CDK6 antagonizes p53-induced responses during tumorigenesis

Abstract

Tumor formation is a multi-step process during which cells acquire genetic and epigenetic changes until they reach a fully transformed state. We show that CDK6 contributes to tumor formation by regulating transcriptional responses in a stage-specific manner. In early stages CDK6 kinase induces a complex transcriptional program to block p53 in hematopoietic cells. Cells lacking CDK6 kinase function are required to mutate p53 to achieve a fully transformed immortalized state. CDK6 binds to the promoters of genes including the p53-antagonists PRMT5, PPM1D and MDM4. The findings are relevant to human patients: tumors with low levels of CDK6 have mutations in p53 significantly more often than expected.

Introduction

The cell-cycle kinase CDK6 has attracted considerable attention over the past decade. CDK6 has been recognized not only as a typical cyclin-dependent kinase but as a transcriptional regulator with properties distinct from those of its close homologue CDK4 (1–5). CDK6 regulates the transcription of a number of genes and its effects may be dependent on or independent of its kinase activity (6). The transcriptional function of CDK6 is crucial for its role in promoting myeloid and lymphoid malignancies, including AML and ALL (1,3) and is important in maintaining hematopoietic and leukemic stem cells (2,7). Recently, CDK6 activity was shown to regulate metabolic functions in T-ALL contributing to the transformed phenotype (8).

In line, the potential of CDK4/6 inhibitors is widely acknowledged and CDK4/6 inhibitors are considered to represent a major breakthrough in cancer therapy (9). A number of clinical trials are starting and CDK4/6 inhibitors are being examined for possible use in patients with hematological disorders.

One of the key factors in determining therapeutic outcome is the status of the p53 pathway (10). p53 is among the most commonly mutated or deleted genes in human cancers and aberrations of the p53 pathway are frequently associated with rapid disease progression and a poor prognosis (11). Nevertheless, the molecular networks that favor the development of p53 aberrations are not fully understood. When considering possible therapeutic options, it is vital to avoid doing anything that might cause the emergence of p53 mutations. Precision medicine currently considers p53 status but could be improved by incorporating knowledge of factors that affect the mutational status of cancer cells. If a therapeutic approach is liable to provoke mutations, this point must be borne in mind when designing combinatorial or sequential approaches.

We now report that CDK6 counteracts p53-induced responses. Of note, high as well as low CDK6 expression levels have been shown to be of bad prognostic value which is currently not understood. High levels of CDK6 are frequently found in malignant lymphoid diseases (12–15). In contrast mono-allelic loss of CDK6 (via 7q deletions or monosomy 7) in ALL, MDS and AML is associated with a poor prognosis (16–20). Beside 7q deletions CDK6 is a target of various miRNAs; a recent study described the downregulation of CDK6 by miR-145, which confers resistance to chemotherapy in lung cancer cell lines (21). Our study sheds light in these apparent contradictions; we show that CDK6 expression levels correlate with the status of the p53 pathway in murine and human tumors. CDK6 suppresses p53 responses upon oncogenic stress, inducing the transcription of a number of genes such as PRMT5, PPM1D and MDM4, which negatively regulate p53 (22). Tumors with low or absent CDK6 expression are pressured to mutate p53 to overcome oncogenic stress. Our findings imply that any therapy that interferes with CDK6 activity may be associated with a higher risk of acquiring p53 mutations.

Results

CDK6 is required to support the outgrowth of malignant cell lines

Recent evidence highlights the role for CDK6 in malignant cells for tumor maintenance and progression and has revealed the importance for CDK4/6 inhibitors in the therapeutic landscape. In contrast only limited information is available on its function during the transformation process and for tumorigenesis. We thus retrovirally infected primary bone marrow cells isolated from Cdk6^+/+ and Cdk6^-/- mice with a pMSCV-BCR-ABL^p185-IRES-GFP vector and monitored leukemic cell outgrowth using the empty vector as control. The initial response of Cdk6^+/+ and Cdk6^-/- cells to transformation and exposure to the oncogene is comparable as the numbers of colonies forming in growth factor free methylcellulose were superimposable 10 days after pMSCV-BCR-ABL^p185-IRES-GFP infection (Figure S1A). In line viable cells were present at equal numbers after 10 days in liquid culture and no differences in apoptosis were detectable at that time point (Figure 1A). Immune-phenotyping by FACS confirmed identical surface marker expression (CD19⁺CD43⁺B220⁺) in both cultures (Figure S1B). Irrespective of the genotype we observed a pronounced initial increase in apoptotic cell numbers that rapidly ceased in Cdk6^+/+ cells allowing the outgrowth of cell lines (Figure 1A, 1B, S1C). In Cdk6^-/- cultures, the percentage of apoptotic cells remained high; GFP⁺ Cdk6^-/- cells largely failed to escape the apoptotic response which resulted in a delayed outgrowth and reduced numbers of BCR-ABL⁺ cell lines with a slightly increased BCR-ABL expression (Figure 1A to 1C and S1D). These observations indicated that CDK6 antagonizes apoptotic responses during transformation to allow for immortalization and outgrowth of primary transformed cells. Similar observations were made when we tried to establish cell lines from NPM-ALK transgenic mice on either a Cdk6^+/+ or a Cdk6^-/- background. Whereas cell lines were readily established from wild type tumors, Cdk6^-/- cell lines were established at a drastically lower frequency (data not shown).

Figure 1. CDK6 supports the outgrowth of malignant cell clones.

A) Cdk6^+/+ and Cdk6^-/- bone marrow cells were transduced with BCR-ABL. Numbers of apoptotic (red line) and living cells (black line) in liquid cultures are summarized in the left panels. The numbers represent the mean±SD (n=3 different biological replicates).

