (1, 3). Notably, these three inhibitors have recently received accelerated FDA approval for treatment of hormone receptor–positive (HR⁺), Her2-negative breast cancer in combination with endocrine therapy based on the PALOMA-2 (palbociclib), the MONALEESA-2 (ribociclib), and the MONARCH-2 (abemaciclib) phase III clinical trials (4).

Despite evidence of clinical response, development of resistance is common, and the underlying mechanisms that lead to resistance are poorly understood. Consistent with the known functions of CDK4/6 in promoting cell-cycle progression through phosphorylation and inactivation of the retinoblastoma tumor-suppressor protein (RB), CDK4/6 inhibitors require an active RB pathway to elicit antitumor effects. As such, resistance to CDK4/6 inhibitors can occur through disruption of the RB pathway, as mediated by loss of the retinoblastoma tumor suppressor (RB), cyclin D1 and CDK6 amplification, and cyclin E-CDK2 activation (3, 5, 6). Other mechanisms of resistance have also been reported that are independent of RB pathway alteration, including alterations or activation of the PI3K/AKT/mTOR pathway (7, 8). Given the promise of CDK4/6 inhibitors in the clinical setting, there is an increasing need to discern mechanisms of bypass and identify mechanisms to anticipate and prevent therapeutic resistance.

Therapeutic opportunities for implementation of CDK4/6 inhibitors in the clinical setting are widest in tumors for which RB pathway disruption is infrequent or occurs late in tumor progression. As such, studies herein were conducted using prostate adenocarcinoma as a tumor paradigm, in which a functional RB pathway is largely intact in early stage disease. Prior to androgen deprivation therapy, which is the first line of therapeutic intervention for disseminated disease, almost all tumors retain RB function, with 5% RB loss observed in primary tumors (9). By contrast, RB loss is enriched in metastatic disease, occurring in 37% of metastatic cases in a retrospective cohort, and in 21% of metastatic castration-resistant prostate cancer (CRPC; ref. 10), and is found to be causative for the transition to CRPC (11), the lethal stage of disease. Thus, CDK4/6 inhibitors are under clinical testing in both hormone-naïve metastatic prostate adenocarcinoma (NCT02059213) and metastatic CRPC (NCT02555189) as adjuvant therapy for first- and second-line hormone therapy, respectively.

As shown herein, acquired palbociclib resistance not only rendered cancer cells broadly resistant to CDK4/6 inhibitors, but also promoted aggressive phenotypes, including accelerated growth in vitro and in vivo, as well as enhanced invasion and clonogenic capacity. Unbiased global gene expression and phosphoproteomic profiling were utilized to interrogate the molecular alterations of acquired resistance to CDK4/6 inhibition in multiple models of palbociclib resistance. These integrated approaches revealed a reduction of RB function resulting in CDK4/6 inhibitor resistance. In addition, acquired CDK4/6 resistance was associated with activation of the MAPK signaling pathway, which conferred sensitization to MEK inhibitors. In sum, these studies demonstrate partially retained, hyperphosphorylated Rb in acquired CDK4/6 inhibitor resistance and nominate MEK inhibitors as a new treatment strategy for advanced cancers upon developing CDK4/6 inhibitor resistance.

LNCaP- and LAPC4-derived cells were cultured in improved minimum essential media (IMEM; Corning) or Iscove's modified Dulbecco media (Corning), respectively, supplemented with 5% heat-inactivated FBS (HyClone), 1% l-glutamine, and 1% penicillin/streptomycin (11). Unless otherwise described, cells were plated overnight and treated with 0.5 μmol/L PD0233991 (PD, palbociclib, SelleckChem), LEE011 (LEE, ribociclib, Novartis), or U0126 (Promega). Palbociclib-resistant cell models were generated from LNCaP and LAPC4 cells by sustained treatment with 0.5 μmol/L PD (schematic in Fig. 1A) and maintained under selection when resistance, measured via bromodeoxyuridine (BrdUrd) incorporation, was achieved after approximately 3 months. Cells were authenticated by the ATCC and checked for mycoplasma upon thawing and at termination of maintenance after <20 passages (ATCC 30-1012K).

Cells were treated with a dose range of indicated doses of PD, LEE, or U0126 for up to 6 days, with drugs refreshed every other day. Cells were trypsinized, counted twice on a hemacytometer using the Trypan blue exclusion method, and normalized to a drug-free control. Experiments were performed at least twice in quadruplicate.

