Targeting CDK4 overcomes EMT-mediated tumor heterogeneity and therapeutic resistance in KRAS mutant lung cancer


 Lack of sustained response to therapeutic agents in patients with K-Ras mutant lung cancer poses a major challenge and arises partly due to intratumor heterogeneity that defines phenotypically distinct tumor subpopulations. To attain better therapeutic outcomes it is important to understand the differential therapeutic sensitivities of tumor cell subsets. Epithelial-to-mesenchymal transition (EMT) is a biologic phenomenon that can alter the phenotypic state and cause transcriptional rewiring to produce distinct tumor cell populations. We utilized functional shRNA screens, in vitro and in vivo models to identify and confirm an increased dependence of mesenchymal tumor cells on CDK4 for survival, as well as a mechanism of resistance to MEK inhibitors. High ZEB1 levels in mesenchymal tumor cells repressed p21, leading to perturbed CDK4 pathway activity. Increased dependence on CDK4 rendered mesenchymal cancer cells particularly vulnerable to selective CDK4 inhibitors. Co-administration of CDK4 and MEK inhibitors in heterogeneous tumors effectively targeted different tumor subpopulations, subverting the resistance to either single agent treatment.


Introduction
Activating KRAS mutation is one of the most frequent oncogenic events in lung cancer, occurring in about 30% of lung adenocarcinoma patients (1)(2)(3). Despite the identification of the oncogene over 20 years ago and significant efforts to treat this subset of patients, 5-year survival rates remain dismal (4). Unlike EGFR mutant lung cancer, KRAS oncoproteins are largely undruggable, with the very recent exception of the KRAS G12C allele (5,6). Pharmacological inhibitors of the MAPK pathway (e.g., MEK), such as selumetinib and trametinib are available, but preclinical and clinical trials have demonstrated poor responses to MEK inhibitors (7).
Combination of MEK inhibitors with conventional chemotherapy did not demonstrate any added benefit to progression free survival (8). Resistance to MEK inhibitors may be intrinsic (de novo) due to tumor cell heterogeneity or acquired due to tumor evolution as an adaptive response to pharmacological agents. In either case, the presence of phenotypically distinct tumor cell subpopulations with reprogrammed cellular machinery makes it difficult to effectively eliminate the broader tumor cell population. To address this, we need to understand the differences in the tumor cell subpopulations within a heterogeneous tumor.
Genetically identical tumor cells possess the ability to undergo transcriptional reprogramming to activate alternate survival pathways and evade therapeutic targeting. Research from our group and others has demonstrated that epithelial-to-mesenchymal transition (EMT) is a central phenomenon occurring in KRAS mutant lung cancer, which contributes to intracellular tumor heterogeneity, increased metastatic potential, therapeutic resistance to pharmacological agents and poor patient outcomes (9)(10)(11). Murine lung cancer models driven by Kras and p53 mutations recapitulate EMT-mediated tumor cell heterogeneity with the ZEB1/miR-200 double negative feedback loop playing a central role in dynamically altering the cellular phenotype (10).
Our previous research highlighted the reliance of KRAS mutant epithelial-like lung cancer cells with high miR-200 expression on activated MAPK signaling pathway and increased susceptibility to MEK inhibitors. On the other hand, tumors demonstrating a mesenchymal-like phenotype due to elevated expression of ZEB1 remained largely unresponsive to MEK inhibitors. Moreover, after an initial response to MEK inhibition, epithelial-like tumors acquired therapeutic resistance by undergoing EMT (11,12). The study identified an unmet need to develop therapeutic approaches to target distinct tumor subpopulations within heterogeneous KRAS mutant lung tumor to achieve a robust therapeutic response. Utilizing multiple loss-of-function shRNA screens, we analyzed the effects on phenotypically different tumor subpopulations and identified CDK4-RB as a major survival pathway in mesenchymal-like tumor cells.
CDK4 acts as a master integrator of mitogenic/oncogenic signaling cascades by inactivating the central tumor suppressor RB and cell cycle commitment at the restriction point allowing cells to transition to S phase (13). The CDK4 axis is altered in many cancers, with clinically approved pharmacologic inhibitors showing promising antitumor activity (14). Some studies have shown that CDK4 and cyclin D1 expression is correlated with the presence of KRAS mutation in lung tumors (15) and a synthetic lethal interaction occurs between KRAS and CDK4 in lung cancer tumor progression (16,17). We found that the differential activation of the CDK4 pathway in epithelial-like and mesenchymal-like cells was determined by ZEB1-mediated p21 regulation. As an intrinsic regulator of CDK4, p21 levels in cells determine the downstream CDK4 pathway activity. p21 is transcriptionally regulated by direct binding of the transcription factor ZEB1 to the promoter region. Our study demonstrates in multiple preclinical models that intrinsic and acquired MEK inhibitor resistance is associated with a rewired kinome in tumors by which the mesenchymal phenotype activates the CDK4 pathway as a common occurrence across models.
This dependence on the CDK4 pathway resulted in a potential therapeutic approach to combine MEK and CDK4 inhibitors to target different tumor subpopulations along the EMT spectrum and combat resistant outgrowth of epigenetic subsets in a heterogeneous tumor.

