Dual-Mechanism ERK1/2 Inhibitors Exploit a Distinct Binding Mode to Block Phosphorylation and Nuclear Accumulation of ERK1/2

This article is featured in Highlights of This Issue, p. 323

The RAS-RAF-MEK1/2-ERK1/2 signaling pathway drives cell survival and proliferation (1). Activation of the RAS GTPases results in the dimerization and activation of RAF kinases (2, 3), which then phosphorylate and activate MEK1/2, which subsequently phosphorylate threonine and tyrosine residues within the T-E-Y motif of the ERK1/2 activation loop. This promotes ERK1/2 activation and release from MEK1/2, enabling ERK1/2 to phosphorylate cytoplasmic substrates and promoting its nuclear translocation to phosphorylate transcription factors to regulate gene expression and drive cell-cycle progression (1, 4). The magnitude and duration of ERK1/2 activity is controlled by intrinsic negative feedback systems including the direct inhibitory phosphorylation of upstream pathway components (5) and the de novo expression of MAP kinase phosphatases (MKP/DUSPs; ref. 6) and the Sprouty proteins (7).

ERK1/2 signaling is frequently deregulated in cancer due to activating mutations in receptor tyrosine kinases (RTK), RAS, or BRAF, resulting in constitutive pathway activation (8) and small-molecule RAF and MEK1/2 inhibitors (RAFi and MEKi) are now approved and used in the clinic (9, 10). Lessons from the use of RAFi and MEKi have prompted interest in targeting the pathway at the level of ERK1/2 for two reasons. First, innate resistance to RAFi or MEKi involves the relief of negative feedback, resulting in the restoration of ERK1/2 activity in the presence of drug, validating the use of ERK1/2 inhibitors (ERKi) in combination to target tumors that are refractory to RAFi or MEKi monotherapy (9, 11). Second, acquired resistance to RAFi or MEKi emerges through mechanisms (KRAS or BRAF amplification, BRAF splice variants, and MEK mutation) that reinstate ERK1/2 signaling, validating the use of ERKi to overcome or forestall acquired resistance to RAFi or MEKi (12–15).

The first selective ERKi are undergoing clinical evaluation and include: BVD-523 (ulixertinib; refs. 16, 17), GDC-0994 (18), LY-3214996 (19), MK-8353 (clinical derivative of SCH772984; refs. 20, 21), ASTX029 (22), LTT462 (23), and KO-947 (24). Furthermore, multiple preclinical compounds have been disclosed (25–31). The majority of these ERKi target ERK1/2 catalytic activity in a reversible, ATP-competitive manner (catalytic ERKi or catERKi). However, dual mechanism ERKi (dmERKi), including SCH772984 and Compound 27 (a potent and selective ERKi developed using fragment-based drug discovery) can additionally antagonize ERK1/2 T-E-Y phosphorylation by MEK1/2, preventing the formation of the active conformation of ERK1/2 (20, 21, 25). These distinct mechanisms of action (MoA) could have important consequences for how cells respond and adapt following ERKi treatment.

MEKi that inhibit both the phosphorylation of MEK1/2 by RAF and MEK1/2 catalytic activity are proposed to delay pathway rebound following feedback relief, causing a more durable inhibition of ERK1/2 and cell proliferation compared with purely catalytic MEKi (32–34). DmERKi act similarly to these “feedback buster” MEKi, so might also delay pathway rebound relative to catERKi. Furthermore, by inhibiting ERK1/2 T-E-Y phosphorylation by MEK1/2 dmERKi should inhibit ERK1/2 nuclear translocation; this could promote more robust suppression of ERK1/2-dependent gene expression relative to catERKi (11). In contrast, catERKi treatment may promote accumulation of p-ERK1/2 (16, 21), which would be expected to drive its nuclear localization. This could facilitate accelerated ERK1/2-dependent gene expression when compound efficacy is lost, resulting in cells recovering more rapidly from treatment with catERKi compared with dmERKi. Accumulation of nuclear p-ERK1/2 following catERKi treatment may also maintain the proposed kinase-independent functions of ERK1/2, including interactions with topoisomerase II (35), PARP1 (36), and DUSP6 (37). Furthermore, binding of nuclear ERK1/2 to lamin A can displace the retinoblastoma (RB) protein, facilitating RB phosphorylation by cyclin-dependent kinases, release of E2F transcription factors, and cell-cycle entry (38). Finally, ERK2 acts as a transcriptional repressor of IFN-responsive genes by directly binding DNA in their promoter regions (39). Most proposed kinase-independent functions of ERK1/2 occur in the nucleus and could persist with nuclear accumulation of ERK1/2 following catERKi treatment. The impact of this on the relative efficacy of catERKis, dmERKis, or MEK inhibitors that prevent ERK1/2 phosphorylation and nuclear import has not been addressed to date.

In this study, we characterized the binding mode and ability to modulate ERK1/2 phosphorylation and nuclear accumulation of five ERKi. We also examined their efficacy, their suppression of pathway rebound, and effects on ERK1/2-dependent gene expression. We demonstrate that dmERKi exploit a distinct binding mode to block ERK1/2 phosphorylation by MEK1/2 and inhibit the nuclear translocation of ERK1/2. Consequently, dmERKi exhibit increased potency and an improved ability to delay pathway rebound in RAS-mutant cell lines, resulting in a more robust suppression of ERK1/2 activity and ERK1/2-dependent gene expression compared with catERKi.

