Point Mutations Can Inactivate in Vitro and in Vivo Activities of P16ink4a/CDKN2A in Human Glioma

Point mutations can inactivate in vitro and in vivo activities of p16INK4a/CDKN2A in human glioma

Wadih Arap1,5, Erik S Knudsen3, Jean YJ Wang3,4, Webster K Cavenee1,2,4 and H-J Su Huang1,2

1Ludwig Institute for Cancer Research, 2Department of Medicine, 3Department of Biology, 4Center for Molecular Genetics, University of California-San Diego 9500 Gilman Drive, La Jolla, California 92093-0660; and 5Cancer Biology Program, Stanford University, Stanford, California 94305, USA

Deletions of chromosomal region 9p21 are among the the most common structural abnormality observed in most common genetic alterations observed during the malignant gliomas (James et al., 1993; Olopade et al., clonal evolution of high grade malignant gliomas. 1992). The common occurrence of deletions of this Structural and functional evidence has suggested that region has prompted several studies to determine the homozygous deletion involving CDKN2A (the genetic frequency of alteration of the CDKN2A (MTS1/ locus encoding the cyclin-dependent kinase inhibitor p16INK4a) gene (Kamb et al., 1994; Nobori et al., p16INK4a) is a mechanism of inactivation of this gene 1994; Serrano et al., 1993) in the genetic progression and that it can be a growth suppressor in human gliomas. of glial tumors (Gianni and Finocchiaro, 1994; He et However, the presence of other potential suppressor al., 1994, 1995; Jen et al., 1994; Kyritsis et al., 1996; genes in the 9p21 region and the relatively large sizes of Li et al., 1995; Moulton et al., 1995; Nishikawa et al., the deletions has made it dicult to be certain that the 1995; Schmidt et al., 1994; Ueki et al., 1994; Walker CDKN2A gene is their actual target. Here, we tested et al., 1995). The accumulated structural evidence this hypothesis by determining the growth suppressive implicates the CDKN2A gene as a target of 9p21 eects, cell cycle inhibitions, and the activities of seven deletions in gliomas. However, the presence of other naturally occurring glioma-derived CDKN2A alleles potential suppressor genes within the 9p21 region such carrying point mutations and found that two of them as the interferon gene cluster, the p15INK4b gene were functionally compromised. To resolve discrepancies (MTS2/CDKN2B) (Hannon and Beach, 1994) and among the dierent existing functional assays, we an alternative transcriptional reading frame of the developed an assay for p16INK4a function that allowed us CDKN2A gene (Duro et al., 1995; Mao et al., 1995; to demonstrate that the expression of wild-type Stone et al., 1995; Quelle et al., 1996); in addition to CDKN2A, but not alleles with inactivating mutations, the relatively large sizes of the deletions, has made it prevents pRB phosphorylation in vivo in human glioma dicult to establish that CDKN2A is their actual cells. These data suggest that CDKN2A is a critical target (Cairns et al., 1995; Jen et al., 1994). First, the target for mutational inactivation in human malignant search for chromosome structural abnormalities has gliomas. been hampered by technical diculties with PCR analysis of the CDKN2A region which has proven Keywords: tumor suppressor gene; CDKN2A; p16INK4a; unreliable and dicult to quantitate due to contam- RB; mutation; glioma inating normal tissue, poor ampli®cation of GC-rich sequences, and the need for multiple closely spaced polymorphic markers to detect small deletions (Cairns et al., 1995). Second, structural analysis of 9p21 likely Introduction underestimates the importance of CDKN2A as a glioma suppressor gene since it does not address Malignant tumors originate and progress through a epigenetic mechanisms of p16INK4a inactivation such as multistep process in which growth-advantageous transcriptional silencing due to aberrant methylation genetic events accumulate (Nowell, 1976). This clonal (Costello et al., 1996; Merlo et al., 1995; Nishikawa et evolution model has undergone rigorous evaluation al., 1995) or modulation by other genes in the p16INK4a during the progression of malignant gliomas, leading to pathway such as CDK4 ampli®cation and overexpres- the identi®cation of multiple speci®c genetic defects sion (He et al., 1994, 1995; Nishikawa et al., 1995; such as loss of heterozygosity (LOH) of chromosome Schmidt et al., 1994). Third, a further level of 17p, mutation of the p53 gene, overexpression of the complexity is created by examples of mutant platelet derived growth factor-a (PDGFa), LOH of CDKN2A alleles derived from familial cases of chromosomes 22q, 19q, 9p and 13q including the malignant melanomas and pancreatic carcinomas retinoblastoma (RB) gene, as well as ampli®cation and which do not seem to segregate with the tumors and in-frame truncation of the epidermal growth factor appear to have only a minor impact on p16INK4a receptor (EGFR) gene (for reviews see Furnari et al., function (Goldstein et al., 1995; Ranade et al., 1995). 1995; Louis et al., 1994). Although the predominant mode of inactivation of Among these genetic lesions, hemi- and homozy- CDKN2A seems to be homozygous deletions, the gous deletions involving chromosome region 9p21 are sequence of the CDKN2A gene has been examined and a number of intragenic point mutations have also been Correspondence: H-JS Huang found in primary malignant gliomas (Gianni and Received 14 August 1996; revised 7 October 1996; accepted 8 October Finocchiaro, 1994; Jen et al., 1994; Kyritsis et al., 1996 1995, 1996; Li et al., 1995; Moulton et al., 1995; CDKN2A mutations in glioma WArapet al 604 Nishikawa et al., 1995; Schmidt et al., 1994; Ueki et length wild-type CDKN2A cDNA. After completion of al., 1994; Walker et al., 1995). Their functional drug selection, the growth suppression was quanti®ed signi®cance, however, has not been elucidated. These by comparing the number of drug resistant cells in include 13 alleles containing missense (A73T, A76V, cultures transfected with CDKN2A wild-type or A85T, T93R, H98Y, A102V, V106M, R107C, P114L, mutant alleles to those of cultures transfected with P114S, A127V, R128W, and G136D) and one allele empty vector (Figure 1). The growth suppressive eect containing nonsense (W110*) mutations. All are of the A73T, H98Y, V106M, R107C, and P114S alleles located in exon 2 of CDKN2A, where they are was marked and not signi®cantly dierent from the distributed in the second (A73T), between the second eect of wild-type allele replacement (less than 30% of and third (A76V), third (A85T, T93R, H98Y, A102V, the relative cell number seen with empty control V106M and R107C) and fourth (W110*, P114L, vector), while a relative relief of growth suppressive P114S, A127V, R128W and G136D) ankyrin repeats eect was observed with the W110* and P114L alleles of p16INK4a. The W110* mutation has also been (more than 70% of the relative cell number seen with observed in one melanoma cell line (Kamb et al., empty control vector). 1994) and two pancreatic adenocarcinoma xenografts The expression of proteins from the various (Caldas et al., 1994) while the P114L mutation was also CDKN2A alleles was assessed in each case by Western reported in two melanoma cell lines (Kamb et al., blot analysis of p16INK4a; the data shown in Figure 2a 1994). are from analyses of U-251MG cells. No p16INK4a was It has been postulated that CDKN2A is the ®rst detectable in vector transfected U-251MG cells. The example of a tumor suppressor gene in which each mutant CDKN2A transfectants showed p16INK4a levels allele is most frequently inactivated by two indepen- similar to those of other glioma cells which express dent deletion events (Kamb, 1995; Liu et al., 1995), endogenous wild-type CDKN2A alleles (Arap et al., consistent with a two-hit model (Knudson et al., 1975) 1995). The exception was the nonsense mutation and with the low incidence of point mutations W110*, in which no protein was detected (Figure 2a), observed in primary tumors (Cairns et al., 1995). although the levels of CDKN2A W110* mRNA were Moreover, we have previously demonstrated that indistinguishable from those of CDKN2A wild-type replacement of a wild-type CDKN2A allele in mRNA by Northern blot analysis (Figure 2b). More- CDKN2A null human glioma cells suppresses their over, a band corresponding to the product of the growth (Arap et al., 1995). This work provided W110* truncated p16INK4a protein was detected in an in evidence that the CDKN2A gene could function as vitro transcription and translation assay (Figure 2c) a glioma growth suppressor, but the overexpression of demonstrating the translatability of the mRNA derived this cell cycle regulator could have downstream eects, from the pCDKN2W110* construct. Empty control despite the care with which expression levels approx- vector and the construct pCDKN2WT were used as imating those of the endogenous levels were main- negative and positive controls, respectively, in this in tained (Arap et al., 1995). Following on these data, we vitro assay. Thus, since a translatable W110* reasoned that a demonstration that the rare naturally transcript, but not its protein, can be detected in vivo, occurring intragenic point mutations caused loss of it is likely that this premature termination aects function would provide strong evidence that the p16INK4a protein stability. CDKN2A gene is a critical target of the 9p21 In order to determine if the growth suppressive deletions in malignant gliomas. Here, we tested this eects of CDKN2A and lack of suppression by W110* hypothesis by determining the growth suppressive and P114L were due to a speci®c block in the cell cycle, eect, cell cycle inhibition, and the in vitro and in we transiently cotransfected the CD20 cell surface vivo activities of a series of seven naturally occurring marker and the various CDKN2A alleles (Koh et al., glioma-derived CDKN2A alleles carrying point mutations and found that two of them were functionally impaired. We also developed a novel in vivo assay that allowed us to demonstrate that the expression of wild- type CDKN2A, but not alleles with inactivating mutations, exert their eects by preventing pRB phosphorylation in human glioma cells.

