Cyclin-dependent kinase 4 (CDK4) and CDK6 are key cell-cycle regulators that are frequently dysregulated in human malignancies. CDK4/6 inhibitors are clinically approved for the treatment of hormone receptor–positive, HER2–negative (HR+/HER2) breast cancer, but improved specificity and reduced toxicity might expand their use to other indications. Through analysis of publicly available genome-wide loss-of-function data combined with single and dual-targeting CRISPR assays, we found differential cell proliferation vulnerability of cell lines to either CDK4 deletion alone, CDK6 deletion alone, combined CDK4/CDK6 deletion, or neither. CDK6 expression was the best single predictor of CDK4 (negatively correlated) and CDK6 (positively correlated) dependencies in the cancer cell lines, with adenocarcinoma cell lines being more sensitive to CDK4 deletion and hematologic and squamous cancer cell lines being more sensitive to CDK6 deletion. RB–E2F signaling was confirmed as a main downstream node of CDK4/6 in these experiments as shown by the survival effects of RB1 deletion. Finally, we show in a subset of cancer cell lines not dependent on CDK4/6 that CDK2–CCNE1 is an important alternative dependency for cell proliferation. Together, our comprehensive data exploration and functional experiments delineate the landscape of pan-cancer CDK4/6 gene dependencies and define unique cancer cell populations that might be sensitive to CDK4-selective or CDK6-selective inhibitors.

Significance:

This study provides functional genomic insight toward understanding the scenarios in which cancer cells are differentially sensitive to CDK4 or CDK6 inhibition and their implications in current treatment strategies.

The cell-cycle machinery is tightly regulated to ensure cell proliferation in normal tissues (1). Cyclin-dependent kinases 4 and 6 (CDK4/6) are important in governing the cell-cycle progression from the gap phase (G1) to the DNA synthesis (S) phase (2–4). In the presence of mitogenic signals, CDK4/6 bind to D-type cyclins (cyclin D1, D2, and D3) and form catalytically active complexes. One key function of these complexes is to phosphorylate the retinoblastoma tumor-suppressor protein (RB1; refs. 5, 6), leading to the activation of the E2F transcriptional program to enter the G1 phase of the cell cycle. Besides CDK4/6, CDK2 can also phosphorylate RB upon activation through binding to E-type cyclins (cyclin E1 and E2) and further drive cell-cycle progression (7).

Hyperactivation of the CDK4/6-cyclin D complex, which provides sustained signaling for cell proliferation, is commonly found in human cancers (8, 9). Numerous mechanisms have been documented, such as copy-number gain of CDK4, CDK6, and/or D-type cyclin genes, loss or inactivation of CDK inhibitor family genes, including CDKN2A, CDKN2B, and CDKN2C, and transcriptional activation of D-type cyclins due to upstream genomic/epigenomic events, including promoter rearrangements (10–13). Three CDK4/6 inhibitors (palbociclib, ribociclib, and abemaciclib) are now approved to treat patients with hormone receptor–positive (HR+), HER2-negative (HER2) advanced breast cancer in combination with hormone therapy (14–16). Abemaciclib is also approved as a single agent, second-line therapy in relapsed, hormone-refractory patients (17). In addition to breast cancer, CDK4/6 inhibitors have also shown activity in different preclinical in vitro and in vivo tumor models and have been actively under clinical investigation in many cancer types (18).

CDK6 is known to play a critical role in hematopoiesis, and therefore, systemic CDK4/6 inhibition also decreases the propagation of hematopoietic cells in the bone marrow (19, 20). As a result, neutropenia and leukopenia are commonly reported in clinical trials of CDK4/6 inhibitors (16, 21, 22). More importantly, neutropenia is also the most common type of serious (grade 3 or 4) adverse effects, affecting 25% to 65% of patients from multiple clinical trials (16, 21, 22). Although mild side effects associated with CDK4/6 inhibition can be adequately managed with standard supportive care, severe hematological disorders often result in dose reduction or treatment interruption. Phase III clinical trials of palbociclib (PALOMA-2 and PALOMA-3) and phase III clinical trial of ribociclib (MONALEESA-2) reported that dose-reduction and permanent treatment discontinuation due to hematological toxicities occurred in about 30% and 10% of participants, respectively (22–24). Roughly, similar percentages of patients underwent dose-reduction or dose-omission due to neutropenia in the phase III MONARCH-2 and MONARCH-3 trials of abemaciclib (25). These lines of evidence suggest that CDK4/6 inhibitor-associated hematological toxicity can potentially limit drug efficacy.

Despite the success of CDK4/6 inhibition in patients with HR+ breast cancer, resistance to CDK4/6 inhibitors is inevitable in most patients. The identification of biomarkers to predict how much patients would respond to CDK4/6 inhibitors remains incomplete. Interestingly, about 10% of patients experience innate resistance to CDK4/6 inhibitors, and clinically useful biomarkers to identify such patients before treatment would facilitate a prompt switch to a more effective treatment. Loss of RB1 has been implicated as a mechanism of both innate and acquired resistance in both preclinical models and clinical settings (26, 27). For instance, Li and colleagues (27) reported that the progression-free survival of 9 patients with HR+ breast cancer with RB1 loss who were subsequently treated with CDK4/6 inhibitors was only 3.6 months compared with 10.1 months for 329 patients with intact RB1. Cyclin E1 (CCNE1) upregulation serves as another CDK4/6 bypass mechanism and has been shown to attenuate the response to CDK4/6 inhibition (28, 29).

In this study, we aim to delineate a CDK4/6 “essentiality map” in human cancers to help guide future therapeutic strategies. We analyzed publicly available genome-wide loss-of-function screen datasets to classify human cancer cell lines based on their CDK4 or CDK6 gene–selective dependencies. In parallel, we assessed CDK4/6 requirements for cell viability using CRISPR-knockout (CRISPR-KO) for either single gene or dual gene. Through this analysis, we have defined cancer cell lines that are uniquely sensitive to CDK4 or CDK6 deletion as well as jointly sensitive cell lines and have also defined a CDK4/CDK6-independent population of cell lines with a requirement for CDK2/cyclin E1 activity.

Cell lines

The following cell lines were purchased from the ATCC: Human lung cancer cell lines A549, NCI-H1915, NCI-H1568, NCI-H23, NCI-H1048; human breast cancer cell lines HCC38, HCC1428; human pancreatic cancer cell line: MIAPACA-2. The following cell lines were obtained from the Cancer Cell Line Encyclopedia project (30): Human lung cancer cell lines NCI-H2030, NCI-H3122, NCI-H1792, HARA, human acute myelogenous leukemia cell line MOLM13. Human breast cancer cell lines T47D and MCF7, and esophageal cancer cell line OE21 were kindly provided by Dr. Adam Bass (Dana-Farber Cancer Institute). All cell lines were grown at 37°C and 5% CO2 and were confirmed to be Mycoplasma negative. Cell lines were cultured in RPMI supplemented with 10% FBS and 1% penicillin/streptomycin, except for MCF7 and MIAPACA2 cell lines that were maintained in DMEM containing 10% FBS and 1% penicillin/streptomycin.

CRISPR-Cas9 gene knockout

Single guide (sg) RNA sequences, designed using the sgRNA Designer tool (http://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design), are shown in Supplementary Table S1. For single and combinatorial CDK4/6 gene knockout experiments, DNA fragments containing two sgRNA sequences were synthesized by Twist Bioscience (https://www.twistbioscience.com/) and were then cloned into the pLC108 lentiviral vector, which was constructed from the backbone of lentiGuide-puro (Addgene #52963) and the mU6 promoter from pDonor-mU6 (Addgene #69350). For lentivirus production, individual lentiviral vectors were introduced along with packaging vectors into 293T cells using the TransIT-LT1 transfection reagent (Mirusbio, #MIR2304). Lentivirus was then harvested according to protocols from the Broad Institute GPP Web Portal (https://portals.broadinstitute.org/gpp/public/resources/protocols). To generate constitutive Cas9-expressing cell lines, cells were infected with the Cas9 lentivirus and selected in 10 μg/mL blasticidin for at least 7 days before use. Cas9-expressing cell lines were then infected with sgRNA lentivirus and selected in 1–2 μg/mL puromycin for at least 2–3 days. Thereafter, protein lysates were collected from the transduced cells, and protein levels of the targeted gene(s) were assessed by immunoblotting.

For RB1 deletion model, sgRB1 RNA sequence was cloned into Cas9-expressing lentiCRISPR v2-Blast (Addgene #83480). Lentivirus was harvested as described before. Transduced cell lines were selected in 10 μg/mL blasticidin for at least 7 days before use in assays. Thereafter, protein lysates were collected, and RB1 gene knockout efficiency was assessed by immunoblotting. For CDK2 gene knockout experiment, sgRNAs that target CDK2 and control were cloned into lentiCRISPRv2-Puro (Addgene #98290). Lentivirus was harvested as described before. Transduced cell lines were selected in 1–2 μg/mL puromycin for at least 2–3 days before use in assays. Thereafter, protein lysates were collected from the transduced cells, and protein levels of the targeted gene(s) were assessed by immunoblotting.