B) Cell-cycle profiles of BCR-ABL-transduced bone marrow cells analyzed at day 22. Cdk6^+/+ cells display only a small sub-G1 fraction whereas Cdk6^-/- cells have a reduced number of cycling cells and an increased fraction of cell in sub-G1 phase. The gating strategy is indicated in the histogram plots; bargraphs summarize our experiments (n=3 different biological replicates).

C) Statistics on the number of successfully established cell lines from BCR-ABL-transduced Cdk6^+/+ and Cdk6^-/- bone marrow cells (n=13 individual experiments).

D) qPCR analysis of the p53-target genes p21, PUMA and NOXA in ex vivo γ-irradiated pre-pro-B cells isolated from Cdk6^+/+ and Cdk6^-/- mice at 4 hours post-treatment (n=3 different biological replicates).

E) Western blot analysis of p53 and CDK6 in Cdk6^+/+ and Cdk6^-/- cell lines upon treatment with DMSO or etoposide for 4 hours. HSC-70 was used as loading control (n=3 different biological replicates).

F) qPCR analysis of the p53-target genes p21, PUMA and NOXA in BCR-ABL⁺ Cdk6^+/+ and Cdk6^-/- cell lines upon treatment with NCS for 4 hours (n=3 different biological replicates).

G) Kaplan–Meier plot showing overall survival of NSG mice after intravenous injection of Cdk6^+/+ (upper panel) or Cdk6^-/- (lower panel) BCR-ABL–transformed cell lines and subsequent γ-irradiation with 1.25 Gy or 2.5 Gy. Statistical differences were calculated using the Log-rank test (**p<0.01; ns: not significant; UT: Untreated).

H) Drug screening of Cdk6^+/+ and Cdk6^-/- cell lines with 272 FDA approved drugs. Relative viability is indicated as % DMSO control (n=3 cell lines per genotype).

Cdk6^-/- cells acquire a defective p53 response during transformation

Immortalization requires counterbalancing oncogenic stress via the p53 – p19^ARF pathway (23). To test whether Cdk6^+/+ and Cdk6^-/- cells harbor a normal p53 response we first used γ-irradiation or the radiomimetic neocarzinostatin (NCS) in primary non-transformed lymphoid cells. We readily induced apoptosis in pre-pro-B cells ex vivo and in vitro as well as thymocytes ex vivo irrespective of the genotype (Figure S2A to S2D). In both cell types, the irradiation-induced apoptosis was preceded by increased expression of the p53 target genes p21, NOXA and PUMA (Figure 1D, S2E and S2F). Differences became apparent in transformed cells: BCR-ABL⁺ Cdk6^-/- cells displayed high p53 protein expression levels already in the absence of any stimulus (Figure 1E). Accordingly, they failed to induce p21, NOXA and PUMA upon NCS or etoposide treatment (Figure 1F, S2G and S2H) and to undergo apoptosis upon γ-irradiation, NCS or etoposide treatment (Figure S2I, S2J and S3A to S3C), whereas BCR-ABL⁺ Cdk6^+/+ cells readily underwent apoptosis upon any stimulus used. Similar results were obtained with nutlin-3 that directly stabilizes p53 by preventing its MDM2-dependent degradation (Figure S3D to S3G). Re-expression of CDK6 in immortalized cells did not reconstitute the ability to induce apoptosis (Figure S3H to S3J). To test the therapeutic significance of our observation we transplanted BCR-ABL⁺ Cdk6^+/+ and Cdk6^-/- cells into NSG mice and subjected them to radiotherapy five days post-transplantation. In this model, the absence of CDK6 significantly delayed disease development. As expected irradiation with 2.5 Gy or 1.25 Gy significantly delayed disease progression in mice that had been transplanted with BCR-ABL⁺ Cdk6^+/+ cells. In contrast no changes in disease latency were detected in mice that had received BCR-ABL⁺ Cdk6^-/- cells (Figure 1G). Similarly, the analysis of chemo-sensitivity of BCR-ABL⁺ Cdk6^+/+ and Cdk6^-/- cells using a library of 272 FDA-approved drugs uncovered a drastically diminished response to standard chemotherapeutic drugs requiring an intact p53 response (Figure 1H and S4A). This included 5-FU, 6-mercaptopurine, 5-azacytidine, docetaxel and doxorubicin; drug responses were recapitulated in a panel of individually derived BCR-ABL⁺ Cdk6^+/+ and Cdk6^-/- cell lines (Figure S4B). Importantly, differences in drug responses were uncoupled from changes in cell-cycle distribution (Figure S4C to S4H). Analysis of the p53 mutational status uncovered p53 mutations in all Cdk6^-/- cell lines including cell lines harboring a point mutation in the kinase domain of CDK6 rendering them kinase inactive (Cdk6^K43M) (24). In contrast, only one out of eight Cdk6^+/+ lines investigated had acquired a p53 mutation (Figure 2A, S5A, Cdk6^+/+ versus Cdk6^K43M p=0.01; Cdk6^+/+ versus Cdk6^-/- p=0.005). All p53 mutations localized within the DNA-binding domain and have been described to disrupt p53 responses (25–27). Apoptosis could be induced following addition of PRIMA-met, which reactivates p53, showing that the inability of Cdk6^-/- cell lines to undergo apoptosis stems solely from the p53 mutations (Figure 2B and S5B). In summary these findings led us to conclude that cells are forced to mutate p53 to compensate for the absence of CDK6 during transformation. Accordingly transformants derived from p53^-/-;Cdk6^-/- double deficient animals were able to escape the oncogene-induced apoptosis in contrast to GFP⁺ p53^+/+;Cdk6^-/- cells (Figure 2C and 2D). These data indicate that the major function of CDK6 during transformation and immortalization is to counteract oncogene-induced p53-mediated apoptotic responses.

