Combination Therapies with CDK4/6 Inhibitors to Treat KRAS-Mutant Pancreatic Cancer

Pancreatic cancer is the third-highest cause of cancer-related mortality in the US (1). The majority (>85%) of pancreatic malignancies are of the pancreatic ductal adenocarcinoma (PDAC) type. The current standards-of-care for PDAC are minimally effective and comprised of conventional cytotoxic drugs (2). Despite a now well-delineated genetic profile, to date, no clinically effective targeted therapies for PDAC have been developed.

The genetic basis of PDAC is well-established (3). Mutational activation of the KRAS oncogene (∼95%) is the initiating step and is essential for maintenance of PDAC growth. The recent approval of the first direct KRAS inhibitor for lung cancer (sotorasib), selective for one KRAS mutant (G12C; ref. 4), supports the potential therapeutic impact of anti-KRAS therapies for PDAC. However, as KRAS^G12C mutations comprise only 2% of KRAS mutants in PDAC, distinct strategies are required for the majority of patients.

A second key genetic event in PDAC development (occurring in greater than 90% of pancreatic cancers) involves genomic deletion or promoter hypermethylation and silencing of the CDKN2A tumor-suppressor gene (5). In mouse model studies, Kras mutation alone initiated premalignant pancreatic ductal lesions at complete penetrance, with delayed low frequency formation of metastatic pancreatic cancer (<10; ref. 6) However, Kras mutation coupled with Ink4a loss induced rapid onset of metastatic PDAC (7).

The primary consequence of CDKN2A loss in PDAC is loss of expression of the cell-cycle inhibitory protein p16^INK4A, which functions through inhibition of the highly related cyclin-dependent kinases CDK4 and CDK6 (8). Binding of the D-type cyclins (e.g., cyclin D1) to CDK4/6 relieves this inhibition, allowing CDK4/6 to phosphorylate and inactivate the retinoblastoma tumor suppressor. Once hyperphosphorylated, retinoblastoma can no longer sequester and inactivate the E2F family of transcription factors, permitting E2F-driven transcriptional activation of genes that drive the G₁–S transition and cell growth.

KRAS also regulates CDK4/6 activity through the canonical RAF–MEK–ERK–MAPK effector signaling network to stimulate transcriptional activation of CCND1 (encodes cyclin D1; ref. 9). KRAS additionally blocks cyclin D1 degradation through AKT-mediated inactivation of GSK3β via the PI3K–AKT effector signaling network. Therefore, KRAS effector signaling and CDKN2A inactivation mechanistically converge to inactivate retinoblastoma.

The critical role of CDKN2A loss of function during PDAC tumorigenesis supports the therapeutic value of pharmacologic agents that phenocopy the effect of p16^INK4A to inhibit the growth of KRAS-driven PDAC tumors. Currently, three CDK4/6 inhibitors (palbociclib, abemaciclib, and ribociclib) have been approved for the treatment of estrogen receptor (ER)–positive, HER2 receptor tyrosine kinase (RTK)–negative breast cancer in combination with ER-targeted therapies (8). A fourth CDK4/6 inhibitor (trilaciclib) has been approved as a myeloprotective agent combined with chemotherapy in small-cell lung cancer, whereas another (lerociclib) is currently under clinical evaluation for multiple indications. Preclinical evaluation of CDK4/6 inhibitors in KRAS-mutant PDAC cell lines (10–12) and patient-derived xenograft tumors (13) has shown single-agent antitumor activity. Growth inhibitory activity was significantly improved when combined with inhibitors of specific components of KRAS effector signaling (10, 14, 15).

Despite promising findings from preclinical studies, the outcomes of clinical evaluation of CDK4/6 inhibitors in PDAC have been disappointing. In a phase II trial (NCT02981342), abemaciclib treatment alone or in combination with the PI3K inhibitor LY3023414 did not show significant improvement in overall survival. A phase I evaluation of ribociclib in combination with the mTOR inhibitor everolimus (NCT02985125) also resulted in no significant clinical response.

In this study, we further assessed CDK4/6 as an effective therapeutic target in PDAC, determined that ERK inhibition effectively enhances CDK4/6 inhibitor cytotoxicity, and conducted a CRISPR-Cas9 loss-of-function screen to identify novel synergistic combinations that further sensitize pancreatic cancer to CDK4/6 inhibition that may improve clinical outcomes.

Pa01C, Pa02C, Pa03C, Pa04C, Pa14C, and Pa16C cell lines were provided by A. Maitra (MD Anderson Cancer Center, Houston, TX). The remaining PDAC cell lines were obtained from the ATCC and were maintained in recommended media (DMEM or RPMI-1640 supplemented with 10% FBS and penicillin/streptomycin). Previously established organoids derived from pancreatic cancer were provided by David Tuveson (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). No patient samples were used in this study. All cell lines were maintained in a humidified chamber with 5% CO₂ at 37°C. Pancreatic organoid cells were seeded in growth factor–reduced Matrigel (Corning) domes and fed with complete human feeding medium: Advanced DMEM/F12-based WRN condition medium (L-WRN, ATCC CRL-3276), 1 × B27 supplement, 10 mmol/L HEPES, 0.01 μmol/L GlutaMAX, 10 mmol/L nicotinamide, 1.25 mmol/L N-acetylcysteine, 50 ng mL–1 hEGF, 100 ng mL–1 hFGF10, 0.01 μmol/L hGastrin I, 500 nmol/L A83–01, 1 μmol/L PGE2 and 10.5 μmol/L Y27632. Colorectal cancer organoids were maintained as described previously (16). All cell lines were validated using short tandem repeat profiling and confirmed regularly as Mycoplasma-negative.

Chemical reagents used were Calcein AM, Caspase-Glo 3/7, and AlamarBlue (Invitrogen), ATP-Lite (PerkinElmer), MTT (Sigma). Antibodies for total retinoblastoma, phospo-RB (780; C84F6), phospho-RB (S807/S811; D20B12), total retinoblastoma (4H1), total ERK (L34F12), phospho-ERK (T202/204; D13.14.4E XP), phospho-p90RSK (T359/S363; 9344), CDK4 (D9G3E), and CDK6 (D4S8S) from Cell Signaling Technologies; BAP1 from Santa Cruz Biotechnology; β-tubulin from Sigma. Each was used at a 1:1,000 dilution, excepting Santa Cruz Biotechnology antibodies at a 1:100 dilution. Blots were probed with species-appropriate horseradish peroxidase–conjugated secondary antibodies (Sigma) at a 1:2,000 dilution and measured on the Bio-Rad Chemidoc MP. Quantification performed using the Bio-Rad Image Lab software. The pLentiPGK Blast DEST ERKKTRmRuby2 plasmid was a gift from Markus Covert (Addgene plasmid # 90231; http://n2t.net/addgene:90231; RRID:Addgene_90231). The pBABE-puro mCherry-EGFP-LC3B retroviral expression vector was a gift from J. Debnath (Addgene plasmid # 22418; http://n2t.net/addgene:22418; RRID:Addgene_22418; ref. 17). Palbociclib, ribociclib (LEE001), quisinosat, saracatenib, hydroxychloroquine, ABT-263, erlotinib, bosutinib, digitoxin, KU60019 were obtained from SelleckChem. Lerociclib was synthesized by G₁ Therapeutics. Abemaciclib (LY2835219), PF-06873600, R428, and BMS303141 were obtained from MedChemExpress. Methotrexate was purchased from Sigma. AMG510 was provided by Deciphera Pharmaceuticals. SCH772984 was provided by Merck. Ulixertinib/BVD523 was provided by Biomed Valley Discoveries.