Immunoblotting

Cells were lysed in RIPA lysis buffer (Millipore Sigma #R0278) in the presence of 1× protease and phosphatase inhibitor cocktails (Thermo Fisher Scientific #78441). Protein concentrations were obtained using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific #23225). Protein extracts were then resolved on 4%–12% NuPAGE Bis-Tris protein gel (Thermo Fisher Scientific) before transferring onto nitrocellulose membrane (Bio-Rad #1620112) for overnight incubation with the following primary antibodies: CDK4 (#12790), CDK6 (#3136), phospho-RB Ser780 (#8180), cyclin A2 (#4656), CDK2 (#2546), α/β-Tubulin(#2148) from Cell Signaling Technology. All primary antibodies were used at a dilution of 1:1,000. Membranes were then incubated with anti-rabbit (LI-COR: 926-32211, 1:10,000) or anti-mouse (LI-COR: 926-68020, 1:10,000) secondary antibodies for 1 hour at room temperature before imaging on the LI-COR digital imaging system.

Cell proliferation

Cell counting was performed using a Vi-Cell XR Cell Counter (Beckman-Coulter) every two days, and the cell population doubling was calculated after four counts. For crystal violet staining, cells were plated at a density ranging from 1,000 to 25,000 cells per well in 12-well tissue culture plates according to cell doubling time. After 10–14 days, cells were fixed in cold methanol for 15 minutes on ice, stained with 0.5% crystal violet solution (made in 25% methanol) for 10 minutes at room temperature, and washed three times with water. Crystal violet quantifications were performed by measuring the absorbance of crystal violet destained by 10% acetic acid at 590 nm. All cell viability assays were performed with two technical replicates.

Cell-cycle analysis

Cell-cycle progression was quantified by flow cytometry as described previously (31). Cells were harvested 2–3 days after lentiviral transduction selection, washed with PBS, fixed in cold 70% ethanol, and stained with propidium iodide (PI). DNA content was measured using the LSR II flow cytometer and analyzed with FCS Express software (De Novo software).

RB1 status identification

Mutation, copy number, and expression data from the CCLE (30) were used to determine cell line RB1 status. A cell line was characterized with an RB1-null status if it (i) harbors homozygous RB1 copy-number loss; (ii) carries homozygous or bi-allelic damaging RB1 mutations (nonsense mutation, splice site mutation, frameshift deletion, frameshift insertion, de novo start out of frame mutation); or (iii) expresses RB1 mRNA at a lower level than that of cell lines reported to harbor homozygous RB1 copy-number loss, this group representing examples such as undetected homozygous deletion, disruptive rearrangement, and/or promoter deletion; or (iv) has been reported to harbor infection with human papillomavirus (HPV) or Merkel cell polyomavirus (32, 33). Identified RB1-null cell lines were manually cross-referenced with independent databases from the ATCC (https://www.atcc.org/cell-products/human-cells/genetic-mutants#t=productTab&numberOfResults=12) and Cellosaurus (34). The rest of the cell lines were annotated as RB1-proficient.

CDK4/6 gene dependency classification

The Broad Institute Dependency Map group has analyzed gene dependency for each gene and cell line based on multiple CRISPR guides for each gene, providing two major output values: A gene dependency or “CERES'' score (lower CERES scores are correlative with more negative log-fold changes while also correcting for copy number, gene_effect_unscaled.csv) and a score for the probability of the dependency value being robust, which also takes into account the estimated data quality for a given cell line (gene_dependency.csv; refs. 35, 36). Here, we classified cancer cell lines according to their probabilities of dependency and their relative dependencies on CDK4 and CDK6.

A given cell line is classified as selectively CDK4-dependent if (i) the probability of dependency for CDK4 > 0.5; and (ii) the probability of dependency for CDK6 < 0.5; and (iii) CERES CDK4 − CERES CDK6 < −x, where −x is the median CERES score for pan-essential genes, −1.79 for the Broad analysis and −1 for the Sanger analysis. A cell line is classified as selectively CDK6-dependent if (i) the probability of dependency for CDK6 > 0.5; and (ii) the probability of dependency for CDK4 < 0.5; and (iii) CERES CDK6 − CERES CDK4 < −x, where −x is the median CERES score for pan-essential genes, −1.79 for the Broad analysis and −1 for the Sanger analysis.

A cell line is classified as CDK4/6 nondependent if (i) the probability of dependency for CDK4 < 0.5; and (ii) the probability of dependency for CDK6 < 0.5; and (iii) CERES CDK4 + CERES CDK6 > CERESCDK4 + CERESCDK6 for the RB1-null cell line with the strongest knockout effect, where −0.8 for Broad analysis and 0.02 for Sanger analysis.

All other cell lines are classified as CDK4/6 jointly dependent. For these cell lines, (i) the probability of dependency for CDK4 > 0.5; and/or (ii) the probability of dependency for CDK6 > 0.5; and/or (iii) CERESCDK6 − CERESCDK4 > −x, where −x is the median CERES score for pan-essential genes, −1.79 for the Broad analysis and −1 for the Sanger analysis; and/or 4) CERES CDK4 + CERES CDK6 < CERES CDK4 + CERES CDK6 for the RB1-null cell line with the strongest knockout effect.

Drug treatment

For the palbociclib treatment experiment, 1–5 × 104 cells per well were plated in 12-well plates. Cells were treated with palbociclib at different concentrations and the drug was replenished every 3 days. After 6 days, cells were fixed in methanol for 15 minutes on ice and then stained with 0.5% crystal violet solution (made in 25% methanol) for 10 minutes at room temperature, and washed three times with water.

Data availability

The Broad institute CRISPR-KO screen data [21Q2 version, Achilles_gene_effect_(unscaled).csv], the Sanger institute CRISPR-KO screen data [Sanger CRISPR CERES: gene_effect_(unscaled).csv], and RNAi DEMETER2 data (D2_combined_gene_dep_scores.csv) were accessed from the DepMap web portal (https://depmap.org/portal/). Cancer cell line data used in this study, including cell line annotation and gene expression determined by RNA sequence, were retrieved from the Broad Institute CCLE database and downloaded from the DepMap web portal (https://depmap.org/portal/download/). RNA sequencing-based gene expression data and tumor histology data from The Cancer Genome Atlas (TCGA) consortium were downloaded from the Broad Institute GDAC firehose web portal (http://gdac.broadinstitute.org/). Palbociclib data from the Sanger Institute Genomics of Drug Sensitivity in Cancer and the Broad Institute PRISM repurposing screen 19Q4 were retrieved from DepMap web portal (https://depmap.org/portal/).

Data analysis

The unpaired Wilcoxon test was used in Fig. 1A, B, and E, Fig. 5BD, Fig. 6D and E, Supplementary Fig. S1A and S1B, Supplementary Fig. S3D, Supplementary Fig. S6A, S6E, and S6F, Supplementary Fig. S7B, Supplementary Fig. S8A and S8C. Pearson correlation analysis was performed on variables extracted from gene dependency data and gene expression data in Fig. 4AD (https://depmap.org/portal/: 21Q2 version, Custom Analyses function), Fig. 6B, Supplementary Fig. S1C–F, Supplementary Fig. S5A–D (https://depmap.org/portal/: Custom Analyses function), Supplementary Fig. S5G, Supplementary Fig. S8B, Supplementary Fig. S10 (https://depmap.org/portal/: Custom Analyses function). Two-class comparisons of gene dependency and gene expression were performed with a limma-based linear model using Custom Analyses function of Depmap portal (21Q2 version) for Fig. 5A and Fig. 6A and C. Kruskal-Wallis test followed by a post-hoc test with a two-step set-up method of Benjamini, Krieger and Yekutieli was performed in Fig. 2BE, Fig. 3AC, Fig. 5E and F, Supplementary Fig. S2A, S2C and D, Supplementary Fig. S3C, Supplementary Fig. S4, Supplementary Fig. S6C, S6D, and S6H, Supplementary Fig. S7D. Two-way ANOVA test was performed in Supplementary Fig. S2B. Fisher exact test was performed in Supplementary Fig. S7A, Supplementary Fig. S8D. Drug response area under the curve (AUC) from the current study was calculated with Prism 9 software and data was normalized against the largest calculated AUC. P values less than 0.05 were considered statistically significant. All statistical tests were performed using R version 4.0.3 or GraphPad Prism 9 software.

