Transcriptional Antagonism by CDK8 Inhibition Improves Therapeutic Efficacy of MEK Inhibitors

Abstract

Aberrant RAS/MAPK signaling is a common driver of oncogenesis that can be therapeutically targeted with clinically approved MEK inhibitors. Disease progression on single agent MEK inhibitors is common, however, and combination therapies are typically required to achieve significant clinical benefit in advanced cancers. Here we focused on identifying MEK inhibitor-based combination therapies in neuroblastoma with mutations that activate the RAS/MAPK signaling pathway, which are rare at diagnosis but frequent in relapsed neuroblastoma. A genome-scale CRISPR-Cas9 functional genomic screen was deployed to identify genes that when knocked out sensitize RAS-mutant neuroblastoma to MEK inhibition. Loss of either CCNC or CDK8, two members of the mediator kinase module, sensitized neuroblastoma to MEK inhibition. Furthermore, small molecule kinase inhibitors of CDK8 improved response to MEK inhibitors in vitro and in vivo in RAS-mutant neuroblastoma and other adult solid tumors. Transcriptional profiling revealed that loss of CDK8 or CCNC antagonized the transcriptional signature induced by MEK inhibition. When combined, loss of CDK8 or CCNC prevented the compensatory upregulation of pro-growth gene expression induced by MEK inhibition. These findings propose a new therapeutic combination for RAS-mutant neuroblastoma and may have clinical relevance for other RAS-driven malignancies.

Introduction

The RAS/MAPK signaling pathway is a critical regulator of cell growth, and uncontrolled RAS signaling is a common occurrence in cancer. The RAS pathway can be aberrantly activated in a variety of ways, including activating mutations in any of the three RAS isoforms (NRAS, HRAS, and KRAS); loss of function mutations and deletions of the RAS-GTPase activating protein (RAS-GAP) NF1, a negative regulator of RAS; or activating mutations in downstream pathway members such as BRAF. RAS itself has proven difficult to drug directly, although the recent clinical approval of sotorasib has allowed direct targeting of RAS for the first time in the setting of KRAS^G12C mutations (1–4). Hyperactivation of RAS/MAPK signaling, irrespective of the specific alteration, can be therapeutically targeted through inhibition of MEK, a downstream effector of RAS signaling. Multiple MEK inhibitors are now FDA-approved and are effective in combination with RAF inhibitors in BRAF mutant cancers, including melanoma and non-small cell lung cancer (5–9). However, use of MEK inhibitors as single agents has largely yielded cytostatic responses of short duration in advanced cancers (10), and therefore MEK inhibitors have not yet impacted the clinical outcome of many RAS-driven cancers. Here, we focus on identifying MEK inhibitor based combination therapies in RAS-mutant neuroblastoma, where single agent MEK inhibitor treatment is insufficient (11,12).

Neuroblastoma arises in the developing peripheral nervous system and is the most common extracranial solid tumor in children (13). Children with low- and intermediate-risk neuroblastoma frequently benefit from therapy, but more than half of children have metastatic disease at the time of diagnosis (12,13). Despite aggressive treatment of high-risk disease with surgery, radiation, cytotoxic chemotherapy, immunotherapy, autologous stem cell transplantation, and differentiation therapy, the majority of these children experience relapse (14). Moreover, there are limited treatment options for children who relapse, and these tumors are typically incurable (15). As such, neuroblastoma causes a disproportionately high number of cancer-related deaths in children (16).

Neuroblastomas have few oncogenic mutations at initial presentation relative to adult cancers (17). However, recent sequencing efforts have uncovered an enrichment for oncogenic mutations in the setting of relapsed disease (11,18–20). Some of these mutations are therapeutically tractable in other cancers (16,21). The most frequent of these alterations is in ALK, with an estimated 26% of relapsed neuroblastoma tumors harboring predicted driver mutations, and efforts investigating the use of ALK inhibitors in the treatment of neuroblastoma are underway (22–25). In addition to ALK mutations, a variety of other alterations leading to hyperactivation of the RAS/MAPK pathway are also enriched at relapse in neuroblastoma (11). Hypothesis-driven approaches have identified several potential combinations that might improve response to MEK inhibitors in neuroblastoma with hyper-activated RAS signaling, including YAP inhibitors and SHP2 inhibitors (26,27). To complement and expand upon these efforts, we decided to take an unbiased functional genomic approach to uncover sensitizers to MEK inhibition in RAS-mutant neuroblastoma and potentially other RAS-driven cancers.

Materials and Methods

Cell Lines and Reagents

SK-N-DZ (RRID:CVCL_1701), SK-N-FI (RRID:CVCL_1702), SK-N-AS (RRID:CVCL_1700), CFPAC-1 (RRID:CVCL_1119), and NCI-H23 (RRID:CVCL_1547) were obtained from the American Type Culture Collection (ATCC). NGP (RRID:CVCL_UF75), GI-M-EN (RRID:CVCL_1232), and KELLY (RRID:CVCL_2092) were obtained from DSMZ. KP-N-SI9s (JCRB Cat# IFO50433, RRID:CVCL_1340) was obtained from the Japanese Collection of Research Bioresources (JCRB) Cell Bank. NB-Ebc1 (RRID:CVCL_E218) was obtained from the Childhood Cancer Cell Line Repository through the Children’s Oncology Group. Cell line identities were confirmed by short tandem repeat (STR) profiling (LabCorp). Cell lines were confirmed negative for mycoplasma and tested regularly using Lonza MycoAlert, with a most recent test date of February 10, 2022. Cell lines were used within 10–15 passages of thawing. SK-N-AS, SK-N-DZ, and SK-N-FI were cultured in DMEM supplemented with 10% FBS and non-essential amino acids. CFPAC-1 and NGP were grown in DMEM supplemented with 10% FBS. NCI-H23, KP-N-SI9s, and KELLY were cultured in RPMI supplemented with 10% FBS. Nb-Ebc1 was cultured in IMDM supplemented with 10% FBS and Insulin-Transferrin-Selenium. Both male and female cell lines were used in consideration of sex as a biological variable. BI-1347 and BI-1374 were generously provided by Boehringer Ingelheim through OpnMe.com. JH-XII-136 was generously provided by Dr. John Hatcher and Dr. Nathanael Gray. Trametinib and selumetinib were purchased from Selleck Chemicals for in vitro use. Selumetinib for in vivo use was purchased from LC Labs.

CRISPR Cas-9 Drug Modifier Screen and Analysis

SK-N-AS constitutively expressing Streptococcus pyogenes Cas9 has been previously generated and described (28). Cells were infected with the Avana-4 sgRNA library (29,30) at an MOI of ~30%. Cells were selected and then divided into two treatment arms, DMSO or 3 nM Trametinib, with three replicates each. At 7 and 15 days after treatment, cell pellets with at least 500X representation were collected. Genomic DNA was extracted and the sgRNA barcode was PCR amplified and this region was submitted for standard Illumina sequencing as previously described (28). sgRNA abundance was compared to the plasmid pool to determine the normalized log₂ fold change in guide abundance. Averages were calculated using a hypergeometric mean. One replicate in the trametinib arm at the 7-day time point failed to sequence and was excluded from the analysis. All other replicates succeeded and were included in the analysis. Gene-level perturbation scores and statistical significances were computed based on the gene-ranking algorithm for perturbation screens STARS (29), available at the Genetic Perturbation Platform (GPP) web portal (https://portals.broadinstitute.org/gpp/public/). STARS takes the list of perturbation barcodes and the associated log₂ normalized scores as input and computes a gene-level score using the probability mass function of a negative binomial distribution (with replacement) for the top barcodes that rank above a user defined threshold (top 10% of perturbations). A null distribution estimated based on permutation testing on the list of all barcode perturbations in the experiment is used for the computation of the adjusted p-values. Significant gene hits were identified based on the cut-offs abs(STARS score) > 2.5 and adjusted p-value < 0.01.