Cells in anchorage-dependent assays were plated in flat 96-well plates at a density of 1,500 cells/well (or 2,000 cells/well for Pa16C). For anchorage-independent assays, cells were mixed with SeaPrep Agarose (final concentration 1% agar in complete media) and plated on a basement of solid bacto-agar (0.6%) at a final count of 5,000 cells/well. A liquid layer of media were added on top of the solidified agar layers to accept dispensed drugs. Compounds were added using a Tecan D300 automated dispenser (HP) 24 hours after plating. Compounds were generally plated in 3-fold dilution schemes at concentrations to encapsulate the entire dose–response curve. For measuring cell viability using direct cell counting, plates were washed once with 1x PBS, and then incubated with a 1:5,000 dilution of live cell dye Calcein AM in PBS for 30 minutes at 30°C. Cells were then counted using the SpectraMax MiniMax (Molecular Devices). The Spectramax software automated counting of live cells. Viability assays using other methods were used as a comparison with direct cell counting, where indicated. For MTT assays, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (Sigma) was added at 0.4 mg/mL in PBS to cell media, incubated at 37°C for 2 hours, and then the wells rinsed with PBS. To read, 100-μL DMSO was added per well, and absorbance read at 590 nm. For Alamar Blue (Invitrogen), reagent was added at a 1:10 dilution and incubated for 3 hours at 37°C, and then read at 535-nm excitation and 590-nm emission fluorescence. A third metabolic-based viability assay, using ATP Lite (PerkinElmer), detects ATP levels as a proxy for cell viability, and was used according to the manufacturer's recommendations. Caspase activation in organoids was measured using Caspase 3/7 Glo (Promega) using the manufacturer's recommended concentrations and carefully mixing with the Matrigel/organoid layer by manual pipetting. GraphPad Prism was used to calculate EC₅₀ values and display dose–response curves. BLISS scores were calculated using the BLISS equation (18). Three-dimensional heatmaps were generated using the R package plot.ly.

For anchorage-dependent colony formation assays, cells were plated in 6-well plates at a count of 500 cells/well. Compound was added after 24 hours, and cells were allowed to grow for 14 days. To visualize colonies, wells were washed with PBS, stained, and fixed with 4% formaldehyde and crystal violet for 15 minutes, washed extensively in water, and then read on a Typhoon FLA 9500 (GE). Colonies were quantified using ImageJ by calculating the percentage of area covered. Organoids were plated and drugged as previously described (19), grown for 10 days, and imaged on the SpectraMax i3x MiniMax 300 imaging cytometer. Cell viability was measured using CellTiter Glo 3D Cell Viability assay (Promega) according to the manufacturer's instructions on the SpectraMax i3x plate reader.

Before siRNA transfection, 2 × 10⁵ PANC-1 cells, 2.5 × 10⁵ Pa14C cells, or 3 × 10⁵ MIA PaCa-2 cells were seeded into individual wells of a 6-well plate. The next day, culture medium was replaced. To suppress CDK4 or CDK6, 30 pmol of siRNA were added to 200 μL of Opti-MEM medium followed by 2.5 μL of RNAiMax. Two siRNAs targeting CDK4 (siCDK4 1, Dharmacon J-003238–13, and siCDK4 2, Dharmacon J-003238–15) and CDK6 (siCDK6 1, Dharmacon J-003240–10, and siCDK6 2, Dharmacon J-003238–12) and a nontargeting siRNA (siNT D-001810–01–20) were used in all experiments. To concurrently suppress CDK4 and CDK6, 30 pmol of siCDK4 1 and 30 pmol of siCDK6 1 were combined. The mixtures were incubated for 8–10 minutes before being added dropwise to cells. The next day, cells were collected and seeded into proliferation assays at 3 × 10⁵ cells per well of a 96-well plate, or 2 × 10⁵ cells were plated into a 6-well plate for immunoblot analyses. For combination experiments, cells were treated with ERKi the next day. Proliferation assays and immunoblot analyses were performed as described previously after an additional 72 hours.

Reverse phase protein array (RPPA) experiments were conducted as previously described (19). Briefly, cells were grown in 6-well plates and subdivided so to not exceed 80% confluency. Drugs were added 24 hours after initial plating and replenished after each subdivision. At the indicated time points, cells were harvested, lysate collected, and analyzed as previously described (19). All unique conditions were collected in quadruplicate. The median of all replicates was calculated, and then samples were normalized as the log₂ fold-change between drug or control treatment.

Cell lysates were collected in RIPA buffer [50 mmol/L TRIS-HCL pH 7.4, 50 mmol/L NaCl, 2 mmol/L EDTA, 0.1% SDS supplemented with protease inhibitor (Complete EDTA-Free, Roche)] and phosphatase inhibitor (Phosphatase Inhibitor Cocktail Sets 1 and 2, Millipore). Equal amount of lysate (18 μg) was boiled in SDS protein loading buffer supplemented with 10 mmol/L DTT, separated by SDS-PAGE, transferred to Immobilon-FL PVDF membrane (Millipore, Billerica), and blotted for indicated protein.

Senescence was detected using the β-galactosidase–based Senescence Detection Kit (ab65351, Abcam), using the recommended protocol. Cells were imaged and photographed on an Evos M7000. Separately, the CellEvent Senescence Green Detection Kit (Thermo Fisher Scientific) was used to stain senescent cells. Cells were plated in 6-well plates, treated with compound for 7 days and passaged as necessary to maintain under 80% confluency, and then replated into 96-well plates and stained using the manufacturer's protocol.

ERK phosphorylation was detected using the Advanced phospho-ERK (Thr202/Tyr204) cellular kit (Cisbio/PerkinElmer) using the manufacturer's protocol. Briefly, 10,000 cells/well were plated into a 96-well plate, treated the following day with compound, and incubated for 48 hours. Media were then removed and cells lysed with 50 μL of the provided lysis buffer. Ten μL of lysate was then transferred to a white 384-well plate, the pre-mixed antibody solution added, and the plate was sealed and rocked at room temperature overnight. The next morning, the plate was read at 620 and 665 nm using the Biotek Cytation 5 plate reader with a TR-FRET red filter cube. The TR-FRET signal is the ratio of 665/620 nm. To normalize for lower overall cell count caused by inhibitor treatment, cells were separately plated and dosed with compound were counted for cell viability using the cell counting protocol described above, and the pERK signal for each treatment was normalized to total cell count for that treatment.