The source and RRID of all reagents and cell lines utilized are detailed in Supplementary Table S1. Cells were grown in DMEM (CO115, DLD-1, HCT116, LoVo, A375), Leibovitz's L-15 (Sw480), McCoy's 5A (HT29), MEMα (RKO), or RPMI1640 (COLO205, SK-MEL-30) media supplemented with 10% (v/v) FBS, penicillin (100 U/mL), streptomycin (100 mg/mL), and 2 mmol/L glutamine. Cells were incubated in a humidified incubator at 37°C and 5% (v/v) CO₂. All cell lines were authenticated by short tandem repeat profiling and confirmed negative for Mycoplasma prior to experiments' commencing. Experiments were performed within 2 months of thawing, except for the generation of drug-resistant cells lines for which the culture time is indicated in the respective figures.

The care and treatment of experimental animals were in accordance with the United Kingdom Coordinating Committee for Cancer Research guidelines (40) and the United Kingdom Animals (Scientific Procedures) Act 1986 (41). Mouse studies were performed with mice allowed access to food and water ad libitum.

COLO205 xenografts were prepared by subcutaneously injecting 5 × 10⁶ cells suspended in serum-free medium mixed 1:1 with Matrigel (BD Biosciences) into the right flank of each male BALB/c nude mouse. A single dose of compound was administered orally to the mice. Tumors were excised and flash-frozen in liquid nitrogen at indicated time points. Tumor lysates were prepared by grinding with a mortar/pestle under liquid nitrogen prior to addition of ice-cold lysis buffer (Meso Scale Discovery), and incubated at 4°C for 30 minutes.

Cell lysis, SDS-PAGE, and Western blotting were performed as described previously (42), with the modification to use fluorescently tagged secondary antibodies to enable band visualization and quantification on a Li-Cor Odyssey Imaging System (LI-COR Biosciences). Membranes were cut to allow probing for multiple molecular weight proteins. Where appropriate, blots were probed with different species of antibodies, using multiple colors to detect the same molecular weight on the same membrane. If necessary multiple independent blots were performed using the lysate from each experiment. Quantification of the protein of interest was normalized to an appropriate loading control. Antibodies are detailed in Supplementary Table S1.

Immunofluorescence staining and high content microscopy (HCM) were performed as described previously (42, 43), with reagents used detailed in Supplementary Table S1. Briefly, cells were seeded in 96-well Imaging Plates (CellCarrier-96, Perkin Elmer) and treated 24 hours later as indicated in the figure legends. For 5-ethynyl-2-deoxyuridine (EdU) incorporation analysis, cells were incubated with 10 μmol/L EdU for the last hour of treatment, except in background control wells where no EdU was added. Cells were then harvested and fixed with 4% formaldehyde/PBS, washed once with PBS, and then permeabilized with 100% methanol for 10 minutes at −20°C. Cells were washed in PBS and EdU click reaction performed following the manufacturer's instructions (Click-iT EdU AlexaFluor 647 HCS Assay, Thermo Fisher Scientific). For detection of p-ERK1/2, ERK1/2, and p-RSK, cells were blocked for 1 hour with 2% BSA/PBS at room temperature, followed by incubation with primary antibody diluted in 2% BSA/PBS at 4°C overnight. For background control wells, 2% BSA/PBS without primary antibody was added. Cells were washed three times with PBS, and then incubated with appropriate secondary antibodies diluted 1:500 in 2% BSA/PBS containing 1 μg/mL of DAPI (Sigma-Aldrich) for 1 hour at room temperature. Cells were washed four times with PBS and stored in 100 μL PBS before imaging. Cells were imaged using an INCell Analyzer 6000 Microscope (GE Healthcare) using a 10 × objective lens, and typically imaging 1,000–15,000 individual cells (in six fields) per well. Image analysis to determine the mean signal intensity per cell or nuclear:cytoplasmic ratio was performed using INCell Analyzer software. To compensate for nonspecific staining by p-RSK (T359) immunofluorescence in some cell lines, 1 μmol/L trametinib was utilized as a negative control as this treatment fully abolished RSK phosphorylation by immunoblot in every cell line tested (Supplementary Fig. S1A and S1B).

Cells were treated as described and 1 hour prior to analysis incubated with 4 μmol/L Sytox green (dead cell stain, Thermo Fisher Scientific) and 1.6 μmol/L Hoechst (live cell stain). Cells were imaged live on a InCELL 6000 High-Content Microscope (GE Healthcare) and the total cell number/condition and the percentage dead cells (Sytox-positive) determined by high-content image analysis using InCELL Analyzer software.