Cells were treated with 0.5 μmol/L PD or LEE for 24 hours, and after a 2-hour pulse labeling of BrdUrd (Amersham, RPN201), adherent cells were harvested and fixed with 100% ethanol. Proliferation was measured by bivariate flow cytometry using BrdUrd and propidium-iodide (PI) staining. A Millipore Guava EasyCyte flow cytometer captured 10,000 BrdUrd/PI events, and the Millipore InCyte software was used to gate for the percentage of BrdUrd incorporation. Experiments were performed at least twice in biological triplicate.

Cells were treated with 0.5 μmol/L PD or CTRL for 24 hours and processed to assess RB1 expression, as described previously (12), performed at least twice in biological triplicate.

Protein harvesting and immunoblotting were performed as previously described (12), quantified in ImageJ. Antibodies used are mouse-α-RB (BD-Biosciences; 554136), rabbit-α-GAPDH (Santa Cruz Biotechnology; SC-25778), rabbit-α-CyclinA (Santa Cruz Biotechnology; SC-596), rabbit-α-ERK1 (Santa Cruz Biotechnology; SC-94), rabbit-α-phospho-p44/p42 (T202/Y204; Cell Signaling Technology, 4370S), and goat-α-LaminB (Santa Cruz Biotechnology; SC-6217). Phospho blots were blocked and immunoblotted in 2.5% BSA, and all other blots with 5% milk in PBS-Tween.

Corning FluoroBlok 24-Multiwell and Corning BioCoat Tumor Invasion System plates were utilized for migration and invasion assays, according to protocols provided by the company, performed at least twice in quadruplicate. Bottom wells were filled with IMEM/20%FBS. A total of 50,000 cells (passed 2x, 40 μm cell strainers) were seeded in the top compartment of each well in serum-free IMEM and incubated for 48 hours (migration) or 72 hours (invasion). Cells were stained with Corning Calcein AM Fluorescent Dye and measured on a BioTek SynergyHT plate reader.

The bottom layer of 1% agar/PBS was mixed 1:1 with culture media supplemented with 20% FBS and poured into 6-well plates, after which a second layer with 0.6% agar/FBS mixed with a cell-media/20%FBS suspension was poured on top. Where mentioned, drugs were mixed into both agar layers. Note that 200 μL media ± drugs were added on top and refreshed twice a week. A total of 5,000 cells were seeded per well in triplicate. Plates were incubated for 3 weeks, after which each well was photographed. Images were analyzed with FIJI (ImageJ) as follows: select each well transform to gray scale, adjust the threshold to remove background, and detect colonies. Use the Analyze Particles algorithm to count total colonies and pixel sizes of each colony. Experiments were performed at least twice in triplicate.

Xenograft studies were performed in accordance with NIH Guidelines, and animal protocols were approved by Institutional Animal Care and Use Committee at Thomas Jefferson University. Cells (3 × 10⁶ per injection) suspended in PBS were combined 1:1 with Matrigel (BD Biosciences; 354234) and injected s.c. into the flanks of 5- to 6-week-old, intact-male athymic nude mice (Charles River Laboratories). Tumor development was monitored over time by palpation. Where indicated, mice received AIN-76A diet laced with 6.7 mg/kg trametinib or control (kindly provided by the laboratory of Dr. Andrew Aplin, Thomas Jefferson University).

RNA was extracted with the RNeasy Kit (Qiagen) from PDR and parental LNCaP or LAPC4 cells pre-treated 24h with 0.5 μmol/L PD or vehicle (CTRL). Note that 100 to 200 ng of total RNA was used to prepare RNA sequencing (RNAseq) libraries using the TruSeq RNA Sample Prep Kit V2 (Illumina), following the protocol described by the manufacturer. High-throughput sequencing (HTS) was performed using an Illumina HiSeq2500 with each sample sequenced to a minimum depth of approximately 50 million reads. A paired end 2 × 125 cycle sequencing strategy was used. Data were subjected to Illumina quality control (QC) procedures (>80% of the data yielded a Phred score of 30). Secondary analysis was carried out on an OnRamp Bioinformatics Genomics Research Platform (OnRamp Bioinformatics; ref. 13). OnRamp's advanced Genomics Analysis Engine utilized an automated RNAseq workflow to process the data (13, 14), including (1) data validation and quality control, (2) read alignment to the human genome (hg19) using TopHat2 (15), which revealed >90% mapping of the paired end reads, (3) generation of gene-level count data with HTSeq, and (4) differential expression analysis with DEseq2 (15), which enabled the inference of differential signals with robust statistical power. (Genomics Research Platform with RNAseq workflow v1.0.1, including FastQValidator v0.1.1a, Fastqc v0.11.3, Bowtie2 v2.1.0, TopHat2 v2.0.9, HTSeq v0.6.0, DEseq v1.8.0.)