Drug Response Curves

Cells were seeded in 384-well tissue culture plates (Corning #3570) at a density of 200–300 cells/50μL/well. Compounds were added alone or in combination to the wells, with four replicates for each dose or dose pairing with a HP D300 Digital Dispenser. For single compound dose curves, we performed six replicates of single agent concentration. DMSO was normalized to the highest DMSO volume added for the entire plate, not to exceed 0.02%/media/well. Unless otherwise indicated, cells were analyzed for cell viability after 7 days of treatment using the CellTiter-Glo luminescent assay (PerkinElmer) following the manufacturer’s instructions.

CRISPR knockouts

For all low throughput CRISPR knockout experiments, sgRNAs were cloned into the pLenti-CRISPRv2 backbone, a gift from Feng Zhang (Addgene plasmid # 52961; http://n2t.net/addgene:52961; RRID:Addgene_52961) (31) Cells were infected with lenti-virus, selected with puromycin, and used for the indicated experiment four days after infection. All experiments were performed with polyclonal cell populations; no single cell cloning was performed after infections. For CCNC knockouts sgCCNC-1 (ACATGTTGCTTCCCCTTGCA) and sgCCNC-3 (GCTCCAGTATGTGCAGGACA) were used. For CDK8 knockouts, sgCDK8–3 (GAGGACCTGTTTGAATACGA) and sgCDK8–4 (AGTGACTTCACCATTCCCCG) were used. Negative control sgRNAs were sgLACZ (AACGGCGGATTGACCGTAAT) and sgChr2–2 (GGTGTGCGTATGAAGCAGTG).

Population Doubling Assay

Cells were seeded in a 6-well plate at a density of 250,000 cells (NB-Ebc1 and CFPAC-1) or 125,000 cells (all other cell lines) per well. Inhibitors were added in triplicate at the indicated concentrations. Every 3–4 days cells were trypsinized, resuspended and counted by trypan blue-exclusion on a Countess II Automated Cell Counter (Thermo Fisher). Cells were then replated at their original density with fresh inhibitors. If insufficient cells remained to achieve initial plating density, all remaining cells were plated for that condition. Population doubling was calculated by log₂ of the fold change from the plated population, and these numbers were added to reach the cumulative population doublings for that time point.

Western Blots

Cells were lysed in Cell Signaling Lysis Buffer (9803) supplemented with protease (Roche #11836170001) and phosphatase inhibitors (Roche #04906845001). Lysates were quantified using a BCA assay (Pierce) and normalized. SDS-PAGE gels were used to separate proteins, and proteins were transferred to a PVDF membrane. The following primary antibodies were used: CDK8 (CST # 4106, RRID:AB_1903936), CCNC (Bethyl #A301989AM), pSTAT1 S727 (CST #8826, RRID:AB_2773718), STAT1 (CST #9176, RRID:AB_2240087), GAPDH (Santa Cruz #sc-47724, RRID:AB_627678), ERK (CST #4696, RRID:AB_390780), and pERK (CST #4370, RRID:AB_2315112). Membranes were incubated with secondary antibodies (Licor #926–68070 and #926–32211) and imaged on a Licor Odyssey. Quantification was performed using the Image Studio Software.

In Vivo Therapeutic Studies

This study was approved by the Institutional Care and Use Committee (IACUC) of Dana-Farber Cancer Institute and performed under protocol 15–029. IACUC guidelines on the ethical use and care of animals were followed. SK-N-AS cells were verified to be free of mouse pathogens, and then 1.0e6 cells in 30% matrigel were injected subcutaneously into the flanks of nude mice. Tumors were measured using a caliper, and tumor volume was estimated using the standard formula. When tumors reached 90–200 mm³ animals were randomized and assigned to a treatment group. Selumetinib was prepared in a vehicle of 0.5% HPMC 0.1% Tween 80 administered BID at 25 mg/kg per dose (32). BI-1347 was administered PO at 10 mg/kg daily as previously described (33). Animals were euthanized when tumors reached a humane endpoint (i.e., tumor 2 cm in any direction). Significance of survival curves was determined with a Mantel-Cox test.

RNA-sequencing

SK-N-AS cells were infected with a single vector containing Cas9 and the indicated sgRNA. Cells were then selected with puromycin. Nine days after infection, cells were treated with either DMSO or 3 nM trametinib in triplicate. After four days of treatment, RNA was harvested and purified with the Qiagen RNeasy Plus Mini Kit (Qiagen # 74134). Preparation of RNA library and transcriptome sequencing were performed by Novogene Co., LTD.

RNA-Seq Analysis

Quality control tests for unmapped reads were performed based on the FastQC v.0.11.9 software (FastQC, RRID:SCR_014583 Babraham Bioinformatics, http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and summarized with multiQC v1.7 (34). The ERCC RNA Spike-In Mix (ThermoFisher #4456740, https://tools.thermofisher.com/content/sfs/manuals/ERCC92.zip) was used as an external RNA control to estimate the variability in transcriptional performance (Bioconductor erccdashboard) (35).

The spiked human reads were mapped to the hg38/gencode v30 human genome using STAR v2.6.1d (RRID:SCR_004463) (36) with standard parameters --outSAMtype BAM SortedByCoordinate --outSAMunmapped None --outSAMattributes NH HI NM MD AS XS --outReadsUnmapped Fastx --outSAMstrandField intronMotif --quantMode TranscriptomeSAM GeneCounts --quantTranscriptomeBan IndelSoftclipSingleend. Quality control for the mapped reads and for replicate reproducibility were performed using SARTools v1.7.3 (RRID:SCR_016533) (37). Gene level reads were summarized by counting the reads that overlapped the hg38/gencode v30 annotated gene exons, by using the featureCounts v1.6.3 method implemented in the Subread v2.0.0 package (http://subread.sourceforge.net) (RRID:SCR_009803) (38) and rsem v1.3.1 (RRID:SCR_013027) (http://deweylab.github.io/RSEM/) (39).

Gene counts were normalized and used to quantify differential genes between the experimental and control conditions using the DESeq2 v1.32.0 method (RRID:SCR_015687) (40). Reads were normalized based on DESeq2 scaling factors. Gene expression was estimated based on log₂(TPM+1) scores for normalized reads (41). Pair-wise genome-wide expression scatter dot plots, PCA, hierarchical clustering dendrograms and volcano plots were used to estimate the correlation of the replicate samples and the transcriptional dynamics across experimental phenotypes (SARTools v1.7.4, and Morpheus visualization platform, https://software.broadinstitute.org/morpheus) (37). Genes with max(log2(TPM+1)) ≥ 1 across all samples were annotated as “expressed”.

Genome-wide differential transcription was estimated at gene level based on two class comparative methods:

1. eBayes limma v.48.3 applied to the log2(TPM+1) expression data (42) with the significance cutoffs abs(fold change expression) ≥1.5 and adjusted P-value ≤ 0.10.