Results

Glioma-derived cells whose p16INK4a-RB pathways have been characterized were used as recipients for CDKN2A gene transfer. In order to test for loss-of- function eects, we introduced seven dierent glioma- derived CDKN2A alleles (A73T, H98Y, V106M, Figure 1 Growth suppressive eect of CDKN2A allele replace- R107C, W110*, P114L and P114S) separately into ment in human glioma cells. Ability of glioma-derived CDKN2A each of three human glioma cell lines: U-87MG and U- alleles to induce growth suppression in a stable transfection assay, 251MG, which are both endogenously p16INK4a7/RB+, following drug selection. The nomenclature indicates the location and LN-319 which is p16INK4a+/RB7 (Arap et al., 1995). and identity of the amino acid changes. Viable cells were counted after G-418 selection. Results are normalized in terms of We then compared the eect of these alleles on cell percentage of vector transfection with the vector control set to growth to that observed when the cells were transfected 100% in each case. A typical experiment is shown. These results either with empty vector sequence or with the full were reproduced in three independent experiments CDKN2A mutations in glioma WArapet al 605 1995; Lukas et al., 1995; van der Heuvel and Harlow, 1993). Two days after the cotransfections, the cell cycle distribution of the population of CD20-positive cells INK4a a was examined in each group. In both of the p16 negative glioma cell lines, the alleles A73T, H98Y, V106M, R107C and P114S induced a signi®cant increase in the G population (to levels similar to VECTOR WT A73T H98Y V106M R107C W110* P114L P114S 1 those achieved by wild-type CDKN2A) and a reciprocal decrease in their G and S populations. p16INK4a ¨ 2 W110* and P114L elicited only a slight G1 increase (Figure 3). The cell cycle distribution of LN-319 cells, which contain endogenous wild-type CDKN2A alleles (Arap et al., 1995) and are pRB negative, was not signi®cantly altered by any of the constructs (Figure 3). b Thus, these two alleles (W110* and P114L) appeared U-87MG U-251MG to encode severely compromised p16INK4a proteins, while the ®ve others (A73T, H98Y, V106M, R107C and P114S) encoded proteins that were not signi®cantly dierent from the wild-type protein in their ability to VECTOR WT W110* VECTOR WT W110* — 1.3 Kb induce G1 arrest. The capacity of the wild-type and mutant p16INK4a CDKN2A RNA ¨ — 0.9 Kb proteins to prevent progression through the G1 phase of the cell cycle generally correlates with their inhibition of CDK4 or CDK6 and D-cyclin activities — 18 S (the only known targets of p16INK4a) in vitro (Ranade et al., 1995; Lukas et al., 1995; Koh et al., 1995). In order to test the eects of the glioma-derived mutations on these speci®c kinase activities, GST- p16INK4a fusion proteins were tested for their ability to c inhibit CDK4/cyclin D1 or CDK6/cyclin D1, using the retinoblastoma protein (pRB) as a substrate. VECTOR pCDKN2 WT pCDKN2 W110*

— 21.0 kD

INK4a p16 WT ¨ — 14.4 kD p16INK4a W110* ¨ — 6.5 kD

Figure 2 Expression of the CDKN2A alleles. (a) Immunoblot analysis of wild-type and mutant p16INK4a. Western blot of wild- type and mutant p16INK4a proteins in an endogenous CDKN2A null background (U-251MG) after stable transfections. The protein expression of wild-type and all mutant alleles, except Figure 3 Eect of CDKN2A alleles on cell cycle progression. W110*, was detected with an anti-p16INK4a polyclonal antibody. Ability of glioma-derived CDKN2A alleles to induce cell cycle G1 Similar results were also observed in U-87MG cells transfected arrest in a transient transfection assay. CDKN2A wild-type, with CDKN2A constructs (data not shown). (b) Comparison of mutant, or vector alone were transfected by the calcium RNA expression of CDKN2A WT and W110* alleles. Upper phosphate method in duplicate 100 mm dishes. Two days after panel. Northern blot analysis of CDKN2A wild-type and W110* transfection, the eect of CDKN2A alleles on cell cycle transcripts in the glioma cells U-87MG and U-251MG, both progression, in a cell population identi®ed by cotransfection of with a CDKN2A null background. A 0.95 Kb band correspond- limiting amounts of the surface marker CD20 (pCMV- ing to the transfected CDKN2A alleles was detected using a CD20 : pCDKN2A molar ratio 1 : 4), was plotted as a function p16INK4a exon 1a speci®c probe as described (Arap et al., 1995). of DG1%. DG1% represents the percent change in the proportion In both cell lines, the levels of WT and W110* messages were of cells in G1 relative to the vector control sample, and is de®ned expressed at comparable levels. Lower panel. Membrane staining by the equation (Koh et al., 1995): of the 18S RNAs with methylene blue (Herrin and Schmidt,