Cells were treated as described and 1 hour prior to harvest incubated with 10 μmol/L EdU (Click-iT EdU Flow Cytometry Kit, Thermo Fisher Scientific). Cells were harvested by trypsinization and fixed with 4% paraformaldehyde/PBS for 10 minutes at room temperature. EdU was detected following the manufacturer's instructions, and cells were resuspended in 1 μg/mL DAPI/PBS (Sigma-Aldrich). DAPI and EdU staining was assessed with a FACS LSRII (BD Biosciences), counting 10 × 10³ cells per sample. Data were analyzed using FlowJo Software (FlowJo), and G₁, S, and G₂–M cell-cycle phases gated.

Cells were treated as described in the figure legends, culture medium was collected, adherent cells trypsinized, and cells and media then recombined. Cells were pelleted by centrifugation (500 × g, 4°C, 5 minutes), resuspended in 1 mL PBS, and then centrifuged again before being washed in 1 mL annexin V binding buffer [10 mmol/L HEPES/NaOH (pH 7.4), 140 mmol/L NaCl, and 2.5 mmol/L CaCl₂]. Cells were then resuspended in 0.2 mL annexin V binding buffer containing 1 μg/mL DAPI (Sigma-Aldrich) and 0.1 μg/mL annexin V-FITC (BioLegend). Annexin V/DAPI staining was assessed using an LSR II flow cytometer (BD Biosciences) and counting 10 × 10³ cells per sample. Data were analyzed using FlowJo (FlowJo) to quantify annexin V- and/or DAPI-positive cells.

Cell proliferation assays were carried out using Alamar Blue (Thermo Fisher Scientific) as described previously (44). Briefly, 5 × 10³ cells were seeded in culture medium into 96-well plates, 24 hours before the drug treatment. Cells were incubated with compound in 0.1% (v/v) DMSO for 96 hours before viability was assessed using Alamar Blue.

A375 cells were seeded at 1.5 × 10⁴ cells per well into 96-well plates and allowed to recover for 16 hours, prior to the addition of compounds for a further 4 hours. Cells were lysed by adding cell lysis buffer (Cell Signaling Technology) and incubating at room temperature for 20 minutes. MSD plates (Meso Scale Discovery) precoated with anti-pRSK antibody (Cell Signaling Technology) were blocked for 1 hour at room temperature, prior to incubation with equivalent amounts of protein lysate for 3 hours at room temperature. After washing, plates were incubated for 1 hour at room temperature with sulfo-tag–conjugated anti-RSK detection antibodies (R&D Systems). Plates were washed and read buffer added before reading on a QuickPlex SQ120 (Meso Scale Discovery).

Total RNA was isolated using QiAshredder and RNeasy Kits (Qiagen) according to the manufacturer's instructions. RNA (200 ng) was reverse-transcribed in 50 μL using TaqMan reverse transcription reagents (Thermo Fisher Scientific). Thermal cycle: 25°C for 5 minutes, 48°C for 30 minutes, and 95°C for 5 minutes. The cDNA sample was diluted 1:3 in RNase-free water. cDNA (4 ng) was analyzed by qRT-PCR using TaqMan prevalidated probes (Supplementary Table S1) and Universal Mastermix (Thermo Fisher Scientific). A Bio-Rad CFX96 system was used with the following cycling conditions: 50°C for 2 minutes, 95°C for 10 minutes, 95°C for 3 seconds, and 60°C for 30 seconds, with the final two steps repeated 40 times. Fluorescence output was considered directly proportional to the input cDNA concentration and was normalized against β-actin or 18S expression.

For microarray data analysis, RNA samples were profiled from A375, COLO205, and HCT116 cells treated with SCH772984, GDC-0994, or DMSO for 24 hours (three independent biological replicates per treatment and four DMSO replicates). The whole-genome expression profiling was carried out using Illumina HumanHT-12v4 expression beadchip platform. The raw un-normalized data were exported from GenomeStudio and analyzed using the limma R package (45). The probe intensities were background corrected using negative control probes and quantile normalized using negative and positive control probes using the limma neqc function. After normalization, probes were then filtered according to their annotation quality and selected only those with an inter quartile range > 0.5. The function lmFit was used to fit linear models on expression values of genes. The function eBayes was used to calculate differential expression between untreated and treated samples using moderated t statistics. P values were corrected for multiple testing using the Benjamini and Hochberg method (46). Following this correction, genes with more than 2-fold expression change and adjusted P < 0.01 were considered significant. Enrichment analysis of gene ontologies of differentially expressed genes was performed using the clusterProfiler R package (47). Microarray data are available in the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-7959.

ERK2 kinase activity was determined using a time-resolved fluorescence (TRF) activity assay. ERK2 (0.25 nmol/L) was incubated with the substrates ATF2-GFP (0.4 μmol/L) and ATP (20 μmol/L) in 50 mmol/L Tris pH 7.5, 10 mmol/L MgCl₂, 1 mmol/L EGTA, 0.01% Triton X100, 1 mmol/L DTT, and 2.5% dimethyl sulfoxide (DMSO), with shaking at room temperature for 30 minutes. Reactions were stopped by the addition of stop and detection mix, containing 25 mmol/L EDTA and 2 nmol/L Tb-pATF2 antibody in TR-FRET dilution buffer (Life Technologies), and the plate was incubated with shaking at room temperature for 1 hour. Upon excitation at 340 nm, fluorescence was measured at 520 nm and 495 nm using a Pherastar plate reader (BMG Labtech).