2. DESeq2 v1.32.0 based on the robust shrunken log2 fold change scores and the approximate posterior estimation for GLM coefficients method for effect size apeglm v1.6 (40) with the significance cutoffs abs(fold change expression) ≥1.5 and adjusted P-value ≤ 0.10.

Heatmaps for transcriptional data visualization were created by using the Morpheus software platform (https://software.broadinstitute.org/morpheus/) based on the log₂(1+ TPM) normalized counts data. The RNA-Seq data for this study are available for download from the Gene Expression Omnibus (GEO, RRID:SCR_005012) repository (GSE190178).

GSEA of transcriptome data

GSEA v4.1.0 software (43,44) was used to identify functional associations of the molecular phenotypes induced by MEK inhibitor and cyclin knock-out treatments with the collections of 50 hallmark gene sets (h), 6,290 curated pathways and experimental gene sets (c2), and 7,481 Gene Ontology/Biological Processes gene sets (c5), available from the MSigDB v7.4 database (43,45), as well as with the van Groningen et al. neuroblastoma lineage differentiation signatures (46).

For each experimental comparison, the hg38/gencode v30 expressed genes were ranked based on the expression fold change in treated vs. untreated phenotypes. The goal of GSEA was to identify the gene sets that are distributed at the top or at the bottom of the ranked list of genes based on the Kolmogorov-Smirnov enrichment test. Gene sets with absolute Normalized Enrichment Score (NES) ≥1.3, a nominal P ≤ 0.05 and an FDR ≤ 0.25 for the Kolmogorov-Smirnov test were considered significant hits. The results were visualized on volcano plots for the normalized enrichment score (NES) vs. -log10(P) and on GSEA plots.

In addition, the functional associations of the molecular phenotypes were explored with the single sample GSEA (ssGSEA) method (47) based on the Bioconductor GSVA v1.40.1 implementation (48). ssGSEA is an extension of Gene Set Enrichment Analysis that calculates separate enrichment scores for each pairing of a sample and gene set. Each ssGSEA enrichment score represents the degree to which the genes in a particular gene set have coordinately increased or decreased expression within a sample.

Cell Cycle Analysis

SK-N-AS and KP-N-SI9s cells were plated in 10 cm plates (2.0e6 cells/plate for SK-N-AS, 1.0e6 cells/plate for KP-N-SI9s). The cells were treated with DMSO, 3 nM trametinib, 10 nM BI-1347, or a combination of the compounds the following day. Four days after treatment, SK-N-AS and KP-N-SI9s cells were incubated with 10 μM EdU for 90 minutes using Click-iT^® Plus EdU Alexa Fluor^® 647 Flow Cytometry Assay Kit (Thermo Fisher Scientific #C10424) at 37°C and 1.0e6 cells were harvested. The cells were fixed, permeabilized, and stained following the manufacturer’s instructions. Propidium Iodide (PI) (CST #4087) was used for DNA staining. Flow cytometry was performed on a BD FACSCelesta and analyzed using FlowJo software.

Statistical Analysis

The statistical test used and p-value calculated are described in the relevant figure legends. All t tests are unpaired and two-sided unless otherwise indicated. A P value of 0.05 was used as the cutoff for significance unless otherwise indicated. Statistical tests were calculated in GraphPad Prism 9 (RRID:SCR_002798) or R 3.63. Error bars represent SD unless otherwise indicated. All duplicate measures were taken from distinct samples rather than repeated measures of the same sample.

Data Availability Statement

All large datasets produced in this study are made publicly available. The CRISPR screening data is available at https://figshare.com/s/c4dec2c76e7170ed84b1 and the RNA-seq data has been deposited in the Gene Expression Omnibus (GEO) (GSE190178). The Dependency Map and Project SCORE data analyzed in this study is available to the public at depmap.org.

Results

Genome-scale CRISPR-Cas9 screen identifies sensitizers to MEK inhibition

To discover genes whose loss sensitizes RAS-mutant neuroblastoma to the FDA-approved MEK inhibitor trametinib, we performed a genome-scale CRISPR-Cas9 depletion screen in the neuroblastoma cell line SK-N-AS, which harbors an NRAS^Q61K activating mutation. We determined that a concentration of 3 nM trametinib reduced cell viability and effectively inhibited MEK, as evidenced by loss of ERK phosphorylation (Supplemental Figure S1A and B). We also confirmed that Cas9 over-expression did not alter the response to trametinib (Supplemental Figure S1C). We then transduced SK-N-AS cells expressing constitutive Cas9 with the genome-scale AVANA-4 library, which contains 74,687 unique sgRNAs with ~4 sgRNAs per gene (29). Cells were selected and then treated with either DMSO or trametinib. We next performed sequencing to assess the relative abundance of sgRNAs in trametinib-treated and DMSO-treated cells after 7 and 15 days of treatment (Figure 1A).

Figure 1. CRISPR screen identifies members of the mediator kinase module as sensitizers to MEK inhibitor treatment.

A, Diagram of genome-scale CRISPR-Cas9 MEK inhibitor modifier screen. B, Correlation between the differential log₂ fold change in sgRNA abundance between DMSO and trametinib treatment at day 7 (x-axis) and day 15 (y-axis). The Pearson correlation coefficient and P value are shown, and the best fit line is plotted. C, Volcano plot showing the STARS score for differential effect in the trametinib vs. DMSO condition (x-axis) and the - log10 p-value (y-axis) at day 15. Each dot indicates an individual gene. Dashed lines indicate cutoffs for significance for -log₁₀ p-value and STARS score. CCNC and CDK8 are highlighted in red. The number of genes reaching these cutoffs is shown. D, Diagram of the mediator kinase module (adapted from (30)). E, Guide level log₂ fold change relative to plasmid pool for each member of the mediator kinase module including mutually exclusive homologs. Each dot represents a different sgRNA targeting the indicated gene. Trametinib treatment is shown in red and DMSO treatment is shown in gray. p-values indicate the p-value calculated using the STARs algorithm for significant difference in the screen, n.s. indicates>0.05.

We were interested in discovering genes that modify response to MEK inhibition and selectively inhibit growth in trametinib treated cells compared to DMSO treated cells. We calculated the log₂ fold change of the abundance of each sgRNA in the trametinib-treated and the DMSO-treated condition relative to the plasmid pool. We then assessed the differential sensitivity in log₂ fold change for each sgRNA in the trametinib treated arm compared to the DMSO treated arm and then calculated the average for each gene. There was a significant correlation in differential log₂ fold change between day 7 and day 15 (r= 0.3831311, p <2.2 e-16, Figure 1B), which provided confidence in the robustness of the screening results as similar genes depleted at both time points. We then used the STARS algorithm to identify genes that had significantly different effects in the MEK inhibitor arm compared to the DMSO arm. As expected, there was a greater dynamic range at day 15; more genes reached our cutoff for significance as sensitizers (Supplemental Table S1). While our screen was designed to identify sensitizers, we also identified genes whose depletion made the cells more resistant to trametinib treatment. TSC1 and TSC2, two of the top genes on the resistance side, are negative regulators of the mTOR complex, and loss of these genes drives hyperactivation of mTOR signaling. This finding is consistent with previous work that demonstrated that mTOR inhibitors improve response to MEK inhibition in NRAS mutant neuroblastoma and other RAS-driven cancers (49–52), providing further confidence that the screen data were robust.