Figure 1.

Landscape of CDK4/6 gene dependencies across human cancer cell lines. A, Boxplot of CDK4 dependency (CERES) score of RB1-proficient and RB1-null cell lines from the Broad CRISPR-KO screen data. The lower the CERES score, the higher the gene dependency. The unpaired Wilcoxon test was performed to compare the two groups. B, Boxplot of CDK6 dependency (CERES) score of RB1-proficient and RB1-null cell lines from the Broad CRISPR-KO screen data. The unpaired Wilcoxon test was performed to compare the two groups. C, Scatter plot showing the predicted CDK4/6 dependency classes of cell lines included in the Broad CRISPR-KO screen data (see Supplementary Table S2 for details). D, Scatter plot showing the predicted CDK4/6 dependency classes of cell lines included in the Sanger CRISPR-KO screen data (see Supplementary Table S3 for details). E, Bar graph plotting the distribution of CDK4/6 gene dependencies across major tumor lineages (excluding RB1-null cell lines) included in the Broad CRISPR-KO screen data. The unpaired Wilcoxon test followed by Benjamini–Hochberg correction was performed to compare the tumor lineage of interest and all lineages. Bold represents tumor lineages that are statistically more dependent on CDK4 (or CDK6) than all, and italic denotes tumor lineages that are statistically less dependent on CDK4 (or CDK6) than all (FDR q value <0.01).

Figure 1.

Landscape of CDK4/6 gene dependencies across human cancer cell lines. A, Boxplot of CDK4 dependency (CERES) score of RB1-proficient and RB1-null cell lines from the Broad CRISPR-KO screen data. The lower the CERES score, the higher the gene dependency. The unpaired Wilcoxon test was performed to compare the two groups. B, Boxplot of CDK6 dependency (CERES) score of RB1-proficient and RB1-null cell lines from the Broad CRISPR-KO screen data. The unpaired Wilcoxon test was performed to compare the two groups. C, Scatter plot showing the predicted CDK4/6 dependency classes of cell lines included in the Broad CRISPR-KO screen data (see Supplementary Table S2 for details). D, Scatter plot showing the predicted CDK4/6 dependency classes of cell lines included in the Sanger CRISPR-KO screen data (see Supplementary Table S3 for details). E, Bar graph plotting the distribution of CDK4/6 gene dependencies across major tumor lineages (excluding RB1-null cell lines) included in the Broad CRISPR-KO screen data. The unpaired Wilcoxon test followed by Benjamini–Hochberg correction was performed to compare the tumor lineage of interest and all lineages. Bold represents tumor lineages that are statistically more dependent on CDK4 (or CDK6) than all, and italic denotes tumor lineages that are statistically less dependent on CDK4 (or CDK6) than all (FDR q value <0.01).

Close modal
Figure 2.

Validation of CDK4/6 gene dependency classes using combinatorial CRISPR-Cas9 KO. A, Schematic diagram of combinatorial CRISPR-Cas9 KO plasmid. B, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4-selective dependent cell line, NCI-H1915. Cell proliferation was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM. C, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK6-selective–dependent cell line, NCI-H1568. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM. D, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4/6-joint–dependent cell line, NCI-H2030. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM.E, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4/6 nondependent cell line, NCI-H1048. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05; **, P < 0.01; ****, P < 0.0001, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Figure 2.

Validation of CDK4/6 gene dependency classes using combinatorial CRISPR-Cas9 KO. A, Schematic diagram of combinatorial CRISPR-Cas9 KO plasmid. B, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4-selective dependent cell line, NCI-H1915. Cell proliferation was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM. C, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK6-selective–dependent cell line, NCI-H1568. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM. D, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4/6-joint–dependent cell line, NCI-H2030. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from three independent biological replicates are shown. Data are shown as mean ± SEM.E, Left, immunoblots showing protein levels of CDK4, CDK6, and tubulin (loading control) in representative CDK4/6 nondependent cell line, NCI-H1048. Cell proliferation was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05; **, P < 0.01; ****, P < 0.0001, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Close modal
Figure 3.

The RB–E2F pathway is essential in both CDK4/6-selective and CDK4/6-joint–dependent cell lines. A, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK4-selective–dependent cell line, NCI-H1915, after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. B, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK6-selective–dependent cell line, NCI-H1568, after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. C, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK4/6-joint–dependent cell line NCI-H2030 after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. D, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H1915 (CDK4-selective dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H1915 was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. E, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H1568 (CDK6-selective dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H1568 was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. F, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H2030 (CDK4/6-joint dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H2030 was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Figure 3.

The RB–E2F pathway is essential in both CDK4/6-selective and CDK4/6-joint–dependent cell lines. A, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK4-selective–dependent cell line, NCI-H1915, after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. B, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK6-selective–dependent cell line, NCI-H1568, after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. C, Top, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative CDK4/6-joint–dependent cell line NCI-H2030 after control or CDK4/6 KO with CRISPR-Cas9. Bottom, the percentage of G1 of the cell cycle was assessed by flow cytometry after control or CDK4/6 KO. Two sgRNAs of the same KO condition were combined for statistical analysis. Three independent biological replicates were performed for each cell line. Data are shown as mean ± SEM. D, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H1915 (CDK4-selective dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H1915 was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. E, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H1568 (CDK6-selective dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H1568 was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. F, Immunoblots (left) showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in control or RB1-KO NCI-H2030 (CDK4/6-joint dependent) after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation of RB1-KO NCI-H2030 was assessed by crystal violet staining 12–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Close modal

Human cancer cell lines show differential CDK4/6 gene dependencies

To characterize CDK4/6 gene dependencies across human cancer cell lines, we analyzed CRISPR dependency data from the Broad Institute Dependency Map and the Sanger Institute Dependency Map (https://depmap.org/portal/). Consistent with previous reports, RB1-null cell lines are generally not dependent on CDK4 or CDK6, corroborating that RB1 loss is an intrinsic resistance mechanism to CDK4/6 inhibition [Fig. 1A and B (Broad data) and Supplementary Fig. S1A and S1B (Sanger data; refs. 26, 37)]. We observed a modest (R = −0.14 for Broad data) but statistically significant (P = 2.9e−5 for Broad data) inverse correlation between CDK4 and CDK6 gene dependencies (Supplementary Fig. S1C); this correlation became stronger when RB1-null cells were removed from the analysis (Supplementary Fig. S1D).

We then sought to identify cell lines with differential CDK4/6 dependencies according to CERES score (35, 36) and probability of dependency (Supplementary Table S2; ref. 35). As described in the Methods and Materials (CDK4/6 gene dependency classification section), we defined cell lines as selectively CDK4-dependent if they showed a probability of CDK4 dependency >0.5, a probability of CDK6 dependency <0.5, and a difference in CERES scores (magnitude of the depletion effect on cell proliferation) between CDK4 and CDK6 that are above a set threshold. Selective CDK6 dependence was defined the same way but with the gene identities reversed. Using this approach for the Broad dataset, we identified 73 CDK4-selective–dependent cancer cell lines and 52 CDK6-selective–dependent cancer cell lines (Fig. 1C; Supplementary Table S2). For the remaining 734 cell lines analyzed using the criteria in the Methods and Materials section, 576 cell lines were classified as CDK4/6-jointly dependent, and 158 were classified as CDK4/6 nondependent cell lines (Fig. 1C; Supplementary Table S2). For the Sanger dataset, we identified 51 CDK4-selective–dependent cancer cell lines, 35 CDK6-selective–dependent cancer cell lines, 147 CDK4/6-jointly dependent cancer cell lines, and 84 CDK4/6 nondependent cancer cell lines (Fig. 1D; Supplementary Table S3). Consistent with previous studies (38), we found a largely concordant CDK4/6 dependency classification (∼65%) between the two CRISPR-KO screen datasets, considering 187 cell lines screened independently by both institutes. In addition, we also observed strong correlation of CDK4/6 gene dependency scores (R = 0.62–0.72) between complementary CRISPR-KO and RNAi screens as reported before (Supplementary Fig. S1E and S1F; ref. 39).

Analysis of tumor lineages for CDK4/6 gene dependencies suggested that CDK4/6 requirement varies by cancer lineages. Of note, and consistent with previous reports, tumor cells of breast and bone lineages are more reliant on CDK4 compared with CDK6 (40–42). In contrast, also as previously reported (43, 44), CDK6 appears to be a stronger gene dependency than CDK4 in cancer cells derived from blood and lymphoid lineages [Fig. 1E (Broad data) and Supplementary Fig. S1G (Sanger data)].