Mesenchymal lung cancer cells exhibit increased dependency on CDK4 for growth
In order to effectively target the mesenchymal tumor subpopulations within heterogeneous tumors, we sought to identify the survival dependencies of these tumor cells. A loss of function screen with a barcoded, pooled small hairpin RNA (shRNA) library targeting about ~500 genes with known kinase activity (Kinome) was conducted. Each gene was targeted with 10 unique shRNA sequences to limit false hits due to off-target effects. This library of shRNAs was transduced into representative non-metastatic, epithelial-like (393P) and metastatic, mesenchymal-like (344P) murine lung cancer cell lines derived from a previously described Kras/p53 mutant (KP) genetically engineered mouse model (GEMM) (10). The cell lines stably expressing the shRNAs from the Kinome library were either cultured in vitro or implanted subcutaneously in nude mice ( Figure 1A). Tumors were harvested, shRNA barcodes were quantified by deep sequencing and referenced with the respective in vitro cell population and quality control measures were completed to ensure sufficient barcode coverage across the library was maintained in vivo ( Figure S1A, Table S1). The phenotypic impact of gene knockdown was inferred by the redundant shRNA activity (RSA) algorithm, where a lower rank of the shRNA barcodes signified dropout from the population and greater dependency on the gene for tumorigenesis (Table S2). Although both cell line models have activating Kras G12D and p53 R172H mutations, comparison of the results of the Kinome screen revealed that the mesenchymal-like cells (344P) and the matched syngeneic tumors were more reliant on Cdk4 for in vitro and in vivo growth ( Figure 1B). We also compared these results to our previously published FDAome shRNA screen (11) and identified Cdk4 as the most consistent hit across in vitro and in vivo conditions in both screens ( Figure 1B, S1A, Table S3, Table S4). CDK4 mRNA expression showed a positive correlation with a previously reported 76-gene EMT signature (18) in 118 human NSCLC cell lines ( Figure 1C). When sub-classified based on mutational status, 41 KRAS-mutant NSCLC cell lines also showed positive correlation of CDK4 mRNA with the EMT signature ( Figure S1B). A panel of epithelial-like and mesenchymal-like murine lung cancer cells were tested and the mesenchymal-like cells demonstrated higher Cdk4 mRNA levels ( Figure S1C). We employed a genetic approach to confirm the dependency of mesenchymal-like tumor cells on CDK4 for survival. Mesenchymal-like cells with an inducible shRNA targeting CDK4 showed a greater reduction in tumor cell growth ( Figure S1D,E), with suppression of phospho-RB ( Figure 1D) compared to epithelial-like 393P cells. In fact, 393P tumor cells appeared to have slightly greater growth rate with CDK4 knockdown than the control cells ( Figure S1E) and continued RB phosphorylation ( Figure 1D). We did not observe a significant difference in baseline proliferation rate between 393P and 344SQ murine lung cancer cells ( Figure S1F).
Next, we functionally validated the shRNA screen and determined whether response to CDK4 inhibitors is dependent on the EMT status of tumor cells. We treated a panel of human and murine lung cancer cells, stratified as epithelial-like or mesenchymal-like based on previous profiling (10,18), with CDK4 inhibitors. Both human and murine mesenchymal-like lung cancer were more sensitive to CDK4 inhibitors (palbociclib, abemaciclib and ribociclib) ( Figure 1E, Table   1). As previously noted (10), EMT status is tightly regulated by the ZEB1/miR-200 double negative feedback loop and manipulation of this axis can induce an epithelial or mesenchymal shift in tumor cells. We therefore utilized isogenic pairs of human (H441) and murine (393P) epithelial-like cell lines with ZEB1 expression to produce a mesenchymal phenotype (19) and isogenic pairs of human (H1299, H23) and murine (344SQ) mesenchymal-like cells with miR-200 expression or ZEB1 knockdown to push the cells to an epithelial state (20). Comparisons across the different cell line pairs revealed that sensitivity to the CDK4 inhibitors was determined by the EMT status ( Figure S2A-C, Table 1).
The downstream targets of CDK4, RB and FoxM1, are important readouts for CDK4 kinase activity, whereas phospho-CDK4 may continue to be present for another 24 hours post inhibitor treatment. Suppression of RB and FoxM1 was observed in mesenchymal-like cells upon treatment with abemaciclib or ribociclib for 24 and 48 hours ( Figure S2D). 393P cells showed an initial suppression of CDK4 targets, but it was not a sustained response. 344SQ cells showed a more robust response to the inhibitor palbociclib over a range of concentrations and at shorter treatment times compared to 393P in terms of suppression of downstream signaling and induction of apoptosis ( Figure 1F).

