The p16 Status of Tumor Cell Lines Identifies Small Molecule Inhibitors Specific for Cyclin-dependent Kinase 4¹

The p16 gene (also known as CDKN2A) encodes p16^INK4A, which inhibits the CDK4⁵:cyclin D and CDK6:cyclin D complexes (1). These complexes mediate phosphorylation of the Rb protein and allow cell cycle progression beyond the G₁-S-phase checkpoint (2). Alterations of p16 have been described in a wide variety of histological types of human cancers including astrocytoma, melanoma, leukemia, breast cancer, head and neck squamous cell carcinoma, malignant mesothelioma, and lung cancer. Alterations of p16 can occur through homozygous deletion, point mutation, and transcriptional suppression associated with hypermethylation in cancer cell lines and primary tumors (reviewed in Refs. 3, 4, 5).

Whereas the Rb gene is inactivated in a narrow range of tumor cells, the pattern of mutational inactivation of Rb is inversely correlated with p16 alterations (6, 7, 8), suggesting that a single defect in the p16/CDK4:cyclin D/Rb pathway is sufficient for tumorigenesis. Genetic alteration or overexpression of CDK4 and cyclin D1 has also been observed in various tumor cells, which supports the model that all tumor cells must circumvent this tumor suppressor pathway (1, 9).

Transfection of the p16 gene into cultured cell lines with p16 alterations (biallelic deletion or transcriptional suppression) causes G₁ arrest and growth suppression in a range of tumor cell types including osteosarcoma, esophageal carcinoma, mesothelioma, and head and neck squamous carcinoma, whereas transfection of this gene does not induce G₁ arrest in Rb-negative cells (10, 11, 12). In addition, p16 expression mediated by an adenovirus vector induces G₁ arrest and inhibits tumor cell proliferation in NSCLC cell lines with homozygous deletion of the p16 gene, but not in NSCLC cell lines expressing functional p16^INK4A (13). These data suggest that an agent possessing p16-like inhibitory activity against the CDK4:cyclin D kinase complex might have selective antitumor activity in patients with p16-altered tumors.

The CDK4 kinase is a member of the evolutionarily conserved family of CDKs, which includes CDC2 and CDK2. However, alterations of CDC2, CDK2, and their associated cyclins and inhibitors are not common in human cancers (14). The frequent defects in p16 with deregulated CDK4 activity suggest that pharmacological inhibitors specific for CDK4 may be more promising as anticancer agents than nonspecific CDK inhibitors. To date, several families of chemical inhibitors with specificity against different CDK activities have been described (15, 16, 17), all of which are ATP competitors (15, 16, 18). In addition, selective peptide inhibitors of CDK2 and CDK4 have been synthesized and evaluated (19, 20). However, no chemical inhibitors specific for CDK4/CDK6 have been reported.

We hypothesized that a small molecule with specific inhibitory activity for CDK4:cyclin D kinase would induce greater growth suppression among p16-altered cell lines than among p16-normal cell lines. The NCI drug screen program has determined the growth-inhibitory properties of over 50,000 compounds of diverse molecular structure against 60 human tumor cell lines of nine histological groups (21). We determined the p16 status of these 60 cell lines to identify pharmacological agents that preferentially inhibited the growth of p16-altered cell lines. Using this method, we identified several novel CDK4 inhibitors, some of which exhibit marked selectivity for CDK4 kinase as compared to CDC2 and CDK2 kinases.

Exponentially growing cultures of the 60 cell lines used in the NCI drug screen panel (21) were generously provided by Dr. A. Monks. All compounds were obtained from the Drug Synthesis and Chemistry Branch, NCI. In vitro antitumor activity (GI₅₀) of compounds was determined as described previously in the 60 human cancer cell lines of the NCI drug screen panel (21). Compounds structurally related to 3-ATA (NSC 680434) or BTD (NSC 645787) were selected by a substructure search of the entire NCI database (approximately 500,000 compounds) for structures with two benzene rings fused to a middle ring of any size or with a sulfone similar to BTD, respectively. This substructure search identified 77 structurally related compounds; 45 of these 77 compounds (24 for 3-ATA and 21 for BTD) were available for in vitro kinase assay. These 45 compounds have not been tested for growth-inhibitory activity in the NCI drug screen panel.

