A revised model of long-term resistance to sustained ERK/MAPK inhibition
We previously assessed long-term ERK/MAPK inhibition across various KRAS-mutant cancers, and consistently observed only transient sensitivity, followed by acquired resistance that develops over several weeks15 (Figure S1A, B). Novel inhibitors of RAF and MEK, despite their proposed mechanisms, nonetheless appear vulnerable to traditional pathway reactivation2,29,30. Alternatively, while dual-mechanism ERK inhibition also activates well-described endogenous feedback events, compensatory activation of MEK and ERK ultimately abates on the timescale at which stable resistance develops (Figure S1C), suggesting that distinct events permit resistance to more sustained ERK/MAPK inhibition.
To confirm these findings in a more translationally relevant system, we utilized a well-credentialed, KrasG12D/Tp53-/- orthotopic, syngeneic mouse model of pancreatic cancer31 to assess resistance to dual-mechanism ERK inhibition (Figures 1A-C). Tumors demonstrated initial drug sensitivity, yet began to progress within two weeks of treatment initiation (Figure 1A). Initial drug sensitivity was characterized by ineffective pathway reactivation, evidenced by increased MEK phosphorylation at residues S217/S221 without commensurate recovery of cell proliferation. Conversely, as stable resistance was achieved, phosphorylation of MEK and ERK was relinquished, coupled by cell cycle resumption of tumor proliferation (Figure 1B,C). We observed similar findings in additional in vitro models of KRAS-mutant pancreatic and lung cancers (Figure S1D). Notably, these signaling events differ from those induced by first generation, single-mechanism ERK inhibitors, where resistance is characterized by hyperphosphorylation of ERK at residues T202/Y204 (Figure S1E). These findings suggest that while sustained ERK inhibition is susceptible to adaptive resistance, the programs driving ultimate resistance are qualitatively and temporally distinct from traditional pathway reactivation. Moreover, these observations provoke questions regarding the long-term role of MAPK feedback events, which to date have been described on the scale of hours to days2,5.
To more comprehensively delineate the impact of long-term ERK/MAPK blockade on signaling networks, we utilized reverse phase protein array (RPPA)32 to serially profile KRAS-mutant pancreatic cancer cells exposed to dual-mechanism ERK inhibition over a time course ranging from one hour to eight weeks (Figure 1D, Table S1). This array probed diverse RTKs and their associated survival pathways, as well as markers of translational control, pro- and anti-apoptotic regulation, cell cycle control, cytoskeletal dynamics, autophagy, transcription factor activation, and histone modifications. Unsupervised clustering revealed three distinct patterns of protein expression characterizing acquired resistance. Cluster 1 was upregulated during an initial period of stunted growth, but then downregulated as stable resistance developed, and was highly enriched for nodes within the MAPK and PI3K/AKT/mTOR signaling pathways. Cluster 2 demonstrated reciprocal downregulation during the initial period of drug sensitivity, followed by a delayed return to baseline expression as resistance developed; this cluster was enriched for markers of cell cycle entry and cap-dependent mRNA translation—which represent an established convergence point of integrated ERK/MAPK and PI3K/AKT/mTOR signaling33—and the asymmetric expression patterns between these clusters suggest that intrinsic feedback events are rendered ineffective in the setting of sustained pathway inhibition. Cluster 3 contained most RTKs and their alternative downstream signaling proteins, which were generally unperturbed by initial drug exposure, and only modestly altered as resistance emerged. Taken together, these findings suggest that pathway reactivation is an endogenous response to ERK/MAPK blockade that may support early survival, but is insufficient to confer stable resistance to sustained inhibition. Furthermore, among diverse alternative signaling pathways, none obviously replaced MAPK signaling to drive growth.