At day 7, the top scoring sensitizer gene was CCNC, which encodes for Cyclin C, a non-cycling cyclin protein (53) (Supplemental Figure S1D). CCNC binding is required for the serine/threonine kinase activity of the transcriptional cyclin dependent kinase CDK8 (54). Strikingly, CDK8 was the third most significant sensitizer gene at day 15 and CCNC remained the most significant sensitizer (Figure 1C, Supplemental Table S1). CCNC and CDK8 are part of the mediator kinase module, which transiently associates with the larger mediator complex, a key transcriptional coactivator(55). This kinase module consists of four members, three of which have paralogs that are mutually exclusive in the module: CCNC, MED12/MED12L, MED13/MED13L, and CDK8/CDK19 (Figure 1D). Notably, the sgRNAs targeting CCNC and CDK8 are only depleted in trametinib-treated conditions and not in the DMSO-treated conditions suggesting that they are not required for normal growth in this cell line, but instead only impair cell fitness in the context of MEK inhibition (Figure 1E, Supplemental Figure S1E). Because CDK8 and CCNC are members of a larger module with multiple paralog members, we were also interested in whether loss of other members of the mediator kinase module could sensitize cells to MEK inhibition. MED12 depleted in both treatment arms, which is consistent with the Dependency Map dataset where MED12 is a dependency in this cell line (Supplemental Figure S1F). MED13 depleted more in the trametinib arm, consistent with pattern for CCNC and CDK8, and was among the 175 genes that reached the significance cut-off for sensitizers at day 15 (Figure 1E and Supplemental Table S1). None of the paralogs, including CDK19, the paralog of CDK8, sensitized cells to MEK inhibition when lost in the screen. As CDK8 knockout did sensitize cells to MEK inhibition, this suggests that CDK8 and CDK19 may not be functionally redundant in this context. Therefore, we focused on CCNC and CDK8 as potential genes that when lost sensitize neuroblastoma cells to MEK inhibitors.

CCNC or CDK8 knockout renders RAS-mutant neuroblastoma cell lines more sensitive to MEK inhibition

We next sought to validate whether loss of CCNC sensitizes RAS mutant neuroblastoma cell lines to MEK inhibitor treatment. To that end we knocked out CCNC in two neuroblastoma cell lines with RAS mutations: SK-N-AS and KP-N-S19s. SK-N-AS harbors an NRAS^Q61K activating mutation and KP-N-SI9s harbors a KRAS^G12A activating mutation. We found that the knockout of CCNC alone did not affect growth compared to the control when treated with DMSO but did improve the response to trametinib in both cell lines (Figure 2A–D). This is consistent with the Broad Institute’s Cancer Dependency Map and the Sanger Institute’s Project SCORE datasets, in which most neuroblastoma cell lines are not dependent on CCNC or CDK8 at baseline (Supplemental Figure S2A and B), as well as our own data from this screen (Supplemental Figure S1F). To establish that CCNC knockout was truly sensitizing the cells to MEK inhibition, and not an off-target effect of trametinib, we confirmed that CCNC knockout sensitized the SK-N-AS to selumetinib, another clinical grade MEK inhibitor (Figure 2E and F). Furthermore, we found that CCNC knockout shifted both the trametinib and selumetinib dose response curves (Figure 2G and H).

Figure 2. Genetic knockout of CDK8 or CCNC sensitizes RAS-mutant neuroblastoma cell lines to MEK inhibition.

A, SK-N-AS cells were infected with a negative control targeting an intergenic region (sgChr2–2, gray) and two different sgRNAs targeting CCNC (red). Parental uninfected cells also served as a negative control (black). Cumulative population doublings (y-axis) over time (x-axis) for SK-N-AS cells with indicated sgRNA treated with 3 nM trametinib (Tram, dotted line) or DMSO (solid line). p-values indicate a one-way ANOVA followed by Tukey’s multiple comparison test for sgChr2–2 + trametinib compared to sgCCNC-1 + trametinib (p<0.0001) and sgCCNC-3 + trametinib (p<0.0001). B, Western blot for CCNC and phosphorylated ERK in indicated genetic knockout 24 hours after 3 nM trametinib (+) or DMSO (–) treatment in SK-N-AS cells. Total ERK and GAPDH serve as loading controls. C, KP-N-SI9s cells were treated, and population doublings were measured, as in A. p-values indicate a one-way ANOVA followed by Tukey’s multiple comparison test for sgChr2–2 + trametinib compared to sgCCNC-1 + trametinib (p<0.0001) and sgCCNC-3 + trametinib (p=0.0003). D, Western blot as in B, in KP-N-SI9s cells. E, As in A, cumulative population doublings (y-axis) over time (x-axis) for SK-N-AS cells with indicated sgRNA treated with 250 nM selumetinib (Sel, dotted line) or DMSO (solid line). p-values indicate a one-way ANOVA followed by Tukey’s multiple comparison test for sgChr2–2 + selumetinib compared to sgCCNC-1 + selumetinib (p<0.0001) and sgCCNC-3 + selumetinib (p<0.0001). F, Western blot for CCNC, and phosphorylated ERK in indicated genetic knockout 24 hours after 250 nM selumetinib (+) or DMSO (–) treatment in SK-N-AS cells. Total ERK and GAPDH serve as loading controls. G, SK-N-AS cells were infected with negative control (sgChr2–2, gray) and two different sgRNAs targeting CCNC (red). Parental uninfected cells also served as negative control (black). 7 days after treatment, viability relative to DMSO was determined using CellTiter-Glo, and the mean ± SD is shown on the y-axis. The log of the nM dose of trametinib shown on the x-axis. H, SK-N-AS cells were infected with negative control (sgChr2–2, gray) and two different sgRNAs targeting CCNC (red). Parental uninfected cells also served as negative control (black). 7 days after selumetinib treatment, viability relative to DMSO was determined using CellTiter-Glo and the mean ± SD of replicates are shown on y-axis. The log of the nM dose is shown on the x-axis. I, Cumulative population doublings as in A, in KP-N-SI9s cells after CDK8 knockout (sgCDK8-3 or sgCDK8-4, red) or controls (uninfected, black, and sgChr2–2, grey) with 3 nM trametinib treatment (Tram, dotted lines) or DMSO (solid lines). These data are from the same experiment as in C, and controls are re-plotted here for clarity. p-values indicate a one-way ANOVA followed by Tukey’s multiple comparison test for sgChr2–2 + trametinib compared to sgCDK8–3 + trametinib (p=0.042) and sgCDK8–4 + trametinib (p=0.0195). J, Western blot demonstrating CDK8 and phosphorylated ERK (p-ERK) levels in indicated genetic condition, 24 hours after treatment with DMSO (–) or 3 nM trametinib (+) in KP-N-IS9s cells. Vinculin and total ERK serve as loading controls.

While CCNC loss caused a more significant sensitization in our screen, CDK8 is the catalytic component of the mediator kinase module. We were therefore interested in confirming that knockout of CDK8 is sufficient to sensitize cells to MEK inhibition. Indeed, when we knocked out CDK8 in RAS mutant neuroblastoma cells we observed increased sensitivity to MEK inhibition (Figure 2I and J, Supplemental Figure S2C and D). Together these data confirmed the screen findings that loss of the mediator kinase module members CCNC and CDK8 sensitizes RAS mutant neuroblastoma cells to MEK inhibition.