PCR-SSCP, DNA sequencing, and reverse transcription-PCR analysis of p16 in the 60 cell lines of the NCI drug screen panel were performed as described previously (22, 23). For Southern blot hybridization analysis, reverse transcription-PCR products were separated by agarose gel electropheresis, transferred to a nylon membrane, and hybridized with a 388-bp p16 exon 1 genomic fragment defined by oligonucleotides 2F and 1108R (24). Expression of the glyceraldehyde-3-phosphate dehydrogenase gene (GAPDH) was examined to assure the presence of intact mRNA in each sample. For immunoblot analysis, 1 × 10⁷ cells were washed with PBS, resuspended in 0.4 ml of lysis buffer [50 mm Tris-HCl (pH 7.4), 250 mm NaCl, 5 mm EDTA, 0.1% NP40, 50 mm NaF, and 1 mm phenylmethylsulfonyl fluoride], and centrifuged at 14,000 rpm for 20 min at 4°C. Total protein (50 μg) was subjected to SDS-PAGE, followed by electroblotting to nitrocellulose. The nitrocellulose membranes were incubated overnight at 4°C with a 1:1000 dilution of polyclonal antihuman p16 antiserum (PharMingen, San Diego, CA) in blocking solution (1× PBS, 5% powdered milk, and 1% BSA). The membranes were then incubated with a mixture of 40 μl of ¹²⁵I-protein A (>30 mCi/mg) in 20 ml of blocking solution and subjected to autoradiography.

The COMPARE algorithm was performed as described previously (25, 26). For the identification of agents with differential activity, GI₅₀s of 0 and 1 were used for p16-normal and for p16-altered cell lines, respectively. p16-altered cell lines were those with biallelic deletion, intragenic mutation, or transcriptional suppression of p16, and p16-normal cell lines were those without these abnormalities. Pearson correlation coefficients were calculated by the SAS procedure PROC CORR (SAS Institute, Inc., Cary, NC).

Active CDK:cyclin complexes were produced in Sf9 cells coinfected with baculoviruses encoding human CDK (CDK1, CDK2, or CDK4) or cyclin (cyclin A, cyclin D1, or cyclin E) gene (the generous gifts of D. Beach) at a multiplicity of infection of 10, and cell lysates were prepared as described previously (27, 28). CDK:cyclin complexes were purified by immunoprecipitation using each cyclin antibody (cyclin A, BF-683; cyclin D1, M-20; cyclin E, C-19; Santa Cruz Biotechnology). The purity of immunoprecipitated complexes was estimated by silver staining and Coomassie Blue staining, followed by Western blotting. The concentration of CDK subunit in the holoenzyme immunoprecipitated by each cyclin antibody was estimated by Coomassie Blue staining of electrophoretically separated proteins in comparison to protein standards of known concentrations (29). After quantitation of the CDK subunits, CDK:cyclin complexes were titrated for Rb kinase activity using 300 ng of GST-Rb to determine the optimal amount of the enzyme for each reaction. The estimated amount of CDK used in each assay was 25, 20, 16, and 60 ng for CDK1:cyclin A, CDK2:cyclin A, CDK2:cyclin E, and CDK4:cyclin D1, respectively.

Crude lysate (5 μl) containing CDK:cyclin or the optimized amount of purified CDK:cyclin complexes was mixed with test compounds in 30 μl of kinase buffer [20 mm Tris-HCl (pH 8.0), 10 mm MgCl₂, and 1 mm EGTA] and incubated at 30°C for 30 min. The kinase reaction was started by adding 300 ng of GST-Rb protein and 5 μCi of [γ-³²P]ATP to the mixture and incubation at 30°C for 30 min (30). Reactions were stopped by adding 7.5 μl of 5× SDS sample buffer [312.5 mm Tris-HCl (pH 6.8), 50% glycerol, 10% SDS, 12.5% 2-mercaptoethanol, and 0.0125% bromphenol blue]; samples were separated on 8–16% Tris-glycine denaturing gels (Novex), and radioactivity incorporated into labeled substrate was measured by liquid scintillation of the excised bands of GST-Rb. To examine the effect of compounds on the initial velocity of the enzyme, reactions were performed for 5 min without preincubation. To assess the effect of ATP on the inhibitory effect of a compound, 20–100 μCi of [γ-³²P]ATP were added to the reaction containing 3–200 μm ATP.