Next, we aimed to validate the CDK4/CDK6 differential selectivity classification by performing CRISPR-KO of CDK4 and/or CDK6 in multiple cell lines predicted to have different CDK4 and CDK6 gene dependencies. To achieve homogenous and simultaneous perturbation of CDK4 and CDK6, we engineered our working vectors to incorporate two sgRNAs driven by two promoters from opposite directions (Fig. 2A). CDK4/6 knockout efficiency was assessed by immunoblotting, and cell viability was evaluated by crystal violet staining and cell counting. Knocking out CDK4 in cell lines predicted to be CDK4-selective–dependent (Breast: T47D, NSCLC: NCI-H1915, NCI-H3122) significantly reduced cell viability (Fig. 2B; Supplementary Fig. S2A). As expected, CDK4 KO resulted in comparable cell growth inhibition with CDK4/6 double KO in these cell lines, whereas CDK6 KO did not impair cell proliferation (Fig. 2B; Supplementary Fig. S2A). For predicted CDK6-selective–dependent cell lines (NSCLC: NCI-H1568, AML: MOLM13), depletion of CDK6 rather than CDK4 substantially decreased cell proliferation; note that cell counting rather than colony formation is used for MOLM13 suspension cells (Fig. 2C; Supplementary Fig. S2B). This CDK6-driven cell growth inhibition was comparable with dual CDK4/6 KO in these cell lines (Fig. 2C; Supplementary Fig. S2B). We also observed that in non–small cell lung cancer (NSCLC) cell lines NCI-H2030 and A549, double KO of CDK4/6 resulted in more substantial cell proliferation defects than KO of CDK4 alone or CDK6 alone, suggesting that these cell lines need either CDK4 or CDK6 to proliferate (Fig. 2D; Supplementary Fig. S2C). Finally, we confirmed that CDK4/6 depletion did not cause any noticeable cell proliferation arrest in predicted CDK4/6 nondependent cell lines (small-cell lung cancer, SCLC; RB1-null, NCI-H1048; ovarian: OVCAR8; Fig. 2E; Supplementary Fig. S2D).

In addition, we evaluated the response of the tested cell lines to the CDK4/6 inhibitor, palbociclib. CDK4/6-dependent cells exhibited variable responses to CDK4/6 inhibition, whereas CDK4/6 nondependent cells were resistant to palbociclib (Supplementary Fig. S3A), with some variation between datasets (Supplementary Fig. S3B). Of note, cell lines with CDK4-selective dependency, CDK6-selective dependency, and joint CDK4/CDK6 dependency all have similar palbociclib sensitivity (Supplementary Fig. S3C), as expected given that palbociclib has similar potency against CDK4 and CDK6 (45, 46). Last, CDK4/6-dependent cell lines are overall more sensitive to palbociclib than CDK4/6 nondependent cell lines (Supplementary Fig. S3D).

RB-E2F pathway is essential in both CDK4/6-selective and CDK4/6-joint–dependent cell lines

We next examined whether CDK4/6 KO-induced cell proliferation arrest is associated with RB–E2F pathway modulation, as evaluated by RB1 phosphorylation, the expression of E2F target gene cyclin A, and the fraction of cells in the G0–G1 phase of the cell cycle, and whether it is specific for CDK4 or CDK6 when the dependency is kinase-specific. As expected, for CDK4-selective–dependent cell lines, CDK4 but not CDK6 KO resulted in significant suppression of RB1 phosphorylation and cyclin A expression and a higher fraction of cells in the G1 phase of the cell cycle. The degree of modulation of these components of the RB–E2F pathway by CDK4 KO in these cell lines is comparable with dual CDK4/6 KO (Fig. 3A; Supplementary Fig. S4A). For CDK6-selective–dependent cell lines, CDK6 KO alone achieved a strong RB–E2F pathway perturbation comparable with double CDK4/6 KO, whereas CDK4 KO did not perturb RB–E2F signaling (Fig. 3B; Supplementary Fig. S4B). Consistently, double CDK4/6 KO resulted in a more significant RB–E2F suppression than single paralog KO in CDK4/6-joint–dependent cell lines (Fig. 3C; Supplementary Fig. S4C). Furthermore, RB1 deletion rescued the cell lethality induced by CDK4/6 KO in all cell lines tested, consistent with the view that RB1 is the major downstream target of CDK4/6 in cells that are dependent on CDK4, CDK6, or both (Fig. 3DF). These data suggest that the cellular phenotype upon CDK4/6 KO is related to the kinase function of CDK4/6 as it impacts RB1 phosphorylation and function, regardless of individual gene essentiality. However, rescue experiments with kinase-dead compared with wild-type kinases would be necessary to further elucidate this potential dependency on enzymatic activity.

CDK6 expression is a potential biomarker for CDK4/6 gene dependencies

To explore molecular correlates of CDK4/CDK6 gene dependency, we compared both pan-transcriptome mRNA expression data and pan-proteomic protein abundance data to these dependencies. In expression versus dependency correlation analyses on all CDK4/6-dependent cell lines using the Broad CRISPR-KO screen data, we found that the strongest anti-correlates for CDK4 gene dependency are CDK6 gene expression and CDK6 protein abundance (Fig. 4A and C). The strongest positive correlates for CDK6 dependency are likewise CDK6 gene expression and CDK6 protein abundance (Fig. 4B and D). CDK6 expression is also the most correlated molecular feature for both CDK4 and CDK6 gene dependencies from the Sanger CRISPR-KO screen data (Supplementary Fig. S5A–S5D). Furthermore, we confirmed with immunoblotting that CDK6 protein expression is variable with the highest level in CDK6-selective–dependent cells, whereas the protein expression of CDK4 is less variable across the panel of CDK4/6-dependent cell lines included in Fig. 2 and Supplementary Fig. S2 (Supplementary Fig. S5E). Together, these data suggest that cell lines with higher CDK6 expression are more dependent on CDK6 and that cell lines with lower CDK6 expression tend to rely more on CDK4 for cell proliferation.

Figure 4.

CDK6 expression is predictive for CDK4 and CDK6 gene dependencies. A, Analysis of correlation between CDK4 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and gene expression from CCLE RNA-seq data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. B, Analysis of correlation between CDK6 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and gene expression from CCLE RNA-seq data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. C, Analysis of correlation between CDK4 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and protein abundance from CCLE proteomics data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. D, Analysis of correlation between CDK6 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and protein abundance from CCLE proteomics data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. E, Violin plot of CDK6 gene expression across tumor lineages from CCLE RNA-seq data.

Figure 4.

CDK6 expression is predictive for CDK4 and CDK6 gene dependencies. A, Analysis of correlation between CDK4 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and gene expression from CCLE RNA-seq data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. B, Analysis of correlation between CDK6 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and gene expression from CCLE RNA-seq data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. C, Analysis of correlation between CDK4 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and protein abundance from CCLE proteomics data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. D, Analysis of correlation between CDK6 gene dependency (-CERES score) from the Broad CRISPR-KO screen data and protein abundance from CCLE proteomics data. Data are plotted by −log10 (q value) on the y-axis versus Pearson correlation coefficients on the x-axis. D-type cyclins (CCND1, CCND2, and CCND3), CDK4, and CDK6 are labeled. E, Violin plot of CDK6 gene expression across tumor lineages from CCLE RNA-seq data.

Close modal

In concordance with our earlier analysis, tumor lineages with higher overall CDK4 dependency (breast and bone cells) have the lowest CDK6 gene expression. In contrast, tumor lineages with higher CDK6 dependence (hematopoietic and lymphoid, plasma cells) have among the highest CDK6 gene expression (Fig. 4E). Consistently, CDK6 gene expression analysis from TCGA patient samples showed a similar association with cancer type (Supplementary Fig. S5F). We further performed correlation analysis for specific tumor lineages. We found that CDK6 gene expression is predictive for both CDK4/6 dependencies in many lineages (bile duct, bone, breast, gastric and esophageal, hematopoietic and lymphoid, lung, ovary, pancreas, soft tissue, and urinary tract). In other lineages, however, CDK6 gene expression was either predictive for only one gene dependency (colorectal, kidney, liver, peripheral nervous system, and urinary tract) or not predictive (central nervous system, head and neck, plasma cell, skin, and uterus; Supplementary Fig. S5G).