Full length human ERK2 (hERK2) was cloned into pET30a with a noncleavable MAHHHHHH N-terminal tag. hERK2 was expressed in E. coli BL21(DE3) and nonphosphorylated hERK2 (confirmed by LC-MS) was purified using sequential Ni-HiTRAP, desalt, Resource-Q, and S75 26/60 column steps. hERK2 was crystallized under conditions adapted from Aronov and colleagues (30) and crystals were soaked in a solution equivalent to the crystallization solution but also containing 0.1–100 mmol/L ligand, 10 mmol/L DTT, and 10% DMSO. Crystals were cryo-protected using crystallization solution containing 35% 2KMPEG final. X-ray diffraction data were collected using both in-house and synchrotron X-ray sources. X-ray crystal structures are available in the wwPDB (www.wwpdb.org) under the indicated PBD ID codes.

We studied two dmERKi, Compound 27 and SCH772984 (21, 25), two catERKi, GDC-0994 and BVD-523 (16, 18), and LY-3214996 (ref. 19; Fig. 1A). Crystal structures of these ERKi bound to ERK2 revealed distinct binding modes between the dmERKi and catERKi (Fig. 1B). GDC-0994 and BVD-523 behaved as typical ATP-competitive inhibitors, binding to the active form of ERK2 and occupying the ATP-binding pocket (pdb: 4nif). GDC-0994 exploited a 2-amino-pyrimidine scaffold to bind to the ERK2 pocket. Its donor-acceptor motif formed a double H-Bond pattern with the “hinge” region residue, Met108. The molecule ended with a terminal 4-chloro-3-fluorophenyl ring sitting under the P-Loop. The Tyr36 phenol ring here was in an “out” conformation and formed a pi−pi stacking interaction with Tyr64 on the C-α helix (18, 25). BVD-523 exhibited a very similar binding mode, with the Tyr36 phenol ring in an “out” conformation. LY-3214996 bound to ERK2 in a similar manner to BVD-523 and GDC-0994, prompting us to predict that it would act as a catERKi. In contrast, while Compound 27 and SCH772984 also occupied the ERK2 ATP binding site, they imposed a conformational change upon the Tyr36 side chain such that it folded beneath the P-loop (Tyr36 “in”), in the pocket occupied by the terminal rings of GDC-0994 and BVD-523. Thus, catERKi and dmERKi have distinct binding modes, consistent with a report that occupancy of the second pocket, displacing Tyr36, correlated with modulation of p-ERK1/2 levels (25).

We monitored the effects of these ERKi on ERK1/2 T-E-Y phosphorylation (p-ERK1/2) and ERK1/2 catalytic activity (phosphorylation of RSK, an ERK1/2 substrate) following 2 hours treatment. In KRAS^mut HCT116 (Fig. 1C; Supplementary Fig. S2B) and Capan-1 (Fig. 1C; Supplementary Fig. S3B) cells the catERKi BVD-523 and GDC-0994 increased p-ERK1/2 levels, reflecting loss of ERK1/2-mediated negative feedback. LY-3214996 treatment also promoted p-ERK1/2 accumulation, validating our prediction that it acts as a catERKi. In contrast, the dmERKi SCH772984 and Compound 27 induced a dose-dependent inhibition of ERK1/2 phosphorylation, comparable with the MEKi (Fig. 1C; Supplementary Fig. S2B, S3B). In contrast to KRAS^mut cells, BRAF^V600E-mutant cells lines display little ERK1/2 rebound following ERK1/2 pathway inhibition, as BRAF^V600E activity is insensitive to ERK1/2-mediated negative feedback (34, 48). Consequently in BRAF^V600E COLO205 (Fig. 1C; Supplementary Fig. S2A) and A375 (Fig. 1C; Supplementary Fig. S3A) cells catERKi did not drive accumulation of p-ERK1/2, instead inhibiting ERK1/2 phosphorylation to varying degrees, although not to the extent of dmERKi. Thus, ERKi exhibit a spectrum of abilities to antagonize ERK1/2 phosphorylation. The mechanistic differences between dmERKi and catERKi were retained in vivo, where Compound 27, but not GDC-0994, inhibited ERK1/2 phosphorylation in COLO205 xenografts (Fig. 1D).

Utilizing p-RSK as a biomarker of ERK1/2 activity revealed that dmERKi exhibited enhanced potency compared with catERKi across both KRAS- and BRAF^V600E-mutant cell lines (Fig. 1C; Supplementary Figs. S2 and S3). This increased potency could be a property of the binding mode of dmERKi, or could reflect their ability to block the active conformation of ERK1/2, in addition to inhibiting catalysis.