In vitro-translated, ³⁵S-labeled CDK4 and cyclin D1 were synthesized using plasmids containing the human CDK4 gene or cyclin D1 gene, a coupled transcription-translation system (TNT lysate; Promega), and [³⁵S]methionine (Amersham). For p16 binding, 1 μg of each GST-p16 or GST protein was mixed with 5 μl of in vitro-translated CDK4 in 100 μl of kinase buffer. After incubation at 30°C for 30 min, GST fusion proteins were separated by glutathione-Sepharose, resolved on an 8–16% Tris-glycine gel, and stained with Coomassie Blue to observe the recovery of GST-p16 fusions from each binding reaction. Quantitation of the binding reactions was then carried out by phosphorimaging. To test the effect of compounds on CDK4 binding of wild-type p16, up to 300 μm of each compound was premixed with CDK4 before adding GST-p16 protein. To examine the compound effect on CDK4:cyclin D1 binding, in vitro-translated cyclin D1, instead of GST-p16, was added and incubated at 30°C for 30 min, and then cyclin D1 was recovered by immunoprecipitation using cyclin D1 antibody (M-20; Santa Cruz Biotechnology).

The 60 cell lines of the NCI drug screen panel were examined for alterations of p16. To detect genetic alterations, PCR-SSCP analysis was performed for exons 1 and 2 of the p16 gene using genomic DNA (data not shown). Of the 60 cell lines, 29 were found to lack amplifiable genomic sequences of one or both exons, indicative of a biallelic deletion involving p16 (Table 1). Seven of the 60 cell lines contained abnormally migrating SSCP bands on repeated analyses that were not previously reported polymorphisms by DNA sequence analysis. The functional effects of these sequence variants were assessed by measuring the binding of GST-p16 fusion proteins to CDK4. Binding of mutant GST-p16 fusion proteins (I1+2^T-C, P81L, and D84Y) to CDK4 was 13%, 14%, and 13% of the binding ability of normal p16, respectively (Fig. 1). Thus, 36 of 60 (60%) cell lines of the NCI drug screen panel contained a genetic alteration (homozygous deletion or intragenic mutation) of p16 that disrupted the function of p16^INK4A (Table 1). To detect epigenetic alterations associated with loss of p16 function, p16 mRNA and protein expression were examined, which revealed 11 additional cell lines expressing neither p16 mRNA nor protein (Table 1). In total, 47 of the 60 (78%) cell lines of the NCI drug screen panel had an alteration of p16 (Table 1).

To identify compounds that are selectively cytotoxic or cytostatic for p16-altered cells compared to p16-normal cells, the p16 status of the 60 cell lines was matched to the GI₅₀ of the compounds of the NCI drug screen program and ranked according to Pearson correlation coefficients using the COMPARE algorithm (25, 26). The GI₅₀ of cephalostatin 1, a disteroidal alkaloid extracted from the marine worm Cephalodiscus gilchristi (31), correlated best with p16 status (r = 0.599; Table 2). The GI₅₀s of five related compounds (cephalostatins 7, 9, 8, 4, and 3) were also positively correlated with p16 status (r = 0.504, 0.493, 0.491, 0.461, and 0.458, respectively; Table 2). Bryostatin 1, a protein kinase C activator isolated from the marine bryozoan Bugula neritina (32), had a correlation coefficient of 0.469 (Table 2).

Aliquots of 26 of the 40 compounds with the highest Pearson correlation rankings were available for biochemical analysis. These compounds were assessed for CDK4:cylin D kinase inhibitory activity using crude Sf9 insect cell lysate containing baculovirus-expressed CDK4:cyclin D1 and a GST-Rb fusion protein as substrate. Six of the 26 compounds examined inhibited CDK4:cyclin D1-mediated phosphorylation of Rb protein with IC₅₀ (50% kinase inhibition) values ranging from 6.8 to more than 100 μm(Table 2). No inhibition of GST-Rb phosphorylation by CDK4:cyclin D1 was observed in the presence of the other 20 compounds at concentrations of up to 100 μm. The most potent inhibitor was 3-ATA, with an IC₅₀ of 6.8 μm, which shows moderate growth-inhibitory activity with a mean GI₅₀ of 30 μm in the 2-day growth assay of the NCI drug screen. Cephalostatin 1, which has potent antitumor activity in vitro (ED₅₀, 10⁻⁷ to 10⁻⁹ μg/ml; Ref. 31), had an IC₅₀ for CDK4:cyclin D1 of 20 μm, and bryostatin 1 had no inhibitory activity at the highest concentration examined (Table 2).