CRISPR screen hit validation using a plate-based siRNA counterscreen was conducted using pooled on-targetPLUS motifs for indicated proteins from Horizon Discovery. Cells were plated in antibiotic-free medium at 5,000 cells/well in a 96-well plate. After 24 hour, medium was replaced with siRNA-containing medium consisting of 0.01% Dharmafect II transfection reagent and 25 nmol/L siRNA sequence specific to the target or nonsilencing control motif. Control wells received normal media. The CDK4/6 inhibitor palbociclib was added in a 5-point dose–response curve at the same time as siRNA addition. Cells were grown for five days, and then viability was read using direct cell counting as described above.

Cells were plated in a 6-well dish at 100,000 cells/well. Cells were dosed with compound for five days. Apoptosis assays were performed with the TACS Annexin V–FITC kit (Trevigen) according to the manufacturer's protocol. For cell-cycle assays, cells were fixed in 70% ethanol for three hours, and then stained for three hours at 37°C with 10 μg /mL propidium iodide and 100 μg/mL RNAse in 1x PBS. Cells were analyzed on either a BD LSRFortessa flow cytometer or Beckman Coulter CytoFLEX, and cells were analyzed with FCS Express 7 (De Novo Software).

To analyze, pericentrin and γH2AX cells were plated on glass bottom dishes (MatTEK Corporation) and imaged using an EVOS M7000 wide-field microscope with a ×63, 1.4 NA objective. Following inhibitor or vehicle (DMSO) treatment cells were fixed with 4% formaldehyde for 20 minutes at room temperature, permeabilized with 0.1% Triton for 5 minutes, rinsed with PBS, and blocked using 3% BSA-PBS for 1 hour at room temperature. The cells were incubated in antibodies targeting pericentrin and γH2AX (1:200 in 3% BSA-PBS) overnight at 4°C or at room temperature for 2 hour, cells were then washed with PBS followed by incubation in Alexa-561 and Alexa-488 secondary antibodies at room temperature for 1 hour. To visualize nuclei, cells were stained with DAPI (1:10,000) for 5–10 minutes in PBS and subsequently rinsed with PBS. For γH2AX, the integrated density per cell was measured and for pericentrin the number of foci were counted per cell, both were done using FIJI. For each cell line a total of 200 cells were analyzed from two independent biological replicates. To calculate the relative intensity of γH2AX, all values were divided by the mean of the corresponding vehicle (DMSO) control and then transformed by log₂. Pa16C cells stably expressing ERK–KTR–mRuby were plated, treated, fixed, stained for DAPI, and imaged as described above. Images were analyzed using FIJI, the DAPI images were used to outline the nucleus and the ERK–KTR–mRuby images were used to outline the whole cell. The integrated density per nucleus and cell were each measured using the ERK–KTR–mRuby channel. The total cell integrated density was divided by the nuclear integrated density for each cell. A total of 50–150 individual cells were analyzed for each condition. Statistical significance was determined using a one-way ANOVA and Dunnett multiple comparison test.

Cells were plated in glass-bottom dishes (MatTEK Corporation) and imaged on a Zeiss 700 confocal microscope with a ×63, 1.4-numerical aperture objective. Cells expressing mCherry–EGFP–LC3 were treated as indicated and imaged live. The 488 and 561-nm laser lines were used to excite EGFP and mCherry, respectively. Images were analyzed in ImageJ, where images were thresholded to isolate mCherry⁺ and/or EGFP⁺ autophagic vesicles and autophagic flux was calculated by dividing the total mCherry⁺ vesicle area to EGFP⁺ vesicle area.

Raw sequence counts were deconvolved to separate treatment groups from total sequence reads, and then single guide RNA (sgRNA) reads were normalized and analyzed for each gene as previously described (20). We used two methods to pick hits: MAGeCK (Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout)-MLE (maximum likelihood estimation) automates sample normalization and comparison between T0, DMSO- and drug-treated samples to determine a beta score, where negative beta score indicate sensitivity to loss of a gene (21), whereas Redundant siRNA Analysis (RSA) scores genes based on the collective fold-change activity of the entire sgRNA set for each gene and assigns a P value for statistical significance (22). Using either method, depletion of an sgRNA construct relative to the initial T0 condition for that cell line indicates the loss-of-fitness impact of a gene over time (a measure of gene “dependency”). In contrast, a selective decrease in counts for a gene in the CDK4/6i-treated group compared with vehicle control indicates a CDK4/6i-specific dependency. For analysis using MAGeCK (Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout), we ran MAGeCK using MLE with treatment-deconvolved FASTQ files, normalized to control (non-targeting) sgRNA constructs (21, 23). To analyze samples using RSA, the number of counts for each construct was normalized to the total number of counts in the same sample and averaged across both replicates. By comparing the number of cells containing each sgRNA in the palbociclib versus DMSO-treated cells, we determined the selectivity effect on viability that each sgRNA had on palbociclib cotreatment. Genes were ranked and assigned significance using published RSA analysis software (24). Genes were selected as hits using RSA if multiple sgRNA targeting that gene potentiated drug activity, and a log(p) for that gene, calculated using the RSA method, was ≤ −3 (24).

Heatmaps were built using the R packages pheatmap and ComplexHeatmap (25). The hits were used as input for STRING visualization (26). ORA pathway analysis was conducted using Webgestalt (27).

Protein interaction networks were generated using STRING-DB. Associations were drawn using “high confidence” settings. Figures redrawn using Gephi were generated by downloading node and edge data from STRING, importing into Gephi 0.9.2, and drawing using a radial layout arranged by modularity, sorted by degree, and colored by RPPA signal as indicated.

Data were analyzed by GraphPad Prism built-in tests (specifics are indicated in figure legends). Data are presented relative to their respective controls. For all graphs, error bars indicate mean or median SD for n  ≥  3 independent experiments (except where noted) and P values on graphs are denoted within each figure panel or in the Supplementary Information (RPPA data). Number of samples analyzed per experiment and whether data are the mean of multiple experiments, or selected as representative, are indicated in figure legends.

The "druggable genome" CRISPR knockout library was described previously (28, 29). The pooled library consists of the lentiviral plasmids, each encoding an sgRNA, the CRISPR-associated nuclease (Cas9) and a gene conferring resistance to puromycin. Five unique sgRNAs for each of the 2,240 genes comprising major cellular and cancer signaling networks, including most of the human kinome, were included in the library. Also included were 50 nontargeting constructs and 750 constructs targeting 150 control genes with previously reported dispensable or essential cellular functions. Library preparation, viral production, and viral titering was as we previously described, except that lentivirus was produced using a mixture of Fugene 6, 12 μg Pax2 packaging plasmid, 3 μg pMD2G envelope plasmid, and 12 μg of the CRISPR library in Optimem.