The resulting BAM files were sorted and input into the Python package HTSeq to generate count data for gene-level differential expression analyses. To infer differential signal within the data sets with robust statistical power, DEseq2 was utilized (15). Transcript count data from DESeq2 analysis of the samples were sorted according to their adjusted P value or q value and the smallest FDR at which a transcript is called significant (q < 0.1). FDR is the expected fraction of false-positive tests among significant tests and was calculated using the Benjamini–Hochberg multiple testing adjustment procedure. LNCaP (LN) and LAPC4 (L4) sequencing data are deposited using NCBI's Gene Expression Omnibus (16), accessible through GEO Series accession number GSE99675.

PDR and parental LNCaP or LAPC4 cells were treated for 24 hours with 0.5 μmol/L PD or CTRL, scraped, pelleted, and snap frozen. Protein digestion and phosphopeptide enrichment were performed as previously described (17–19) with minor modifications. Briefly, cells were lysed in 6 mol/L guanidinium hydrochloride buffer (6 mol/L guanidinium chloride, 100 mmol/L Tris pH 8.5:10 mmol/L Tris (2-carboxyethyl) phosphine, 40 mmol/L 2-chloroacetamide, 2 mmol/L vanadate, 2.5 mmol/L NaPyrophosphate, 1 mmol/L Beta-glycerophosphate, 10 mg/mL N-octyl-glycoside). Lysates were sonicated and cleared, and protein was measured. Note that 5 mg of protein was digested with trypsin, and the resulting phosphopeptides were subjected to phosphotyrosine antibody-based enrichment via immunoprecipitation. The immunoprecipitate was washed, and phospho-Tyrosine (pY) peptides were eluted. The supernatant from the pY immunoprecipitations was kept for phospho-serine/threonine (pST) peptide enrichment. pST peptides (2.5 mg) were desalted using C18 columns and then separated using strong cation exchange chromatography. In separate reactions, the pY and pST peptides were then further enriched using titanium dioxide columns to remove existing nonphosphorylated peptides. The pY and pST peptides were then desalted using C18 tips prior to submission on the mass spectrometer. Samples were analyzed by LC-MS/MS using a dual pump nanoRSLC system (Dionex) interfaced with a Q Exactive HF (ThermoFisher). Samples were run in technical duplicates, and data were searched using MaxQuant Andromeda version 1.5.3.30 (20) against the Uniprot human reference proteome database with canonical and isoform sequences (downloaded September 2016 from http://uniprot.org). MaxQuant Andromeda parameters were set as previously described (21). Data are deposited in the ProteomeXchange Consortion via the PRIDE partner repository, accessible through dataset identifier PXD006561 (22).

MS data analysis was performed as previously described (23). For clustering, pY data were filtered using an FDR-corrected ANOVA P value of 0.2, and pS/pT data were filtered using an FDR-corrected ANOVA P value of 0.05. Hierarchical clustering was performed using the Cluster version 3.0 with the Pearson correlation and pairwise complete linkage analysis (24). Java TreeView version 1.1.6r4 was used to visualize clustering results (25).

Kinase substrate enrichment analysis (KSEA) was performed as previously described (18). Briefly, phosphopeptides were rank-ordered by average fold change between PDR and parental cells. The enrichment score was calculated using the Kolmogorov–Smirnov statistic. Statistical significance was calculated via permutation analysis. The normalized enrichment score (NES) was calculated by taking the enrichment score and dividing it by the mean of the absolute values of all enrichment scores from the permutation analysis. The Benjamini–Hochberg procedure was utilized to calculate FDR for each kinase. For pY analyses, cutoffs of FDR < 0.05, hits > 4, and NES > 1.3 were used. For pST analyses, cutoffs of FDR < 0.02, hits > 5, and NES > 2 were used.

Although CDK4/6 inhibitors have shown promise in clinical trials for cancer treatment, acquired resistance is common. Thus, the present study interrogated the underpinnings of therapeutic resistance, to identify markers for therapeutic outcome and develop new strategies when resistance develops.