Association of CDK6 expression with CDK4/6 gene dependencies in cancer subtypes

We next sought to determine the potential implication of CDK6 gene expression in CDK4/6 gene dependency stratification within cancer subtypes. Earlier studies have reported that CDK6 expression is highly variable among cancer types; several studies have shown that cancers of squamous cells have higher CDK6 expression than adenocarcinomas (47–49). Thus, we hypothesized that adenocarcinoma and squamous cell carcinoma (SCC) would be differentially dependent on CDK4/6 for cell proliferation. Analysis of the Broad and the Sanger CRISPR-KO screen data showed that adenocarcinoma cell lines were more dependent on CDK4, whereas cell lines of SCC were reliant more on CDK6 (Fig. 5A; Supplementary Fig. S6A). Gene expression analysis on cell lines (CCLE) and clinical samples (TCGA) also showed a higher CDK6 gene expression in squamous cells (Fig. 5B). To reduce the possibility that the preferential CDK4/6 gene dependency between adenocarcinoma and SCC is due to sampling bias, we further focused on NSCLC because adenocarcinoma and SCC subtypes of NSCLC are well represented in the Broad CRISPR-KO screen data (Supplementary Table S4) and TCGA. Lung adenocarcinoma cell lines generally are more dependent on CDK4 than lung SCC cell lines (Fig. 5C) and lung adenocarcinoma cell lines generally express less CDK6 mRNA than lung SCC cell lines (Fig. 5D). Finally, after confirming the relative expression of CDK4 and CDK6 (Supplementary Fig. S6B), we performed CRISPR-KO to show that CDK4 KO substantially decreased cell proliferation in three adenocarcinoma cell lines (MIAPACA-2, NCI-H23, and NCI-H1792; Fig. 5E; Supplementary Fig. S6C). In comparison, CDK6 KO decreased cell proliferation in two aerodigestive SCC cell lines more significantly (HARA and OE21; Fig. 5F; Supplementary Fig. S6D).

Figure 5.

Association of CDK6 expression with CDK4/6 gene dependencies in cancer subtypes. A, Analysis of the Broad CRISPR-KO screen data showing differential gene dependency CERES scores in adenocarcinoma cell lines versus squamous cell carcinoma cell lines as illustrated in the volcano plot. Each dot represents a gene and is plotted by −log10 (q value) on the y-axis and effect size explaining the mean difference between the two groups on the x-axis. B, Gene expression of CDK6 in adenocarcinoma and squamous cell carcinoma samples from CCLE and TCGA RNA-seq data. The unpaired Wilcoxon test was performed to compare the two groups. C, Gene dependency CERES scores of CDK4 and CDK6 in NSCLC adenocarcinoma cell lines and NSCLC squamous cell carcinoma lines from the Broad CRISPR-KO screen data. The unpaired Wilcoxon test was performed to compare the two groups. D, Gene expression of CDK6 in NSCLC adenocarcinoma and squamous cell carcinoma samples from CCLE and TCGA RNA-seq data. The unpaired Wilcoxon test was performed to compare the two groups. E, Left, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative adenocarcinoma cell line, MIAPACA-2, after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation cell line was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. F, Left, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative squamous cell carcinoma cell line, HARA, after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Figure 5.

Association of CDK6 expression with CDK4/6 gene dependencies in cancer subtypes. A, Analysis of the Broad CRISPR-KO screen data showing differential gene dependency CERES scores in adenocarcinoma cell lines versus squamous cell carcinoma cell lines as illustrated in the volcano plot. Each dot represents a gene and is plotted by −log10 (q value) on the y-axis and effect size explaining the mean difference between the two groups on the x-axis. B, Gene expression of CDK6 in adenocarcinoma and squamous cell carcinoma samples from CCLE and TCGA RNA-seq data. The unpaired Wilcoxon test was performed to compare the two groups. C, Gene dependency CERES scores of CDK4 and CDK6 in NSCLC adenocarcinoma cell lines and NSCLC squamous cell carcinoma lines from the Broad CRISPR-KO screen data. The unpaired Wilcoxon test was performed to compare the two groups. D, Gene expression of CDK6 in NSCLC adenocarcinoma and squamous cell carcinoma samples from CCLE and TCGA RNA-seq data. The unpaired Wilcoxon test was performed to compare the two groups. E, Left, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative adenocarcinoma cell line, MIAPACA-2, after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation cell line was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. F, Left, immunoblots showing protein levels of CDK4, CDK6, phospho-RB(S780), cyclin A, and tubulin (loading control) in representative squamous cell carcinoma cell line, HARA, after control or CDK4/6 KO with CRISPR-Cas9. Cell proliferation was assessed by crystal violet staining 10–14 days after control or CDK4/6 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (right) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05, as calculated by the Kruskal–Wallis test followed by the post hoc test with two-step set-up method of Benjamini, Krieger, and Yekutieli. Sg control was used as reference in the test.

Close modal

This differential CDK4/CDK6 dependency and expression were also observed in breast cancer. Although most breast tumors express a low level of CDK6, triple-negative breast cancers (TNBC) were shown to have higher CDK6 expression than HR+ luminal breast cancer (50, 51). Because CDK6 gene expression strongly correlates with CDK4/6 gene dependencies in breast cancer (Supplementary Fig. S5G), we hypothesized that breast cancer of HR+ and TNBC would be differentially dependent on CDK4/6. As expected, analysis of the Broad CRISPR-KO screen data showed that HR+ breast tumors are generally more dependent on CDK4 but less dependent on CDK6 than TNBC cell lines (Supplementary Fig. S6E). Consistently, CCLE and TCGA gene expression analysis showed elevated CDK6 gene expression in TNBC breast cancers compared with HR+ breast cancers (Supplementary Fig. S6F). Finally, we examined the protein expression and validated the predicted differential CDK4/6 dependencies on two HR+ cell lines (HCC1428 and MCF7) relative to two TNBC breast tumor lines (HCC1806 and HCC38; Supplementary Fig. S6G and S6H).

The CDK2-CCNE1 axis represents a main pathway dependency in CDK4/6 nondependent, RB1-proficient cells

Different mechanisms have been reported in tumors to bypass the requirement of CDK4/6 for cell-cycle entry. RB1 loss due to either genomic alteration or protein degradation is a contributing factor in the development and progression of SCLC, cancers that are HPV-positive, and Merkel cell virus-positive (32, 33, 52). Interestingly, our analysis above suggested that RB1 loss is found in only a fraction of cell lines predicted to be CDK4/6 nondependent, suggesting possible alternative gene dependencies in RB1-proficient, CDK4/6 nondependent cell lines (Fig. 1C and D). Differential gene dependency analysis showed that the CDK2–CCNE1 axis represents major gene dependencies for RB1-proficient, CDK4/6 nondependent cell lines (Fig. 6A). CDK2 and CCNE1 gene dependencies are highly correlated in RB1-proficient, CDK4/6 nondependent cells (Fig. 6B) and high CCNE1 expression was identified as a main molecular property of these cell lines (Fig. 6C). Next, we found that cell lines of the uterus and ovary lineages were overrepresented in CDK4/6 nondependent, RB1-proficient cells, and both lineages showed overall higher gene dependencies of CDK2 and CCNE1 (Supplementary Fig. S7A and S7B; refs. 53, 54). Consistently, we found that ovarian and uterine tumors have the highest CCNE1 expression of all cancer types in CCLE and TCGA (Supplementary Fig. S7C). Furthermore, we performed CRISPR-KO of CDK2 and CDK4/6 to validate the predicted gene dependencies (Fig. 6D and E; Supplementary Fig. S7D, please refer to Supplementary Figs. S2A and S2D, and Fig. S6C for CDK4/6 KO of NCI-H3122, OVCAR8, NCI-H1792, respectively). Last, our analysis of the Sanger CRISPR-KO screen data also demonstrated that a higher pathway dependency of the CDK2–CCNE1 axis, an elevated CCNE1 expression, and an overrepresentation of ovarian tumor cell lines in CDK4/6 nondependent, RB1-proficient cell lines versus CDK4/6-dependent cell lines, which are in line with our analysis above (Supplementary Fig. S8A–S8D).

Figure 6.

The CDK2–CCNE1 axis represents an important pathway dependency in CDK4/6 nondependent cell lines. A, Volcano plot showing differential gene dependency CERES score in CDK4/6-dependent cell lines versus RB1-proficient, CDK4/6 nondependent cell lines (see Fig. 1C; Supplementary Table S2) using the Broad CRISPR-KO screen data. Each dot represents a gene and is plotted by log10 (q value) on the y-axis and effect size explaining the mean difference between the two groups on the x-axis. B, Scatter plot showing the correlation between CDK2 CERES score and CCNE1 CERES score in RB1-proficient, CDK4/6 nondependent cell lines using the Broad CRISPR-KO screen data. Pearson correlation analysis was performed to measure the correlation coefficient. The trend line represents linear regression line and shade represents 95% confidence interval. C, Analysis of differentially expressed genes between CDK4/6-dependent and RB1-proficient, CDK4/6 nondependent cancer cell lines using CCLE RNA-seq data. Individual genes are plotted by log10 (q value) on the y-axis versus effect size on the x-axis. Horizontal dotted line indicates a q value of 0.05. D, Top, immunoblots showing protein levels of CDK2 and tubulin (loading control) in representative CDK4/6 nondependent, CDK2–CCNE1-dependent cell lines, OVCAR8 and OVSAHO, after control or CDK2 KO with CRISPR Cas9. Cell proliferations of these cell lines were assessed by crystal violet staining 10–14 days after control or CDK2 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (bottom) from two independent biological replicates are shown. Data are shown as mean ± SEM. E, Top, immunoblots showing protein levels of CDK2 and tubulin (loading control) in representative CDK4/6-dependent cell lines, NCI-H3122 and NCI-H1792, after control or CDK2 KO with CRISPR Cas9. Cell proliferations of these cell lines were assessed by crystal violet staining 10–14 days after control or CDK2 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (bottom) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05, as calculated by the Wilcoxon test between sg CDK2 and sg control.