The screen was conducted by first plating 1 × 10⁶ cells in 2 mL of recommended media per well into 25 six-well plates. The following day, virus and polybrene (8 μg/mL) were added to the cells. Plates were centrifuged at 800 × g for one hour, and then incubated overnight at 37°C. The next day, transduction medium was replaced with the fresh medium containing puromycin at 2 μg/mL. As expected, significant cell death developed within two days. All subsequent cell culture was done in presence of puromycin. The remaining cells were trypsinized and re-plated on 500 cm² square culture plates, with 1.2 × 10⁷ cells frozen as a T0 sample. Seven days later, a second sample of 1.2 × 10⁷ cells was frozen (T7 timepoint), and cells were separately plated into drug and replicate treatment groups, to receive continuous treatment with either vehicle (DMSO) control or palbociclib (200 nmol/L) for 4 weeks. Cells were collected after two and four weeks of treatment and flash-frozen in liquid N₂. Two technical replicates for each condition were prepared, excepting Pa16C where only a single replicate was collected. MIA PaCa-2 cells were collected only at two weeks of treatment. Cells were always maintained at no fewer than 1.2 × 10⁷ cells to maintain at least ×1,000 coverage of the library in the population. Genomic DNA was isolated and prepared for sequencing as previously described (20). Sequencing was performed on an Illumina NextSeq 500 with 75bp, single-end reads. A final concentration of three pmol DNA was loaded with a PhiX spike of greater than 15%.

The data generated in this study are available within the article and its Supplementary Data files or upon request from the corresponding author.

Given the lack of clinical efficacy seen with single-agent CDK4/6 inhibitor treatment in PDAC (30), we sought to identify inhibitor combination strategies to enhance their antitumor activity. First, we established a baseline of CDK4/6 dependency and addressed the relative importance of the two highly related CDK4 and CDK6 kinases in PDAC cell lines. Interestingly, we found that acute (72 hours) siRNA suppression of either CDK4 or CDK6 alone in some cell lines caused a compensatory increase in the expression of the other protein (Fig. 1A). We found that although single-gene suppression did not cause significant reductions in growth, concurrent CDK4/6 suppression was much more effective (Fig. 1B). Our findings agree with data from Project Achilles (DepMap), where a panel of 25 PDAC cell lines showed little dependence on CDK4 or CDK6 for proliferation after RNAi-mediated knockdown (Supplementary Fig. S1A). These findings are also consistent with other studies that support the functionally redundant activities of CDK4 and CDK6 (11, 31). All clinically approved inhibitors targeting CDK4 and CDK6 (palbociclib, ribociclib, and abemaciclib), as well as the clinical candidate lerociclib, have dual activity against both kinases, with abemaciclib also inhibiting several transcriptional CDKs, including CDK9 (32).

Because CDK4/6 inhibitors have been reported to increase oxidative cellular metabolism through on-target mechanisms (14, 33), we first determined whether these activities would confound metabolism-based cell viability assays. We found that metabolism-based viability assays (MTT, alamar blue and ATPLite) dramatically underestimated the antiproliferative activity of the CDK4/6 inhibitor palbociclib (but not the ERK1/2-selective inhibitor SCH772984) compared with plate-based automated direct cell counting, presumably due to the palbociclib-induced upregulation of mitochondrial respiration and ATP production (Supplementary Fig. S1B; ref. 14). Therefore, we used direct cell counting in all viability assays to avoid falsely underestimating CDK4/6 inhibitor activity.

We found that 13 of 15 PDAC cell lines were sensitive to palbociclib, the first clinically approved CDK4/6 inhibitor, in an anchorage-dependent proliferation assay (Fig. 1C, Supplementary Fig. S1C). PDAC cells were sensitive to palbociclib and other approved CDK4/6 inhibitors in anchorage-dependent (Supplementary Fig. S1D and S1E), and anchorage-independent clonogenic colony growth formation (Supplementary Fig. S1F and S1G). Palbociclib reduced the level of phosphorylated (inactivated) retinoblastoma (pRB), a well-validated marker for CDK4/6 inhibition, at an IC₅₀ value of approximately 100 nmol/L (Supplementary Fig. S1H), comparable with what has been described previously (34). As expected, CDK4/6 inhibition induced a strong G₁ arrest, with treatment increasing the G₁ population by 80% (Supplementary Fig. S1I). The strong similarity in mechanistic and phenotypic activity seen with chemically distinct CDK4/6 inhibitors suggests on-target growth suppression.

Strong short-term efficacy of protein kinase inhibitors is frequently limited by long-term treatment-induced kinome reprogramming and the onset of resistance mechanisms (35, 36). For example, although retinoblastoma phosphorylation was strongly inhibited at 24 hours of CDK4/6i treatment, pRB returned to pretreatment levels by 3 to 5 days (Supplementary Fig. S1J). To identify further compensatory signaling changes caused by CDK4/6 inhibition, we treated six PDAC cell lines with palbociclib for 5 days and applied RPPA analyses to monitor alterations in protein phosphorylation or expression of cancer signaling proteins (37). As anticipated given the genetic heterogeneity of PDAC (5), there was considerable cell line variation in signaling activities detected (Supplementary Fig. S1K and S1L). However, the six cell lines responded similarly to palbociclib overall (Fig. 1D). CDK4/6 inhibition consistently reduced pRB and the cell proliferation marker Ki-67, as expected. Palbociclib also induced strong compensatory modulation of key signaling pathways, including upregulation of proteins in the antiapoptotic and PI3K–AKT–mTOR prosurvival signaling pathway (Fig. 1E). Surprisingly, the single largest upregulation was increased phosphorylation and activation of ERK (pERK), indicating dramatically increased upstream proproliferative signaling through the ERK–MAPK pathway. We further validated treatment-induced ERK activation using the ectopically expressed ERK kinase translocation reporter (KTR), which consists of the ERK substrate recognition motif from ELK1 tagged with mRuby2 red fluorescent protein (38). We observed significantly higher ERK activity after CDK4/6i treatment compared with vehicle control, as indicated by higher cytoplasmic levels of the ERK–KTR fluorescence (Fig. 1F and G; Supplementary Fig. S1M). To confirm this finding, we measured pERK activity after palbociclib, abemaciclib, or ribociclib treatment using a TR-FRET assay and saw a strong dose-dependent increase in pERK levels (Supplementary Fig. S1N). In totality, we found that three chemically distinct CDK4/6 inhibitors increase pERK across a panel of six cell lines, as assayed using RPPA, pERK TR-FRET, Western blots, and an ERK activity reporter cell line. Multiple lines of evidence, therefore, suggest that CDK4/6i-driven ERK activation is a major mechanism by which cancer cells compensate to counter CDK4/6 suppression and the concomitant antiproliferative signaling.