We next assessed the effects of ERKi on p-ERK1/2 levels and pathway rebound. To differentiate MoA from compound potency, compound concentrations were normalized for potency, by selecting the lowest concentration that inhibited >90% of RSK phosphorylation (Fig. 1C; Supplementary Figs. S2 and S3; Supplementary Table S2). In HCT116 (Fig. 2A and B; Supplementary Fig. S4A) and Capan-1 (Supplementary Fig. S4B and S4C) cells, catERKi promoted p-ERK1/2 accumulation over time, including a 5- to 6-fold increase with BVD-523 and LY-3214996. In contrast, dmERKi caused a rapid, strong inhibition of ERK1/2 phosphorylation, which then recovered from 4–8 hours onwards. Despite this restoration of p-ERK1/2, p-RSK levels were still robustly suppressed by dmERKi (Fig. 2B; Supplementary Fig. S4C). Thus, dmERKi were more effective at inhibiting ERK1/2 catalytic activity than ERK1/2 phosphorylation.

Feedback relief after ERK1/2 inhibition enables pathway rebound over time. Indeed, the “feedback buster” MEKi trametinib delayed and reduced pathway rebound (p-RSK levels) compared with selumetinib, which does not block MEK1/2 phosphorylation by RAF (Fig. 2B; Supplementary Fig. S4C; refs. 32–34). Interestingly, both dmERKi and catERKi prevented rebound in a comparable manner with trametinib, however there were no distinct trends by MoA (Fig. 2B; Supplementary Fig. S4C). BVD-523 was the only ERKi that totally prevented pathway rebound at the normalized concentration utilized (Fig. 2B; Supplementary Fig. S4C).

We also examined all compounds across multiple concentrations using dose/time matrices for each compound in HCT116 and Capan-1 cells (Fig. 2C and D; Supplementary Fig. S5). DmERKi displayed superior potency relative to catERKi across all treatment times in both cell lines (Fig. 2C; Supplementary Fig. S5A). Compound 27, SCH772984, GDC-0994, LY-3214996, and trametinib displayed similar pathway rebound following the initial loss of p-RSK, whereas selumetinib or PD184352 displayed more dramatic pathway rebound. BVD-523 was unique in fully abolishing pathway rebound (Fig. 2C; Supplementary Fig. S5A). Consistent with previous data, catERKi induced a strong accumulation of p-ERK1/2, while dmERKi caused an initial reduction in p-ERK1/2 followed by gradual recovery over time (Fig. 2D; Supplementary Fig. S5B). Together these data demonstrate that when utilized at comparable concentrations all ERKi prevent ERK1/2 pathway rebound as effectively as trametinib; however, dmERKi display increased potency relative to catERKi (Figs. 1C and 2C), most likely reflecting their novel binding mode. Interestingly, BVD-523–induced p-ERK1/2 accumulation peaked between 2–8 hours before declining (Fig. 2D; Supplementary Fig. S5B); this decline correlated with a progressive loss of total ERK1/2 (>75% reduction; Supplementary Fig. S5C and S5D), which could explain the apparent durability of BVD-523–mediated ERK1/2 pathway inhibition. The cause of this loss of total ERK1/2 is currently under investigation.

T-E-Y phosphorylation induces ERK1/2 nuclear translocation (4), therefore catERKi which induce the accumulation of p-ERK1/2 (Figs. 1C, 2B–D) should promote the nuclear accumulation of inhibited p-ERK1/2; we tested this by immunofluorescence and HCM. In HCT116 cells all catERKi promoted a striking nuclear accumulation of p-ERK1/2, and more subtle nuclear redistribution of total ERK1/2 (Fig. 3A and B). In contrast, while dmERKi or MEKi abolished p-ERK1/2 levels, only MEKi treatment induced a strong cytoplasmic redistribution of total ERK1/2 (Fig. 3A and B), suggesting that preventing both MEK1/2 activity and ERK1/2 phosphorylation is essential to prevent the release of ERK1/2 from MEK1/2 and retain ERK1/2 in the cytoplasm. Consistent results were observed in COLO205, A375, and Capan-1 cells, with changes in p-ERK1/2 localization varying in proportion with the level of p-ERK1/2 accumulation (Supplementary Fig. S6A and S6B). To confirm that the nuclear p-ERK1/2 that accumulated following catERKi treatment was inhibited, we quantified levels of the ERK1/2-dependent transcripts DUSP6 and SPRY2; both were suppressed to comparable levels by dmERKi, catERKi, or MEKi demonstrating that catERKi-driven nuclear accumulation of p-ERK1/2 was not able to promote ERK1/2-dependent gene expression (Fig. 3C).