Figure 6.

The CDK2–CCNE1 axis represents an important pathway dependency in CDK4/6 nondependent cell lines. A, Volcano plot showing differential gene dependency CERES score in CDK4/6-dependent cell lines versus RB1-proficient, CDK4/6 nondependent cell lines (see Fig. 1C; Supplementary Table S2) using the Broad CRISPR-KO screen data. Each dot represents a gene and is plotted by log10 (q value) on the y-axis and effect size explaining the mean difference between the two groups on the x-axis. B, Scatter plot showing the correlation between CDK2 CERES score and CCNE1 CERES score in RB1-proficient, CDK4/6 nondependent cell lines using the Broad CRISPR-KO screen data. Pearson correlation analysis was performed to measure the correlation coefficient. The trend line represents linear regression line and shade represents 95% confidence interval. C, Analysis of differentially expressed genes between CDK4/6-dependent and RB1-proficient, CDK4/6 nondependent cancer cell lines using CCLE RNA-seq data. Individual genes are plotted by log10 (q value) on the y-axis versus effect size on the x-axis. Horizontal dotted line indicates a q value of 0.05. D, Top, immunoblots showing protein levels of CDK2 and tubulin (loading control) in representative CDK4/6 nondependent, CDK2–CCNE1-dependent cell lines, OVCAR8 and OVSAHO, after control or CDK2 KO with CRISPR Cas9. Cell proliferations of these cell lines were assessed by crystal violet staining 10–14 days after control or CDK2 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (bottom) from two independent biological replicates are shown. Data are shown as mean ± SEM. E, Top, immunoblots showing protein levels of CDK2 and tubulin (loading control) in representative CDK4/6-dependent cell lines, NCI-H3122 and NCI-H1792, after control or CDK2 KO with CRISPR Cas9. Cell proliferations of these cell lines were assessed by crystal violet staining 10–14 days after control or CDK2 KO with CRISPR-Cas9. A representative image of crystal violet staining (middle) and quantitation of cell viability (bottom) from two independent biological replicates are shown. Data are shown as mean ± SEM. ns, not significant; *, P < 0.05, as calculated by the Wilcoxon test between sg CDK2 and sg control.

Close modal

Although current CDK4/6 inhibitors have provided significant clinical benefits for patients with HR+ breast cancer, the safety and efficacy aspects of these small molecules remain to be improved. Neutropenia, the primary cause for dose reduction and treatment discontinuation in approximately 40% of patients in clinical trials, is an on-target toxicity driven by CDK6 inhibition of hematopoietic cells (22–24). Several studies have shown that cyclin D1–CDK4 and cyclin D3–CDK6 complexes are differentially critical for breast cancer and leukemia (40–44). These results suggest that CDK4 and CDK6 may be selectively essential for cell growth in different tumor contexts. Therefore, identifying the CDK4/6 gene dependencies and their predictive molecular characteristics across tumor types would provide important insights into drug innovation and improve current treatment strategies. Indeed, a reported selective CDK4 inhibitor (CDDD2–94) has recently been developed and tested in multiple ovarian cancer models (55, 56). Notably, CDDD2–94 was shown to have better safety profiles than palbociclib in preclinical in vitro and in vivo models (55).

In the present study, we have performed data exploration and functional validation to characterize CDK4/6 gene dependencies across human cancer cell lines. To elaborate, we have categorized cancer cells into four major classes due to different CDK4/6 essentialities for cell proliferation. We predicted a functional overlapping, paralog buffering relationship between CDK4 and CDK6 for most cell lines (57). Interestingly, our data also reveal that about 15% of human cancer cell lines may display selective gene dependency of CDK4 or CDK6. Notably, our results share a high degree of consistency with two recent studies describing CDK4/6 plasticity in selected human cancer models demonstrated by either siRNA- or shRNA-mediated approach (58, 59). In line with previous reports, we further demonstrate the RB–E2F pathway is a vital downstream node of CDK4/6 regardless of the individual contribution of CDK4/6 in cell proliferation.

Next, it is hypothesized that the substantial paralog gene expression variation among tumor cell lines could contribute to the variable dependency of paralogs (57). For the CDK4/6 pair, CDK6 exhibits much more variable expression, whereas the expression of CDK4 remains stable among tumor lineages (Supplementary Fig. S9). Consequently, we show that CDK6 gene expression is a major predictor for gene dependencies of this paralog pair, which is also in line with the prediction from a previous large-scale RNAi screen (Supplementary Fig. S10; ref. 60). As the CDK6 gene expression pattern is highly similar between cell lines and patient samples, we, therefore, believe that CDK6 expression may potentially serve as a useful biomarker of CDK4/6 essentiality for patients.

Furthermore, it emerges from our study that RB1-loss represents a common but not exclusive mechanism responsible for non-CDK4/6–mediated cell proliferation. We, therefore, uncover that the CDK2–CCNE1 axis is an important gene dependency for the RB1-proficient, CDK4/6 nondependent cell lines, and tumor lineages with high CCNE1 expression such as ovary and uterus are most susceptible to CDK2 loss (60). Consistent with earlier reports (29, 58), our data suggest that tumors with hyperactive CDK2–CCNE1 are likely intrinsically resistant to CDK4/6 inhibition and there is evidence for a similar mechanism in acquired resistance to CDK4/6 inhibition (28, 61, 62). Thus, the development of selective CDK2 small-molecule inhibitors or degraders could be helpful for patients of this class. Notably, a few CDK2-targeting compounds have recently been discovered with improved CDK2 specificity (63–65) and one of them (PF-07104091) is now under clinical investigation (NCT04553133).

L. Golomb reports grants from Lung Cancer Research Foundation–AstraZeneca outside the submitted work. M. Meyerson reports grants, personal fees, and other support from Bayer, as well as personal fees from Interline and Isabel, and grants from Janssen, other support from Labcorp., and grants from Ono outside the submitted work. No disclosures were reported by the other authors.

Z. Zhang: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft. L. Golomb: Conceptualization, resources, supervision, investigation, methodology, writing–review and editing. M. Meyerson: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing.