Figure 2. Cells lacking CDK6 kinase function are required to mutate p53.

A) cDNA of the protein-coding p53 transcript was amplified by PCR and analyzed by Sanger-sequencing. The substituted amino acids are as well as the corresponding human mutations are indicated.

B) Annexin V/7-AAD staining of Cdk6^-/- cells treated with etoposide, PRIMA-met or the combination of both drugs for 24h hours. Numbers represent the mean±SD (n=3 cell lines per genotype, one representative example is depicted).

C) Cell-cycle profiles analyzed by PI-staining of BCR-ABL-transduced bone marrow cells at day 19. The gating strategy is indicated in the histogram plots; bargraphs summarize our experiments (n=3 different biological replicates).

D) p53^+/+;Cdk6^+/+, p53^+/+;Cdk6^-/-, p53^-/-;Cdk6^+/+ and p53^-/-;Cdk6^-/- bone marrow cells were transduced with BCR-ABL. The panels show representative Annexin V/7-AAD stainings at day 19. The numbers represent the mean±SD (n=3 different biological replicates).

CDK6 induces a transcriptional program to antagonize p53 during transformation

To investigate how CDK6 counteracts p53 we decided for an unbiased approach. We transduced Cdk6^+/+, kinase-dead Cdk6^K43M mutant and Cdk6^-/- primary bone marrow with BCR-ABL followed by cloning of the cells in growth-factor-free methylcellulose. Individual colonies were picked at day 10 as this time point represents the critical phase where CDK6 is required to counteract p53. Gene expression analysis by microarrays of single Cdk6^+/+, Cdk6^K43M and Cdk6^-/- colonies (n=4 of each genotype) revealed a highly consistent pattern and identified 4471 genes that are regulated in a kinase-dependent manner that showed an altered regulation in Cdk6^-/- and Cdk6^K43M colonies (Figure 3A, S6A, S6B). Gene ontology analysis of enriched biological processes identified execution phase of apoptosis, DNA damage response and signal transduction by p53 class mediator as the main deregulated processes (Figure 3B, Table S1). Several of the deregulated genes have been reported to promote tumorigenicity (exemplified in Figure 3B) including the BCL2-regulator RRM2 and prominent p53 antagonists such as MDM4, PPM1D and the methyltransferase PRMT5. PPM1D, PRMT5, and MDM4 are part of a tightly controlled negative feedback regulator network that ensures limitation of the p53 response after cell-cycle arrest and subsequent DNA damage repair (28). In total we were able to identify 22 out of 59 well-established p53 regulators by comparing our list of deregulated genes to the PID p53 regulation pathway gene set (29). Validation by qPCR in individually derived single colonies and western blot in outgrowing BCR-ABL⁺ cells verified the fidelity of our microarrays (Figure 3C, S6C and S6D). Besides regulating p53 itself, PRMT5 has been shown to interfere with MDM4 mRNA splicing (30) which was also evident in colonies lacking kinase-active CDK6 that displayed increased levels of aberrantly spliced MDM4 mRNA (MDM4-S) (Figure S6E). At this time point the cells were sensitive to CDK6 re-expression: when we performed a rescue experiment and co-transduced CDK6 and BCR-ABL in Cdk6^-/- bone marrow cells we found a positive correlation between CDK6 and MDM4 and between CDK6 and PRMT5 mRNA in single colonies (Figure S6F). Cells that do express CDK6 were capable to induce the p53 antagonists MDM4 and PRMT5.

Figure 3. CDK6 induces a transcriptional program to antagonize p53.

A) Heatmaps of transcripts deregulated in Cdk6^+/+, Cdk6^K43M and Cdk6^-/- colonies 10 days after BCR-ABL transduction (n=4 colonies per genotype).

B) Gene ontology analysis of genes regulated in a kinase-dependent manner in colonies 10 days after BCR-ABL transduction (upper table). Examples of deregulated genes involved in the p53-reponse (lower table).

C) qPCR analysis of the p53 regulators PRMT5 and MDM4 in individual colonies isolated from methylcellulose.

D) Heatmaps of transcripts deregulated in BCR-ABL⁺ Cdk6^+/+, Cdk6^K43M and Cdk6^-/- cell lines (n=2 cell lines per genotype).

E) Venn diagram showing the number of transcripts regulated in colonies and cell lines (blue circles) and the numbers of gene promoters with ChIP-Seq peaks (red circle). The numbers in the intersecting areas show the overlap between the two datasets and the genes having a ChIP-Seq peak in the promoter region.

F) Representative examples of ChIP-Seq peaks in the promoter regions of the p53 antagonists Prmt5 and Mdm4.

G) Pie charts showing the functional classification of gene ontology terms identified in the gene sets that are bound by CDK6 and show expression changes in colonies or cell lines.

Different transcriptional responses are induced by chromatin – associated CDK6 during and after transformation