1988) showing equal loading of total RNA. (c) In vitro q 7

transcription and translation of CDKN2A WT and W110* I

7 ells in q Y test smpleÿ 7 ells in q Y ontrol

I constructs. A coupled in vitro transcription and translation assay I

IHH

INK4a

Y ontrol 7 ells in q followed by immunoprecipitation with an anti human p16 I polyclonal antibody was performed. Empty control vector and pCDKN2WT were used as negative and positive controls Although the cell-cycle distributions of vector control samples respectively. In the lane corresponding to the W110* construct, displayed some inter-experiment variation, the relative eect a band of corresponding to the predicted size of the p16INK4a (dierences between vector control and testing samples) observed truncated product at position 110 (approximately 11.5 kD) was in one particular experiment was consistent in three independent detected, demonstrating the translatability of the W110* RNA experiments. A typical experiment is shown. DG1% results are transcript presented with the vector control set to zero in each case CDKN2A mutations in glioma WArapet al 606 Titration of the in vitro inhibitory activity for each wild-type in vitro. In contrast, a comparison of the control vector, wild type, and for each mutant inhibitory eect of GST-p16INK4a wild-type and P114L construct was performed, using the CDK4/cyclin D1 proteins demonstrated that increasing the P114L kinase. Increasing amounts of each of the GST- product to more than three times the amount of p16INK4a fusion proteins (ranging from 100 to 800 ng) wild-type still resulted in an inability to inhibit pRB were used and, under the conditions of this assay, phosphorylation in vitro (Figure 4b). A positive none of the alleles were able to inhibit pRB control assay showed that none of the p16INK4 phosphorylation when up to 800 ng of the each of proteins could inhibit pRB phosphorylation by the GST-16INK4a proteins were used (negative data not CDK2/cyclin A kinase (Figure 4a, lower panel), while shown). Thus, the lack of inhibitory ability of the a negative control assay with an irrelevant extract mutant alleles is unlikely to be due to concentration expressing large T antigen showed no pRB phos- dierences. When 1 mg of the fusion GST-p16INK4a phorylation (negative data not shown). Therefore, the was used, the product of each of the alleles, except only discrepancy between our in vivo and in vitro P114L but including W110*, was able to prevent results was the mutation W110*, which behaved as a pRB phosphorylation by CDK4/cyclin D1 kinase severely defective protein in both growth suppression (Figure 4a, upper panel) or by CDK6/cyclin D1 and cell cycle assays, but which was still similar to kinase in vitro (Figure 4a, middle panel). Thus, the in wild-type in its ability to inhibit pRB phosphoryla- vitro activities of the constructs which behaved tion in vitro. comparably to the wild-type allele in the transfection In order to address this problem, we devised an assay studies (A73T, H98Y, V106M, R107C and P114S) for in vivo p16INK4a inhibition of pRB phosphorylation in also had essentially the same inhibitory activity as which glioma cells were transiently cotransfected with limiting amounts of pCMV-RB large pocket (LP, residues 379-928) and either control vector or each of the CDKN2A (wild-type and mutant) constructs. a Cotransfected glioma cells were then used to test the eects of the various CDKN2A alleles on RB phosphorylation in vivo using immunoprecipitation, VECTOR WT A73T H98Y V106M R107C W110* P114L P114S — 64 kD electrophoresis, and analysis of the exogenous pRB/LP pRB/SE ¨ CDK4/Cyc D1 by immunoblotting. The ppRB/LP to pRB/LP ratio — 36 kD (ppRB/LP : pRB/LP) revealed the relative eects of the cotransfection of each of the CDKN2A alleles on the phosphorylation of the transfected RB/LP product, — 64 kD setting the ppRB/LP : pRB/LP ratio in the contransfec- ¨ pRB/SE CDK6/Cyc D1 tion of empty vector control to 100%. The eects of the — 36 kD CDKN2A alleles A73T, H98Y, V106M, R107C and P114S were comparable to those observed upon transfer