Because CDK4/6i treatment induced compensatory pro-proliferative ERK upregulation, we reasoned that combined therapy would cooperatively block PDAC cell growth. We found that concurrent treatment with the ERK1/2-selective inhibitor (ERKi) SCH772984 and CDK4/6i synergistically reduced the proliferation of PDAC cell lines in anchorage-dependent (Fig. 2A–C) and colony formation assays (Fig. 2D; Supplementary Fig. S2A). Lower concentrations of ERKi and CDK4/6i were required to cause the same loss of proliferation when used in combination than when used separately, and combined treatment sensitized PDAC lines that were initially resistant to CDK4/6i (e.g., MIA PaCa-2). Similar results were seen with the combination of lerociclib with SCH772984 (Supplementary Fig. S2B) or with a second ERK1/2-selective inhibitor (ulixertinib/BVD-523) together with CDK4/6i (Supplementary Fig. S2C). Recent studies determined that concurrent CDK4/6 inhibition enhanced the activity of direct KRAS G12C mutant-selective inhibitors (39). We found that KRAS^G12C-mutant MIA PaCa-2 cells were resistant to CDK4/6i, but concurrent treatment with the KRAS^G12C-mutant selective inhibitor sotorasib/AMG510 synergistically suppressed MIA PaCa-2 growth (Supplementary Fig. S2D and S2E).

Treatment with ERKi or CDK4/6i alone caused a partial reduction in pRB, but combined treatment caused more complete loss of pRB and reduction of phosphorylation of the ERK cytoplasmic substrate p90RSK (pRSK; Fig. 2E). These results were phenocopied by siRNA silencing of CDK4 and CDK6 concurrently with ERKi treatment (Supplementary Fig. S2F). Notably, although treatment with CDK4/6i alone caused a transient reduction in pRB, concurrent ERKi and CDK4/6i treatment caused sustained reduction in pRB (Fig. 2F). Taken together, our results strongly suggest that concurrent CDK4/6i and ERKi synergistically decrease viability in PDAC.

We next determined how concurrent CDK4/6 and ERK inhibition synergistically reduced cell viability. Combined CDK4/6i or ERKi induced G₁ cell-cycle arrest as anticipated on the basis of the strong arrest caused by each agent individually (Supplementary Fig. S2G and S2H; refs. 40, 41). Both palbociclib single-agent and combination treatment with ERKi induced senescence (Supplementary Fig. S2I–S2K). However, although ERKi or CDK4/6i treatment alone caused limited apoptosis, the combination markedly increased apoptosis in the majority of cell lines tested (Fig. 3A and B; Supplementary Fig. S3A).

We next measured a system-wide profile of temporal signaling changes caused by single and combination inhibitor treatment using RPPA analyses by measuring total and phosphoprotein levels at times ranging from one to 14 days. We found that signaling responses to single-agent CDK4/6i or ERKi treatment differed strongly at 24 and 72 hours (Supplementary Fig. S3B; Supplementary Table S1). Whereas ERKi caused both down- and upregulation of protein signaling activities that were consistent over time, 24-hour treatment with CDK4/6i was almost exclusively inhibitory and not correlated to the response in ERKi-treated cells. However, after three days, a significant fraction of signaling in CDK4/6i-treated cells was reactivated, and by day 14, CDK4/6i and ERKi signaling compensations were highly correlated (Fig. 3C; Supplementary Fig. S3C). The correlated long-term response was not an indirect consequence of G₁ arrest because both ERKi and CDK4/6i cause G₁ arrest by 24 hours (Supplementary Fig. S1I). In contrast, CDK4/6i-driven signaling changes are highly correlated with CDK4/6i + ERKi treatment at all time points (Supplementary Fig. S3D), suggesting that CDK4/6i effects supersede those caused by ERKi. Ultimately, the cellular response at longer time points to all three inhibitors was largely similar, though the magnitude of the response to combination treatment was generally greater than that seen with single-agent treatment.

Kinome alteration extended far beyond the cell-cycle regulatory proteins (Supplementary Fig. S3E). In particular, RTK-mediated activation of RAS effector signaling increased sharply after five and 14 days of CDK4/6i treatment, as did activation of the PI3K–AKT–mTOR-S6 kinase signaling axis, and AMPK, PAK1, and IGF1R protein kinases. These results are consistent with our previous finding that ERKi-mediated changes involved not only downregulation of signaling activities that drive growth suppression but also upregulation of RTK-mediated compensatory signaling changes (29, 42). Interestingly, the 5-day response to CDK4/6i, ERKi, or the combination resembled the signaling alterations caused by acute siRNA knockdown of KRAS (72 hours; Supplementary Fig. S3F; ref. 43). Thus, the direct consequences of KRAS suppression were phenocopied by concurrent inhibition of ERK and CDK4/6.

The combination of CDK4/6i and ERKi suppressed signaling activity more strongly than either inhibitor alone. Retinoblastoma dephosphorylation and activation was decreased more substantially by the combination, as were cell-cycle activating proteins (e.g., CDK1, CDK2, cyclin A and B1, Aurora A, FOXM1; Fig. 3C and D). ERK activity itself, as measured by phosphorylated ERK protein or ERK substrate RSK, was blocked by the combination despite increasing with CDK4/6i alone (Supplementary Fig. S3G).

We recently showed that MYC loss was essential for the antitumor activity of ERKi in PDAC, and that direct MYC suppression impaired PDAC tumorigenic growth (42). Whereas CDK4/6i treatment alone only weakly reduced MYC protein levels, concurrent treatment with CDK4/6i and ERKi strongly ablated MYC protein levels (Fig. 3E). Thus, potent MYC reduction may be a key basis for the synergistic growth suppression caused by the combination. To address this possibility, we performed a rescue experiment by ectopic overexpression of MYC and measured cell viability upon treatment with CDK4/6i and ERKi in a dose–response matrix. Overexpression of MYC antagonized the response to CDK4/6i, ERKi, or combination treatment (Supplementary Fig. S3H and S3I), supporting a role for MYC loss in the growth-suppressing activity of these inhibitors. In MIA PaCa-2 cells, the EC₅₀ value for CDK4/6i alone or in combination with ERKi increased over 10-fold compared with empty vector control, whereas the single-agent ERKi EC₅₀ value increased 2-fold.

We next evaluated the combination in KRAS-mutant PDAC patient-derived organoid models that may better model patient response (44). Organoid cell viability was synergistically reduced by the combination (Fig. 4A–C; Supplementary Fig. S4A and S4B), and apoptosis was increased as determined by caspase activation (Fig. 4D). The combination also synergistically suppressed the growth of KRAS-mutant colorectal cancer organoids (Supplementary Fig. S4C–S4H), suggesting that the effectiveness of this combination is not limited to PDAC.

Most clinical and preclinical evidence suggests that CDK4/6 inhibitors are substantially more effective in combination with other targeted therapies (45). To identify additional combinations that have not been described previously, we applied a CRISPR-Cas9 loss-of-function screen to identify genes whose loss sensitized cells to CDK4/6i treatment and therefore represent potential combination strategies with CDK4/6i. We applied a focused library targeting 2,240 genes that comprise major cellular and oncogenic signaling networks (Fig. 5A; Supplementary Fig. S5A; refs.28, 29). We screened six KRAS-mutant PDAC cell lines treated with a sublethal dose (GI₃₀ or below) of CDK4/6i (Fig. 5B; Supplementary Fig. S5B). Similar time points, treatment groups, and cell lines clustered together by using principal component analysis (Supplementary Fig. S5C and S5D) and in a Pearson correlation matrix (Supplementary Fig. S5E). Individual genes were scored for their effect on proliferation using MAGeCK (Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout)-MLE (maximum likelihood estimation), where negative beta scores indicate sensitivity to loss of a gene (21). Essential and nonessential genes closely agreed with prior studies with excellent precision/recall (Supplementary Fig. S5F; ref. 46).