If the nuclear accumulation of ERK1/2 following catERKi treatment is sustained in the absence of drug, this could alter the kinetics of pathway reactivation, potentially accelerating ERK1/2-dependent gene expression. To address this, HCT116 cells were treated with ERKi for 24 hours, and then switched to drug-free media for a 4-hour time course. Following wash-off of GDC-0994 or LY-3214996, p-ERK1/2 accumulation and nuclear localization rapidly decreased to basal levels within 2 hours (Fig. 3D and E). This correlated with a rapid reactivation of ERK1/2, inducing peak p-RSK 1 hour following drug withdrawal (Fig. 3F and G; Supplementary Fig. S7A). In contrast, p-ERK1/2 nuclear accumulation was retained following BVD-523 wash-off (Fig. 3E); however, this correlated with a very slow recovery in p-RSK levels (Fig. 3G; Supplementary Fig. S7B and S7C). Withdrawal of dmERKi or MEKi had varying effects on pathway reactivation, with selumetinib displaying a rapid increase in p-RSK levels, comparable with GDC-0994 and LY-3214996, whereas the withdrawal of Compound 27, SCH772984, or trametinib elicited a more delayed response (Fig. 3F and G; Supplementary Fig. S7C). Wash-off of selumetinib induced a more rapid reexpression of DUSP5 than GDC-0994 or LY-3214996 (Fig. 3F and G; Supplementary Fig. S7A). This, coupled with the rapid loss of nuclear p-ERK1/2 upon GDC-0994 or LY-3214996 withdrawal, indicates the catERKi are unlikely to facilitate accelerated ERK1/2-dependent gene expression upon drug withdrawal, due to feedback controls rapidly restoring homeostatic ERK1/2 phosphorylation and cellular localization. With the exception of BVD-523, the kinetics of pathway reactivation following compound withdrawal correlated with compound potency, not compound target or MoA. This could be due to more potent compounds having a slower off-rate, thereby remaining bound to ERK1/2 or MEK1/2 for longer following the withdrawal of drug-containing media (49).

We next investigated whether, by preventing ERK1/2 from entering the nucleus (Fig. 3A and B; Supplementary Fig. S6), dmERKi exerted a more robust effect on ERK1/2-dependent gene expression. We treated HCT116, COLO205, and A375 cells with concentrations of SCH772984 or GDC-0994 that induced a comparable growth arrest (Supplementary Table S3) and performed gene expression profiling using microarrays. SCH772984 induced a more comprehensive inhibition of 10 well-established ERK1/2 target genes across all cell lines (Fig. 4A). We also compared global gene expression changes (Supplementary Fig. S8A); while both compounds altered the expression of a common set of genes, SCH772984 selectively altered the expression of a significant number of further genes (Fig. 4B). Gene ontology (GO) analysis of common or SCH772984-specific–downregulated gene signatures revealed that many of the most significantly downregulated processes were involved in DNA replication or cell-cycle progression (Fig. 4C). Many GO terms identified for SCH772984-specific–downregulated genes were the same as those identified for the common downregulated genes, indicating that SCH772984 was inhibiting the same processes as GDC-0994 but in a more comprehensive manner (Fig. 4C).

ERK2 is proposed to act as a kinase-independent transcriptional repressor of IFN signaling by directly binding DNA (39). The “response to type I IFN” GO term was far more significantly upregulated in SCH772984-treated HCT116 cells (P = 1.08 × 10⁻⁸) compared with either GDC-0994 or DMSO (Fig. 4D); within this signature were multiple genes that ERK2 has been shown to bind directly to and repress in a kinase-independent manner (Fig. 4E; ref. 39). Treatment of HCT116 cells with dmERKi or MEKi consistently caused a greater upregulation of these ERK2-repressed genes relative to catERKi, at concentrations that induced comparable downregulation of the established ERK1/2 target genes DUSP6 and SPRY2 (Fig. 4F). Furthermore, the upregulation of these genes did not correlate with the ability of ERKi to repress ERK1/2 target genes (Supplementary Fig. S8B), indicating that this property of dmERKi was likely due to their MoA, and ability to retain ERK1/2 in the cytoplasm, not their enhanced potency relative to catERKi. The ability of some ERKi, notably BVD-523, to reduce total ERK1/2 (Fig. 2B) could also help to facilitate the derepression of these ERK2 bound genes. Together these data indicate that SCH772984, a dmERKi, differentially regulates a larger pool of genes and processes than GDC-0994, a catERKi, even when doses are normalized for pathway inhibition. This likely reflects the ability of SCH772984 to prevent MEK1/2-catalysed phosphorylation-dependent conformational changes and nuclear-localization, compared with simple inhibition of catalytic activity of ERK1/2 by GDC-0994. However, the preferential upregulation of IFN-induced genes by SCH772984 and Compound 27 suggests that only dmERKi have the potential to inhibit proposed nuclear kinase-independent functions of ERK1/2 (35, 36, 38, 39).

To determine whether the increased potency (loss of p-RSK) of dmERKi relative to catERKi (Figs. 1C and 2C; Supplementary Fig. S5A) translated into increased biological efficacy, we assessed proliferation of a panel of eight colorectal cancer cell lines (Fig. 5A and B; Supplementary Figs. S9 and S10) using high-content imaging to detect EdU incorporation and p-RSK or p-ERK1/2 in the same cell population following a 24-hour compound treatment. DmERKi demonstrated superior antiproliferative potency across all cell lines tested, irrespective of BRAF- or KRAS-mutant status (Fig. 5B). All ERKi displayed reduced efficacy in KRAS^mut cell lines relative to BRAF^mut (Fig. 5C), reflecting innate resistance by additional KRAS effector signaling pathways. Relating p-RSK and EdU incorporation revealed that the antiproliferative effects of both catERKi and dmERKi correlated with their ability to inhibit ERK1/2 catalytic activity (Fig. 5D). In contrast, ERK1/2 phosphorylation or subcellular localization did not correlate with compound efficacy (Fig. 5A; Supplementary Figs. S9 and S10). Consistent with the effects of feedback relief and pathway rebound, in almost all cases the 72-hour dose-response curves were right-shifted relative to the 24 hour, indicating that a greater compound concentration was required to achieve comparable inhibition (Supplementary Figs. S11A, 5A, S9, and S10). Regardless, the antiproliferative effects of all ERKi or MEKi correlated with their ability to inhibit ERK1/2 catalytic activity, and there were no major differences between dmERKi and catERKi (Supplementary Fig. S11B). Thus, proposed kinase-independent functions of ERK1/2 appeared not to play crucial roles in regulating proliferation in response to ERKi treatment, although they could mediate other cellular phenotypes.