— 64 kD pRB/SE ¨ CDK2/Cyc A

— 36 kD R MOCK VECTO WT A73T H98Y V106M R107C W110* P114L P114S

b   P114L(µg) WT (µg) ppRB/LP ¨  pRB/LP ¨ — 66kD 0 1 2 3 0 1 — 64 kD ppRB/LP pRB/SE ¨ CDK4/Cyc D1 % 100% 24% 28% 29% 26% 12% 122% 99% 23% pRB/LP Figure 4 In vitro activities of the wild-type and mutant p16INK4a Figure 5 In vivo inhibition of RB phosphorylation assay. U- products. (a) In vitro inhibition of RB phosphorylation assay. 251MG glioma cells were cotransfected with limiting amounts of GST-p16INK4a fusion proteins encoded by each of the indicated pCMV-RB large pocket (LP, residues 379 ± 928) and either empty alleles were assayed for their ability to inhibit CDK4/cyclin D1, control vector, wild-type, or mutant CDKN2A alleles. Cells CDK6/Cyclin D1, or CDK2/Cyclin A, using a recombinant GST- cotransfected with a molar ratio of 1 : 15 (pCMV-RB/ RB SspI-End (SE, residues 768 ± 928) as the substrate for the LP : pCDKN2A) were used test to test the CDKN2A con- three CDK/cyclin kinases. Here, a total of 1 mg of GST-p16INK4a structs. pRB/LP was immunoprecipitated, resolved by 8.5% was used. All experiments were repeated at least three times. In SDS ± PAGE, and detected by immunoblotting. Shown below the experiments shown, samples were resolved by 4 ± 20% are the relative ppRB/LP to pRB/LP ratios, setting the vector gradient SDS ± PAGE. Upper panel. CDK4/cyclin D1 phosphor- control to 100%. The prevention of phosphorylation of the INK4a ylation of RB assay. All fusion proteins except GST-p16 cotransfected native RB/LP construct observed after introduction P114L are able to inhibit pRB/SE phosphorylation. Middle panel. of the CDKN2A A73T, H98Y, V106M, R107C and P114S alleles, CDK6/cyclin D1 phosphorylation of RB assay. All fusion which contain p16W mutations, is similar to that observed upon proteins except GST-p16INK4a P114L are able to inhibit pRB/ transfer of the CDKN2A wild-type allele in the CDKN2A null U- SE phosphorylation. Lower panel. CDK2/cyclin A phosphoryla- 251MG glioblastoma cells. In contrast, the introduction of either tion of RB assay. None of the fusion GST-p16INK4a constructs the W110* or the P114L allele is unable to prevent the are able to inhibit pRB/SE phosphorylation. (b) Titration of the phosphorylation of pRB/LP in vivo. Thus, phosphorylation of in vitro activity of the P114L allele. The fusion protein GST- the cotransfected pRB/LP product was similar to that observed in INK4a p16 P114L which contains an inactivating (p16M) mutation cells transfected with the empty control vector. Likewise, the is unable to inhibit pRB phosphorylation even if 3 mg of its relative ratio of ppRB/LP to pRB/LP is less than 30% in the product are used. One mg of p16INK4a WT fusion protein was wild-type, A73T, H98Y, V106M, R107C and P114S alleles, used as control. Thus, even threefold the amount of the mutant whereas the ratios in the alleles containing the loss-of-function protein (relative to the wild-type protein) does not recapitulate the mutations W110* and P114L are similar to that in the empty inhibitory eect of the wild-type product vector lane CDKN2A mutations in glioma WArapet al 607 of the CDKN2A wild-type allele (ppRB/LP : pRB/ cies between the ability of CDKN2A alleles to arrest LP530%), while the introduction of either W110* or cell cycle progression and their inhibition of CDK4/ P114L allele was unable to prevent the phosphorylation cyclin D1 kinase activity by recombinant p16INK4a in of pRB/LP, and was comparable to the eect observed in vitro have also been reported (Koh et al., 1995) glioma cells transfected with the control vector (Figure emphasizing the need for a more reliable and 5). The results of this assay clearly demonstrate that accurate assay to assess the in vivo biochemical W110* and P114L are loss-of-function alleles, contain- function of p16INK4a. ing p16INK4a inactivating mutations. We designed a novel in vivo assay of inhibition of pRB phosphorylation (Figure 5), which overcomes some of the limitations of the in vitro assays, such as Discussion the possibility of protein instability, and it has the advantage of simultaneously assessing both known A modest, yet signi®cant number of single base (CDK4 or CDK6/cyclin D1 or D2 or D3) and alterations in CDKN2A have been detected in unknown p16INK4a driven pathways to RB phosphory- primary glial tumors. Until now, the relevance of lation. The results of this assay unequivocally con®rm these putative CDKN2A point mutations in the that the alleles W110* and P114L indeed contain loss- biology of human gliomas has remained largely of-function mutations, in complete accordance with the unknown. Here, we present a functional analysis of a results of our growth arrest and cell cycle assays series of glioma-derived mutant CDKN2A alleles. A (Figures 1 and 3), suggesting that this in vivo assay novel in vivo assay of inhibition of RB phosphorylation provides a more