The CDK4 pathway is dynamically regulated by the EMT status of tumor cells
Immunofluroscent staining of tumor cells demonstrated an activated CDK4/RB axis with a higher percentage of 344SQ tumor cells with positive nuclear staining for phospho-CDK4 and phospho-RB and a stronger staining for total CDK4 (Figure 2A,B). Using reverse phase protein arrays (RPPA) to analyze changes in cell signaling proteins in a high-throughput manner, we screened a panel of previously characterized isogenic murine epithelial-like and mesenchymallike lung cancer cell lines and observed an increase in CDK4 axis-related molecules, phospho-RB and Cyclin D1 in cells with a mesenchymal phenotype ( Figure 2C). A subcellular fractionation assay also showed higher levels of phospho-RB, Cyclin D1 and CDK4 in mesenchymal-like cells ( Figure S3A). We next tested the effects of altering the EMT status of tumor cells on the CDK4 signaling pathway using the previously described isogenic cell line pairs. ZEB1 overexpression in H441 and 393P cells produced higher levels of CDK4 and phospho-RB ( Figure 2D). Conversely, miR-200 expression in H1299 and 344SQ cells caused a suppression of the CDK4 axis ( Figure 2E).
Immunohistochemistry on 344SQ syngeneic tumors revealed higher phospho-CDK4 and phospho-RB staining with absent phospho-Erk, and the reverse was observed in 393P syngeneic tumors ( Figure 2F). Additionally, we have previously observed that epithelial-like 393P tumors initially respond to MEK inhibitors, however, long-term exposure produces acquired resistance with the acquisition of a mesenchymal phenotype (393P-AZD R ) (11). 393P-AZD R tumors also showed higher phospho-CDK4 and phospho-RB staining, with suppressed phospho-Erk, an observation similar to the de novo 344SQ mesenchymal-like tumors ( Figure 2F). Cell lines derived from 393P-AZD R tumors showed higher phospho-CDK4 and ZEB1 expression, with generally lower levels of phospho-Erk ( Figure S3B). Resistant cells were no longer sensitive to AZD6244, and instead became sensitive to palbociclib with an IC50 similar to 344SQ cells ( Figure 2G), and greater suppression of phospho-RB and phospho-CDK4 in 393P-AZD R than in 393P-vehicle cell lines. In contrast there was an accumulation of phospho-CDK4 in 393P-vehicle cells ( Figure S3C). We next tested if there were phenotypic differences in the manner which epithelial-like and mesenchymal-like cell lines undergo cell cycle progression. Upon serum starvation for up to 48 hours, 393P cells almost completely (~90% of the cells) arrested in the G0-G1 phase of the cell cycle with a complete suppression of the CDK4 pathway ( Figure S3D,E). In contrast, 344SQ cells resisted cell cycle arrest in serum-free conditions, with ~80% of the cells in G0-G1 state but 20% of cells continuing to cycle through S or G2/M ( Figure S3D). This observation corresponded to higher levels of CDK4, Cyclin D1 and phospho-RB in the cells when assayed by subcellular fractionation ( Figure S3E), suggesting that CDK4 activity in mesenchymal tumor cells could be uncoupled from extrinsic mitogenic signals. After release of cells from the arrested state by addition of serum-containing media, 344SQ cells transitioned into S phase more readily (within 20 hours) than 393P cells, which remained arrested in G1 phase up to 36 hours before returning to the baseline cycling state ( Figure S3D). Although cell cycle arrest in G1 phase with palbociclib was essentially similar between 393P and 344SQ cells, a significant increase in the percentage of 344SQ cells in apoptotic cells was detected ( Figure S3F,G), corresponding to increased cleaved caspase-3 ( Figure 1F). We conclude from the above findings that tumor cells with a mesenchymal-like phenotype, either due to intrinsic factors or arising from epithelial cells undergoing EMT as an adaptive resistance mechanism, have rewired survival pathways to activate CDK4 signaling that is independent of mitogenic signals.

ZEB1 regulates p21 expression and causes differential CDK4 pathway activation
To identify the mechanistic basis of the differential dependency on the CDK4 pathway between the phenotypic epithelial and mesenchymal cancer cells we investigated the canonical upstream survival pathways, including AKT, PIK3CA, MAPK11, and FGFR, but did not observe differential regulation between epithelial-like and mesenchymal-like cells. We then focused on the intrinsic regulators of CDK4 activity, p21 (WAF1/CIP1) and p27 (KIP1) and transiently knocked down each in epithelial-like and mesenchymal-like cells to determine the effects on the CDK4 pathway ( Figure S4A). p21 knockdown had a more significant impact on the phosphorylation of CDK4 and RB compared to p27 ( Figure S4B). The much higher phosphorylation of RB in 393P cells with p21 knockdown indicated that p21 maintains a check on the CDK4-RB pathway in the epithelial-like cells and when disrupted activates CDK4 pathway. Loss of p21 in mesenchymallike cells only modestly increased phosphorylation of RB compared to the control cells, suggesting that an intrinsic deficiency of p21 protein in the mesenchymal cells could lead to a dysregulated CDK4 pathway. Immunofluorescence assays on 393P, 393P-AZD R and 344SQ cells revealed that a higher percentage of epithelial-like cells had nuclear p21 than mesenchymal-like cells, along with higher co-localization of CDK4 and p21 in epithelial-like tumor cells ( Figure S4C,D).