Palbociclib-resistant (PDR) prostate adenocarcinoma cell models generated from hormone therapy–sensitive prostate adenocarcinoma cells (LNCaP and LAPC4) demonstrated retained BrdUrd incorporation after PD treatment (LNCaP-parental: 5.3%; LN-PDR1: 32.3%; LN-PDR2: 34.2%), indicative of a G₁–S cell-cycle checkpoint bypass (Fig. 1A; Supplementary Fig. S1A). Retained cyclin A (an RB/E2F target gene) protein levels after 24-hour treatment with 0.5 μmol/L PD confirmed acquired resistance (Fig. 1B; Supplementary Fig. S1B). Ribociclib (LEE011, LEE) is similar to PD in chemical structure and likewise targets the ATP-binding pocket, and thus it was not surprising that the PDR models were resistant to LEE, as observed via retained BrdUrd incorporation (LNCaP-parental: 9.8%; LN-PDR1: 35.4%, LN-PDR2: 37.9%) and cyclin A (Fig. 1A and B; Supplementary Fig. S1A and S1B). Dose–response curves confirmed that PDR models continued to proliferate, whereas the parental cells were arrested at submicromolar doses of PD or LEE (Fig. 1C and D). These findings indicate that acquired resistance to palbociclib confers broad resistance to this class of agents, suggestive of common mechanisms of bypass.

To identify transcriptional alterations underlying CDK4/6 inhibitor resistance, PDR and parental cells were treated for 24 hours with 0.5 μmol/L PD or control (CTRL), and RNAseq was performed. The MA plots in Supplementary Figs. S2A (right) and S3A represent the log ratio (M) of the PD versus CTRL values over the average log intensity (A) of each transcript, visualizing a global reduction of differentially expressed genes in PDR models compared with the parental cells after PD treatment. Strikingly, the large cluster of downregulated genes in the parental model (indicated by the blue arrow) was absent in the resistant cells, confirming that PD is unable to significantly inhibit this gene cluster. When comparing PDR1 and PDR2 profiles with parental cells, in both PD-treated and CTRL conditions, vast transcriptomic changes were observed in both the total number of differentially expressed transcripts and log ratio amplitudes (Fig. 2A, left; Supplementary Fig. S3B). Although PD treatment showed limited impact on the transcriptome of the PDR lines compared with parental cells, these models demonstrated extensive alteration in the transcriptome to adapt to extended PD exposure. The majority of transcripts (>1.5-fold, q value < 0.1) were common between the two independently generated LNCaP PDR models in both CTRL and PD conditions (Fig. 2A, right). Although the PDR models showed a few distinctions, deregulated signaling pathways common between PDR models are most likely to contribute to resistance.

To identify common pathways enriched in the PDR models, complete transcriptional profiles were utilized in Gene Set Enrichment Analyses (GSEA: www.broadinstitute.org/GSEA) interrogating the predefined Oncogenic Signatures and Hallmarks from the Molecular Signature Database (MSigDB; Fig. 2B; Supplementary Figs. S2B and S3D). Gene sets enriched in both PDR models under at least one condition were selected to highlight pathways that are most likely to contribute to acquired CDK4/6 inhibitor resistance. Enriched hallmark gene sets included G₂–M checkpoint, supporting the contention that resistant cells bypass the G₁–S checkpoint and progress through cell cycle. Enrichment of E2F targets in PDR lines suggests increased E2F activity, potentially mediated through a bypass of RB. Genomic RB loss has been observed in approximately 30% of advanced prostate adenocarcinoma across different patient cohorts, and as RB is the main target of CDK4/6-mediated phosphorylation, loss of this gene has been attributed to resistance to CDK4/6 inhibitors (5, 26). Concordantly, RB knockdown Oncogenic Signatures were also enriched, further highlighting that RB function may be reduced (Fig. 2B). RT-PCR and immunoblotting demonstrated that total RB mRNA and protein were retained in all PDR lines, albeit reduced to 29% to 56% (Supplementary Figs. S2C and S3C, top). Retained RB protein in the PDR models remained hyperphosphorylated [Supplementary Figs. S2C and S3C, bottom, upper band total RB (tRB)] in the presence of a CDK4/6 inhibitor, which attenuates RB function, confirming aberrant inactivation of RB in the presence of CDK4/6 inhibitors that likely cause bypass of the G₁–S checkpoint and activation of E2F transcription factors. As the kinome forms an intricate network of interactions (27), it was postulated that bypass of CDK4/6 inhibition likely causes kinome rewiring. Notably, RNAseq uncovered enrichment of numerous kinase-related signatures (Fig. 2B and C; Supplementary Fig. S4A). These findings served as the impetus for further exploration to identify specific kinase pathway(s) that could present novel targets for therapeutic intervention.