We evaluated which gene knockouts selectively affected cell proliferation in combination with two or four weeks of CDK4/6i at sublethal concentrations. As expected, loss of RB1 potently antagonized CDK4/6i growth-suppressive activity in all cell lines (Supplementary Fig. S5G–S5I). Because our PDAC cell lines are very heterogeneous, we averaged the selectivity scores across the six cell lines at both time points to find the consensus-sensitizing hits (Fig. 5C and D; Supplementary Tables S2 and S3). To increase confidence in our hits, we separately picked hits using RSA (22), an orthogonal approach that largely agreed with MAGeCK to identify genes that enhanced growth suppression in a CDK4/6i-selective manner (Fig. 5E, Supplementary Fig. S5J; Supplementary Table S4).

As expected, there was significant heterogeneity among the six cell lines, even among the 50 most CDK4/6i-sensitizing proteins (Fig. 5D). These genes encode proteins centered around diverse signaling nodes, including PI3K–AKT–mTOR signaling, cell-cycle regulation and mitosis, SRC family kinase signaling, cell metabolism and biosynthesis, chromosome regulation and maintenance, and DNA damage and repair pathways. Genes related to metabolism and DNA damage were highly enriched. Several hits are known to reduce CDK4/6 activity when lost, such as RUNX1 (47) and HDACs (48), immediately suggesting a mechanistic basis for their identification.

Our RPPA analysis found that CDK4/6 inhibition resulted in activation of signaling proteins with known driver functions in cancer, including PAKs and components of the PI3K–AKT pathway. That these pathways were also found as sensitizing CRISPR screen hits further emphasizes their importance as compensatory mechanisms to overcome loss of CDK4/6 function. Using a random walk network propagation analysis on CDK4/6i-sensitizing hits based on the PPI-BIOGRID database (27, 49), we found that HDAC2 and MYC displayed the highest random walk probability of the MAGeCK-MLE and CRISPR hits, respectively, followed closely by epigenetic regulators, DNA-damage response genes, and CDK2 (Supplementary Fig. S5K–S5M). Suppression of MYC is a mechanistic basis for the synergistic growth inhibitory activity caused by concurrent suppression of CDK4/6 and ERK (Fig. 3E), and supporting this mechanism, genes that regulate MYC expression were also identified in the CRISPR screen. These included the tumor-suppressor FBXW7, which encodes an E3 ligase that directly causes MYC protein degradation driven through the ERK-regulated phosphodegron sequence (50).

We also identified genes whose loss imparted a survival advantage under CDK4/6i treatment, which as expected includes RB1 (Supplementary Fig. S5I and S5N, S5O). Loss of many of these genes have an antiproliferative effect by themselves (e.g., CDK1, CHEK2), but the antiproliferative effect in combination with CDK4/6i was less than additive.

Genes whose loss increased sensitivity to CDK4/6i were broadly involved in regulation of RNA Pol II transcription, the cell cycle, and DNA damage (Supplementary Fig. S6A). To validate individual-sensitizing genes, we applied siRNA oligos, whose acute and partial suppression of expression better mimics the partial inhibition seen with pharmacologic inhibitors (pooled with four constructs per gene). Consistent with their identification in the CRISPR screen, suppression of these genes caused a sensitizing and frequently synergistic shift in the CDK4/6i EC₅₀ value in one or more cell lines (Fig. 6A). For example, suppression of CDK2, BAP1 or EZF3 expression synergistically increased sensitivity to CDK4/6i (Fig. 6B). We observed synergism between BAP1 loss and CDK4/6 inhibition in our CRISPR screen and validation (Fig. 6A and B, Supplementary Fig. S6B and S6C). The deubiquitylase BAP1 is classically (though not exclusively) described as a tumor-suppressor driving cancer progression when lost through germline or somatic loss-of-function mutations (51). However, in this context, our results support a pro-growth function for BAP1 in pancreatic cancer and a synergistic antiproliferative effect with CDK4/6i.

To further validate the screen, we evaluated small-molecule inhibitors of some of the hits and determined that many synergistically decreased cell proliferation with CDK46i (red, Fig. 6C). For example, the SRC inhibitor bosutinib and HDAC inhibitor quisinostat synergistically increased CDK4/6i activity (Fig. 6D and E). Some of the combinations were either additive (white) or less-than-additive (blue), though all combinations resulted in greater overall reduction in viability as compared with single-agent treatment in at least one cell line.

Autophagy, a tightly regulated process whereby cells break down intracellular components, is often upregulated as a stress-tolerance response (52). It has been reported that CDK4/6 inhibition promoted compensatory upregulation of autophagy (53), and we recently determined that autophagy is elevated in response to ERK inhibitor-mediated suppression of glycolysis, resulting in enhanced sensitivity to autophagy inhibition (19). Interestingly, our CRISPR screen identified multiple genes that encode autophagy-blocking proteins as sensitizers to CDK4/6i, including CSNK2A/B (CK2) and PHF23. We used a fluorescence-based reporter to confirm that CDK4/6 inhibition caused increased autophagic flux (Supplementary Fig. S6D and S6E), supporting the conclusion that elevated autophagy upon CDK4/6i is an adaptive response to promote PDAC cell survival. Providing a mechanistic basis for CDK4/6i-induced autophagy, our RPPA dataset identified phosphorylation and activation of AMPK by combination CDK4/6i and ERKi treatment (Fig. 3C). AMPK activation is a key initiating step in the induction of the autophagy pathway (52). To exploit the role of autophagy as a compensatory and protective response to CDK4/6i treatment, we evaluated concurrent treatment with CDK4/6i and the autophagy inhibitor chloroquine and observed synergistic loss of viability after five-day treatment (Supplementary Fig. S6F and S6G).

CDK2 was the strongest hit from the CRISPR screen (Figs. 5D and 6A and B; Supplementary Fig. S5K). CDK2 is activated by cyclin E, the cyclin E–CDK2 complex cooperates with cyclin D–CDK4/6 to promote G₁ progression, and CDK2 can partially compensate for CDK4/6 loss to promote G₁–S transition (31). CDK2 additionally helps maintain cells in S phase, regulates apoptosis by stabilizing antiapoptotic MCL-1 levels, activates AKT through direct phosphorylation, and of particular importance, replicates ERK activity by stabilizing MYC through direct phosphorylation on S62 (54). Each of these roles would support concurrent CDK2 inhibition in combination with CDK4/6i. To address this possibility, we used the potent and selective CDK2/4/6 inhibitor (CDK2/4/6i) clinical candidate PF-06873600 (55). We found that CDK4/6i and CDK2/4/6i showed comparable EC₅₀ concentrations in blocking cell proliferation in anchorage-dependent viability assays (Supplementary Fig. S7A). However, they caused distinct consequences on cell-cycle progression. Although CDK4/6i caused G₁ arrest, CDK2/4/6i treatment caused minimal increase in cells in G₁ and instead, the percentage of cells in S phase increased from 27% to 36% on average (Fig. 7A). Furthermore, CDK2/4/6i induced levels of apoptosis between 3.5- and 8.7-fold higher than CDK4/6i (Fig. 7B). CDK2/4/6 inhibition also caused a much larger increase in DNA damage than CDK4/6i (Supplementary Fig. S7B and S7C), and caused strong increases in polycentrism, leading to anaphase collapse and apoptosis (Supplementary Fig. S7B and S7D).