Alterations in the tumor suppressor TP53 is one of the most commonly occurring co-mutation events in KRAS driven lung cancer and p21 is a direct target of p53. Therefore, we investigated if there was any effect of p53 on CDK4 pathway in the epithelial-like and mesenchymal-like tumor cells. With transient knockdown of p53, there was no significant difference in downstream CDK4 signaling ( Figure S4E,F). We also utilized previously published Kras G12D mutant (K1) and Kras G12D /p21 -/-(KC3 and KC4) murine tumor cells (21). Absence of p21 in tumor cells sensitized KC cell lines to palbociclib with an increase in CDK4 signaling ( Figure S4G,H), emphasizing that p21-mediated CDK4 dysregulation was independent of p53 control.
Next, we inquired whether EMT status of tumor cells could directly regulate the expression of p21. Our previously published microarray datasets that interrogate differential gene expression  Figure 3D) (18). Immunohistochemistry analysis of 393P, 344SQ and 393P-AZD R tumors also showed an inverse correlation between p21 and ZEB1 levels ( Figure 3C).
Pathologic analysis of human NSCLC samples for ZEB1 and p21 by IHC staining revealed an inverse correlation between nuclear ZEB1 and p21 H-scores ( Figure 3D). We also grouped the samples based on low ZEB1 (<4 H-Score) or high ZEB1 (>4 H-score) staining and found a significant difference in p21 H-score ( Figure S5B).
We further tested this observation by inducing EMT or MET via overexpression of ZEB1 or miR-200, respectively, in human and murine isogenic cell line pairs. We observed p21 mRNA and protein repression with ZEB1 expression. Conversely, with miR-200 expression, there was an upregulation of p21 ( Figure 3E,F , S5C,D). Transient and stable knockdown of ZEB1 in human and murine cells, respectively, caused p21 expression ( Figure S5E-G). Cells treated with the HDAC inhibitor mocetinostat undergo an MET by upregulation of the miR-200 family and ZEB1 suppression (11,22). Treatment also produced increased expression of p21 ( Figure S5H-J).

Induction of miR-200 in tumor cells by different means pushes the cells to a more epithelial state,
which is generally considered a less aggressive phenotype for tumor cells and more akin to a "normal" cell state, with expression of p21 and restoration of the cell cycle checkpoint that is lost or blunted in mesenchymal-like tumor cells with high ZEB1 activity.
Luciferase reporter assays were utilized to investigate ZEB1-mediated transcriptional regulation of p21. The promoter region of p21 was cloned upstream of a luciferase reporter and transfected into human lung cancer cells with either ZEB1 or miR-200 expression. H358 and H441 cells expressing ZEB1 led to a decrease in relative luciferin signal confirming transcriptional repression of the p21 promoter in the presence of high ZEB1. Conversely, an increase in luciferin signal was detected in H1299 cells with miR-200 induction, which suppresses the endogenous cellular ZEB1 expression and relieves transcriptional repression of the p21 promoter ( Figure 3G).
Binding of ZEB1 to the endogenous p21 promoter was confirmed by ChIP qPCR assays in cells with inducible ZEB1 or miR-200 expression, using previously published primer pairs (23). Using GAPDH as the negative control and miR-200c as a positive control, we confirmed direct binding of ZEB1 to the p21 promoter ( Figure 3H). Altogether the data support the EMT-dependent regulation of p21 in tumor cells by specific and direct ZEB1 binding.

Suppression of p21 in mesenchymal cells regulates CDK4 pathway
We next explored the effect of p21 on CDK4 activity in epithelial and mesenchymal lung cancer cells. With transient knockdown of CDK4, phosphorylation of RB was continuously suppressed in 344SQ cells for 48 hours ( Figure S6A , 4A). On the contrary, CDK4 knockdown in 393P cells appeared to have only slightly muted downstream signaling, which coincided with a surprising accumulation of phospho-CDK4, even with very low levels of total CDK4 protein. This also corresponded to a continued presence of p21 protein in 393P cells ( Figure 4A). These intriguing findings recapitulated previous data by Bisteau et al (24), which showed that a sustained presence of p21 protein in cells was able to maintain the phosphorylation status of CDK4 (and hence the stability of the complex), but still inhibit its kinase activity. A similar observation was made in our epithelial-like, but not mesenchymal-like, model where the presence of p21 maintained CDK4 in a phosphorylated state. To further demonstrate this point, we coimmunoprecipitated CDK4 and p21 from mesenchymal-like and epithelial-like cancer cells ( Figure   4B). In 344SQ and 344SQ_vector cells, lower amounts of CDK4-p21 complex coimmunoprecipitated compared to 393P and 344SQ_miR-200 cells, where an increased binding of CDK4 and p21 was detected. We also observed the seemingly contradictory presence of phospho-RB in epithelial-like cells alongside p21 expression. An explanation for this observation is the sequestration of p21 into the CDK4 complex, alleviating the repression from CDK2-Cyclin E complex, which can phosphorylate RB to maintain cell cycle progression. In fact, we observed that epithelial cells are more sensitive to the CDK2 inhibitor (miciclib) than mesenchymal cells We next tested the long-term effect of inhibiting CDK4 on the EMT status of tumor cells.