Phosphoproteomic mass spectrometry profiling provided an unbiased approach to investigate the PDR-associated kinome. Cells were treated for 24 hours with either 0.5 μmol/L PD or control, lysed, and digested (Fig. 3A). pY peptides were immunoprecipitated with an anti-pY antibody, whereas pST peptides remained in the supernatant, allowing analysis of both peptide populations via LC-MS/MS (23). Hierarchical clustering revealed altered phosphorylation levels of peptides across models and treatments, whereas duplicate samples cluster together (Fig. 3A; Supplementary Fig. S4B). Phospho-peptide profiles were subjected to KSEA. The pY peptides unveiled multiple motifs targeted by Src and its SH2 domain (Fig. 3B; Supplementary Fig. S4C), which is linked to cancer progression (28). The pST peptides were strongly enriched for motifs targeted by ERK1 and ERK2 (aka MAPK3 and MAPK1; Fig. 3C; Supplementary Fig. S4D). To prioritize changes in the phosphoproteome most likely to link to CDK4/6 inhibitor resistance, results were compared with the gene expression data. Intriguingly, KRAS and RAF kinases, both of which were highly represented in the Oncogenic Signatures from the RNAseq data, regulate ERK1/2 via mitogen-activated protein kinase kinase, or MEK (29). These data suggest that kinase-signaling cascades play an important role in acquired CDK4/6 inhibitor resistance in these models. Moreover, the MAPK pathway is targetable with clinically tested pharmacologic agents, such as MEK inhibitors that are already FDA approved for some cancers (e.g., melanoma; ref. 29), nominating this pathway for further study. Together, these data indicate that acquired resistance to CDK4/6 inhibitors was associated with induced MAPK pathway activity.

As described above, oncogenic signatures representative of altered KRAS and RAF signaling were highly enriched as a function of acquired CDK4/6 inhibitor resistance. KRAS acts upstream of RAF family kinases, regulating their activity, and in turn, the RAF family kinases directly activate ERK1/2 kinases through phosphorylation (29). Indeed, although overall ERK1/2 levels were unchanged, ERK1/2 kinases were hyperphosphorylated in the resistance models (Fig. 4A; Supplementary Fig. S5A), which would explain the observation that ERK1/2 motifs are differentially phosphorylated in the phosphoproteomics analysis. Although canonical MAPK-activating growth factor receptors (EGFR and FGFR families; ref. 30)) displayed no consensus in differential expression, EGF transcripts were consistently upregulated in the PDR models, which was associated with altered EGFR activity in the phosphoproteomic data (Fig. 3B; Supplementary Fig. S4C). Elevated EGF protein levels were confirmed via Western blotting, which likely causes MAPK activation in these models (Fig. 4B; Supplementary Fig. S5B). Combined, these data strongly suggest that acquired CDK4/6 inhibitor resistance results in increased EGF production, leading to hyperactivation of the MAPK pathway.

As activated MAPK signaling is known to induce proliferation via induction of D-type cyclins and prevent apoptosis, and thus can drive cancer progression (29, 31–33), the PDR models were further characterized biologically. Baseline growth (off PD selection) was compared with the parental cells, demonstrating an almost 3-fold higher growth rate for the LNCaP-derived models (Fig. 4C; no growth difference was observed for LAPC4 cells, data not shown). Transwell chemotaxis assays from 0% to 20% serum conditions revealed an increased migratory capacity and enhanced invasion in the PDR models (Fig. 4D; Supplementary Fig. S5C). Consonantly, transcriptome data revealed enrichment of the GSEA hallmark for epithelial–mesenchymal transition (EMT) in the PDR lines (Supplementary Figs. S2D and S3E), which could underlie the aggressive, migratory, and invasive characteristics these models have obtained upon acquiring PD resistance. To assess clonogenic growth capacity, cells were suspended at low density in agar-containing culture media and incubated for 3 weeks to allow colony formation. HRAS-transformed mouse adult fibroblasts (MAF cells) served as a positive control due to previously described capacity to grow three-dimensional (3D) colonies (34). Representative images of culture wells for each cell line are shown in Fig. 4E (left). To utilize an unbiased approach for colony formation analyses, a macro was developed for FIJI (ImageJ), which counts total colonies per well and measures the pixel size of each colony. Although there was no significant difference in total colonies formed by the LN-PDR or parental LNCaP cells (Supplementary Fig. S5D; note that LAPC4 and derivatives did not form colonies after >1 month), median colony sizes in the LN-PDR1 and LN-PDR2 models were significantly larger than the parental cells (P = 0.0008 and P = 0.0001, respectively, one-way ANOVA; Fig. 4E, right). Combined, these findings reveal that acquired resistance to CDK4/6 inhibitors is associated with phenotypes linked to aggressive tumor behavior.