Because CDK2/4/6i and CDK4/6i treatments caused different phenotypic consequences, we applied RPPA analyses to determine system-wide inhibitor-induced temporal changes in signaling protein activation or expression. CDK2/4/6i sharply downregulated protein expression and phosphorylation in a time-dependent manner, with the strongest effect seen after 14 days (Fig. 7C; Supplementary Fig. S7E). Proliferation and mitosis markers were strongly inhibited (Ki67 and histone H3 pS10), and CDK2/4/6i more potently suppressed MYC expression than did CDK4/6i (Fig. 7D). Retinoblastoma and cyclins A2, B1, D2, E1, and E2 were strongly blocked (Fig. 7E). As we saw with CDK4/6i, CDK2/4/6i treatment also caused a compensatory increase in ERK activation (Fig. 7F, Supplementary Fig. S7F) that strongly suggested that a CDK2/4/6 combination might also synergize with ERKi. Indeed, combining CDK2/4/6i with ERKi blocked pERK activation and further reduced cell viability in cell lines (Supplementary Fig. S7G) and PDAC patient-derived organoids, with strong or partial synergy observed in each of the six organoid models tested (Fig. 7G; Supplementary Fig. S7H). Finally, to assess the on-target activity of CDK2/4/6i, we used siRNA to silence CDK2 and observed that the combination of CDK2 siRNA and CDK4/6i led to a cooperative dose-dependent decrease in antiproliferative activity as well as to increased apoptosis compared with either alone (Supplementary Fig. S7I–S7K). Thus, concurrent CDK2 suppression together with CDK4/6i treatment phenocopied the activities seen with CDK2/4/6i.

Despite the well-validated role of CDKN2A tumor-suppressor loss in PDAC development, pharmacologic restoration of CDKN2A function with CDK4/6 protein kinase inhibitors has shown limited clinical efficacy in PDAC (30). To identify combination strategies to enhance CDK4/6 inhibitor activity in KRAS-mutant PDAC, we applied RPPA to profile CDK4/6 inhibition-induced signaling changes that may drive de novo resistance, and we performed a CRISPR-based genetic loss-of-function screen to identify genes that modulate PDAC sensitivity to CDK4/6 inhibition. Together, these analyses identified multiple combinations that synergistically enhanced CDK4/6 inhibitor activity in preclinical models of KRAS-mutant PDAC and KRAS-mutant colorectal cancer. Our CRISPR screen expanded the field of known synergistic combinations with CDK4/6i, with the highest enrichment consisting of RAS pathway proteins (including ERK2), and encompassing metabolic glycolysis pathway; DNA synthesis, replication, and damage pathways; other cell-cycle regulatory proteins; antiapoptotic proteins; and epigenetic regulators. The broad range of these hits suggests that CDK4/6i combinations may have clinical utility outside of the limited range of combinations currently evaluated.

A critical limitation seen with essentially all protein kinase inhibitors involves treatment-induced compensatory signaling activities that drive de novo and acquired resistance (35, 36). Our system-wide analyses of signaling changes in response to prolonged CDK4/6 inhibition determined that KRAS-mutant PDAC cells attempt to overcome the deleterious consequences of CDK4/6 suppression by upregulating several key oncogenic pathways, including antiapoptotic proteins and the PI3K–AKT–mTOR and ERK-MAPK-RAS effector signaling networks. These findings were conducted in anchorage-dependent in vitro cellular systems. It is probable that a similar study conducted in a tumor model with intact tumor and stromal microenvironment (TME) would identify additional inhibitor bypass mechanisms, given the strong role the TME plays in driving resistance, as has been shown specifically for CDK4/6 inhibition in other contexts (56)

The rational combination of CDK4/6i with ERKi to block this compensatory activation was effective in both KRAS-mutant PDAC and colorectal organoids. Interestingly, whereas the combination with MEKi showed enhanced induction of senescence but not apoptosis, we found that concurrent CDK4/6i and ERKi treatment showed strong induction of apoptosis and a lower required concentration of each inhibitor in combination to elicit the same loss of viability seen when used as a single-agent. This difference may reflect the refractoriness of ERKi to ERK reactivation mediated through loss of ERK-dependent negative feedback inhibitory mechanisms (57).

Our findings provide the rationale for our initiation of a phase I trial evaluating the ERK inhibitor ulixertinib in combination with palbociclib in patients with metastatic pancreatic cancer (NCT03454035). Our observations agree with other studies that determined a similar enhancement of CDK4/6i activity when combined with MEK inhibitor (MEKi) treatment (10, 15) in PDAC, providing the rationale for a Phase I perioperative analysis of the MEKi binimetinib and palbociclib in KRAS-mutant PDAC and other cancers (NCT04870034).

RAS-addicted PDAC cells are dependent on autophagy for survival, in particular when the RAS–MAPK pathway is inhibited, and blockade of autophagy is synergistically toxic in this context (19). CDK4/6i has been reported to increase autophagy, so we were not surprised that hits from the CRISPR screen with CDK4/6i block autophagy. We tested this hypothesis with the general autophagy inhibitor hydroxychloroquine and found strong synergy with CDK4/6i to block growth.

From our CRISPR screen, our most consistent sensitizer to CDK4/6i was CDK2, a kinase that can compensate for CDK4/6 loss to promote G₁–S transition and is also important in regulating S-phase transition, cell division, and other pathways (31). Comparing the activities of a CDK2/4/6 inhibitor with CDK4/6 inhibition alone, we observed significantly greater induction of apoptotic death and DNA damage upon treatment with CDK2/4/6i, supporting an enhanced therapeutic potency of concurrent CDK2 inhibition. However, as with CDK4/6i, ERK reactivation remained a limitation for CDK2/4/6i, and concurrent ERK inhibition to block this shared compensatory response further enhanced CDK2/4/6i growth inhibitory activity. We also determined that MYC loss provided one basis for the more potent growth suppression seen with CDK2/4/6i that was further enhanced with concurrent ERK inhibition. The importance of MYC loss for combination inhibitor efficacy is supported by a recent determination that MYC is a driver of resistance to CDK4/6 inhibition (55).