To elucidate how CDK6 regulates transcription we decided again for an unbiased approach and performed RNA-Seq and ChIP-Seq analysis in BCR-ABL⁺ Cdk6^+/+, Cdk6^K43M and Cdk6^-/- cell lines. RNA-Seq analysis identified 1785 genes that are regulated in a CDK6-dependent manner (Figure 3D). Analysis of the gene ontology terms among the deregulated genes identified biological processes including regulation of the immune response, cytoskeleton and wound healing as major enriched (Table S2) in established cell lines. In the initially transformed cells, responses to stress were the dominating biological process (Figure S7A). ChIP-Seq analysis was used to define directly transcriptionally controlled genes using antibodies against CDK6. We identified CDK6 bound to the promoter regions of 2468 genes deregulated in CDK6 deficient versus Cdk6^+/+ BCR-ABL⁺ colonies and to the promoter regions of 706 genes in cell lines (Figure 3E). Importantly, 16 of 22 differentially expressed p53 regulators including Mdm4, Prmt5 and Ppm1d display a peak in their respective promoter region (exemplified in Figure 3F, S7B). The binding of CDK6 to the Mdm4 and Prmt5 promoters was validated by ChIP followed by qPCR analysis (Figure S7C). ChIP-Seq results were verified in cells expressing an HA-tagged version of CDK6 (Figure S7D). Gene ontology analysis focusing on deregulated genes with a CDK6 peak identified responses to stress signals such as execution phase of apoptosis and signal transduction in response to DNA damage as major deregulated pathway in the initially transformed cells highlighting the role of CDK6 in this process (Figure 3G, S7E and Table S3). In contrast, regulation of cellular component size, embryonic morphogenesis (developmental processes), regulation of growth, locomotion and intracellular signal transduction was predominant in established cell lines (Table S4). As expected, regulation of cell-cycle was identified as common CDK6-dependent pathway during initial transformation and in established cells. Furthermore, we found intracellular signal transduction pathways including IκB kinase/NF-κB, MAPK and JAK-STAT signaling as commonly regulated (Figure S7E and Table S5).

NFY and SP1 as mediators of CDK6 dependent transcription

CDK6 as co-factor does not contain a DNA binding domain, thus we performed a motif enrichment analysis to understand how CDK6 interacts with chromatin. NFY and SP1 were identified as common denominators in cell lines and colonies underlying ChIP-Seq peaks associated with altered gene transcription (Figure 4A). Further evidence for a role of NFY and SP1 was obtained in an assay where we analyzed differentially phosphorylated chromatin-bound proteins (Phospho-Chromatome) in bone marrow cells 10 days after BCR-ABL transduction either expressing or lacking CDK6 (Figure 4B). This assay analyzes the phosphorylation status by quantitative mass spectrometry and uncovered NFYA and SP1 among the top differentially phosphorylated proteins in the critical phase when differences between Cdk6^+/+ and Cdk6^-/- cells became evident (Figure 4C). Significant differences were also uncovered for the transcription factors ZBTB7A, EBF1, YY1, ELF2 and ATF1 (Figure 4C). Changes in the non-phosphorylated chromatome fraction confirmed reduced levels of PRMT5, MDM4 and PPM1D upon CDK6 deficiency (Figure S8A). To test whether the identified transcription factors are subject to direct phosphorylation by CDK6 we performed in vitro kinase assays. Indeed, active CDK6/CyclinD3 complexes phosphorylate recombinant NFYA, ZBTB7A and EBF1 in a dose-dependent manner, a response which was suppressed by addition of the CDK4/6 kinase inhibitor palbociclib (Figure 4D, 4E and S8B). Co-immunoprecipitation, ChIP and ChIP-ReChIP experiments in BCR-ABL⁺ cell lines verified the interaction between NFYA and CDK6 (Figure 4F, 4G, S8C and S8D). Similar to CDK6 knockdown experiments, knockdown of NFYA and SP1 in initially transformed cells induced apoptosis that was partially counteracted in the absence of p53 (Figure S8E).

Figure 4. Differential phosphorylation of transcriptional regulators upon CDK6 deficiency.

A) Motif analysis of ChIP-Seq promoter peaks underlying CDK6-regulated genes.

B) Schematic workflow of the technique used to detect phosphopeptides on chromatin (phospho-chromatome; LCMS: Liquid chromatography–mass spectrometry).

C) Schematic representation of SP1 and NFYA proteins showing annotated domains and the position of residues that were found differentially phosphorylated in Cdk6^+/+ versus Cdk6^-/- cells. Transcriptional regulators showing reduced phosphorylation in the absence of CDK6 are listed (n=3 different biological replicates).

D) Quantification of ADP formation in a luminescent kinase assay using recombinant NFYA and increasing amounts of active CDK6 protein. Photometrically acquired relative luminescence units (RFU) are shown (n=2 technical replicates).

E) Luminescent kinase assay in the presence or absence of ATP or CDK4/6 inhibitor palbociclib (500nM) (n=2 technical replicates).

F) NFYA chromatin immunoprecipitation (NFYA ChIP) followed by qPCR analysis of target gene promoters. Fold enrichment over the established Cd19 negative region is shown.

G) ChIP of NFYA with subsequent Re-ChIP of CDK6. qPCR analysis shows the concomitant presence of CDK6 and NFYA at the indicated promoters. Fold enrichment over no-antibody control is shown.

As SP1 and NFY were reported to cooperate with p53 at promoter regions we next overlaid the CDK6 ChIP-Seq peaks with p53 ChIP-Seq data performed in untreated or γ-irradiated B-cells (31). In untreated cells we identified a low number of promoter regions bound to p53 and CDK6 (4%, Figure 5A). In γ-irradiated cells this situation changed dramatically with more than 50% of all CDK6 peaks showing an overlay with p53 peaks (9304 common peaks). The perfect alignment of the peak positions suggested common binding sites for CDK6 and p53 (Figure 5A) which again were enriched for NFY and SP1 motifs (Figure S8F). As we failed to obtain any evidence for a direct interaction between CDK6 and p53 (Figure S9A and S9B), we speculated that p53 competes with CDK6 for binding to the promoters of p53 antagonists PPM1D, PRMT5 and MDM4 via NFY or SP1. In line with the p53 ChIP-Seq data we observed a nutlin-3-induced increase in p53 bound to the promoters of Prmt5, Mdm4 and Ppm1d which was paralleled by a decrease of promoter-bound CDK6 (Figure 5B). Global analysis of binding sites for p53 revealed that hardly any direct p53 binding sites were detected in the overlap of p53/CDK6 (Figure 5C and 5D). CDK6 and p53 seem to predominantly interact via SP1 and NFY to regulate transcription as evident when analyzing the p53 negative regulators PRMT5, PPM1D and MDM4 (Figure 5E). In contrast, the direct p53-targets and apoptosis-inducers PUMA and NOXA were not altered by the absence of CDK6 (Figure S9C). Based on these data we propose a model where CDK6 directs p53-dependent transcription by phosphorylating NFY and SP1 (Figure 5F).