Recent reports have postulated that kinase inhibition induces complex changes to the transcriptional and enhancer landscape permitting a drug-tolerant state, and that early changes (within one week) reflect the necessary adaptations for stable resistance16,17. However, we have consistently observed stunted cell growth beyond this period, and have found that outgrowth is a gradual rather than immediate process during which cell populations display increasing fitness as stable resistance evolves.15,34 This suggests that additional transcriptional evolution may be required to permit the terminally resistant state. In fact, our RPPA analysis demonstrated that all histone markers included in our panel underwent dynamic changes throughout the adaptive resistance process (Figure S1F), further supporting longer-term transcriptional reprogramming as a necessary component of the terminally resistant phenotype.
To test this hypothesis, we performed RNA sequencing in parallel with chromatin immunoprecipitation-DNA sequencing (ChIP-seq) for acetylated histone 3, lysine 27 (H3K27ac) in treatment-naïve KRAS-mutant pancreatic cancer cells, as well as at one week of dual-mechanism ERK inhibition and following the development of stable resistance. H3K27ac represents a histone modification that is associated with transcriptional activation and marks active enhancers; thus, changes in H3K27ac density may broadly signify epigenomic remodeling as cells adapt to environmental stress35.To that end, we found that the majority of of H3K27ac peaks gained or lost in stable resistance differed from those within the early response, contrasting with previous models suggesting that enhancer remodeling plateaus within 72 hours of drug exposure16,17 (Figure 1E, Table S2).And while RNA sequencing revealed a broad early transcriptional response—including well established transcription factors (e.g. EGR1, JUN, FOS), MAPK regulators (e.g. SPRY1/2/4, SPRED1/2), and pro-survival genes in NF-κB/interferon, TGF-β, and alternative tyrosine kinase families16,17 (Figure S1G)—extensive further transcriptional changes were present in stably resistant cells, and the intersection of dysregulated transcripts between these time points was quite limited (Figure 1F, Table S3). This pattern was consistent among even the most highly up- or downregulated transcripts, including a subset of gene expression changes with opposing directionality at the two time points (Figure S1H). Gene set enrichment analysis (GSEA) further confirmed that there were very few enriched pathways shared between the early response and stable resistance (Figure S1I, Table S4). Taken together, these findings demonstrate an adaptive transcriptional process that is buttressed by parallel remodeling of active enhancers, and that counter to prior models, achieving stable resistance may require chromatin and gene expression changes that evolve beyond the intrinsic early response to drug exposure. This model is also consistent with recent work proposing that even treatment-naïve cells upregulating key resistance markers require longer-term transcriptional adaptations to cultivate stable resistance36.
Transcriptionally-mediated resistance programs are broadly vulnerable to manipulation of the epigenetic machinery
To confirm the generalizability of this long-term transcriptional response, we developed and profiled two additional models of evolved resistance to dual-mechanism ERK inhibition in KRAS-mutant lung and colon cancer cells. These cells similarly underwent broad early transcriptional changes followed by extensive further adaptations during the development of stable resistance (Figure 2A, Table S3). Notably, the transcriptional programs giving rise to stable resistance demonstrated limited overlap between cells of different tissue origin (Figure 2B). Correspondingly, a complete absence of enriched or suppressed GSEA pathways was observed across resistant models (Figure 2C, Table S4). To test whether these differences were due simply to tissue type, we developed stable resistance in an additional KRAS-mutant pancreatic cancer cell line; indeed, the intersection of differentially expressed genes was no greater among the pancreatic cancer lines (Figure S2A, Table S3), nor were the associated gene annotations using GSEA (Figure S2B, Table S4). Collectively, these findings demonstrate that diverse, KRAS-mutant cancer cells undergo large-scale transcriptional changes during sustained ERK/MAPK inhibition, yet the transcriptional programs associated with terminal resistance are heterogenous and model-specific.
Given the obvious challenge of developing strategies to target diverse terminal resistance programs, we instead sought to test whether combining ERK/MAPK inhibition with drugs targeting the epigenetic and transcriptional machinery might broadly perturb heterogeneous responses to drug exposure. To accomplish this, we performed long-term pharmacologic screens assessing a panel of drugs targeting diverse transcriptional processes on the development of resistance to second-generation RAF, MEK, and ERK inhibitors.