With doxycycline-mediated induction of CDK4 shRNA for 7 days, we observed a shift towards an epithelial phenotype indicated by decreased ZEB1 and vimentin levels and an accumulation of phospho-Erk ( Figure 4I). We also generated palbociclib-resistant cells with treatment of 344SQ cells for ~4 weeks. An epithelial phenotype was observed with an increase in E-cadherin and decrease in ZEB1 and vimentin levels ( Figure S7A-C). Therefore, targeting CDK4 allows the tumor population to shift to a more epithelial state, which would prime the tumor cells for MEK inhibitor treatment. To test this, we transiently knocked down CDK4 and treated cells with AZD6244, which sensitized the previously unresponsive mesenchymal-like 344SQ and 344P cells to MEK inhibition ( Figure S7D). We then tested the effect of combination palbociclib and AZD6244 treatment using a series of fixed concentrations at 1:1 ratio and calculated the fraction affected (Fa) values after exposure to the drugs. The Chou-Talalay method (25) was used to determine the combination index (CI) and drug reduction index (DRI). The favorable DRI, shown in yellow ( Figure S7E), was used to confirm the CI data. The drug combinations showed favorable DRI (DRI>1) and evidence of synergism (CI<1) at Fa>0.5 for palbociclib and AZD6244 ( Figure   S7E). Additionally, when tumor cells were treated with single-agent MEK or CDK4 inhibitor, there was a reciprocal activation of the CDK4 or MEK signaling pathways, respectively ( Figure 4J), showing a dynamic switching of signaling pathway activation and survival dependencies in the face of pharmacological treatments.

Co-targeting CDK4 and MAPK pathways targets different tumor cell subsets
To test if the pharmacological inhibitors have differential apoptotic effects on tumor subpopulations, we treated human (H1299 and H358) and murine (393P and 344SQ) tumor cells with AZD6244 and palbociclib and stained the cells with annexin V and propidium iodide.
Mesenchymal-like tumor cells underwent greater apoptosis in response to CDK4 inhibitors, while epithelial-like tumor cells were highly sensitive to MEK inhibition ( Figure 5A,B, S8A,B).
Given that tumors are heterogeneous and consist of subpopulations with distinct phenotypes along a spectrum of EMT, we utilized a previously described sensor model that can detect the epithelial or mesenchymal state of individual tumor cells in real time (12,26). Briefly, the 344SQ_Z-cad cell line expresses dual fluorescent sensors: a destabilized GFP with the ZEB1 3' UTR cloned downstream and E-cadherin promoter driving expression of RFP. This tool exploits the ZEB1/miR-200 double negative feedback loop. In an epithelial state, with high miR-200 and E-cadherin, the cells express RFP and emit red fluorescence and the presence of miR-200 suppresses GFP production by binding the ZEB1 3'UTR to prevent translation. Conversely, in a mesenchymal state with high ZEB1 and low miR-200, cells emit green fluorescence on account of GFP translation, whereas ZEB1 binds to the E-cadherin promoter to suppress transcription of RFP. As seen in Figure 5C, the majority of cells in 2D culture were mesenchymal and GFP + . With mocetinostat treatment, there was an enrichment of RFP + epithelial cells. There was a reduction of epithelial RFP + cells with AZD6244 treatment and mesenchymal GFP cells with palbociclib treatment ( Figure 5C, 5D). With dose escalation of single agent treatment, reciprocal pathway activation occurred, while combination treatment with both drugs suppressed MAPK and CDK4 pathways and enhanced tumor cell killing ( Figure S8C). Since western blots are bulk assays, we wanted to assess which specific populations undergo apoptosis within this heterogeneous dynamic system. We utilized a DNA binding dye that is cleaved by caspases present in the cells undergoing apoptosis to produce blue fluorescence. Co-localization of blue/green fluorescence with palbociclib treatment and blue/red fluorescence with AZD6244 treatment demonstrated the specificity of each individual drug to target specific cell types, whereas the combination of both drugs targeted both subpopulations ( Figure 5D,E, S8D,E).
In vitro three-dimensional assays very closely recapitulate the tumor growth in vivo. An established ex vivo tumor (EVT) model to culture lung tumors that retains tumor cell heterogeneity (27) was utilized to test the therapeutic sensitivity of distinct tumor cell subpopulations ( Figure   5F,G). Similar to the observations in 2D cultures, we found different subpopulations targeted by individual drugs when EVTs were cultured in laminin-rich Matrigel (MG). Since MG is known to promote an epithelial phenotype (10,27,28), AZD6244 effectively eliminated this cell subtype and resulted in an enrichment of GFP + cells. Palbociclib conversely caused a depletion of mesenchymal tumor cells within the heterogeneous EVTs and enrichment of the RFP + . We also noted a change in phenotype of EVTs treated with palbociclib, producing more structures with a central lumen as compared to other groups ( Figure S8F). Lumen formation and organization in a 3D matrix is characteristic of epithelial phenotype. Clearly, treatment with palbociclib not only targets mesenchymal phenotype but also promotes an epithelial phenotype, which makes it ideal to be combined with AZD6244. In combination treatment, both populations were targeted, which produced a net decrease in size and viability of EVTs ( Figure 5F and S8G). EVTs were also cultured in a matrix containing MG/collagen I, and as previously noted collagen promotes a mesenchymal phenotype in tumor cells (20,27), allowing us to test the efficacy of palbociclib on this specific subpopulation. AZD6244 remained ineffective on the GFP + mesenchymal tumor cells, however, there was a significant reduction in the viability of EVTs with palbociclib treatment ( Figure 5G). Combination treatment proved to be significantly better over the individual treatments in both Matrigel and collagen matrices in terms of suppression of viability of tumor cells. In summary, these results demonstrate the efficacy of CDK4 and MEK inhibitors in combination for effective therapeutic targeting of the tumor cell subpopulations.