Figure 5. CDK6 and p53 share common binding sites.

A) Overlay of CDK6 with p53 ChIP-Seq peaks in untreated or γ-irradiated B-cells. The left panel depicts the distribution of peakshifts across CDK6 and p53 ChIP-Seq datasets. The Venn diagram depicts overlapping and distinct peak numbers in γ-irradiated B-cells.

B) HA-CDK6 and p53 ChIP-qPCR experiments performed in BCR-ABL⁺ cell lines upon nutlin-3 pre-treatment (30µM, 4 hours). IgG-ChIP qPCR values were used as control. Upon nutlin-3 treatment, p53 shows increased binding to the Prmt5, Mdm4 and Ppm1d promoters paralleled by decreased CDK6-binding.

C) Motif analysis of p53 ChIP-Seq peaks from γ-irradiated cells that overlap with CDK6 ChIP-Seq peaks (left) or are not found in the CDK6-ChIP-Seq dataset (right).

D) Cumulative position-specific densities of motif found in p53 ChIP-Seq peaks from γ-irradiated cells that overlap with CDK6 ChIP-Seq peaks (left) or are not found in the CDK6-ChIP-Seq dataset (right).

E) Schematic representation of the Prmt5, Ppm1d, Mdm4, Puma and Noxa promoter regions. The NFYA, SP1 and p53 binding sites and the identified motifs are indicated.

F) Proposed model for the CDK6-dependent regulation of p53-responses via NFY/SP1 phosphorylation.

The CDK6-associated gene signature is conserved in human and murine leukemia

To study whether our findings in murine models of leukemia relate to human patients we analyzed publically available gene expression signatures from BCR-ABL⁺ and BCR-ABL^- ALL and MDS patients (International multi-center studies TARGET and MILE). We initiated an unbiased approach: Pairwise Spearman correlation coefficients of mean CDK6 gene expression to all other probes on the array were calculated and biologically informative genes were selected by implementing a permutation-based test. Gene set enrichment analysis was used to identify common pathways under control of CDK6: E2F-dependent transcriptional programs were used as positive control and showed a clear correlation in a human gene list ranked according to the correlation coefficient (Figure S10A and Table S6). When we analyzed the enrichment of p53 regulators and NFY target genes we found a significant positive correlation across all disease entities verifying the global nature of our finding (Figure 6A, S10B and Table S6). Overlapping the CDK6-dependent human gene patterns identified 377 genes that were consistently correlated to CDK6 levels (Figure 6B and S11A to S11D). Among the genes that were consistently deregulated in a CDK6-dependent manner were the p53 negative regulator PRMT5 and the anti-apoptotic gene BCL2 (Figure 6C and S6C).

Figure 6. The CDK6-associated gene signature is conserved in human and murine leukemia.

A) Gene set enrichment analysis of the p53-regulatory pathway in ALL and MDS patients.

B) Heatmap of fRMA-normalized values from probes correlating with CDK6 expression in all of the analyzed leukemia-entities of the individual patients. To visualize co-expression with CDK6, patients within each datasets were sorted for increasing mean CDK6 levels.

C) Scatter plots show the correlation between CDK6 expression and the expression levels of the indicated transcripts. Each dot indicates one patient. fRMA normalized expression values are shown.

D) Percentages of ALL, AML and MDS patients with p53 mutations and mono-allelic loss of 7q. The chi-square test was used to test for statistical significance.

Our data would also predict that low CDK6 levels predispose human tumors to p53 mutations or alternatively that high CDK6 levels may substitute for p53 mutations. We analyzed two datasets that allowed correlating the p53 status to CDK6 levels, a patient cohort suffering from thymoma and one from glioblastoma. In both diseases CDK6 is frequently expressed at high levels (12,32). In both independent datasets a higher incidence of p53 mutations accompanied low CDK6 levels. (Figure S11E). For glioblastoma the analysis reached statistical significance, the cohort of thymoma patients was smaller but showed a clear trend (Figure S11E).

Another means of testing this concept is to study patients with a 7q deletion that includes the Cdk6 gene. We thus investigated two cohorts of patients with hematologic malignancies. The first cohort consisted of 187 AML and 202 ALL patients and the second cohort included 697 cases of MDS. In the first cohort, 17 patients had a genomic loss of the CDK6 locus on chromosome 7 which is proposed to include CDK6 and 48 patients had mutations in p53. Mono-allelic loss of CDK6 resulted in a decreased CDK6 expression in AML and ALL (Figure S11F). Within the AML/ALL patients (n=389), 7 patients (1.8%) were found to harbor mutations in p53 and loss of CDK6 (Figure 6D). Within the 697 MDS patients, 90 patients (12.9%) had mutations in p53 and 93 (13.3%) had losses on chromosome 7 (Figure 6D). 38 cases (5.5%) had both, mutations in p53 and losses on chromosome 7q (Figure 6D). In summary we have observed patients with both anomalies far more frequently than would be expected by chance (p < 0.0001).