reliable method to functionally was also used in our study. Based on the ability to evaluate mutations of the INK4 family (and, perhaps, induce G1 arrest or their long term growth suppression, mutations in other cyclin-dependent kinase inhibitors the CDKN2A alleles which behaved similarly to wild- such as p21WAF1 or p27KIP) than previously reported in type were classi®ed as p16W alleles, whereas the vitro assays (Koh et al., 1995; Lukas et al., 1995; Parry glioma-derived CDKN2A alleles exhibiting impaired and Peters, 1996; Ranade et al., 1995; Yang et al., function were classi®ed as p16M alleles (Goldstein et 1995). al., 1995). The p16W alleles (A73T, H98Y, V106M, Recently, a 20-residue peptide corresponding to R107C and P114S) aected either growth suppression amino acids 84 to 103 of the third ankyrin repeat or cell cycle distribution to levels greater than 70% of of p16INK4a was demonstrated to inhibit pRB the wild-type CDKN2A allele, while the p16M alleles phosphorylation in vitro (FaÊ hraeus et al., 1996). (W110* and P114L) aected each of the parameters by Since the inactivating mutations W110* and P114L less than 30% relative to wild-type. Therefore, both are located outside this functional domain, they are transient and stable CDKN2A gene transfer into likely to disrupt it by distinct molecular mechanisms, human glioma cells revealed consistent results. such as protein instability and allosteric conforma- The ability of p16INK4a proteins to prevent CDK4 tional changes, respectively. However, not only the or CDK6/cyclin D1 phosphorylation of pRB was codon location, but also the speci®c amino acid measured in vitro (Figure 4a) and shown to be changes appear to be relevant for the function of generally correlated with the transfection results, and the p16INK4a protein. For instance, the H98Y and our results for the comparative titration of the eects P114S alleles behave as wild-type (p16W) alleles, as of p16INK4a WT and P114L (Figure 4b) were also opposed to H98P (Koh et al., 1995) and P114L consistent with previously reported in vitro data (Koh (Koh et al., 1995; Lukas et al., 1995), which are et al., 1995; Lukas et al., 1995). Thus, the exception loss-of-function (p16M) alleles containing inactivating was the W110* allele, in which a nonsense mutation mutations. is located in the ®rst residue of the fourth ankyrin As established above, two of seven CDKN2A repeat of the p16INK4a protein, generating a truncated sequence alterations signi®cantly diminish p16INK4a protein which behaved as a loss-of-function allele in function, leaving open the question of the remaining transfection experiments but was able to inhibit pRB ®ve non-inactivating mutations. A prediction of our phosphorylation in vitro (Figure 4a). Since the levels functional analysis is that the p16W alleles, which of WT and W110* RNA are similar by Northern appear to contain non-inactivating mutations, would blot analysis (Figure 2b), these results could be confer no selective growth advantage in relation to the explained either by an inability of the W110* RNA wild-type p16INK4a allele (Sidransky et al., 1992). to be translated or by protein instability of a Indeed, these ®ve alleles (A73T, H98Y, V106M, prematurely truncated protein. An in vitro transcrip- R107C, P114S) were reported in malignant gliomas tion and translation assay demonstrated that the (Kyritsis et al., 1995, 1996) but each of these CDKN2A W110* RNA message is translatable. Taken together, mutations was detected only in a small percentage of one explanation for these observations is that while the tumor cells by individual colony sequencing and the native protein is unstable and undetectable in vivo Southern hybridization, consistent with heterogeneous (Figure 2a), the GST-p16INK4a W110* fusion construct glioma cell populations (Kyritsis et al., 1996), in which is expressed and able to function in vitro (Figure 4a), the mutations do not appear to have been selected. representing a limitation of the latter to model the While the biological signi®cance of these p16W former. These ®ndings are consistent with the mutations remains uncertain, this phenomenon was previous suggestions that the COOH-terminal region observed in other tumors such as malignant melano- of p16INK4a may be involved in the stability of the mas and pancreatic carcinomas (Goldstein et al., 1995; protein (Yang et al., 1995) and that the majority of Ranade et al., 1995). One explanation may lie with the CDKN2A premature terminations are inactivating p16b, a novel human p16INK4a transcript (Duro et al., mutations (Parry and Peters, 1996). Other discrepan- 1995; Mao et al., 1995; Stone et al., 1995) or an CDKN2A mutations in glioma WArapet al 608 unrelated protein from an alternative reading frame of Analysis of CDKN2A RNA and p16INK4a protein expression in the murine p16INK4a gene (p19ARF) which has recently vitro and in vivo