Combination of CDK4 and MEK inhibitors controls syngeneic tumor growth and prevents emergence of EMT-mediated resistance
We next evaluated in vivo tumor response to the combination of CDK4 and MEK inhibitors.
Mesenchymal-like (344SQ) or epithelial-like (393P) tumor cells were subcutaneously implanted in syngeneic wildtype mice. Tumor growth in response to either single agent (palbociclib or AZD6244) or both was monitored over a period of 6-14 weeks. Mice bearing 344SQ tumors remained unresponsive to AZD6244, but responded to palbociclib alone or in combination with AZD6244 ( Figure 6A). This treatment continued for ~6 weeks (short-term) and scored as additive using Bliss effect analysis ( Figure S9D) (29). This promising tumor response in the short term led us to repeat the experiment to determine if there was a durable and sustained response to the combination treatment. Treatment of the cohorts for up to 10 weeks produced the emergence of resistance to palbociclib treatment alone ( Figure S9A, B). The tumors acquired resistance to single agent palbociclib over an extended period of time, which was prevented with combination treatment, and the group initially treated with only palbociclib was re-sensitized upon addition of AZD6244 at week 10, either as measured by tumor growth or fold change of tumor volume ( Figure   S9A,B). We also observed an increase in E-cadherin and a decrease in nuclear ZEB1 with singleagent palbociclib treatment ( Figure S9C), demonstrating the selective outgrowth of an epithelial phenotype. The combination treatment for a period of 14 weeks scored as an additive response ( Figure S9D). The number of lung metastatic nodules in short-and long-term experiments were also significantly lower with palbociclib or combination treatment ( Figure S9E,F).
Epithelial-like 393P tumors that initially respond to AZD644 develop resistance to treatment by undergoing EMT (11). When treated with single agents, 393P tumors were resistant to palbociclib alone and responded to AZD6244 for about 7 weeks ( Figure 6B). However, the combination of both the drugs suppressed tumor growth with a durable response for ~10 weeks.
In the 393P tumor model, combination of CDK4 and MEK inhibitor scored as synergistic using the Bliss effect analysis ( Figure S9D). Since 393P is a non-metastatic model, there were no significant differences in lung metastases ( Figure S9E, F). The previously described 393P-vehicle and 393P-AZD R cells were also implanted in syngeneic wildtype mice to assess the sensitivity to CDK4 and MEK inhibitors. 393P-vehicle tumors retained their sensitivity to AZD6244 and resistance to palbociclib ( Figure S10A), whereas the tumors derived from 393P-AZD R cells were unresponsive to AZD6244, and responsive to palbociclib, with one mouse showing complete tumor regression (FigureS10B).
Primary tumor tissues were collected at the end of the mouse experiments and stained for the CDK4 and MAPK signaling pathway markers. Untreated 344SQ tumors showed higher phospho-CDK4 and phospho-RB compared to untreated 393P tumors, which had higher phospho-Erk ( Figure 6C,D). In 344SQ tumors, treatment with palbociclib led to suppression of phospho-CDK4 and phospho-RB staining, with an activation of MAPK signaling as marked by phospho-Erk; AZD6244 treatment lead to an increase in phospho-CDK4, and the combination drug treatment suppressed both CDK4 and MAPK signaling. Conversely, 393P tumors showed suppression of phospho-Erk when treated with AZD6244, accompanied with an increased expression of phospho-Cdk4. Palbociclib caused an increase in phospho-Erk in 393P tumors as well. Combination drug treatment in both models significantly suppressed both pathways compared to either single agent ( Figure 6C,D).
To determine the effect of single and combination agent treatments on cell proliferation and cell death, we performed Ki67 staining and TUNEL assay on the tumor tissues. 344SQ tumors treated with palbociclib for 6 weeks had fewer proliferating and more apoptotic cells ( Figure 6E,F).
393P tumors treated with AZD6244 had fewer proliferating cells and higher apoptotic cells ( Figure   6G,H). However, combination inhibitor treatment in both models significantly suppressed cell proliferation and produced apoptosis in >60% of the tumor cells. We also compared the cell proliferation and death in the 344SQ tumors treated long term (10 weeks) with the single agents and combination. As noted, the 344SQ tumors acquired resistance to palbociclib alone after 10 weeks. This was reflected in the Ki67 and TUNEL staining, which were similar to the single agent AZD6244-treated 344SQ tumors, which were unresponsive ( Figure S10C Figure   S11A. Histological analyses on lung tumors displayed differences in the signaling pathways dependent on the genetic background (Figure7A). Tumors with mutant Kras alone showed greater MAPK pathway activation compared to KP and KM tumors, which instead showed an activation of CDK4 pathway as demonstrated by phospho-CDK4 and phospho-RB staining ( Figure 7A). We also utilized these models to interrogate if the ZEB1-p21 axis was altered within these tumors and could determine their sensitivity to palbociclib. Tumor regions with high nuclear ZEB1 corresponded to lower levels of nuclear p21 in KP and KM tumors, as compared to Kras tumors alone ( Figure S11B).