In total, we screened 42 combination therapies (Figure 2D, Table S5), revealing, most strikingly, that resistance to sustained ERK inhibition was broadly susceptible to manipulation of the transcriptional machinery, including drugs co-targeting CREBBP/EP300, EZH1/EZH2, BET family bromodomain proteins, and CDK8/19. In keeping with our previous findings of pathway reactivation driving resistance to RAF and MEK inhibition, these targets were generally less vulnerable to combination therapy. We found that only the BET bromodomain inhibitor JQ-1 and the CDK8/19 inhibitor Senexin A delayed resistance to inhibition at all three ERK/MAPK nodes, with the strongest effect in combination with ERK inhibition. The effect of BET bromodomain inhibition on drug response has been broadly reported34,37-39. CDK8 and CDK19 are paralogous proteins that reversibly associate with the multiprotein Mediator complex (Mediator kinase)40, and have never, to our knowledge, been implicated in resistance to ERK/MAPK inhibition. CDK8/19 inhibition alone had no effect on basal growth (Figure S2C), nor did it demonstrate short-term synergy with ERK/MAPK inhibition (Figure S2D), suggesting that the impedance of resistance occurred through later-stage inhibition of the adaptive process. Given the profound effect of Mediator kinase inhibition on the activity of drugs targeting all three ERK/MAPK nodes, we chose to characterize this interaction across the KRAS-mutant tissues types for which we had developed models of transcriptionally-mediated resistance.
Mediator kinase inhibition impedes long-term acquired resistance to ERK/MAPK-targeted therapy
We first validated the interaction of combined ERK/MAPK and Mediator kinase inhibition using three structurally distinct preclinical compounds (Senexin A, Cortistatin A, and CCT251545) in long-term colony forming assays at doses shown to selectively inhibit CDK8/1941-43. All three compounds profoundly inhibited clonal outgrowth during sustained treatment with RAF, MEK, and ERK inhibition, and had minimal effect on basal cell growth (Figure 2E). Most notably, combined ERK and CDK8/19 inhibition completely prevented the emergence of resistant colonies at four weeks, supporting a dominant role for Mediator kinases in transcriptional reprogramming during sustained ERK inhibition.
As our work to this point relied exclusively on pharmacologic inhibition of CDK8/19, we next sought to probe the mechanistic role of each Mediator kinase. CDK8 and CDK19 possess enzymatic activity, but the proteins also serve scaffold functions44, and thus protein depletion and kinase inhibition have distinct cellular effects45,46. To test whether pharmacologic inhibition was in fact specific to Mediator kinase function, we utilized dual CRISPR/Cas9 constructs47 to create knockout derivatives of both CDK8 and CDK19, CDK8 only, CDK19 only, or double-sham control knockouts in three KRAS-mutant cancer cell lines. Each condition was then subjected to treatment with DMSO, ERK inhibition, CDK8/19 inhibition, or combined ERK and CDK8/19 inhibition (Figure 2F, Figure S2E,F). As expected, in basal growth conditions treatment with CDK8/19 inhibition had a limited effect in all four derivatives. In the presence of ERK inhibition, however, CDK8/19 inhibition profoundly suppressed outgrowth of control cells expressing both CDK8 and CDK19, modestly blocked growth in cells with individual knockout of CDK8 or CDK19, yet had no effect in cells depleted of both CDK8 and CDK19 (Figure 2F). This confirmed that CDK8 and CDK19 were indeed the targets of kinase inhibition responsible for the long-term impedance of resistance.
Finally, to evaluate the effect of CDK8/19 inhibition on sustained MAPK suppression within a more clinically meaningful timescale, we utilized an established time-to-progression model15,34,48 to test growth up to eight weeks. Despite the diverse transcriptional resistance programs observed across tissue types, CDK8/19 inhibition completely prevented lung and colon cancer cells from developing resistance to ERK inhibition (Figure 2G, bottom), and markedly delayed the emergence of resistance in pancreatic cancer cells. Of note, combined ERK and CDK8/19 inhibition demonstrated no short-term synergy in any cell line (Figure 2G, top), and CDK8/19 inhibition alone had a negligible effect on long-term basal growth (Figure S2G), positioning this combination as a promising strategy for preventing long-term acquired resistance in KRAS-mutant cancers.