CDK8 inhibitors improve response to MEK inhibition in RAS-mutant neuroblastoma cell lines

CCNC is not currently druggable, but multiple selective small molecule CDK8 kinase inhibitors have been developed (56). We were therefore interested in whether inhibition of CDK8 kinase activity would phenocopy genetic knockout of CCNC or CDK8 and sensitize cells to MEK inhibition. We tested two selective CDK8 inhibitors, BI-1347 and JH-XII-136 (33,57), across three RAS-mutant neuroblastoma cell lines: SK-N-AS, KP-N-S19s, and NB-Ebc1. CDK8 has been reported to phosphorylate STAT1 at serine 727 (58); we verified that both compounds successfully inhibited CDK8 and reduced phosphorylation of STAT1 at this residue (Figure 3A and Supplemental Figure S3A). We then treated neuroblastoma cells with these CDK8 inhibitors and found limited effect on cell viability with single agent treatment (Figure 3B and Supplemental Figure S3B), consistent with the genetic data.

Figure 3. CDK8 kinase inhibitors sensitize RAS-mutant NBL to MEK inhibition.

A, Western blot demonstrating levels of STAT1 phosphorylation on serine 727 (pSTAT1^S727) 24 hours after treatment with indicated concentrations of the CDK8 inhibitor BI-1347 in three RAS-mutant neuroblastoma cell lines. Cell line name is indicated at the top. B, Viability, measured by CellTiter-Glo, relative to DMSO at 72 hours after treatment with indicated concentrations of BI-1347 in three RAS-mutant neuroblastoma cell lines: SK-N-AS (black), NB-Ebc1 (red), and KP-N-SI9s (grey). C, Cumulative population doublings (y-axis) over time (x-axis) after treatment with DMSO (black), 10 nM BI-1347 (dotted grey), 3 nM trametinib (solid grey), or both trametinib and BI-1347 (Combo, red) in three neuroblastoma cell lines. Cell line name is indicated at top. Below, western blots show phosphorylated ERK (p-ERK) and phosphorylated STAT1 at serine 727 (pSTAT1^S727) after treatment with DMSO, 3 nM trametinib, 10 nM BI-1347, or both drugs as indicated. Total ERK and STAT1 levels serve as loading controls. D, As in C, cumulative population doublings for SK-N-AS are shown after treatment with DMSO (black), 100 nM JH-XII-136 (dotted grey), 3 nM trametinib (solid grey), or both (Combo, red). Below, western blots show phosphorylated ERK (p-ERK) and phosphorylated STAT1 at serine 727 (pSTAT1^S727) levels after treatment with 3 nM trametinib, 10 nM BI-1347, or both as indicated. Total ERK and STAT1 levels serve as loading controls. E, Trametinib dose response curves with (red) and without (black) co-treatment with 10 nM BI-1347 in two neuroblastoma cell lines. Trametinib concentration is indicated on the x-axis, viability relative to DMSO assayed by CellTiter-Glo, is shown on the y-axis. F, Cell cycle analysis in two neuroblastoma cell lines after treatment with DMSO, 3 nM trametinib, 10 nM BI-1347, or both, as indicated on the x-axis. Percent of cells in each phase of the cell cycle is shown on the y-axis.

However, the combination of the CDK8 inhibitor BI-1347 and trametinib induced a greater effect on cell viability than trametinib alone in all three cell lines (Figure 3C, Supplemental Figure S3C). Combination with BI-1374, a structural analog of BI-1347 but with >600-fold less activity against CDK8, had no effect in combination with trametinib, supporting that the observed effect with BI-1347 is due to on-target CDK8 inhibition (Supplemental Figure S3D). Consistent with on-target efficacy, combination with JH-XII-136, a structurally distinct CDK8 inhibitor, also improved the effect of trametinib (Figure 3D). Furthermore, addition of either CDK8 inhibitor was sufficient to induce a shift in the trametinib dose response curve (Figure 3E and Supplemental Figure S3E). We then performed cell cycle analysis and found that while CDK8 inhibition alone did not impact cell cycle progression, trametinib treatment induced an accumulation of cells in G0/G1, and this effect was enhanced by the addition of CDK8 inhibitors (Figure 3F). Together, these data support the notion that the combination of CDK8 inhibitors and MEK inhibitors may yield a greater therapeutic response than MEK inhibitors alone.

Combined MEK and CDK8 inhibition improves tumor response in vivo

We were next interested in whether the combination of MEK and CDK8 inhibition would have therapeutic efficacy in an in vivo context. To this end, we evaluated the combination of the CDK8 inhibitor BI-1347 and the MEK inhibitor selumetinib in a xenograft model of RAS-mutant neuroblastoma. SK-N-AS cells were injected into the flanks of nude mice and after tumors were established, mice were randomized to one of four treatment groups: vehicle, BI-1347, selumetinib, or the combination of BI-1347 and selumetinib. We confirmed that BI-1347 and selumetinib effectively inhibited their targets in vivo in tumors using STAT1 and ERK phosphorylation, respectively, as biomarkers (Figure 4A and B). This combination was well tolerated by the mice, and we did not observe any overt signs toxicity or significant weight loss over the course of the experiment (Figure 4C).

Figure 4. Combined CDK8 and MEK inhibition is effective in vivo.

A, On the left, western blot showing phosphorylated STAT1 (pSTAT1^S727) levels after treatment with vehicle or 10 mg/kg BI-1347 in SK-N-AS xenograft tumors. Total STAT1 serves as a loading control. Mice were treated for 24 hours, and tumors were harvested 2 hours after last treatment. On the right, quantification of phosphorylated STAT1 relative to total STAT1 for each treatment group. ** indicates p-value of 0.0054 in an unpaired two-tailed t-test. B, On the left, a western blot showing phosphorylated ERK (pERK) levels after treatment with 25 mg/kg selumetinib or vehicle in SK-N-AS xenograft tumors. ERK serves as a loading control. Mice were treated BID for 24 hours and tumors were harvested two hours after last treatment. On the right, quantification of phosphorylated ERK relative to total ERK for each treatment group. **** indicates a p-value of less than 0.0001 in an unpaired two-tailed t-test. C, Mean body weights (y-axis) for each treatment group at the indicated day. Vehicle is shown in solid black, 10 mg/kg BI-1347 is shown in dotted black, 25 mg/kg selumetinib (BID) is shown in grey, and the combination of BI-1347 and selumetinib is shown in red. Error bars indicate standard deviation. n=8 for all treatment groups until day 10, after which they reflect the number of animals still surviving in each group. D, Mean tumor volume (y-axis) for each treatment group at the indicated day (x-axis). Vehicle is shown in solid black, 10 mg/kg BI-1347 is shown in dotted black, 25 mg/kg selumetinib (BID) is shown in grey, and the combination of BI-1347 and selumetinib is shown in red. Error bars indicate standard error. n=8 for all treatment groups on all days except vehicle (day 14 n=6), and BI-1347 (day 10 n=7 and day 14 n=4) for which some animals had to be euthanized prior to day 14. P-value indicates significance for the combination compared to selumetinib (p= 0.0030) in a Mann-Whitney test. E, Kaplan-Meier survival curve indicates percent of mice still alive (y-axis) at indicated days (x-axis) for each treatment group. Vehicle is shown in solid black, BI-1347 (10 mg/kg QD) is in dotted black line, selumetinib (25 mg/kg BID) is grey, and combination is in red. p-values indicate significance for selumetinib vs. combination (p=0.0029) in a log rank Mantel Cox test.