Three months after induction, lung tumor formation was confirmed and monitored over 6-8 weeks by micro-CT scans for changes in overall lung tumor burden in response to pharmacological agents. Response to AZD6244 alone across all three genotypes was similar to our previous results (11), where Kras tumors showed complete regression upon treatment and only tumor stability or lack of response was achieved in KP and KM tumors ( Figure 7B, S11C, S12A). Palbociclib alone had more significant tumor growth control in KP and KM mice than AZD6244 alone with ~30% of tumors demonstrating regression ( Figure 7B, S11C, S12A).
Histological staining showed that treatment with each single agent led to an activation of the reciprocal signaling pathway in KP and KM tumors ( Figure 7C, 7D). Palbociclib led to suppression of ZEB1 indicating a shift to an epithelial phenotype and AZD6244 led to an accumulation of ZEB1 indicating the presence of mesenchymal-like tumor cells ( Figure 7D, 7E). Combination palbociclib and AZD6244 produced a more significant reduction of tumors over a period of 8 weeks with complete regression in ~80% mice across all three genotypes ( Figure 7B, S11C, S12A). Lack of sufficient tumor burden precluded us from staining the lung sections obtained from combination treatments ( Figure S11C). Separate studies have presented contradictory findings for the correlation of EMT with CDK4 pathway signaling. CDK4 inhibition in triple negative breast cancer reversed the EMT status of cancer cells (33,34), as seen in the mesenchymal-like KP tumors treated with palbociclib in the present study. Within Kras-mutant pancreatic cancer, one study showed that tumor cells underwent EMT with palbociclib monotherapy (35) and MET in another (36). Another study in colorectal cancer noted no difference in EMT status of tumor cells in response to palbociclib (37).
These findings highlight the fact that there are cell-type or context-specific phenomena that warrant further investigation in different cancer types. In our studies, we found that modulation of the EMT status of cancer cells by perturbing the ZEB1/miR-200 axis lead to CDK4 pathway modulation and determined the sensitivity to CDK4 inhibitors both in vitro and in vivo.
Mechanistically, high ZEB1 levels in mesenchymal cancer cells was responsible for transcriptional repression of p21 by direct binding to the promoter region. Conventionally, p21 is described as a suppressor of CDK4 kinase activity and downregulation in patients predicts poor survival (38,39). Studies in recent years have further explored the role of p21 and revealed a dual function of p21, acting in some cases as an activator for CDK4 activity (40). Lower levels of p21 binding are generally required for the assembly and stability of the CDK4-cyclin D complex, which partially accounts for maintaining CDK4 phosphorylation and primes CDK4 for catalysis by releasing the activation segment without affecting kinase function (24). A sustained presence of p21 at higher stoichiometric concentrations can render CDK4 ineffective (24). We found that mesenchymal cancer cells had lower levels of p21 in the CDK4-p21 complex, which explains their increased CDK4 activity. Continued presence of an activated CDK4 rendered the mesenchymal cells highly dependent on CDK4 for survival. With p21 overexpression in mesenchymal cancer cells, we detected increased CDK4-p21 complex, reduced in vitro and in vivo growth of tumors.
Interestingly, 344SQ cells demonstrated reduced sensitivity to palbociclib with p21 overexpression. A previous study had shown that p21 can interfere with the binding of small inhibitors to CDK4 complex, as we observe in wild type epithelial tumor cells and in p21 overexpressing mesenchymal cells (41). Thus, p21 serves as a regulator of CDK4 activity and sensitivity to inhibitors in mesenchymal lung cancer cells.
With an understanding of how lung cancer cells adapt to therapeutic intervention, we interrogated the combination of CDK4 and MEK inhibitors. The success of CDK4 inhibitors in combination with endocrine therapy in breast cancer patients has encouraged investigations into the role of CDK4 inhibitors in other cancer types, including lung cancer (42)(43)(44)(45). Kras driven murine lung cancers were particularly susceptible to CDK4 ablation and a sustained tumor response was achieved with concomitant CDK4 inactivation and RAF1 ablation in Kras/p53 driven murine lung cancers (16,17). A phase II trial in NSCLC patients with inactivated CDKN2A treated with palbociclib monotherapy showed modest response with stable disease in 50% of the patients (46). Partial response to CDK4 inhibitor in a subset of lung cancer patients warranted an exploration of combination with other targeted therapies. Zhou et al demonstrated a synergistic growth inhibition in KRAS and CDKN2A mutant NSCLC xenografts with AZD6244 and palbociclib (47). Ongoing phase I/II clinical trials (NCT03170206 and NCT02022982) in advanced KRASdriven NSCLC patients are investigating the combinatorial effect of MEK and CDK4 inhibitors.