To identify compounds in the NCI drug screen that may have a mechanism of action similar to that of 3-ATA, we compared the pattern of the GI₅₀ of 3-ATA with the GI₅₀ of all compounds tested previously. Six compounds not examined previously for CDK4 kinase inhibitory activity had similar patterns of growth-inhibitory activity, with correlation coefficients greater than 0.6. Among these six compounds, two BTD compounds (NSC 645787 and NSC 645788) inhibited CDK4:cyclin D1 kinase activity in vitro with IC₅₀s of 5.0 and 17 μm, respectively.

Forty-five additional compounds with structural similarity to 3-ATA and BTD were analyzed to identify additional CDK4-specific inhibitors and obtain preliminary structure-activity relationship information. Nineteen of these compounds inhibited CDK4 kinase activity with IC₅₀s ranging from 1.1 to more than 100 μm using crude Sf9 cell lysate containing CDK4:cyclin D1. These candidate compounds for CDK4-specific inhibitors were tested for CDK inhibitory activity using purified enzymes.

To remove other factors in crude lysates that may affect the enzymatic activity or the effect of inhibitor compounds, we purified baculovirus-expressed CDK:cyclin complexes by immunoprecipitation. In addition, we tested flavopiridol (NSC 649890) to compare its inhibitory kinetics with those of the novel compounds we identified. Flavopiridol showed the most potent inhibition in the compounds we tested on CDK4:cyclin D1, CDC2:cyclin A, CDK2:cyclin A, and CDK2:cyclin E, with IC₅₀s of 0.14, 0.1, 0.08, and 0.32 μm, respectively (Fig. 2,A). The mechanism of inhibition of flavopiridol on CDK4:cyclin D1 is proposed to be mediated by competition with ATP (Fig. 2,B), as described previously on CDC2 kinase (15). 3-ATA, BTD, and compounds structurally related to 3-ATA (NSC 625987, NSC 645153, and NSC 521164) inhibited immunopurified CDK4:cyclin D1 with IC₅₀s ranging from 0.2–3.1 μm. These five compounds were significantly less potent inhibitors of CDC2:cyclin A, CDK2:cyclin A, and CDK2:cyclin E ,with IC₅₀s at least 30-fold higher compared to the IC₅₀s for CDK4:cyclin D1 (Table 3, Fig. 2,A). Kinetic studies using purified CDK4:cyclin D1 and GST-Rb protein as a specific substrate showed that 3-ATA does not compete with ATP for the inhibition of CDK4 kinase activity and that BTD has a mixed pattern of inhibition with respect to ATP (Fig. 2 B). The K_i values of 3-ATA, BTD, and flavopiridol on CDK4 kinase against ATP were calculated to be 5.5, 0.73, and 0.076 μm, respectively. To assess the effect of the compounds (3-ATA, BTD, NSC 625987, NSC 645153, NSC 521164, and flavopiridol) on the binding of CDK4 to p16 or cyclin D1, binding assays using in vitro-translated CDK4 and bacterially expressed GST-p16 or in vitro-translated cyclin D1 were performed in the presence of these compounds. No inhibition of CDK4 binding to GST-p16 was observed in the presence of up to 300 μm of these compounds (data not shown). For CDK4-cyclin D1 binding, two compounds (BTD and NSC 625987) inhibited CDK4:cyclin D1 binding only at concentrations of at least 300 μm (data not shown), which is nearly 3 logs higher than their IC₅₀s on CDK4 kinase inhibitory assay.

Genetic alterations of the p16 gene in primary tumors, including homozygous deletions and intragenic mutations, are observed in 35% of NSCLC and 60% of glioma, head and neck squamous carcinoma, and pancreatic cancer (reviewed in Ref. 4). In addition, hypermethylation of the 5′ CpG island of the p16 gene, correlating with complete transcriptional suppression, is observed in 25% of NSCLC, 31% of breast cancer, and 40% of colon cancer (reviewed in Ref. 5). Thus, inactivation of the p16 gene is a frequent event in various histological types of primary tumors and tumor cell lines.