Consistent with our in vitro observations, single agent CDK8 inhibition with BI-1347 did not slow tumor growth in vivo (Figure 4D). MEK inhibition with selumetinib initially had a cytostatic effect, but this was followed by rapid outgrowth of the tumors. However, the combination treatment of CDK8 and MEK inhibition had a sustained and more durable impact on tumor growth relative to single agent MEK inhibition, leading to improved survival (p = 0.0038, Figure 4D and E and Supplemental Figure S4). These data confirm that the enhanced efficacy of combined CDK8 and MEK inhibition is not limited to an in vitro setting but is also observed in vivo.

CDK8 and CCNC knockout antagonize the transcriptional adaptation to MEK inhibition

We next sought to understand the mechanism by which loss of the mediator kinase module activity sensitized cells to trametinib treatment. The mediator complex is a transcriptional co-activating complex that is generally required for gene transcription by RNA pol II (59). The exact role of the kinase module, which is transiently associated with the larger mediator complex, remains somewhat enigmatic. While the mediator kinase module is not thought to have a role in general transcription, it is a key regulator of stem cell maintenance and can phosphorylate sequence specific transcription factors (60,61). We were therefore interested in the consequences of CCNC and CDK8 loss on transcription, particularly in the context of MEK inhibition. To that end, we performed RNA-sequencing in SK-N-AS cells harboring sgRNAS targeting LACZ (control), CCNC, or CDK8 and then treated with either DMSO or trametinib for four days (Supplemental Table S2A).

We first confirmed that trametinib treatment in control cells yielded transcriptional changes consistent with MEK inhibition. Indeed, gene set enrichment analysis (GSEA) demonstrated that RAS activity signatures were downregulated in the trametinib treated cells relative to DMSO (Figure 5A, Supplemental Figure S5A, and Supplemental Table S3). Additionally, we observed downregulation of inflammatory gene sets, and upregulation of gene sets related to cell cycle and elongation (Figure 5A and Supplemental Figure S5A and B). We next considered the transcriptional impact of CCNC or CDK8 knockout compared to control cells. Despite no observed impact on cell viability, knockout of CCNC and CDK8 induced many transcriptional changes. These alterations had a strong positive correlation to one another (Figure 5B and C, Supplemental Figure S5C and D).

Figure 5. Mediator kinase knockout prevents transcriptional adaptation to MEK inhibitor treatment.

A, Volcano plot showing gene sets that are altered in SK-N-AS cells four days after trametinib treatment (MEKi) compared to DMSO (CTRL) in cells infected with a control sgRNA (sgLACZ). The normalized enrichment score for single sample gene set enrichment analysis (ssGSEA) is shown on the x-axis, and the -log₁₀ (P-value + 0.0001) is shown on the y-axis. Gene sets that fall in key altered functional groups are highlighted in black (RAS/MAPK/ERK), red (Cell Cycle) and blue (Elongation). These gene sets are annotated in Supplemental Table S3. All other gene sets are shown in grey. Dotted line indicates p-value cut-off for significance. B, Gene level scatter plot of log₂ fold change in expression relative to sgLACZ (CTRL) for cells infected with sgCDK8 (y-axis) and sgCCNC (x-axis). A best fit line and the Pearson R value is shown. C, Gene set enrichment analysis (GSEA) plot demonstrating enrichment of the top 500 genes upregulated by CDK8 knockout in those upregulated by CCNC knockout. D, GSEA plot demonstrating enrichment of the top 500 genes upregulated by MEK inhibitor treatment in those downregulated by CCNC knockout relative to the sgLACZ + DMSO (CTRL) condition. E, Volcano plot showing the gene sets that are altered in SK-N-AS cells after knockout of CCNC. The normalized enrichment score from an ssGSEA analysis is shown on the x-axis, and the -log₁₀ (P-value +0.0001) is shown on the y-axis. Gene sets that fall in key altered functional groups are highlighted in black (RAS/MAPK/ERK), red (Cell Cycle), and blue (Elongation). F, As in E, Volcano plot showing gene sets that are altered in SK-N-AS cells after CDK8 knockout. G, Heatmap of gene expression for all genes that met the expression cut-off. Treatment groups are indicated at the top, and log₂ fold change relative to the sgLACZ + DMSO condition (CTRL) was calculated. Color scale bar is at bottom, color scaling is done per row. H, Heatmap for normalized enrichment scores for each gene set included in the MSigDB hallmark, c2 and c5 gene set collections in each treatment group in ssGSEA relative to sgLACZ + DMSO (CTRL). Color scale bar is at bottom, color scaling is done per row. I, Volcano plot showing the gene sets that are altered in SK-N-AS cells after knockout of CCNC combined with trametinib treatment relative to infection control (sgLACZ) treated with DMSO (CTRL). The normalized enrichment score calculated by ssGSEA (x-axis) and the -log10 (P-value + 0.0001) (y-axis) are shown. Gene sets that fall in key altered functional groups are highlighted in black (RAS/MAPK/ERK), red (Cell Cycle), and blue (Elongation). J, As in I, Volcano plot showing the gene sets that are altered in ssGSEA in SK-N-AS cells after knockout of CDK8 combined with trametinib treatment relative to infection control (sgLACZ) treated with DMSO (CTRL).

In contrast, the expression changes induced by CDK8 and CCNC knockout had a strong negative correlation to that of trametinib treatment (Figure 5D, Supplemental Figure S5E). Strikingly, we found that the same cell cycle and elongation gene sets that were upregulated by trametinib treatment were downregulated by CCNC and CDK8 loss (Figure 5E and F, Supplemental Figure S5F). When we examined all genes and gene sets, we again saw that CCNC and CDK8 loss largely led to opposing transcriptional changes compared to MEK inhibition: genes and gene sets that increase expression with MEK inhibition decrease with CCNC and CDK8 loss, and genes and gene sets that decrease expression with MEK inhibition increase with CCNC and CDK8 loss (Figure 5G and H, Supplemental Figure S5G–I, Supplemental Table S2B–F). Together, these data suggest that loss of the mediator kinase module would antagonize MEK inhibitor induced transcriptional changes.

To determine if that is indeed the case, we next examined the transcriptional consequences of MEK inhibition in cells with CCNC or CDK8 knockout. Once again, we observed strong correlation in gene level expression changes between CDK8 and CCNC knockout in the context of trametinib treatment (Supplemental Figure S5J). We found that the genes and gene sets downregulated by MEK inhibition alone, including the RAS/MAPK associated gene sets, were still suppressed in the combination despite the upregulation of these gene sets observed with CCNC and CDK8 knockout alone (Figure 5G–J, Supplemental Figure S5G–K). However, the genes and gene sets that were up-regulated by MEK inhibition and down-regulated by CCNC and CDK8 are down-regulated or unchanged when MEK inhibition is combined with CCNC or CDK8 knockout (Figure 5G–J, Supplemental Figure S5G–I and L). Notably, these gene sets are largely related to cell cycle and elongation. These data are consistent with our previous finding that combined CDK8 and MEK inhibition leads to a more profound cell cycle arrest than MEK inhibition alone. Together, these data support a model where mediator kinase activity is required for the upregulation of a pro-growth gene expression program in the context of MEK inhibition and that loss of mediator kinase function prevents this transcriptional adaptation.