MEK1/2 inhibition promotes a cytostatic response in most ERK1/2 pathway-mutant cell lines, due to the loss of ERK1/2-dependent transcription of D-type cyclins promoting a G₁-phase cell-cycle arrest (9, 13). To investigate whether ERKi induces a comparable response, we treated eight colorectal cancer cell lines with ERKi or MEKi for 72 hours and determined the total cell number (Supplementary Fig. S12A) and the proportion of dead cells (Supplementary Fig. S12B). Both dmERKi and catERKi induced a predominantly cytostatic response in the majority of cell lines (Supplementary Fig. S12A and S12B). Where cell death did occur, it was induced to a similar magnitude by all ERKi (Supplementary Fig. S12B), and Annexin V staining revealed this to be apoptotic cell death (Supplementary Fig. S12C). To further characterize the mechanism of ERKi-induced proliferative arrest, we examined the cell-cycle profiles of COLO205 and HCT116 cells following ERKi treatment. While dmERKi promoted a G₁-phase arrest in both cell lines, the catERKi BVD-523 and LY-3214996 promoted a G₂–M-phase arrest in HCT116 cells, but not COLO205 (Supplementary Fig. S13A). This G₂–M-phase arrest correlated with the strong nuclear accumulation of p-ERK1/2 seen in catERKi-treated HCT116, but not COLO205 cells (Supplementary Fig. S13B). ERK1/2 has been suggested to drive G₁-phase progression through a kinase-independent manner, via ERK1/2 displacing Rb from lamin A, to facilitate CDK-dependent Rb phosphorylation (38). Therefore, we hypothesized that nuclear p-ERK1/2 induced by catERKi might act in a kinase-independent manner to promote G₁-phase progression, thus enabling a G₂–M-phase checkpoint arrest due to the inhibition of ERK1/2 catalytic activity. Indeed, BVD-523- and LY-3214996–treated HCT116 cells retained Rb phosphorylation even though CyclinD1 was decreased to levels comparable with that induced by dmERKi or MEKi, whereas Rb phosphorylation was lost following treatment with all ERKi in COLO205 cells (Supplementary Fig. S13C). BVD-523- and LY-3214996–treated HCT116 cells also retained expression of the G₂-phase markers Cyclin A and Cyclin B (Supplementary Fig. S13C), but were p-Histone H3 negative (Supplementary Fig. S13D), consistent with arrest at the G₂–M-phase checkpoint. However, this phenotype was not consistent across a range of other cell lines (Supplementary Fig. S14A–S14C), despite the strong induction of nuclear p-ERK1/2 in some cell lines (Supplementary Fig. S14B).

To model the ability of cells to acquire resistance to ERKi, we treated HCT116 and COLO205 cells with doses of ERKi normalized to induce comparable pathway inhibition and short-term growth arrest, then monitored their growth (Fig. 6A; Supplementary Fig. S15A) and ability to proliferate in drug (Fig. 6B; Supplementary Fig. S15B) over time. In HCT116 cells, resistance emerged slightly more slowly with the five ERKi compared with the MEKi (PD184352) but there was no clear trend in terms of ERKi MoA (Fig. 6A). In all cases HCT116 or COLO205 cells adapted to ERKi treatment by reinstating ERK1/2 signaling and this was associated with an increase in KRAS expression in HCT116 (Fig. 6C and D), or BRAF expression in COLO205 (Supplementary Fig. S15C and S15D). This is consistent with our previous demonstration that HCT116 and COLO205 cells adapt to MEKi by amplifying their driving oncogene (13). Consistent with these mechanistic similarities, all ERKi- or MEKi-resistant cell lines displayed cross-resistance to other ERKi or MEKi (Supplementary Fig. S16A and S16B). However, surprisingly ERKi-resistant HCT116 cells displayed a greater degree of cross-resistance to other ERKi than they did to the MEKi PD184352 (Supplementary Fig. S16A).

In addition to inhibiting catalysis, ERKi entering the clinic possess a range of abilities to modulate the phosphorylation of ERK1/2 by MEK1/2 (16, 20, 21, 25, 28). The biological consequences of these different MoA are largely unknown and prompted this study. DmERKi possess a unique binding mode that mediates a conformational change in the Tyr36 side chain of the EKR1/2 P-loop (Fig. 1B; refs. 25, 49); such inhibitors robustly suppress ERK1/2 T-E-Y phosphorylation in both BRAF- and RAS-mutant cell lines (Fig. 1C). Furthermore, dmERKi consistently exhibited enhanced potency (Fig. 1C) and more durable ERK1/2 pathway suppression (Fig. 2C; Supplementary Fig. S5A). However, in contrast to MEKi (32–34), the ability of dmERKi to inhibit ERK1/2 phosphorylation did not appear to delay or reduce pathway rebound relative to catERKi when used at comparable concentrations (Fig. 2). Instead, both catERKi and dmERKi displayed similar rebound profiles to the “feedback buster” MEKi trametinib (Fig. 2). This ability to induce durable pathway inhibition could be a characteristic of targeting the terminal kinase in the ERK1/2 pathway, and is consistent with reports that greater levels of BRAF amplification are required to generate resistance to ERKi than to MEKi or BRAFi (50).