Antagonization of a conserved response network paralyzes further adaptive potential
Given that KRAS-mutant cancer cells of distinct tissue origin demonstrated distinct stable resistance programs to ERK inhibition, yet that these programs were universally susceptible to CDK8/19 co-inhibition, we next asked whether there might be a common early response to ERK inhibition vulnerable to combination treatment. In fact, we found that in stark contrast to the limited intersection of expression changes seen in stable resistance (Figure 2B), considerable overlap could be appreciated between cell lines of distinct tissue origin following one week of ERK inhibition. This suggested that ERK inhibition induces a conserved early transcriptional response before cells undergo further heterogeneous adaptations that ultimately establish stable resistance programs. This was further revealed by GSEA (Figure 3B), which demonstrated substantial overlap of enriched gene annotations between tissue types within the early response, again contrasting with the lack of shared pathways in stable resistance (Figure 2C). Broadly, GSEA revealed that this initial response resulted in downregulation of major anabolic processes such as DNA replication, cell cycle entry, and protein translation (Table S4), which may be necessary for cells to forego growth and replication as unique transcriptional programs are enacted.
We next asked whether this conserved response to ERK inhibition was particularly vulnerable to combined CDK8/19 inhibition. Specifically, we tested whether CDK8/19 co-inhibition preferentially dysregulated the genes identified within this early conserved response network, or whether CDK8/19 inhibition agnostically caused gene expression changes throughout the transcriptome. We found that genes dysregulated by ERK inhibition were significantly more likely to be further altered by combined CDK8/19 inhibition (Figure 3C). We next asked whether this interaction occurred with specific directionality. A nonspecific effect should cause limited expression changes with near-random directionality; alternatively, cooperation or antagonism of a transcriptional program should preferentially amplify or dampen expression changes. Strikingly, we found that CDK8/19 co-inhibition broadly antagonized the expression changes within the early conserved response (Figure 3D). Notably, despite this highly specific and organized effect, the magnitude of antagonization tended to be modest (Figure S3A). Moreover, at this early time point, cells treated with ERK inhibition alone or in combination with CDK8/19 inhibition were phenotypically similar, with equivalent growth inhibition in each treatment condition (Figure S3B). This led us to ask whether the ultimate consequence of Mediator kinase co-inhibition was downstream impairment of further transcriptional adaptations necessary for stable resistance.
To do this, cells treated with ERK inhibition alone or combined with CDK8/19 inhibition were serially profiled by RNA sequencing over the five-week time course during which resistance developed (Table S6). Notably, we found that individual transcripts underwent dynamic alterations following the early conserved response, including thousands of genes found to be either significantly upregulated and downregulated at different time points throughout this evolved process (Figure S3C). Notably, transcriptome-wide differences between treatment conditions became increasingly pronounced over time (Figure S3D), further suggesting that the phenotype induced by CDK8/19 co-inhibition was caused by disruption of this downstream evolutionary process.