The authors would like to thank Dr. Peter Choi, Dr. Takahiro Ito, and Mr. Michael Young for help on experiment design. This work is supported by grants from the National Cancer Institute to M. Meyerson (P01CA154303 and R35CA197568).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Nurse
P
.
A long twentieth century of the cell cycle and beyond
.
Cell
2000
;
100
:
71
8
.
2.
Meyerson
M
,
Enders
GH
,
Wu
CL
,
Su
LK
,
Gorka
C
,
Nelson
C
, et al
.
A family of human cdc2-related protein kinases
.
EMBO J
1992
;
11
:
2909
17
.
3.
Kato
J
,
Matsushime
H
,
Hiebert
SW
,
Ewen
ME
,
Sherr
CJ
.
Direct binding of cyclin D to the retinoblastoma gene product (pRb) and pRb phosphorylation by the cyclin D-dependent kinase CDK4
.
Genes Dev
1993
;
7
:
331
42
.
4.
Matsushime
H
,
Ewen
ME
,
Strom
DK
,
Kato
JY
,
Hanks
SK
,
Roussel
MF
, et al
.
Identification and properties of an atypical catalytic subunit (p34PSK-J3/cdk4) for mammalian D type G1 cyclins
.
Cell
1992
;
71
:
323
34
.
5.
Meyerson
M
,
Harlow
E
.
Identification of G1 kinase activity for cdk6, a novel cyclin D partner
.
Mol Cell Biol
1994
;
14
:
2077
86
.
6.
Matsushime
H
,
Quelle
DE
,
Shurtleff
SA
,
Shibuya
M
,
Sherr
CJ
,
Kato
JY
.
D-type cyclin-dependent kinase activity in mammalian cells
.
Mol Cell Biol
1994
;
14
:
2066
76
.
7.
Rubin
SM
.
Deciphering the retinoblastoma protein phosphorylation code
.
Trends Biochem Sci
2013
;
38
:
12
9
.
8.
Sherr
CJ
.
Cancer cell cycles
.
Science
1996
;
274
:
1672
7
.
9.
Malumbres
M
,
Barbacid
M
.
Cell cycle, CDKs and cancer: a changing paradigm
.
Nat Rev Cancer
2009
;
9
:
153
66
.
10.
Knudsen
ES
,
Witkiewicz
AK
.
The strange case of CDK4/6 inhibitors: mechanisms, resistance, and combination strategies
.
Trends Cancer Res
2017
;
3
:
39
55
.
11.
Beroukhim
R
,
Mermel
CH
,
Porter
D
,
Wei
G
,
Raychaudhuri
S
,
Donovan
J
, et al
.
The landscape of somatic copy-number alteration across human cancers
.
Nature
2010
;
463
:
899
905
.
12.
Jeselsohn
R
,
Yelensky
R
,
Buchwalter
G
,
Frampton
G
,
Meric-Bernstam
F
,
Gonzalez-Angulo
AM
, et al
.
Emergence of constitutively active estrogen receptor-α mutations in pretreated advanced estrogen receptor–positive breast cancer
.
Clin Cancer Res
2014
;
20
:
1757
67
.
13.
Diehl
JA
,
Cheng
M
,
Roussel
MF
,
Sherr
CJ
.
Glycogen synthase kinase-3beta regulates cyclin D1 proteolysis and subcellular localization
.
Genes Dev
1998
;
12
:
3499
511
.
14.
Finn
RS
,
Crown
JP
,
Lang
I
,
Boer
K
,
Bondarenko
IM
,
Kulyk
SO
, et al
.
The cyclin-dependent kinase 4/6 inhibitor palbociclib in combination with letrozole versus letrozole alone as first-line treatment of oestrogen receptor-positive, HER2-negative, advanced breast cancer (PALOMA-1/TRIO-18): a randomised phase 2 study
.
Lancet Oncol
2015
;
16
:
25
35
.
15.
Hortobagyi
GN
,
Stemmer
SM
,
Burris
HA
,
Yap
Y-S
,
Sonke
GS
,
Paluch-Shimon
S
, et al
.
Ribociclib as first-line therapy for HR-positive, advanced breast cancer
.
N Engl J Med
2016
;
375
:
1738
48
.
16.
Sledge
GW
,
Toi
M
,
Neven
P
,
Sohn
J
,
Inoue
K
,
Pivot
X
, et al
.
MONARCH 2: abemaciclib in combination with fulvestrant in women with HR+/HER2 advanced breast cancer who had progressed while receiving endocrine therapy
.
J Clin Oncol
2017
;
35
:
2875
84
.
17.
Dickler
MN
,
Tolaney
SM
,
Rugo
HS
,
Cortés
J
,
Diéras
V
,
Patt
D
, et al
.
MONARCH 1, a phase II study of abemaciclib, a CDK4 and CDK6 inhibitor, as a single agent, in patients with refractory HR+/HER2 metastatic breast cancer
.
Clin Cancer Res
2017
;
23
:
5218
24
.
18.
Schettini
F
,
De
SI
,
Rea
CG
,
De Placido
P
,
Formisano
L
,
Giuliano
M
, et al
.
CDK 4/6 inhibitors as single agent in advanced solid tumors
.
Front Oncol
2018
;
8
:
608
.
19.
Scheicher
R
,
Hoelbl-Kovacic
A
,
Bellutti
F
,
Tigan
A-S
,
Prchal-Murphy
M
,
Heller
G
, et al
.
CDK6 as a key regulator of hematopoietic and leukemic stem cell activation
.
Blood
2015
;
125
:
90
101
.
20.
Laurenti
E
,
Frelin
C
,
Xie
S
,
Ferrari
R
,
Dunant
CF
,
Zandi
S
, et al
.
CDK6 levels regulate quiescence exit in human hematopoietic stem cells
.
Cell Stem Cell
2015
;
16
:
302
13
.
21.
Verma
S
,
Bartlett
CH
,
Schnell
P
,
DeMichele
AM
,
Loi
S
,
Ro
J
, et al
.
Palbociclib in combination with fulvestrant in women with hormone receptor-positive/HER2-negative advanced metastatic breast cancer: detailed safety analysis from a multicenter, randomized, placebo-controlled, phase III study (PALOMA-3)
.
Oncologist
2016
;
21
:
1165
75
.
22.
Hortobagyi
GN
,
Stemmer
SM
,
Burris
HA
,
Yap
YS
,
Sonke
GS
,
Paluch-Shimon
S
, et al
.
Updated results from MONALEESA-2, a phase III trial of first-line ribociclib plus letrozole versus placebo plus letrozole in hormone receptor-positive, HER2-negative advanced breast cancer
.
Ann Oncol
2019
;
30
:
1842
.
23.
Finn
RS
,
Martin
M
,
Rugo
HS
,
Jones
S
,
Im
S-A
,
Gelmon
K
, et al
.
Palbociclib and letrozole in advanced breast cancer
.
N Engl J Med
2016
;
375
:
1925
36
.
24.
Cristofanilli
M
,
Turner
NC
,
Bondarenko
I
,
Ro
J
,
Im
S-A
,
Masuda
N
, et al
.
Fulvestrant plus palbociclib versus fulvestrant plus placebo for treatment of hormone-receptor-positive, HER2-negative metastatic breast cancer that progressed on previous endocrine therapy (PALOMA-3): final analysis of the multicentre, double-blind, phase 3 randomised controlled trial
.
Lancet Oncol
2016
;
17
:
425
39
.
25.
Rugo
HS
,
Huober
J
,
García-Sáenz
JA
,
Masuda
N
,
Sohn
JH
,
Andre
VAM
, et al
.
Management of abemaciclib-associated adverse events in patients with hormone receptor-positive, human epidermal growth factor receptor 2-negative advanced breast cancer: safety analysis of MONARCH 2 and MONARCH 3
.
Oncologist
2021
;
26
:
e53
65
.
26.
Fry
DW
,
Harvey
PJ
,
Keller
PR
,
Elliott
WL
,
Meade
M
,
Trachet
E
, et al
.
Specific inhibition of cyclin-dependent kinase 4/6 by PD 0332991 and associated antitumor activity in human tumor xenografts
.
Mol Cancer Ther
2004
;
3
:
1427
38
.
27.
Li
Z
,
Razavi
P
,
Li
Q
,
Toy
W
,
Liu
B
,
Ping
C
, et al
.
Loss of the FAT1 tumor suppressor promotes resistance to CDK4/6 inhibitors via the hippo pathway
.
Cancer Cell
2018
;
34
:
893
905
.
28.
Herrera-Abreu
MT
,
Palafox
M
,
Asghar
U
,
Rivas
MA
,
Cutts
RJ
,
Garcia-Murillas
I
, et al
.
Early adaptation and acquired resistance to CDK4/6 inhibition in estrogen receptor–positive breast cancer
.
Cancer Res
2016
;
76
:
2301
13
.
29.
Turner
NC
,
Liu
Y
,
Zhu
Z
,
Loi
S
,
Colleoni
M
,
Loibl
S
, et al
.
Cyclin E1 expression and palbociclib efficacy in previously treated hormone receptor–positive metastatic breast cancer
.
J Clin Oncol
2019
;
37
:
1169
78
.
30.
Barretina
J
,
Caponigro
G
,
Stransky
N
,
Venkatesan
K
,
Margolin
AA
,
Kim
S
, et al
.
The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity
.
Nature
2012
;
483
:
603
7
.
31.
Darzynkiewicz
Z
,
Juan
G
.
DNA content measurement for DNA ploidy and cell-cycle analysis