been reported and shown to be capable of inducing G1 CDKN2A Northern blots Total RNA was prepared from

and also G2 phase arrest independent of CDK4 or G418-resistant glioma cells and the CDKN2A RNA CDK6 (Quelle et al., 1995). Most of the p16INK4a expression in the transfectants was analysed by Northern glioma-derived mutations studied here (A73T, H98Y, blots essentially as described (Arap et al., 1995). V106M, R107C, W110*, P114L, P114S) would be expected to alter dierent codons in the human p16b p16INK4a Western blots After completion of drug selection, reading frame (respectively R87H, A112V, R120H, CDKN2A-transfected glioma cell lysates (20 mlfromavol/ A121V, G125R, A128A, A128V). The P114L allele, vol sample buer/cell pellet solution) were boiled, resolved by a gradient 4 ± 20% SDS ± PAGE and transferred to INK4a which demonstrably contains a p16 loss-of-func- Immobilon-P membranes (Millipore, Bedford, MA) as tion mutation, would correspond to a silent mutation described (Arap et al., 1995; Nishikawa et al., 1995). in the p16b reading frame. Thus, we are currently Sample loading was monitored by Ponceau S staining exploring the possibility that some of the so-called (Sigma, St Louis, MO). After blocking the non-speci®c p16W alleles may actually represent p16b inactivating sites, membranes were probed with anti-p16INK4a polyclonal mutations. antibody (Pharmingen, San Diego, CA). Normal rabbit These considerations caution that the ®nding of a and mouse sera were used as negative controls. Bound CDKN2A sequence alteration may not always indicate antibody was detected by enhanced chemiluminescence a defective p16INK4a protein. Without exhaustive according to the manufacturer's protocol (ECL, Amersham functional assessment, the possibilities of p16W Corp, Arlington Heights, IL). alleles, polymorphisms, additional reading frames, or In vitro transcription and translation assays An in vitro even sequencing errors due to the very high GC-rich transcription and translation assay of empty control vector, nature of the CDKN2A gene still exist. Nonetheless, pCDKN2WT and pCDKN2W110* constructs was performed the data reported here show that the CDKN2A gene using the TNT-coupled system (Promega, Madison, WI). The can be a speci®c target for mutational inactivation in products were immunoprecipitated in vitro using an anti- human glial tumors. p16INK4a polyclonal antibody (Pharmingen, San Diego, CA), resolvedbya 16% SDS ± PAGE and detected by¯uorography.