The observation of growth suppression after expression of p16^INK4A in tumor cells with deletion of the p16 gene (13, 33) has suggested that inhibition of CDK4 kinase activity may be a useful therapeutic strategy for patients whose tumors have p16 defects. We used the existing database of the NCI drug screen program (21, 25) to identify potential pharmacological inhibitors of CDK4 kinase activity. From among the large number of compounds of diverse molecular structure in this database, we selected for further study compounds with greater growth inhibitory activity against p16-altered cells than against p16-normal cells.

Using this approach and further biochemical analyses, we identified five compounds (3-ATA, NSC 625987, NSC 645153, NSC 521164, and BTD) that inhibit CDK4 kinase activity in vitro, with an IC₅₀ of 30-fold to more than 500-fold lower than the IC₅₀ required to inhibit CDC2 and CDK2 kinases. The inhibitory activity of the parent compound, 3-ATA, was not attenuated with increasing concentrations of ATP, unlike other chemical inhibitors of CDKs described to date (15, 16, 17, 18). However, BTD, which has a growth-inhibitory pattern similar to that of 3-ATA, inhibited CDK4 in a linear mixed fashion with respect to ATP. Because the ATP-binding pocket of CDKs is likely to accommodate various structures (34), BTD might partially compete with ATP by binding to the ATP-binding pocket or by interfering with ATP binding.

These five compounds, as well as flavopiridol, did not affect p16 binding to CDK4 in vitro, suggesting that the mechanism of CDK4 inhibition by these compounds is not mimicking the tumor suppressor p16. In addition, only two compounds (BTD and NSC 625987) inhibited in vitro cyclin D1 binding to CDK4 at nearly 1000-fold higher concentration than their IC₅₀ on CDK4 kinase. This observation suggests that direct inhibition of cyclin D1 binding to CDK4 is not the central mechanism of inhibition of CDK4 kinase activity by these compounds. INK4 inhibitors bind next to the ATP binding site of the catalytic cleft of CDK6 and interfere with ATP binding by causing conformational changes (35, 36). Our data suggest that there may be additional mechanisms mediating specific CDK4 inhibition.

Through COMPARE analysis and biochemical screening, the cephalostatins were also found to have greater growth-inhibitory activity against p16-altered cells than against p16-normal cells. However, CDK4 kinase inhibitory activity by cephalostatin 1 occurs at concentrations at least 1000-fold higher than the GI₅₀ of cephalostatin 1, suggesting that the growth-inhibitory activity of the cephalostatins is not predominantly due to CDK4 kinase inhibition. Our analysis also identified bryostatin 1 as being more active against p16-altered cells than p16-normal cells. However, there was no direct inhibition of CDK4 kinase activity in vitro. The addition of bryostatin 1 to cells has been shown to result in decreased CDK2 activity, which is due, at least in part, to dephosphorylation of CDK2 (37). Bryostatin 1 may similarly decrease CDK4 activity through dephosphorylation of CDK4, thus explaining its greater activity against p16-altered cells.

The therapeutic index of agents in vivo is thought to be related to the specificity of their actions on molecular targets. In our initial attempt to improve upon the approximately 10-fold specificity of 3-ATA for CDK4 compared to CDC2 and CDK2, we were able to identify compounds structurally related to 3-ATA with at least 100-fold higher specificity for CDK4 than for CDC2 and CDK2. These compounds may provide a step to develop structure-based chemical libraries to identify or synthesize more potent inhibitors of CDK4 kinase activity. Also, determination of the crystal structure of CDK4 or CDK6 bound to these compounds will allow us to better understand the mechanism of inhibition specific for CDK4 or CDK6.

In conclusion, we have identified specific small molecule inhibitors of CDK4 by comparing the growth-inhibitory activity of more than 50,000 compounds with the p16 status of the cell lines in the NCI drug screen panel. This approach may ultimately lead to the development of a useful therapeutic strategy for patients with p16-altered tumors.

We thank George Johnson, Jill Johnson, Edward Sausville, and Kenneth Paull of the Developmental Therapeutics Program, NCI for advice and assistance in performing COMPARE analyses. We thank Anne Monks and her laboratory for providing cell lines and John Weinstein for helpful discussion. We are grateful to David Beach for providing the baculovirus stocks, Sachiko Kajigaya for advice on propagating baculovirus in Sf9 cells, and Jamie Hui for assistance in preparing and purifying CDKs.