In order to visualize, quantify, and evaluate patterns of whole-transcriptome evolution over time, we utilized an established Dirichlet process Gaussian process (DPGP) mixture model, which facilitates time series cluster measurement of genomic features49. Applying this model to the genes with the greatest variance over the course of acquired resistance, we identified 67 gene sets reflecting diverse expression trajectories (Figure 3E; Table S7). Like the gene-level analysis between treatment conditions, cluster trajectories between treatment conditions demonstrated overall divergence after the early conserved response (Figure S3E). By quantifying and comparing all cluster-level interval changes, we found that Mediator kinase inhibition exerted its dominant effect following this early response, paralyzing transcriptional reprogramming between weeks one and three of treatment, at which point dually-treated cells resumed a trajectory that mirrored cells treated with ERK inhibition alone (Figures 3F-H). By visualizing individual clusters and then shifting the timescale of dually-treated cells to “remove” this period of transcriptional stagnation, the trajectory curves superimposed upon one another (Figure 3F), and we ensured that all cluster trajectories could be realigned with this time shift (Figures 3G,H). We further extrapolated these findings to the entire transcriptome at both the gene-level and using GSEA (Figures S3F-H). Taken together, these findings indicate that the emergence of resistance coincides with transcriptional escape from the early conserved response, a concept that can be observed at the transcript level, using functional annotated gene sets, and using a DPGP mixture model. Co-targeting Mediator kinase therefore antagonizes the conserved response to ERK inhibition, and then exerts its phenotypic effect by preventing further transcriptional changes necessary to establish stable resistance.
Transcriptionally-mediated resistance is driven by distinct terminal mechanisms in different models
Given the magnitude and heterogeneity of gene expression changes observed in each resistant model (Figure 3A,B), yet the ability of CDK8/19 inhibition to broadly curtail the emergence of resistance (Figure 2G), we sought to map out a specific resistance mechanism in one model. To date, efforts to characterize the function of individual components of a transcriptional program have been limited by the scalability of candidate-based approaches. We considered it unrealistic to implicate the functional importance of individual genes based simply on expression changes, as some highly dysregulated transcripts were likely secondarily regulated passengers of alterations to the chromatin architecture, vestigially over- or underexpressed following the early response, or simply dysregulated at random.
In order to test the consequence of gene-level transcriptional events, we designed a specialized loss-of-function library of CRISPR/Cas9 constructs targeting the subset of genes most dysregulated throughout the adaptive process in MIA PaCa-2 cells (Figure 4A, Table S8). We then subjected treatment-naïve and evolved resistant cells transduced with this library to five weeks of either ERK inhibition or control conditions to test the effect of gene-loss at various time points. We validated our approach by comparing final and initial sgRNA pools from treatment-naïve parental cells across replicates, focusing on known essential genes, non-essential genes, and non-targeting controls (Figure 4B). These results revealed that while extensive transcriptional changes were observed as a stably resistant phenotype was established, the vast majority of these genes were not independently necessary for the maintenance of the resistant state, as their knockout failed to re-sensitize resistant cells to ERK inhibition (Figure 4C, Figure S4A). Among the subset of genes demonstrating functional importance in the terminally resistant state, the knockout of one gene upregulated in resistant cells —encoding the ATP-binding cassette protein ABCG2 that functions as notorious multidrug transporter with broad polysubstrate specificity50—dramatically sensitized resistant cells to ERK inhibition (Figures 4C, D). ABCG2-mediated resistance was extensively validated by demonstrating its reproducible upregulation, and subsequently that both gene knockout and pharmacologic inhibition of its ATPase function fully reversed ERK inhibitor resistance, while sensitizing cells to treatment with other known ABCG2 substrates (Figures 4E-H). ABCG2 loss demonstrated no functional consequence in basal growth nor in resistant cells when drug was removed (Figure 4D (insert), Figures S4A, B), confirming it as a bona fide driver of drug resistance in this model. To ensure that ABCG2-upregulation was not simply occurring via the outgrowth of small population of ABCG2 high-expressing “persister” cells, we demonstrated that the vast majority of individual cells were capable of developing resistance across a consistent timescale (Figure S4C). The reproducible resistance mechanism observed in MIA PaCa-2 cells (Figure 4E) was distinct from KRAS-mutant colon and lung cancer models, which evidenced neither ABCG2 upregulation (Figure S4D) nor resensitization to ERK inhibition by pharmacologic ABCG2 inhibition (Figure S4E). Collectively, these findings demonstrate that transcriptional ABCG2 upregulation reproducibly drives acquired resistance to ERK inhibition in MIA PaCa-2 cells, but that differing mechanisms drive resistance in other cell line models, highlighting the notion that strategies targeting transcriptional adaptation are likely to be more broadly effective at blocking resistance than strategies targeting discrete gene expression changes.