Additionally, the combination of CDK4 and MAPK pathway inhibitors have shown tumor regression in xenografts models of cancers with KRAS, NRAS or BRAF mutations, particularly BRAF-and NRAS-mutant melanoma, with promising results from phase I clinical studies (29,(48)(49)(50)(51)(52). Clinical trials are currently investigating BRAF and MEK inhibitors in combination with ribociclib in BRAF-mutant melanoma and other solid tumors with BRAFV600 mutations (54). Not only are the two therapies synergistic, but studies have also shown that CDK4 inhibitors may overcome MEK inhibitor resistance (58) and vice versa (59). These findings are corroborated by our results in the present study demonstrating the efficacy of CDK4 and MEK inhibitors.
Results in our immunocompetent syngeneic models will allow us to further extend our investigation into effects on the immune microenvironment. Evidence from past studies indicated that CDK4 depletion reduced infiltration of CD4+ FoxP3+ Tregs (60) and CDK4 inhibitors increased tumour immunogenicity and cytotoxic T-cell mediated clearance of tumor cells (61).
CDK4 inhibitors also enhanced effector T-cell infiltration and activation (62). Additionally, PD-L1 degradation was shown to be regulated by CDK4 through cullin3-SPOP E3 ligase via proteasome-mediated degradation, which primed the tumors for effective response to combination treatment with CDK4 inhibitor and PD-(L)1 immune checkpoint blockade (63). Other investigations revealed that PD-L1 expression was modulated by the RB-NF-κB axis, which could be exploited to overcome cancer immune evasion triggered by conventional or targeted therapies (64). Combination of CDK4 and MEK inhibitor induced a senescence-associated secretory phenotype (SASP) that provoked a natural killer cell surveillance program and resulted in tumor cell death (65).
The application of combinatorial treatments with MEK and CDK4 inhibitors in multiple preclinical in vitro (dual fluorescent sensor system, 3D assays) and in vivo models (syngeneic and autochthonous mouse models) effectively prevented outgrowth of resistant tumor subpopulations and was significantly better than either monotherapy. Such findings demonstrate that CDK4 and MAPK pathway are intertwined in lung cancer progression and durable response can be attained if these pathways are targeted judiciously. Fighting cancer at two fronts: by interfering with two distinct regulatory networks and targeting tumor subpopulations should benefit patients and help to prevent resistance development.

shRNA screens
Murine lung cancer cell lines (393P and 344P) were infected at a multiplicity-of-infection (MOI) of 0.3 with a pooled shRNA lentiviral library targeting genes associated with known kinase activity (10 shRNA/gene, for target list and shRNA sequences see Table S2 and S4). Parallel in vivo and in vitro screens were performed, and the shRNA-coupled barcodes were detected by high-throughput sequencing technology [for detailed procedures and primer sequences see (66)].
In vivo and in vitro screens were carried out in triplicate and duplicate, respectively. Raw counts for the screen endpoints and a reference population, isolated after transduction, were normalized using the variance stabilizing transformation with the DESeq2 in R. The normalized counts were divided by the reference cells that were isolated immediately following transduction to estimate a fold change in barcode abundance. Four independent shRNAs targeting essential genes (RPL30, PSMA1) or luciferase (LUC) were cloned with 5 unique barcodes each and incorporated in the library as positive and negative controls (20 reagents/control, see Table S1 and S3). One LUC hairpin showed apparent off-target effect, which has been observed over a wide-spectrum of in vitro and in vivo screens. One hairpin for PSMA1 did not show robust drop out, and this pattern was consistent across the 5 barcodes, indicating that this result was not reflective of poor screen performance. The separation of positive and negative controls was evaluated by the robust strictly standardized mean (SSM , Table S1 and S3), excluding the hairpins mentioned above. Fold change distribution was converted to percentiles, and biological replicates were collapsed for RSA analysis. The RSA logP-values and ranks are provided in Tables S2 and S4.

Cell Lines
Human and murine lung cancer cell lines were cultured in RPMI1640 (Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS, Gibco). 293T cells were cultured in DMEM (Gibco) supplemented with 10% FBS. All human cell lines were obtained through ATCC. Murine lung cancer cells were created from Kras LA1/+ /p53 R172H genetically engineered mice as previously described (10). Manipulated human and murine cells cell lines with ZEB1 and miR-200 expression were derived as previously described (11). All cells were cultured at 37 O

Statistical analysis
Statistical analysis was carried out as described in each corresponding figure legend. A p-value of <0.05 was considered statistically significant. Data are presented as mean + SD unless otherwise noted. All analyses were performed in GraphPad Prism software (version 8).

Study approval
All animal experiments were reviewed and approved by the Institutional Animal Care and Tumor sizes were measured weekly. AZD6244 was dissolved at 5 mg/mL in solvent (4% DMSO, 30% PEG 300, 5% Tween 80), and palbociclib was dissolved at 10 mg/mL in solvent (Lactic acid buffer (50 mM, pH 4.0)). Control mice received solvent at a volume equal to the drug dosage at the indicated drug concentrations. After euthanasia by CO2 exposure at 3 L/min, syngeneic primary tumors and/or mouse lungs were formalin-fixed, paraffin-embedded, and sectioned for histological analysis. In vivo combination synergy analysis was done using the method of Bliss Independence as previously described (29).    Data are presented as mean + SD and one-way ANOVA was used for statistical analysis in all the panels. **** p<0.0001; *** p<0.005; ** p<0.001; * p< 0.05; ns: not significant.