By blocking T-E-Y phosphorylation, dmERKi did not elicit the striking nuclear accumulation of p-ERK1/2 that was observed following catERKi treatment (Fig. 3A and B; Supplementary Fig. S6). Changes in p-ERK1/2 localization were coupled with more subtle changes in the localization of total ERK1/2 (Fig. 3A and B; Supplementary Fig. S7), reflecting the small fraction of ERK1/2 known to be phosphorylated at any one time (42) and the large proportion bound in scaffold complexes. This ability of dmERKi to inhibit nuclear localization of ERK1/2 could explain their ability to facilitate more robust suppression of ERK1/2-dependent gene expression than catERKi (Fig. 4) and therefore contribute to the increased antiproliferative efficacy observed with dmERKi (Fig. 5A and B). In contrast, the nuclear accumulation of p-ERK1/2 driven by catERKi would increase the likelihood that any ERK1/2 activity that escapes inhibition would be able to target transcription factors and restore prosurvival and proliferative transcription programs. In addition, catERKi-driven nuclear p-ERK1/2 has the potential to sustain nuclear noncatalytic functions of ERK1/2, such as its role as a transcriptional repressor of IFN responsive genes (39), that appear to be suppressed by dmERKi treatment (Fig. 4E and F). The nuclear accumulation of p-ERK1/2 was rapidly lost following catERKi withdrawal, indicating that the nuclear p-ERK1/2 was inhibitor bound and that this localization does not influence the kinetics of pathway reactivation as compound activity is lost (Fig. 3D–G; Supplementary Fig. S7).

While both dmERKi were consistently more potent we found that the antiproliferative effects of all ERKi ultimately correlated with their ability to inhibit ERK1/2 catalytic activity rather than their distinct MoA (Fig. 5A–D; Supplementary Figs. S9–S11). At concentrations of each ERKi that inhibited the same proportion of p-RSK (a measure of ERK1/2 inhibition) all compounds displayed a similar ability to inhibit proliferation, despite clear differences in p-ERK1/2 levels and localization. Therefore, the differences in regulation of ERK1/2 phosphorylation and localization associated with each ERKi MoA did not influence their antiproliferative activity, although we cannot rule out effects on other cellular phenotypes such as cell motility or survival.

In summary, the dmERKi tested exhibited a distinct binding mode, increased potency, and more durable pathway inhibition. dmERKi also prevented ERK1/2 nuclear localization, thereby phenocopying strong “feedback buster” MEKi such as trametinib. As a consequence, dmERKi exhibited enhanced suppression of ERK1/2-dependent gene expression, both for selected ERK1/2 target genes and in global transcriptomic analysis. Nuclear accumulation of p-ERK1/2 driven by catERKi has the potential to sustain noncatalytic functions of ERK1/2 (the majority of which occur in the nucleus and therefore could be regulated by nuclear translocation). While this seems less important for antiproliferative efficacy, it may contribute to other cancer hallmarks. Together these results suggest that a dual-mechanism profile is likely to be advantageous for ERKi development.

J.M. Munck is an associate director (paid consultant) for Astex Pharmaceuticals. A. Courtin is a senior research associate (paid consultant) for Astex Pharmaceuticals (former employee). B. Graham is a biologist (paid consultant) for Astex Pharmaceuticals. M. O'Reilly is a senior director (paid consultant) for Astex Pharmaceuticals. S.J. Cook reports receiving a commercial research grant from and has unpaid consultant/advisory board relationship with Astex Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.M. Kidger, A. Courtin, S.J. Cook

Development of methodology: A.M. Kidger, B. Graham

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.M. Kidger, J.M. Munck, K. Balmanno, E. Minihane, A. Courtin, B. Graham, M. O'Reilly, R. Odle

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.M. Kidger, H.K. Saini, K. Balmanno, A. Courtin, M. O'Reilly, R. Odle

Writing, review, and/or revision of the manuscript: A.M. Kidger, J.M. Munck, H.K. Saini, S.J. Cook

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Kidger

Other (secured funding for the study): S.J. Cook

We would like to thank past and present members of the Cook laboratory and Babraham Institute Science Services for their support throughout this study, especially Simon Walker, Hanneke Okkenhaug (Imaging), and Matthew Sale. This study was funded by a grant from Astex Pharmaceuticals awarded through the Milner Therapeutics Consortium (to A.M. Kidger and S.J. Cook) and Institute Strategic Programme Grants BB/J004456/1 and BB/P013384/1 from BBSRC (to S.J. Cook, K. Balmanno, and R. Odle).

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.