Discussion

Recent discoveries have shed new light on CDK6. The protein was initially described as a cell-cycle kinase with a function highly redundant to that of its close homologue CDK4. CDK6 is now also known to be a transcriptional regulator whose targets include important proto-oncogenes such as VEGF-A and FLT-3 (1,3). In this study we define CDK6 as a pro-survival factor that is required to counteract p53 responses, which are of utmost importance during the early phases of transformation and immortalization. The CDK6-mediated antagonism of stress responses is mirrored in human hematopoietic malignancies such as ALL, AML and MDS and its significance is not restricted to BCR-ABL-induced transformation, which we used as a model. CDK6’ wide-ranging effects on transcription are highlighted by the numerous specific ChIP-Seq peaks associated with transcriptional changes. The fact that CDK6 regulates slightly distinct transcriptional programs in transformed and in established cells lines confirms that this versatile protein has different functions during the multi-stage process of tumor formation and leukemogenesis.

Although CDK6 is of minor importance under homeostatic conditions, it assumes a prominent role under conditions of stress, such as hematopoiesis (2,7) and oncogene-induced stress. The importance of CDK6 in stress responses is clear in our global analysis; gene expression data and the CDK6 ChIP-Seq analysis reveal “cellular stress responses” as major regulated processes during the transformation phase. A p53 response represents the cell’s reaction to stress, activating pathways that lead to apoptosis or to DNA repair and survival (33). CDK6 may assume particular importance in situations with high levels of MYC and thus a pronounced pro-apoptotic response. It has been shown that activation of the stress kinase p38 translocates CDK6 into the nucleus, where it is available to induce transcriptional responses (4).

Oncogene-induced stress (OIS) represents a particular case. CDK6 is at the intersection of the two major pathways of tumor formation: it regulates RB via p16^INK4A to arrest the cell cycle, enabling DNA repair, while counteracting p53-induced apoptosis and thereby promoting cell survival. Of note, CyclinD1 has also been assigned a role in DNA damage repair, which represents a hallmark of OIS. CyclinD1 is recruited to sites of DNA damage; the role for CDK4/6 in this context remained elusive (34). We observed a divergent regulation of p16^INK4A and p19^ARF in Cdk6^+/+ and Cdk6^-/- cells. Whereas Cdk6^+/+ cells initially up-regulate and then silence p16^INK4A expression upon BCR-ABL transformation (1), Cdk6^-/- cells induce p19^ARF which remains consistently high in CDK6-deficient p53 mutant cells (Figure S12). The cause and consequences of these findings remain to be determined.

Cells with little or no CDK6 frequently show mutations in p53 to maintain a transformed state in mice and men. Transformed Cdk6^-/- cells are insensitive to irradiation and respond poorly to classical chemotherapeutic drugs. Deletion of p53 in the absence of CDK6 enables cells to establish cell lines at normal rates, confirming that CDK6 functions primarily to antagonize p53. The underlying transcriptional program includes many genes whose promoter regions are directly bound by CDK6, including Ppmd1, Prmt5 and Mdm4. Although the initial discovery of a CDK6 function in regulating gene transcription suggested it to be independent of the kinase activity, CDK6 controls the transcription of p53 regulators in a kinase-dependent manner linking cell-cycle control to the regulation of apoptosis. Several lines of evidence point at a central role for the transcriptional regulators NFY and SP1 as integrative elements at the crossroad of p53 and CDK6. Although SP1 and NFY have previously been described to be involved in p53 mediated transcription (35–37), our study puts them in a novel context: We propose that CDK6 phosphorylates NFY and SP1 to allow p53 to act as transcriptional regulator. The absence of CDK6 thereby shifts the delicate balance upon p53 activation towards apoptosis.

The fact that CDK6 protects transformed cells from oncogenic stress and from DNA damage-induced apoptosis may account for the apparent selection for cells with high levels of CDK6 in vivo. CDK6 suppresses p53 mediated stress responses while supporting oncogene-induced replication. Although cells are usually more vulnerable to cell death when proliferating, CDK6 not only promotes cell division but also helps protect proliferating cells from p53-induced cell death. Inhibition of CDK4/6 kinase activity leads to growth arrest and impairs responses to classical chemotherapeutics but our data predict that a specific CDK6 inhibitor, which would not target CDK4 and would thus have little effect on proliferation, should enhance p53-induced drug responses.

p53 is itself subject to complex regulation (38) and we show that CDK6 is part of the network that regulates p53 responses during transformation. Different cell types may react differently to CDK6 inhibition: senescence dominates in epithelial cells, whereas treatment of hematological malignancies with CDK4/6 inhibitors can induce apoptosis (39,40). The molecular explanation has remained obscure but the finding that CDK6 regulates the p53 response provides a tantalizing hint: p53 links senescence and apoptosis, balancing the two pathways and directing a cell type in one or the other direction. We propose that CDK6 regulates p53. The CDK6 mediated transcriptional responses are highly conserved in mouse and man and are relevant in human tumors. We found a correlation between levels of CDK6 and occurrence of p53 mutations. The finding that CDK6 is controlled by a super-enhancer in glioblastoma, multiple myeloma, T-ALL and colorectal cancer suggests that the level of CDK6 may be a predictive factor in several disease entities (41). Our data also suggest that the association of 7q deletions and monosomy 7 with p53 mutations (19,42–45) is related to CDK6 (46,47): 7q deletions and monosomy 7 (which remove CDK6) enhance the risk of a concomitant p53 mutation by a factor of four.