Flow cytometric analysis of the cell cycle Materials and methods Cell cycle analysis was performed using the CD20 system CDKN2A plasmid construction and protein puri®cation with minor modi®cations (Koh et al., 1995; Lukas et al., 1995; van der Heuvel and Harlow, 1993). Brie¯y, quantifica- The mammalian expression construct pCDKN2WT has tion of cell cycle alterations was accomplished by cotransfec- been previously described (Arap et al., 1995) and used for tion of pCMV-CD20, with either vector alone or each of the construction of the mutant CDKN2A alleles. Site-directed CDKN2A constructs. CD20 was cotransfected at a molar mutagenesis by recombinant PCR with Pfu polymerase ratio of 1 : 4 (pCMV-CD20 : pCDKN2A) relative to test (Stratagene, La Jolla) was used to construct the CDKN2A constructs. Forty-eight hours after transfection, cells were alleles containing the point mutations (Arap et al., 1995). harvested, immunostained for CD20 antigen using a FITC- All constructs were cesium chloride puri®ed twice before conjugated anti-CD20 monoclonal antibody (Becton-Dick- use in transfection experiments. Wild-type and mutant inson, San Jose, CA), ®xed with 0.25% paraformaldehyde CDKN2A alleles were sequenced using the dideoxy method (Sigma, St Louis MO) in PBS, and permeabilized with 0.03% (USB, Cleveland, OH). saponin in PBS (Sigma, St Louis MO). Propidium iodide The wild-type and mutant CDKN2A cDNAs were fused DNA staining was then used to determine cell cycle in-frame to the GST gene in the vector pGEX-5X-3 distribution of at least 5000 CD20-positive cells as described (Pharmacia, Piscataway, NJ). The expression of each of the (Koh et al., 1995; Lukas et al., 1995). GST fusion proteins was induced with 1 mM IPTG (Stratagene, La Jolla, CA) and allowed to proceed for 6 h at 308C. GST-p16INK4a and GST-pRB fusion protein In vitro inhibition of RB phosphorylation assay puri®cation was accomplished by gluthatione anity column Sf9 insect cell lysates containing CDK4/cyclin D1 or chromatography according to the manufacturer's instruction CDK6/cyclin D1 were prepared and kinase assays (Pharmacia, Piscataway, NJ). performed as described (Ranade et al., 1995). For each reaction, 2 ml of baculoviral lysate was mixed with CDKN2A transfection and growth suppression assay increasing amounts of GST-p16INK4a fusion protein (100 ng to 1000 ng) in a total volume of 30 mlofkinase The human glioma cell lines selected for use in this study buer (20 mM Tris pH 8.0, 10 mM MgCl2,1mM EGTA) have been previously described (Arap et al., 1995). U- for 30 min at 308C. Following this pre-incubation, the 87MG and U-251MG were derived from glioblastomas and kinase reactions were started by adding 500 ng of the LN-319 from an anaplastic astrocytoma. bacterially expressed GST-Rb-SspI/End (SE, residues 768 ± Transfections were performed by the calcium phosphate 928) retinoblastoma protein (Welch et al., 1993) as a method as previously described (Arap et al., 1995), using substrate and 10 mCi of [g-32P]ATP (3000 Ci/mMol; 20 mg of either empty vector (pcDNA3) or pCDKN2A Amersham), with the mixtures were then incubated for alleles (wild-type or each of the mutant constructs) per 30 min at 308C. Subsequent processing was done essen- 100 mm dish. Transfections were terminated after 12 h and tially as described (Ranade et al., 1995). 48 h post-transfection, cells were either split at a 1 : 5 dilution and maintained for 7 ± 10 days in G418-containing (Geneticin, Gibco/BRL, Gaitherburg, MD) media to In vivo inhibition of RB phosphorylation assay generate stable transfectants or analysed by ¯ow cytome- try. Trypan blue excluding cells were counted using a Cotransfection of pRB/large pocket (LP, residues 379 ± hemocytometer. 928) (Qin et al., 1992) with vector alone or CDKN2A CDKN2A mutations in glioma WArapet al 609 constructs in the CDKN2A-null U-251 glioblastoma cells Diego, CA) and the relative ratios for the dierent was performed as described above, at a molar ratio of 1 : 15 CDKN2A alleles were obtained by setting the control (pRB/LP : pCDKN2A). Forty-eight hours after transfec- vector group at 100%. tion, cells were harvested and lysed in NET-N supplemen- ted with protease and phosphatase inhibitors as previously Acknowledgements described (Welch et al., 1993). Lysates were clari®ed by The authors thank Drs Karen E Knudsen and Suresh centrifugation and subjected to immunoprecipitation with Subramani for use of the digitizer, Shiyuan Cheng for XZ91 monoclonal antibody (Pharmingen, San Diego, CA) assistance with Sf9 cells, Nicholas C Dracopoli, Roger and protein G-sepharose chromatography (Pharmacia, Chammas, Joseph Costello and Renata Pasqualini for Piscataway, NJ). Immunoprecipitates were washed four helpful insights, and the LICR post-doctoral fellows for times with NET-N, the proteins were solubilized by boiling critical reading of this manuscript. pCMV-CD20 was a gift in SDS sample buer and resolved by 8.5% SDS±PAGE. from Dr Ed Harlow, pCMV-RB (379 ± 928) was a gift from The samples were transferrd to Immobilon-P (Millipore, Dr William Kaelin and CDK/cyclin D baculoviruses were Bedford, MA) and pRB was detected using the anti-RB gifts from Drs David Beach, Gregory Hannon, Tony polyclonal antibody 851 as described (Welch et al., 1993). Hunter, Wei Jiang, Robert S Sikorsky and Harold The ratio of ppRB/LP to pRB/LP was determined using an Varmus. WA was supported by the Brazilian National a-Innotech IS-1000 analytical digitizer (Innotech, San Research Council (CNPq).