Combined CDK8 and MEK inhibition is effective in neuroblastoma with other RAS-activating mutations

Multiple mutations can lead to hyper-activation of the RAS/MAPK pathway in neuroblastoma beyond oncogenic mutations in RAS itself. Most notably, activating mutations in ALK and inactivating mutations in the RAS-GAP NF1 can lead to downstream activation of RAS signaling. We were therefore interested in determining whether this combination could be effective in these settings as well, as this would expand the number of patients who might respond to this combination. To that end, we tested combined MEK and CDK8 inhibition in SK-N-FI and GI-M-EN, two NF1-null lines, and found that indeed both cell lines had a profound response to the combination (Figure 6A). We then investigated KELLY and SK-N-SH, two lines that each harbor an ALK^F1147L activating mutation. Consistent with prior reports that ALK mutant neuroblastoma is more resistant to MEK inhibition (12), neither cell line had a substantial response to either CDK8 or MEK inhibitor single agent treatment (Figure 6B). KELLY did not respond to the combination of CDK8 and MEK inhibitors either, while SK-N-SH had a dramatic response to the combination despite limited efficacy of either single agent (Figure 6B, Supplemental Figure S6A). Interestingly, KELLY harbors an amplification of MYCN, while the other RAS active cell lines we tested do not, raising the possibility that high MYCN expression renders cells resistant to the combination effects. To interrogate that question, we over-expressed MYCN in GI-M-EN cells and found that they remained sensitive to the combination, suggesting that high MYCN expression is not sufficient to render resistance (Supplemental Figure S6B). Finally, we tested the combination in two lines that do not harbor known RAS-pathway activating mutations: SK-N-DZ and NGP. SK-N-DZ did not respond to either the single agents or the combination (Figure 6C). Interestingly, in contrast with other neuroblastoma cell lines, the CDK8 inhibitor alone had a significant impact on NGP viability, which was minimally improved with addition of the MEK inhibitor. Together, these data demonstrate that this combination is effective in a subset of neuroblastomas that have RAS pathway activation due to other mutations.

Figure 6. Response of neuroblastoma cell lines with other RAS-pathway mutations to combined MEK and CDK8 inhibition.

A, Cumulative population doublings (y-axis) over time (x-axis) after treatment with DMSO (black), 10 nM BI-1347 (dotted grey), 3 nM trametinib (solid grey), or both trametinib and BI-1347 (Combo, red) in two NF1-null neuroblastoma cell lines SK-N-FI (left) and GI-M-EN (right). Below, a western blot shows phosphorylated ERK (p-ERK) and phosphorylated STAT1 (pSTAT1) levels after treatment with 3 nM trametinib or 10 nM BI-1347 as indicated. Total ERK and STAT1 levels serve as loading controls. B, Cumulative population doublings (y-axis) over time (x-axis) after treatment with DMSO (black), 10 nM BI-1347 (dotted grey), 3 nM trametinib (solid grey), or both trametinib and BI-1347 (Combo, red) in two ALK-mutant neuroblastoma cell lines KELLY (left) and SK-N-SH (right). Below, a western blot shows phosphorylated ERK (p-ERK) and phosphorylated STAT1 (pSTAT1) levels after treatment with 3 nM trametinib or 10 nM BI-1347 as indicated. Total ERK and STAT1 levels serve as loading controls. C, Cumulative population doublings (y-axis) over time (x-axis) after treatment with DMSO (black), 10 nM BI-1347 (dotted grey), 3 nM trametinib (solid grey), or both trametinib and BI-1347 (Combo, red) in two RAS-WT neuroblastoma cell lines SK-N-DZ (left) and NGP (right). Below, a western blot shows phosphorylated ERK (p-ERK) and phosphorylated STAT1 (pSTAT1) levels after treatment with 3 nM trametinib or 10 nM BI-1347 as indicated. Total ERK and STAT1 levels serve as loading controls.

Combined CDK8 and MEK inhibition is effective in other RAS-mutant cancers

Because RAS pathway mutations are common in many cancer types, we next sought to determine whether this combination is effective in other RAS-mutant cancers or if this phenomenon is specific to neuroblastoma. Sulahian and colleagues have previously performed genome-scale MEK inhibitor sensitizer screens in three RAS-mutant lung and pancreatic models (62). We analyzed their data and found that loss of CCNC and CDK8 sensitized two of the three cell lines to MEK inhibition (Figure 7A–C). CCNC and CDK8 met their criteria for inclusion in a more focused library that was screened in seven additional lung and pancreatic RAS-mutant models, for a total of ten cell lines. CCNC or CDK8 loss sensitized five of these lines to MEK inhibition (Figure 7D). We therefore sought to validate whether these models might also respond to combined CDK8 and MEK inhibition. We acquired NCI-H23, a lung cancer line harboring a KRAS^G12C mutation, and CFPAC-1, a pancreatic cell line harboring a KRAS^G12V mutation. Indeed, we found that combined CDK8 and MEK inhibition had greater efficacy than the single agents in both cell lines, although the effect was more profound in CFPAC-1 (Figure 7E and F). These data confirm that our findings may have broad relevance for other RAS-mutant malignancies beyond neuroblastoma.

Figure 7. Combined CDK8 and MEK inhibition is effective in a subset of other RAS-driven cancers.

A, Correlation between the differential trametinib sensitivity observed in our genome-wide CRISPR screen in SK-N-AS (x-axis) and in the lung cancer cell line A549 cells screened by Sulahian et al. (y-axis). Each dot indicates a gene. CCNC and CDK8 are highlighted in red. B, As in A, but for pancreatic cancer cell line CFPAC-1 (y-axis) screened by Sulahian et al. C, As in A, but for lung cancer cell line NCI-H23 screened by Sulahian and colleagues (y-axis). D, Heatmap of differential trametinib sensitivity observed in both focused and genome-wide screens performed in RAS-mutant cell lines by Sulahian et al. Color scale is shown at bottom. E, Cumulative population doublings (y-axis) over time (x-axis) after treatment with DMSO (black), 10 nM BI-1347 (dotted grey), 3 nM trametinib (solid grey), or both trametinib and BI-1347 (Combo, red) in RAS-mutant lung cancer cell line NCI-H23. Below, a western blot shows phosphorylated ERK (p-ERK) and phosphorylated STAT1 (pSTAT1) levels after treatment with 3 nM trametinib or 10 nM BI-1347 as indicated. Total ERK and STAT1 levels serve as loading controls. F, As in E, but in RAS-mutant pancreatic cell line CFPAC-1.

Discussion

Relapsed neuroblastoma is largely incurable, but the enrichment of genetic alterations in the RAS/MAPK pathway in relapsed tumors suggest that at least a subset of these children could be treated with MEK inhibitors. Single agent MEK inhibitor treatment, however, has not been effective in advanced cancers. To address this problem, we performed an unbiased genome-scale screen to identify potential sensitizers that could be combined with MEK inhibitor-based therapy. These screens uncovered a previously unappreciated role for the mediator kinase module in the adaptive response of neuroblastoma cells to MEK inhibition. The power of this screening approach lies in uncovering unexpected biology; however, one limitation is that CRISPR knockout screens have insufficient power to detect increased efficacy when the single gene knockout substantially reduces viability in the DMSO condition. Importantly, CCNC and CDK8 do not have an impact on cell viability as single knockouts, so they were readily detectable with this approach. There may be genes which dramatically impact cell viability as single knockout or have functionally redundant paralogs that could sensitize cells to trametinib treatment that we did not and cannot identify using this approach.