To challenge these concepts in vivo, cells were injected s.c. on the flanks of athymic nude mice (n = 6 per group) and monitored over time for tumor take. Strikingly, the PDR models reached a 50% tumor take after only 13 to 14 days, whereas the first parental tumor formed after 20 days, and a 50% tumor take was not reached until 47 days after injection. Moreover, 100% of the mice injected with PDR cells formed palpable tumors, compared with only 67% in the parental model (Fig. 4F). Combined, acquired CDK4/6 inhibitor resistance is associated with hyperproliferation, enhanced migration and invasion, enlarged colonies, and accelerated, more efficient in vivo tumor take, thus promoting aggressive tumor phenotypes.

With clinical trials testing an MEK inhibitor already underway for prostate adenocarcinoma (trametinib: NCT01990196, NCT02881242), and the observation that the direct downstream targets of MEK (i.e., ERK1/2) are hyperactivated upon CDK4/6 inhibitor resistance, the PDR models were assessed for sensitization to MEK inhibition. Cells treated overnight with a preclinical MEK inhibitor (U0126) showed reduction in phosphorylation of ERK1/2, the primary target proteins for MEK kinases (Fig. 5A; Supplementary Fig. S6A). To assess sensitivity to MEK inhibition, cells were treated for 6 days with U0126, and quantified via Trypan blue exclusion, demonstrating that the CDK4/6 inhibitor–resistant cells are sensitized to MEK inhibition (Fig. 5B; Supplementary Fig. S6B). In addition, MEK inhibition reduced the invasion capacity of the PDR lines, whereas the parental cells were unresponsive (Fig. 5C). The MEK inhibitor hampered the clonogenic capacity in the PDR models, reducing not only the size of the colonies, but also total numbers of colonies formed, whereas the parental cells were not affected (Fig. 5D). Combining MEK and CDK4/6 inhibitors resulted in a cooperative effect (Fig. 5E). Sensitization to MEK inhibition was validated in vivo (Fig. 5F), wherein mice with palpable xenograft tumors received a diet laced with trametinib (6.7 mg/kg AIN-76A diet) or control diet (AIN-76A). These data demonstrated that tumor growth of the CDK4/6 inhibitor–resistant models was reduced, whereas parental tumors were unresponsive. Together, these findings suggest that activation of the MAPK pathway promotes not only the cell growth in the PDR models, but also the more aggressive, invasive phenotypes observed. In addition, the CDK4/6i-resistant cells appear to have become more reliant on MAPK signaling and as such are sensitized to MEK inhibition.

Although KRAS-activating mutations are not commonly observed in prostate adenocarcinoma, the MAPK pathway can be activated via other mechanisms, such as RAF fusions and overexpression (35). To investigate the clinical relevance of potential MAPK activation in prostate cancer, human primary prostate adenocarcinoma [The Cancer Genome Atlas (TCGA); ref. 36] and metastatic CRPC (SU2C/PCF; ref. 10) datasets were interrogated for alterations of the following components of the MAPK signaling pathway (Fig. 6A and B): RAS family (KRAS, HRAS, NRAS), RAF family (c-RAF/RAF1, ARAF, BRAF), MEK family (MAP2K1, MAP2K2), and ERK1/2 (or MAPK3 and MAPK1). The primary prostate adenocarcinoma dataset displayed alterations (mutations, CNA, mRNA z-score:2.0) in any of these kinases in 38% of all patients, the majority of which were mRNA upregulation (Fig. 6A and B; Supplementary Fig. S7). The frequency of MAPK-related upregulation was confirmed in the metastatic CRPC cohort, in which 47% presented with alterations, including RAF amplifications and gene fusions that may lead to MAPK activation. Interestingly, RAF alterations are more common than RAS perturbations in either cohort (Fig. 6B). Taken together, data herein suggest that acquired CDK4/6 inhibitor resistance is associated with reduced RB expression and loss of function, and this coincides with MAPK activation that may bypass CDK4/6-RB signaling to induce phenotypes of aggressive features and metastasis; however, this dependence confers sensitivity to MEK inhibitors that may be exploited therapeutically.

CDK4/6 inhibitors have shown clinical benefit in multiple tumor types, including breast cancer; however, these agents are unlikely to provide a durable cure, and development of resistance is anticipated. Studies herein developed preclinical models of CDK4/6 inhibitor resistance to identify major mechanisms and consequences of CDK4/6 bypass, wherein: (i) Acquired resistance to palbociclib resulted in broad CDK4/6 inhibitor resistance; (ii) Resistance was associated with MAPK activation, which (iii) conferred aggressive in vitro and in vivo phenotypes; and (iv) MAPK-activated, CDK4/6 inhibitor–resistant cancer cells are sensitized to MEK inhibitors. In summary, these studies identify MAPK induction in acquired CDK4/6 inhibitor resistance and nominate MEK inhibitors as a means to either prevent or treat CDK4/6 inhibitor–resistant cancer.