During the course of completion of our studies, two studies reported that concurrent CDK2 inhibition enhanced CDK4/6 inhibitor antitumor activity in vitro and in vivo. First, Kumarasamy and colleagues (58) reported that concurrent CDK2 and CDK4/6 inhibition was more effective in pancreatic cancer due to more efficient restoration of retinoblastoma activity. Second, Freeman-Cook and colleagues (55) determined that CDK2 was a significant driver of resistance to CDK4/6 inhibition, leading to their development of PF-06873600 for concurrent inhibition of CDK2/4/6. They showed that PF-06873600 was active in a large panel of patient-derived PDAC tumors. However, both studies found that CDK2/4/6i caused a greater induction of senescence compared with CDK4/6i, whereas we observed a more potent apoptotic activity with triple CDK inhibition. The basis for the different findings is not clear.

We identified loss of MYC expression as one mechanistic basis for the potent growth suppressive activity of combined CDK4/6 and ERK inhibition. We determined previously that PDAC sensitivity to ERK inhibition correlated with proteasomal degradation of MYC (41, 42). We found that whereas CDK4/6i alone did not reduce MYC expression, it caused a synergistic decrease in MYC when combined with ERKi treatment. Thus, as we have described previously with other ERK inhibitor combinations, despite the highly divergent targets of the combined inhibitors (tubulin, SRC tyrosine, and CHK1 serine/threonine kinases), they converged on driving loss of MYC (59, 60). This consistent role of MYC in diverse ERK inhibitor combinations reflects the key role of MYC in ERK-dependent PDAC growth (29, 42).

In summary, our application of unbiased system-wide signaling and genetic loss-of-function screens identified a diverse spectrum of functionally distinct signaling components that antagonize CDK4/6i efficacy in PDAC. These findings led us to identify and validate inhibitor combinations that enhance CDK4/6 inhibitor cytotoxicity. Among these combinations, concurrent inhibition of CDK2 and ERK represents a logical and promising combination. Currently, the triple CDK2/4/6 inhibitor PF-06873600 is under phase I/IIa clinical evaluation in hormone receptor–positive, HER2-negative breast and ovarian cancer. Our results, together with those of two recent studies (55, 58), support the clinical evaluation of CDK2 inhibitors in combination with CDK4/6i for KRAS-mutant PDAC.

C.M. Goodwin reports grants from NIH during the conduct of the study. K.L. Bryant reports grants and nonfinancial support from SpringWorks Therapeutics and Deciphera Therapeutics outside the submitted work. M. Pierobon reports grants from NIH/NCI during the conduct of the study, as well as personal fees from Theralink Inc. outside the submitted work. A.P. Beelen reports support from G₁ Therapeutics, Inc. during the conduct of the study, as well as support from G₁ Therapeutics, Inc. outside the submitted work, and reports patents 11357779 and 11395821 issued. K.C. Wood reports grants from NIH and DoD during the conduct of the study; grants, personal fees, and support from Tavros Therapeutics and Celldom, and personal fees from Guidepoint Global, Bantam Pharmaceuticals, and Apple Tree Partners outside the submitted work. E.F. Petricoin reports personal fees from Theralink Technologies, Inc., Perthera, Inc., and Ceres Nanosciences, Inc. outside the submitted work. A.J. McRee reports support from Johnson and Johnson outside the submitted work. A.D. Cox reports a patent for US-2019055562-A1 issued. C.J. Der reports grants from National Institutes of Health, American Cancer Society, AACR, Pancreatic Cancer Action Network, Slomo and Cindy Silvian Foundation, Lustgarten Foundation, Julie M. Brown and Christina Gianoplus Colon Cancer Foundation, Deutsche Forschungsgemeinschaft, and Royster Society of Fellows during the conduct of the study; as well as grants and personal fees from Mirati Therapeutics, Deciphera Therapeutics, and grants from SpringWorks, grants and personal fees from Revolution Medicines, Reactive Biosciences, and personal fees from Ribometrix, Turning Point Therapeutics, Jazz Therapeutics, Eli Lilly, and Sanofi outside the submitted work. No disclosures were reported by the other authors.

C.M. Goodwin: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. A.M. Waters: Data curation, formal analysis, investigation, methodology. J.E. Klomp: Conceptualization, data curation, formal analysis, investigation. S. Javaid: Data curation, formal analysis, investigation, methodology. K.L. Bryant: Conceptualization, formal analysis, investigation, methodology. C.A. Stalnecker: Data curation, formal analysis, investigation, methodology. K. Drizyte-Miller: Data curation, formal analysis, investigation, writing–review and editing. B. Papke: Investigation, methodology. R. Yang: Data curation, investigation. A.M. Amparo: Data curation, formal analysis, investigation. I. Ozkan-Dagliyan: Data curation, investigation, methodology. E. Baldelli: Data curation, investigation, methodology. V. Calvert: Data curation, investigation, methodology. M. Pierobon: Data curation, investigation, methodology. J.A. Sorrentino: Data curation, formal analysis. A.P. Beelen: Supervision. N. Bublitz: Data curation, formal analysis, investigation. M. Lüthen: Data curation, formal analysis, investigation. K.C. Wood: Conceptualization, methodology. E.F. Petricoin III: Conceptualization, supervision, investigation, methodology. C. Sers: Supervision, funding acquisition, methodology. A.J. McRee: Conceptualization, writing–review and editing. A.D. Cox: Conceptualization, formal analysis, supervision, funding acquisition, writing–review and editing. C.J. Der: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.

The authors thank Brian Golitz for assistance with Illumina sequencing, and David Tuveson (Cold Spring Harbor Laboratory) and Calvin Kuo (Stanford University) for PDAC organoids. C.M. Goodwin was supported by NCI T32CA009156 and F32CA221005. A.M. Waters was supported by a fellowship from the American Cancer Society (PF-18–061). S. Javaid was supported by a fellowship from the Royster Society of Fellows. J.E. Klomp was supported by NCI T32CA009156, NCI F32CA239328, and American Cancer Society PF-20–069. K.L. Bryant was supported by the 2015 Pancreatic Cancer Action Network-AACR Pathway to Leadership Grant (15–70–25-BRYA), the NCI (P50CA196510 and R37CA251877), and the Sky Foundation. C.A. Stalnecker was supported by NCI T32CA009156 and F32CA232529. K. Drizyte-Miller was supported by NCI T32CA009156 and American Cancer Society (PF-22–066–01). B. Papke was supported by the Deutsche Forschungsgemeinschaft (DFG PA 3051/1–1). I. Ozkan-Dagliyan was supported by the Slomo and Cindy Silvian Foundation. M. Pierobon and E.F. Petricoin III were supported by NIH grant CA203657. A.D. Cox and/or C.J. Der received support from the NCI (R01CA42978, R01CA175747, P50CA196510, U01CA199235 and P01CA203657, and R35CA232113), and from the 2015 Pancreatic Cancer Action Network–AACR Research Acceleration Network Grant (15–90–25-DER), Department of Defense (W81XWH-15–1-0611), the Lustgarten Foundation (388222), and Julie M. Brown and Christina Gianoplus Colon Cancer Foundation.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.