.
Curr Protoc Cytom
2001
;
Chapter 7:Unit 7.5. doi: 10.1002/0471142956.cy0705s00
.
32.
Gonzalez
SL
,
Stremlau
M
,
He
X
,
Basile
JR
,
Münger
K
.
Degradation of the retinoblastoma tumor suppressor by the human papillomavirus type 16 E7 oncoprotein is important for functional inactivation and is separable from proteasomal degradation of E7
.
J Virol
2001
;
75
:
7583
91
.
33.
Borchert
S
,
Czech-Sioli
M
,
Neumann
F
,
Schmidt
C
,
Wimmer
P
,
Dobner
T
, et al
.
High-affinity Rb binding, p53 inhibition, subcellular localization, and transformation by wild-type or tumor-derived shortened Merkel cell polyomavirus large T antigens
.
J Virol
2014
;
88
:
3144
60
.
34.
Bairoch
A
.
The cellosaurus, a cell-line knowledge resource
.
J Biomol Tech
2018
;
29
:
25
38
.
35.
Dempster
JM
,
Rossen
J
,
Kazachkova
M
,
Pan
J
,
Kugener
G
,
Root
DE
, et al
.
Extracting biological insights from the project achilles genome-scale CRISPR screens in cancer cell lines
.
bioRxiv
2019
;
720243
.
36.
Meyers
RM
,
Bryan
JG
,
McFarland
JM
,
Weir
BA
,
Sizemore
AE
,
Xu
H
, et al
.
Computational correction of copy number effect improves specificity of CRISPR–Cas9 essentiality screens in cancer cells
.
Nat Genet
2017
;
49
:
1779
84
.
37.
Ma
CX
,
Gao
F
,
Luo
J
,
Northfelt
DW
,
Goetz
M
,
Forero
A
, et al
.
NeoPalAna: neoadjuvant palbociclib, a cyclin-dependent kinase 4/6 inhibitor, and anastrozole for clinical stage 2 or 3 estrogen receptor-positive breast cancer
.
Clin Cancer Res
2017
;
23
:
4055
65
.
38.
Dempster
JM
,
Pacini
C
,
Pantel
S
,
Behan
FM
,
Green
T
,
Krill-Burger
J
, et al
.
Agreement between two large pan-cancer CRISPR-Cas9 gene dependency datasets
.
Nat Commun
2019
;
10
:
5817
.
39.
McFarland
JM
,
Ho
ZV
,
Kugener
G
,
Dempster
JM
,
Montgomery
PG
,
Bryan
JG
, et al
.
Improved estimation of cancer dependencies from large-scale RNAi screens using model-based normalization and data integration
.
Nat Commun
2018
;
9
:
4610
.
40.
Yu
Q
,
Sicinska
E
,
Geng
Y
,
Ahnström
M
,
Zagozdzon
A
,
Kong
Y
, et al
.
Requirement for CDK4 kinase function in breast cancer
.
Cancer Cell
2006
;
9
:
23
32
.
41.
Zhou
Y
,
Shen
JK
,
Yu
Z
,
Hornicek
FJ
,
Kan
Q
,
Duan
Z
.
Expression and therapeutic implications of cyclin-dependent kinase 4 (CDK4) in osteosarcoma
.
Biochim Biophys Acta Mol Basis Dis
2018
;
1864
:
1573
82
.
42.
Landis
MW
,
Pawlyk
BS
,
Li
T
,
Sicinski
P
,
Hinds
PW
.
Cyclin D1-dependent kinase activity in murine development and mammary tumorigenesis
.
Cancer Cell
2006
;
9
:
13
22
.
43.
Sicinska
E
,
Aifantis
I
,
Le Cam
L
,
Swat
W
,
Borowski
C
,
Yu
Q
, et al
.
Requirement for cyclin D3 in lymphocyte development and T-cell leukemias
.
Cancer Cell
2003
;
4
:
451
61
.
44.
Placke
T
,
Faber
K
,
Nonami
A
,
Salih
HR
,
Sykes
SM
,
Root
DE
, et al
.
Requirement for CDK6 In MLL-rearranged acute myeloid leukemia
.
Blood
2014
;
124
:
13
23
.
45.
George
MA
,
Qureshi
S
,
Omene
C
,
Toppmeyer
DL
,
Ganesan
S
.
Clinical and pharmacologic differences of CDK4/6 inhibitors in breast cancer
.
Front Oncol
2021
;
11
:
693104
.
46.
Marra
A
,
Curigliano
G
.
Are all cyclin-dependent kinases 4/6 inhibitors created equal?
NPJ Breast Cancer
2019
;
5
:
27
.
47.
Timmermann
S
,
Hinds
PW
,
Münger
K
.
Elevated activity of cyclin-dependent kinase 6 in human squamous cell carcinoma lines
.
Cell Growth Differ
1997
;
8
:
361
70
.
48.
Gong
W
,
Wang
L
,
Zheng
Z
,
Chen
W
,
Du
P
,
Zhao
H
.
Cyclin-dependent kinase 6 (CDK6) is a candidate diagnostic biomarker for early non–small cell lung cancer
.
Transl Cancer Res
2020
;
9
:
95
103
.
49.
Poomsawat
S
,
Sanguansin
S
,
Punyasingh
J
,
Vejchapipat
P
,
Punyarit
P
.
Expression of cdk6 in head and neck squamous cell carcinoma
.
Clin Oral Investig
2016
;
20
:
57
63
.
50.
Dai
M
,
Zhang
C
,
Ali
A
,
Hong
X
,
Tian
J
,
Lo
C
, et al
.
CDK4 regulates cancer stemness and is a novel therapeutic target for triple-negative breast cancer
.
Sci Rep
2016
;
6
:
35383
.
51.
Hsu
Y-H
,
Yao
J
,
Chan
L-C
,
Wu
T-J
,
Hsu
JL
,
Fang
Y-F
, et al
.
Definition of PKC-α, CDK6, and MET as therapeutic targets in triple-negative breast cancer
.
Cancer Res
2014
;
74
:
4822
35
.
52.
George
J
,
Lim
JS
,
Jang
SJ
,
Cun
Y
,
Ozretić
L
,
Kong
G
, et al
.
Comprehensive genomic profiles of small-cell lung cancer
.
Nature
2015
;
524
:
47
53
.
53.
Yang
L
,
Fang
D
,
Chen
H
,
Lu
Y
,
Dong
Z
,
Ding
H-F
, et al
.
Cyclin-dependent kinase 2 is an ideal target for ovary tumors with elevated cyclin E1 expression
.
Oncotarget
2015
;
6
:
20801
12
.
54.
Lin
ZP
,
Zhu
Y-L
,
Ratner
ES
.
Targeting cyclin-dependent kinases for treatment of gynecologic cancers
.
Front Oncol
2018
;
8
:
303
.
55.
Bantie
L
,
Tadesse
S
,
Likisa
J
,
Yu
M
,
Noll
B
,
Heinemann
G
, et al
.
A first-in-class CDK4 inhibitor demonstrates in vitro, ex vivo
,
and in vivo efficacy against ovarian cancer
.
Gynecol Oncol
2020
;
159
:
827
38
.
56.
Tadesse
S
,
Bantie
L
,
Tomusange
K
,
Yu
M
,
Islam
S
,
Bykovska
N
, et al
.
Discovery and pharmacological characterization of a novel series of highly selective inhibitors of cyclin-dependent kinases 4 and 6 as anticancer agents
.
Br J Pharmacol
2018
;
175
:
2399
413
.
57.
De Kegel
B
,
Ryan
CJ
.
Paralog buffering contributes to the variable essentiality of genes in cancer cell lines
.
PLoS Genet
2019
;
15
:
e1008466
.
58.
Kumarasamy
V
,
Vail
P
,
Nambiar
R
,
Witkiewicz
AK
,
Knudsen
ES
.
Functional determinants of cell-cycle plasticity and sensitivity to CDK4/6 inhibition
.
Cancer Res
2021
;
81
:
1347
60
.
59.
Wu
X
,
Yang
X
,
Xiong
Y
,
Li
R
,
Ito
T
,
Ahmed
TA
, et al
.
Distinct CDK6 complexes determine tumor cell response to CDK4/6 inhibitors and degraders
.
Nat Cancer
2021
;
2
:
429
43
.
60.
McDonald
ER
III
,
de Weck
A
,
Schlabach
MR
,
Billy
E
,
Mavrakis
KJ
,
Hoffman
GR
, et al
.
Project DRIVE: a compendium of cancer dependencies and synthetic lethal relationships uncovered by large-scale, deep RNAi screening
.
Cell
2017
;
170
:
577
92
.
61.
Guarducci
C
,
Bonechi
M
,
Benelli
M
,
Biagioni
C
,
Boccalini
G
,
Romagnoli
D
, et al
.
Cyclin E1 and Rb modulation as common events at time of resistance to palbociclib in hormone receptor–positive breast cancer
.
NPJ Breast Cancer
2018
;
4
:
38
.
62.
Franco
J
,
Witkiewicz
AK
,
Knudsen
ES
.
CDK4/6 inhibitors have potent activity in combination with pathway selective therapeutic agents in models of pancreatic cancer
.
Oncotarget
2014
;
5
:
6512
25
.
63.
Teng
M
,
Jiang
J
,
He
Z
,
Kwiatkowski
NP
,
Donovan
KA
,
Mills
CE
, et al
.
Development of CDK2 and CDK5 dual degrader TMX-2172
.
Angew Chem Int Ed Engl
2020
;
59
:
13865
70
.
64.
Sabnis
RW
.
Novel CDK2 inhibitors for treating cancer
.
ACS Med Chem Lett
2020
;
11
:
2346
7
.
65.
Zhou
F
,
Chen
L
,
Cao
C
,
Yu
J
,
Luo
X
,
Zhou
P
, et al
.
Development of selective mono or dual PROTAC degrader probe of CDK isoforms
.
Eur J Med Chem
2020
;
187
:
111952
.

Supplementary data