Several previous reports have approached CDK4/6 inhibitor resistance in various cancers from different perspectives, including clinical correlations with genetic alterations (6), correlation of RB loss-of-function gene signature (37), and an siRNA screen identifying PDK1 by targeting the whole kinome (7), which was not observed in the current study. To date, few studies have reported on acquired resistance modeling (8, 26, 31, 38), which largely focused on genomic alterations, gene amplifications, or deletions that could lead to CDK4/6 inhibitor resistance (including RB1 loss and other aberrations in the CDK4/6-RB pathway). Although the observed genetic alterations are poised to affect CDK4/6 inhibitor efficacy, these may not reflect the molecular underpinnings of disease progression in patients that initially respond. Data herein suggest that cancer cells acquired resistance through aberrant inactivation of retained (albeit reduced) RB protein, likely via rewiring of the kinome to bypass CDK4/6 inhibition. Induction of the MAPK pathway was observed in CDK4/6 inhibitor resistance, determined by comparing phosphoproteomic and transcriptomic data and selecting alterations observed across PDR models. Although genomic and functional RB losses are associated with poor cancer outcomes (39) and progression to castration resistance in the context of prostate adenocarcinoma (11, 12, 40), the role for MAPK signaling in regulation of RB remains somewhat elusive. MAPK controls expression and function of D-type cyclins, and CDK4/6 requires direct binding to D-type cyclins to exert kinase function and phosphorylate RB; therefore, it could be speculated that the observed elevated MAPK signaling bypasses CDK4/6 inhibitors via cyclin D (Fig. 6C). Cyclin D1 elevation has been linked to early adaptation to palbociclib in breast cancer models, where it induced a noncanonical cyclin D1/CDK2 complex that restored RB phosphorylation under CDK4/6 inhibition (8). However, the CDK4/6 inhibitor–resistant models presented herein displayed no change in cyclin D1 levels (data not shown), and MEK inhibition did not affect hyperphosphorylation of RB (Fig. 5A; Supplementary Fig. S6A), demonstrating that MAPK activation does not mediate CDK4/6 inhibitor resistance directly, but likely presents a kinome bypass to promote tumor progression. Some studies have demonstrated intrinsic insensitivity to CDK4/6 inhibitors in cancer models with MAPK-activating mutations (41). Whole-exome sequencing revealed no reproducible alterations (data not shown), and none had relevance to MAPK signaling. Taken together, nongenomic MAPK activation, e.g., through upregulation of EGF observed in these models, likely mediates acquired resistance.

MAPK was identified as a major factor in kinome rewiring upon acquired CDK4/6 inhibitor resistance and associated with aggressive tumor phenotypes in vitro and in vivo. Activated MAPK signaling has been previously shown to promote prometastatic signaling and EMT (42). Invasion was attributed to ATF-2–mediated MMP-2 activation across various cancer models, including prostate (43). Moreover, MAPK activation has been associated with advanced stages of solid tumors (e.g. prostate, breast, lung; ref. 43), and therefore, it is not surprising that the PDR models displayed more aggressive phenotypes than the parental cells; thus, it is crucial to identify vulnerabilities of disease that progresses on CDK4/6 inhibitors and to develop alternative treatment regimens. Strikingly, activation of the MAPK pathway sensitized the CDK4/6 inhibitor–resistant cells to MEK inhibition, providing new rationale for testing MEK inhibitors in advanced prostate and other cancers that do not necessarily activate MAPK via classical KRAS mutations or RAF fusions. Moreover, phospho-ERK1/2, indicative of MAPK activity, can be determined via immunolabeling in clinical specimens from clinical trials (NCT02881242). It would be of interest to explore phospho-ERK1/2 as a clinical biomarker for resistance to CDK4/6 inhibitors. Alternatively, CDKs and MAPKs are known to form complex kinase networks that interact and regulate cellular processes involved in cell growth and death (27), and thus targeting these two nodes of the kinome would provide a novel opportunity for therapeutic intervention that could extend CDK4/6 inhibitor response or potentially prevent resistance altogether. These drug combinations are demonstrated to synergize in models of colorectal cancer (33), neuroblastoma (41), and even in MEK inhibitor–resistant melanoma models (32, 44). BRAF inhibitor–resistant melanoma may still respond to CDK4/6 inhibitors combined with mTOR inhibitors (45). One activated RAS melanoma study shows MEK and CDK4/6 inhibitors target complementary downstream networks inducing apoptosis and cell-cycle arrest, causing in vivo tumor regression (46). KRAS-mutant non–small cell lung cancer patients showed improved progression-free survival on palbociclib and an MEK inhibitor (PD-0325901; ref. 47). Considering the cooperative effect of a CDK4/6 and MEK inhibitor observed in the PDR models, these combinations merit prioritization for further preclinical cancer studies with the potential for future clinical trial development, either upfront to delay onset of CDK4/6 inhibitor resistance or when cancer progresses. As trametinib has already entered clinical trials in prostate adenocarcinoma (NCT02881242) and has been FDA-approved for melanoma (44), its clinical application could be considered in the context of CDK4/6 inhibitor resistance.