These findings have consequences for our assessment on the effects of CDK4/6 inhibitors. They may be helpful to eliminate leukemic stem cells; a recent study showed that BCR-ABL⁺ CML stem cells are sensitive to a combination of MYC inhibition and p53 induction. CDK6 inhibition in the light of p53 activation may be exploited in combination therapies to sensitize leukemic stem cells that are resistant to tyrosine kinase inhibitor therapy (48). On the other hand, persistent inhibition of CDK6 may provoke the outgrowth of p53 mutant clones, so patients who receive CDK6 inhibitors may be at risk of developing p53 mutations. CDK6 inhibition may allow premalignant cells and cancer-initiating cells to progress to a more malignant stage, for example in patients with clonal hematopoiesis, who are at a higher risk of developing hematopoietic malignancies such as MDS or AML (49). CDK6 inhibitors should therefore be used at precise doses and times. While this poses a challenge for precision medicine, the availability of CDK6 inhibitors represents a unique opportunity to target the two key pathways in tumor formation, as CDK6 not only regulates RB by direct phosphorylation but also interferes with p53-dependent responses.

Methods

Description of MDM4 mRNA splicing analysis, phospho-chromatome, kinase assay, microarray, motif enrichment analysis and analysis of publically available leukemia datasets is contained in the supplementary section.

Cell culture

Preparation of retroviral supernatant and generation of BCR-ABL⁺ cell lines was performed as described (1). Briefly, single cell suspensions from bone marrow were transduced with a pMSCV-BCR-ABL^p185-IRES-GFP vector and the outgrowth of cell lines was monitored by microscopic inspection and FACS analysis. Cell lines were maintained in RPMI medium supplemented with 10% fetal calf serum, 50 μM 2-mercaptoethanol and 100 U/ml penicillin, 100 μg/ml streptomycin (PAA, Pasching). To preserve integrity, BCR-ABL⁺ cell lines were expanded and frozen down into a large number of cryogenic vials upon generation. Experiments were performed with cells passaged less than 6 months after generation. For the re-expression of CDK6 and HA-CDK6 in Cdk6^-/- cell lines, pMSCV-IRES-puromycin and pMSCV-IRES-GFP plasmids were used. Pre-pro-B cells were derived and cultured as described (50). Knockdown experiments were performed using the pLENC plasmid (pMSCV-miRE-PGK-NeoR-IRES-mCherry) (51). Transient transfection experiments were performed using the Neon Transfection System (Invitrogen) according to the manufacturer's instructions. shRNA sequences are included in the supplementary section.

Mouse strains

All mice were maintained on a C57Bl/6J background (Cdk6^−/− (52), p53^-/- (53)) under specific pathogen-free conditions at the University of Veterinary Medicine Vienna. NOD/SCID/IL-2Rγ^−/− (NSG) mice were purchased from The Jackson Laboratory (USA). In all experiments, 6- to 8-weeks-old mice were used. Animal experiments were performed in accordance with protocols approved by the Austrian law and the Animal Welfare Committee at the University of Veterinary Medicine Vienna (License BMWFW-68.205/0045-WF/V/3b/2016).

Drug screening

Drug screening was performed as described previously (3). Briefly, 50 nL of each compound was spotted into 384-well plates (Corning 3701) and cells were incubated in the presence of drugs for 72 hours. 32 negative control wells (DMSO) and 32 positive control wells (Bortezomib, concentration 10µM) were prepared on each plate. Data from these wells were used to calculate a Z'-factor for each plate individually (54). Cell viability was assessed using CellTiterGlo (Promega). To compare compounds across plates with different signal, a percentage of control (POC) was calculated by linear regression. Thereby, DMSO was set to 100% and Bortezomib to 0% viability. Hits were defined as compounds when inhibition was above 50% compared to the DMSO controls.

Western Blotting and Immunoprecipitation

Cell lysis, immunoprecipitation, blotting and blocking was performed as described (55). For western blot analysis of initially transformed cells, density gradient centrifugation (Histopaque®-1077, Sigma-Aldrich) was performed to isolate viable lymphocytes. NFYA was immunoprecipitated using the NFYA (G-2, Santa Cruz Biotechnology) antibody. For immunoprecipitation of HA-tagged CDK6, Pierce™ Anti-HA Magnetic Beads (Thermo Fisher Scientific) were used. The following antibodies were used for detection: PRMT5 (07-405, Merck-Millipore), GAPDH (ABS16, Merck-Millipore), pY²⁰⁷Crkl (Cell Signaling Technologies), pS⁷⁸⁰Rb (Cell Signaling Technologies), HA (ab9110, Abcam), CDK4 (H-22), CDK6 (H96 and DCS90), NFYA (G-2), β-ACTIN (AC-15), BCL2 (N-19), p53 (DO-1), p21 (F-5) and HSC70 (B-6) all from Santa Cruz Biotechnology.

FACS analysis

Cells were stained using antibodies directed against: CD43-APC, B220-FITC, CD19-APC-Cy7, CD43-PE, Streptavidin-APC-Cy7, IgM-FITC, CD16/CD32 Fc receptor block (all from BD Biosciences), CD19-eFluor450, B220-PerCP-Cy5.5, BP1-biotinylated and IgD-APC (all from eBiosciences). For cell-cycle analysis 1 x 10⁶ cells were stained with propidium iodide (PI, 50 μg/ml) in a hypotonic lysis buffer (0.1% sodium citrate, 0.1% triton X-100, 100 μg/ml RNAse) and incubated at 37°C for 30 minutes. Analysis of apoptotic fractions was performed by staining with Annexin V and 7-AAD from the Annexin V Apoptosis Detection Kit eFluor^® 450 (eBiosciences) following manufacturer’s instructions. Samples were analyzed using a FACS Canto II flow cytometer equipped with 488, 633 and 405 nm lasers (BD Biosciences). FACS-Plots and calculations were done using the FACS Diva software version 6.1.2 (BD Biosciences). High-purity FACS-sorting was performed on a FACS Aria III equipped with a 488 nm laser (BD Biosciences) as described (2).