Co-targeting Mediator kinases delays resistance to ERK inhibition across translational models of KRAS-mutant cancers
To assess the translational generalizability of combined ERK and Mediator kinase inhibition, we evaluated this strategy across multiple in vitro and in vivo models of KRAS-mutant cancer, including additional eight-week time-to-progression models, patient-derived rectal cancer tumoroids, and a pancreatic cancer mouse model.
We first expanded upon the existing time-to-progression model by testing combined ERK and CDK8/19 inhibition across diverse KRAS-mutant pancreatic, lung, and colorectal cancer cell lines. In all models, cotreatment with CDK8/19 inhibition delayed the emergence of resistance, in many cases completely preventing resistance until the assay was terminated at eight weeks (Figure 5A, S5A). Again, combined ERK and CDK8/19 inhibition demonstrated no short-term synergy in any cell line (Figure S5B), and CDK8/19 inhibition alone had a negligible effect on long-term basal growth (Figure S5C). We also tested two lines with distinct oncogenic drivers (both KRAS wildtype; one BRAF mutant melanoma (A375) and one EGFR driven colorectal cancer (LIM1215)) with respective genotype-specific targeted inhibitors. In these two models, CDK8/19 inhibition had no effect on the emergence of resistance (Figures 5A, S5A). This suggests that either resistance in these model systems is not driven by transcriptional reprogramming, or that the transcriptional reprogramming occurring in these settings is not CDK8/19-dependent. Either way, it underscores that across KRAS-mutant model systems of pancreatic, lung, and colorectal cancer, resistance to dual-mechanism ERK/MAPK inhibition occurs secondary to large-scale transcriptional adaptations which rely on Mediator kinase. Whether CDK8/19 inhibition serves as an effective combination strategy in other genomic contexts where resistance develops via large-scale, long-term transcriptional adaptations requires further investigation..
Next, we tested the combination of ERK and CDK8/19 inhibition in three KRAS-mutant, patient-derived tumoroid models of rectal cancer (Figure 5B-D). These tumoroids had previously been shown to retain the molecular, histological, and clinical features of the patient tumors from which they were derived, as well as accurately predict patient responses to chemotherapy and radiation therapy53. Like the cell line time-to-progression models, tumoroids developed resistance to ERK inhibition within five weeks of treatment (Figure 5C). Alternatively, tumoroids treated with combined ERK and CDK8/19 inhibition demonstrated complete growth suppression up to eight weeks (Figure S5D). Again, inhibition of Mediator kinase alone had no effect on tumoroid growth (Figure 5D).
Finally, we tested the combination of ERK and CDK8/19 inhibition in a well-credentialed, orthotopic, syngeneicmouse model of pancreatic cancer. We first tested three independent CDK8/19 inhibitors in vitro using a cell line derived from the same model. Cell growth was unaffected by each inhibitor alone, yet resistance to ERK inhibition was markedly impeded by co-inhibition of Mediator kinase, with similar efficacy between compounds (Figure 5E). Of these three inhibitors, we selected CCT251545, a selective inhibitor of CDK8/1942,54 for in vivo studies (Cortistatin A is not yet commercially available, and Senexin A has known limited oral bioavailability). Treatment with ERK inhibition alone dramatically reduced initial tumor growth, yet resistance to treatment developed within two weeks of drug exposure (as in Figure 1A). In contrast, inhibition of ERK and CDK8/19 delayed the acquisition of resistance (Figure 5F), resulting in a 34% reduction in tumor weight at study endpoint (Figure 5G) without additional toxicity (Figure 5H). As in the cell line and tumoroid models, inhibition of Mediator kinase alone had no effect on in vivo tumor growth or overall toxicity (Figures S5E-G), positioning CDK8/19 co-inhibition as an effective and well-tolerated strategy for preventing resistance to sustained MAPK inhibition.