We find that small molecule inhibition of CDK8 kinase activity is sufficient to improve tumor response to MEK inhibition in vivo, suggesting that this combination could be clinically translatable. A first-generation CDK8/CDK19 inhibitor tested in animal models had significant systemic toxicity, but recent data suggests this was not due to on-target effects and these toxicities can be avoided (63). Indeed, while knockout of murine cdk8 is embryonic lethal, tamoxifen-inducible ubiquitous depletion in adult mice has no overt phenotype (55). Similarly, most human cell lines screened in the Broad Institute’s Dependency Map project and Sanger’s Project SCORE do not depend on CDK8, suggesting that collateral toxicity of inhibiting this target systemically would likely be limited. There are currently no CDK8 inhibitors approved for clinical use, but a first-in-human Phase I trial of SEL-120, a CDK8/19 inhibitor, is on-going in adults with acute myeloid leukemia or high-risk myelodysplastic syndrome (NCT04021368). This trial will yield valuable data on the tolerability of inhibiting CDK8 in humans. We note, however, that neuroblastoma arises in infants and young children and the pediatric toxicity profile could be distinct from that of adults, particularly if CDK8 has a required role in developing tissues.

While we observe improved survival and delayed outgrowth of tumors treated with both MEK and CDK8 inhibitors, we did not observe tumor regression and all animals ultimately succumbed to the disease. Therefore, we do not expect that the combination of MEK inhibitors and CDK8 inhibitors alone will be sufficient for a curative response in humans. However, we note that the efficacy of single agent MEK inhibition was not as profound as has been observed in other studies, perhaps because of differences in the MEK inhibitor, dosing, or mouse strains used (11,12,27). We elected to use selumetinib because it is FDA approved for use in children down to two-years of age. It is possible, however, that the combination effect would be greater with another MEK inhibitor. Additionally, the CDK8 inhibitor we employed is not a clinical candidate, and it is likely that a CDK8 inhibitor optimized for in vivo application would improve response as we did not see complete target inhibition in vivo. CDK8 inhibition has been shown to enhance anti-tumor surveillance of NK cells in other contexts (33,64). There may be extrinsic benefits to CDK8 inhibition in addition to the intrinsic cancer cell response reported here. Importantly, in that study, intermittent treatment with the CDK8 inhibitor was required to avoid a hyporesponsive NK cell state, so alternative dosing strategies of these agents could be explored to better exploit tumor extrinsic effects of CDK8 inhibition, which might improve the therapeutic efficacy of combined MEK and CDK8 inhibition. Our work provides proof of concept that this combination could be effective in vivo, but our data is from a single mouse model that represents a relatively rare subset of neuroblastoma and thus has limitations. Additional studies exploring alternative dosing strategies, using a clinical grade CDK8 inhibitor, and performed in an expanded number of pre-clinical models will be required to clarify whether clinical trials are warranted and to determine the appropriate combination and dosing strategy.

We found that CCNC or CDK8 knockout and MEK inhibition induce opposing transcriptional responses. Rather than enhancing the MEK inhibitor response or inducing new changes in combination that are not observed with either single agent, we observed that CCNC or CDK8 knockout prevents the upregulation of a subset of genes normally upregulated by MEK inhibition while maintaining the downregulation of MAPK signaling. In contrast to other transcriptional CDKs, CDK8 is not required for basal transcription, but is necessary in the context of some stress and developmental stimuli (65–68). This putative role in transcriptional reprogramming in response to stress could explain why in the absence of CDK8 or CCNC we observe a failure of neuroblastoma cells to activate the pro-growth compensatory transcriptional response after MEK inhibition, resulting in delayed progression through the cell cycle. It will therefore be important to determine whether CKD8 inhibitors can be effectively combined with other therapeutic agents where transcriptional adaptation mediates resistance. However, we note that loss of the mediator kinase module has been reported to promote resistance to PARP inhibitors in BRCA2-deficient cells, suggesting that the impact of CDK8 inhibition is likely context dependent (69). More broadly, our data provides robust support for the theory that drugs that induce opposing transcriptional responses can produce additive or synergistic effects on cancer cell viability. This mechanism of increased efficacy mediated by transcriptional antagonism of a pro-survival response has been proposed by others (70) and may be widely applicable to other drug combinations.

Here we demonstrate that the mediator kinase module members CCNC and CDK8 are required for upregulation of pro-survival gene sets, including those involved in cell cycle and elongation after trametinib treatment. Interestingly, Coggins and colleagues identified that YAP1 was regulating similar gene sets in their study where they identified loss of YAP1 as a sensitizer to MEK inhibition, suggesting that these targets could converge on the same downstream pathways (26). Future studies will be required to determine the exact mechanism by which CCNC and CDK8 regulate these gene sets, and whether there is any interaction between the mediator kinase module and YAP1, or if this is an orthogonal mechanism that converges on the same transcriptional output. An additional cellular adaptation to RAF or MEK inhibition that has been described in other settings is the activation of receptor tyrosine kinases (71–76). We did not observe any RTKs as sensitizers in our CRISPR screen, but it is possible that there is a role for RTK activation upstream of CDK8 inhibition that we did not capture here due to functional redundancy between RTKs. It will be important to determine whether the neuroblastoma receptor tyrosine kinome is altered by MEK inhibitor treatment, and if so, whether that contributes to the CDK8 dependency observed here or if it represents a parallel pathway.

In this study, we primarily focused on RAS-mutant neuroblastoma, but our data and that of Sulahian and colleagues suggest that there may be efficacy for this combination in a subset of other RAS-mutant malignancies (62). This effect was not observed in all cell lines screened; a key next step will be to understand what underlies this difference in sensitivity to the combination, and whether any biomarkers can be identified to help stratify which patients might benefit from the addition of a CDK8 inhibitor, and which tumors are unlikely to respond to the combination. Furthermore, in the present study we focused primarily on tumors with RAS pathway activation due to activating mutations in RAS itself, but other alterations in this pathway, including the loss of the RAS-GAP NF1, can cause hyperactivation of this signaling pathway. Our data suggests that the combination may be effective in neuroblastomas with activation of the RAS pathway via different mutational mechanisms and is not restricted to oncogenic RAS mutations. However, the efficacy was not uniform which suggests that some additional factors could be contributing to response, particularly in the setting of ALK mutations. We note that of the ALK mutant lines tested here, KELLY, which is an adrenergic cell line that harbors a MYCN-amplification, did not respond to the combination, while SK-N-SH, which is a mesenchymal cell line that does not have a MYCN-amplification, did respond. While exogenous over-expression of MYCN in an unamplified neuroblastoma cell line did not render resistance, it remains possible that naturally occurring MYCN amplification could impact response to the combination. Further studies will be required to fully elucidate whether MYCN status or mesenchymal/adrenergic cell state modulates the response to the combination, or if other factors explain this difference.

We propose a new treatment strategy for relapsed neuroblastoma harboring RAS-mutations, a tumor type that is generally incurable. We also demonstrate that there could be efficacy for combined MEK and CDK8 inhibition in other cancers, including RAS-mutant pancreatic and lung cancers. Finally, we provide robust evidence for the model that antagonism of a pro-survival transcriptional response can be an effective approach to improve therapeutic response.

Statement of Significance.

Transcriptional adaptation to MEK inhibition is mediated by CDK8 and can be blocked by the addition of CDK8 inhibitors to improve response to MEK inhibitors in RAS-mutant neuroblastoma, a clinically challenging disease.