References

Arap W et al. (1995). Cancer Res. 55, 1351 ± 1354. Louis DN, Seizinger BR and Cavenee WK. (1995). Benign Caldas C et al. (1994) Nature Genet., 8, 27 ± 32. Cerebral Gliomas. Apuzzo MLJ (ed). AANS publications: Cairns P et al. (1995). Nature Med., 11, 210 ± 212. Lebanon, NH, pp. 163 ± 180. Costello JF, Berger MS, Huang HJS and Cavenee WK. Lukas J et al. (1995). Nature, 375, 503 ± 506. (1996). Cancer Res., 56, 2405 ± 2410. Mao L et al. (1995). Cancer Res., 55, 2995 ± 2997. Duro D et al. (1995). Oncogene, 11, 21 ± 29. Merlo A et al. (1995). Nature Med., 1, 686 ± 692. FaÊ hraeus R et al. (1996). Curr. Biol., 6, 84 ± 91. Moulton T et al. (1995). Am. J. Pathol., 146, 613 ± 619. Furnari FB, Huang HJS and Cavenee WK. (1995). Cancer Nishikawa R et al. (1995). Cancer Res., 55, 1941 ± 1945. Surv., 25, 233 ± 275. Nobori T et al. (1994). Nature, 368, 753 ± 756. Gianni C and Finocchiaro G. (1994). Cancer Res., 54, 6338 ± Nowell PC. (1976). Science, 194, 23 ± 28. 6339. Olopade OI et al. (1992). Cancer Res., 52, 2523 ± 2529. Goldstein AM et al. (1995). New Engl. J. Med., 333, 970 ± Parry D and Peters G. (1996). Mol. Cell. Biol., 16, 3844 ± 974. 3852. Hannon GJ and Beach D. (1994). Nature, 371, 257 ± 261. Qin XQ, Chittenden T, Livingston DM and Kaelin WG Jr. He J et al. (1994). Cancer Res., 54, 5804 ± 5807. (1992). Genes Dev., 6, 953 ± 964. He J, Olson JJ and James CD. (1995). Cancer Res., 55, 4833 ± Quelle DE, Zindy F, Ashmun RA and Sherr CJ. (1995). Cell, 4836. 83, 993 ± 1000. Herrin DL and Schmidt GW. (1988). Biotechniques, 6, 196 ± Ranade K et al. (1995). Nature Genet., 10, 114 ± 116. 200. Schmidt EE, Ichimura K, Reifenberger G and Collins VP. James CD et al. (1993). Cancer Res., 53, 3674 ± 3676. (1994). Cancer Res., 54, 6321 ± 6324. Jen J et al. (1994). Cancer Res., 54, 6353 ± 6358. Serrano M, Hannon GJ and Beach D. (1993). Nature, 366, Kamb A et al. (1994). Science, 264, 436 ± 440. 704 ± 707. Kamb A. (1995). Trends Genet., 11, 136 ± 140. Sidransky D et al. (1992). Nature, 355, 846 ± 847. Knudson AG, Hethcote HW and Brown BW. (1975). Proc. Stone S et al. (1995). Cancer Res., 55, 2988 ± 2994. Natl. Acad. Sci. USA, 72, 5116 ± 5120. Ueki K et al. (1994). Hum. Mol. Genet., 3, 1841 ± 1845. Koh J, Enders GH, Dynlacht BD and Harlow E. (1995). van den Heuvel S and Harlow E. (1993). Science, 262, 2050 ± Nature, 375, 506 ± 510. 2054. Kyritsis AP et al. (1995). Proc. Am. Ass. Cancer Res., 36, 582. Walker DG et al. (1995). Cancer Res., 55, 20 ± 23. Kyritsis AP et al. (1996). Oncogene, 12, 63 ± 67. Welch PJ and Wang JY. (1993). Cell, 75, 779 ± 790. Li YJ et al. (1995). Oncogene, 11, 597 ± 600. Yang R, Gombart AF, Serrano M and Koeer HP. (1995). Liu Q et al. (1995). Oncogene, 10, 1061 ± 1070. Cancer Res., 55, 2503 ± 2506.