Clinically, MEK inhibitors are likely to be combined with AR signaling–targeted therapeutics (e.g., enzalutamide). Interestingly, the PDR models herein display a de-enrichment of the androgen response hallmark signature, suggesting a reduced reliance on AR, which could result in castration resistance. Therefore, it is paramount to assess whether acquired CDK4/6 inhibitor resistance can lead to enzalutamide resistance, and whether this can be circumvented with the addition of an MEK inhibitor, to define optimal strategies for clinical implementation.

Although induction of MAPK signaling, associated with aggressive phenotypes, was commonly observed across the PDR models, this does not rule out that other kinases identified in either of the RNAseq and phosphoproteomics datasets may contribute to the development of resistance and cancer progression. Notably, Src motifs were highly enriched in the phosphoproteomic data alone, yet Src kinase is known to impinge on cyclin D/CDK4, and is an actionable target for which clinical agents (i.e., dasatinib) have already been developed (28). Although dasatinib in combination with abiraterone did not improve progression-free survival in metastatic CRPC (48), it merits further preclinical investigation in the context of CDK4/6 inhibitor resistance with the potential for additional therapeutic application.

In summary, the findings presented demonstrate that acquired CDK4/6 inhibitor resistance resulted in a kinome rewiring that not only promotes therapeutic resistance, but also confers aggressive phenotypes associated with tumor cell proliferation and invasion. Mechanistic investigation identified a reliance on MAPK activation and therefore nominates the MAPK signaling pathway as a potential therapeutic target for tumors bypassing CDK4/6 inhibition. These collective observations not only provide insight into the molecular underpinnings of acquired resistance to cell-cycle–targeted therapies, but also provide the basis for the next line of preclinical investigation and a rationale to develop novel combinatorial or sequential therapeutic strategies in the clinic.

No potential conflicts of interest were disclosed.

Conception and design: R. de Leeuw, M.J. Schiewer, M.A. Augello, K.E. Knudsen

Development of methodology: R. de Leeuw, M.A. Augello, Z. Li, J.M. Drake, K.E. Knudsen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. de Leeuw, N. Poudel Neupane, L.J. Brand, M.A. Augello, J.M. Drake, W.K. Kelly

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. de Leeuw, C. McNair, L.J. Brand, M.A. Augello, Z. Li, L.C. Cheng, S.M. Courtney, E.S. Hazard, G. Hardiman, M.H. Hussain, J.M. Drake, W.K. Kelly, K.E. Knudsen

Writing, review, and/or revision of the manuscript: R. de Leeuw, C. McNair, M.J. Schiewer, L.J. Brand, Z. Li, L.C. Cheng, A. Yoshida, S.M. Courtney, E.S. Hazard, G. Hardiman, M.H. Hussain, J.A. Diehl, J.M. Drake, W.K. Kelly, K.E. Knudsen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. de Leeuw, M.J. Schiewer, N. Poudel Neupane, J.A. Diehl

Study supervision: R. de Leeuw, K.E. Knudsen

This study was funded by the NIH R01CA176401 (K.E. Knudsen), R01CA217329 (K.E. Knudsen and W.K. Kelly), Prostate Cancer Foundation Young Investigator Awards (R. de Leeuw: Marjorie Katz Foundation 2016, J.M. Drake: 2015), Prostate Cancer Foundation Challenge Award (W.K. Kelly and K.E. Knudsen), Novartis (J.A. Diehl and K.E. Knudsen), R01CA093237 (J.A. Diehl), National Institute of General Medical Sciences of the NIH T32 GM008339 (L.C. Cheng), Department of Defense Prostate Cancer Research Program W81XWH-15-1-0236 (J.M. Drake), New Jersey Health Foundation grant (J.M. Drake), and Sidney Kimmel Cancer Center Shared Resources (5P30CA056036-17).

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.