Colony formation assay

Bone marrow cells from 6 weeks old donor mice were transduced with retroviral supernatants as described previously (1). 24 hours cells thereafter the cells were harvested and embedded into growth factor-free methylcellulose (MethoCult™) at a density of 5 x 10⁵cells/ml. After 10 days, colonies were counted using inverted microscope device (CKX41, Olympus) and picked for RNA extraction.

RNA isolation, PCR and qPCR analysis

Total RNA was isolated from individual colonies as well as from stable cell lines. RNA was extracted using the RNeasy Micro Kit (Qiagen). Reverse transcription was performed using the iSCRIPT cDNA synthesis kit following manufacturer’s instructions (Bio-Rad). All qPCRs were performed in duplicate with the SsoFastTM EvaGreen®Supermix (Bio-Rad) according to the instructions of the manufacturer. Primer information and detailed thermocycling conditions are available upon request. Levels of mRNAs were normalized to rplp0 mRNA.

For analysis of the p53 mutational status, the p53 coding sequence was amplified using the following primer pairs:

p53_1

Fw: 5’-GTGCTCACCCTGGCTAAA-3’

Rv: 5’-CCAGCTGGCAGAATAGCT-3’

p53 _2

Fw: 5’-AGTGAAGCCCTCCGAGT-3’

Rv: 5’-CCAGTGTGATGATGGTAAGGATAG-3’

p53_3

Fw: 5’-CTTATCCGGGTGGAAGGAAAT-3’

Rv: 5’-GTCTCAGCCCTGAAGTCATAAG-3’

PCR products were purified from a 3% agarose gel using the E.Z.N.A. Gel Extraction Kit (OMEGA) and sanger-sequenced by Microsynth AG.

RNA-Seq Analysis

RNA was isolated from immortalized BCR-ABL⁺ Cdk6^+/+ and Cdk6^-/- cells. Single-end, 50bp sequencing of libraries prepared with the Lexogen SENSE mRNA-Seq library preparation kit was performed on an Illumina HiSeq-2500 sequencer. After quality control of raw data with FASTQC and removal of adapters and low quality reads with trimmomatic (version 0.36), reads were mapped to the Genecode M13 genome using STAR (version 2.5.2b) with default parameters. Then the featureCounts tool of the Subread package (version 1.5.1) was used to obtain gene counts for union gene models. Differentially expressed (p-adjust < 0.05) genes were identified using DESeq2 (version 1.16.1). For heatmaps centered and scaled rlog transformed library size normalized countes were visualized using the heatmap.2 function of the R package gplots version 3.0.1. The RNA-Seq data reported in this article has been deposited in the Gene Expression Omnibus (GEO) database (Accession ID: GSE113752).

Chromatin immunoprecipitation (ChIP)

ChIP and ChIP-ReChIP experiments were performed using homemade anti-CDK6 rabbit polyclonal serum (BioGenes) or antibodies against HA (ab9110, Abcam), NFYA (G-2, Santa Cruz) and p53 (CM5, Novocastra) (1–3). For ChIP-Seq analysis, sequencing reads were quality controlled by FastQC. Quality filtering, trimming of reads and adapter removal was done by using trimmomatic (version 0.36). Quality filtered reads were mapped against the mouse reference genome (Gencode M13) with bwa-mem (version 0.7.15). Blacklisted regions were removed from the analysis using bedtools subtract (version 2.26.0). Multimapping reads and reads with bad mapping quality were removed by removing reads with a mapping quality below 10 using samtools (version 1.3.1). Peak calling was performed by MACS2 (version 2.1.0) using default parameters. For validation, co-immunoprecipitated DNA was analyzed by qPCR (EpiTect HRM PCR Kit, Qiagen) for Prmt5, Mdm4, Ppm1d, Cdk1, Sp1 and Cyclin G1 promoter sites using a MyiQ device (Bio-Rad). Binding to the murine Vegf-A promoter was used as positive control for CDK6 ChIP and the p21 promoter for p53 ChIP (1–3). Antibody specificity was evaluated by including an IgG isotype control (02-6102, Invitrogen). Binding to a negative control region downstream of murine CD19 was used to estimate non-specific binding (1–3). Primer sequences are listed in the supplementary section. The ChIP-Seq data reported in this article has been deposited in the Gene Expression Omnibus (GEO) database (Accession ID: GSE113752).

Motif enrichment analysis

The peaks of the ChIP-Seq analysis were annotated with Homers annotate.pl script and default parameters. Only peaks within promoter regions (promoter-TSS tag) were considered for further analysis. The intersect between closest downstream genes of these promoter peaks and differentially regulated genes found within colonies, RNA-Seq and both was computed. The promoter peaks of these intersected genes were used for a motif enrichment analysis using Homer (v.4.9.0) findMotifsGenome.pl with the default -size 200. The motif analysis was performed 3 times and motifs that i) were not marked as potentially false positives within Homers de novo results and ii) appeared in at least 2 replicates, were counted as present. Density plots were generated by Homers annotate.pl script with a range of -500 to +500 bp from peak summit for SP1, NFY and p53 motifs obtained from the motif enrichment analysis.

Statistical Analysis

Statistical analyses were performed using analysis of variance (ANOVA), Fisher’s exact test for the p53 mutations in the mouse cell lines or the Student t or log-rank tests (GraphPad Prism 5). All data are shown as mean ± standard deviation or ± standard error of the mean. Probabilities of p-value < 0.05 were considered significant. For the analysis of an association between -7/del(7q) and the p53 mutational status in ALL, AML and MDS samples, the Chi-squared test was used.

Significance.

CDK6 acts at the interface of p53 and RB by driving cell-cycle progression and antagonizing stress responses. While sensitizing cells to p53-induced cell death, specific inhibition of CDK6 kinase activity may provoke the outgrowth of p53 mutant clones from premalignant cells.