Initiation of Human Astrocytoma by Clonal Evolution of Cells with Progressive Loss of p53 Functions in a Patient with a 283H TP53 Germ-line Mutation: Evidence for a Precursor Lesion¹

The early events of human tumorigenesis are poorly understood. The relevance of the accumulation of point mutations in single tumor suppressor genes in this process has not been examined. Tumor formation is driven by clonal expansion of cells containing genetic alterations (1). Analysis of these mutations provides important clues in cancer etiology. Whereas most missense point mutations found in tumor suppressor genes result in inactive proteins, some lead to only a partial loss of function. We hypothesized that the latter might allow us to identify the importance, at different stages of tumorigenesis, of distinct functions mediated or controlled by tumor suppressors. Understanding when functions such as loss of apoptosis induction or cell cycle control may occur is important to comprehend the mechanisms that start tumor formation, to help define new stage-specific targets for therapy, and to define the appropriate available treatment at a given progression stage.

Diffusely infiltrating astrocytomas progress from low-grade (WHO grade II) lesions to grade III anaplastic astrocytoma and grade IV glioblastoma (2). No lesion preceding grade II astrocytoma has been identified to date, and the sequence of molecular alterations leading to this tumor are unknown. Although the brain does not allow for easy examination of preneoplastic lesions (such as adenomas in the colon), the presence of clinical symptoms before brain tumor detection suggests that such lesions might exist. Recent studies on animal models for gliomas have identified hyperplastic regions in the brain before tumor establishment (3, 4). We hypothesized that remnants of such precursor lesions might still exist in some low-grade tumors. The finding of a minor cell population in a grade II tumor that contained only part of the genetic changes present in the majority population of the tumor would support the existence of such lesions in humans.

The involvement of p53 mutation in clonal expansion of tumor cells was shown previously (5, 6, 7, 8, 9), and this process occurs early in astrocytomas (10, 11). The importance of p53 in tumor suppression is explained by its multiple functions, including induction of apoptosis, growth arrest, and regulation of angiogenesis (Refs. 10 and 12, 13, 14 and the references therein). p53 accomplishes most of its functions acting as a transcription factor. It transactivates the cell cycle inhibitor gene CDKN1A and proapoptotic genes, including BAX, p53AIP1, and PUMA (reviewed in Refs. 12 and 13). In most tumors, these functions are lost through deletion of one allele (LOH)⁴ and loss of function of the gene product of the other allele via a missense point mutation. However, in some tumors, more than one mutation is present in one TP53 allele. Some missense mutations only partially disrupt the functions of p53, preventing transactivation of some p53 targets (15, 16, 17, 18, 19, 20, 21).

We assumed that tumor initiation and early progression might require loss of different functions of p53. As a result, a partial loss of p53 function might predispose a cell to preneoplastic and early neoplastic development, and additional events eliminating the remaining p53 functions would be required for further progression. To examine this issue, we first analyzed the distribution of TP53 mutations in tumors with single and multiple TP53 alterations using the IARC database of somatic TP53 mutations (22). Secondly, we screened the IARC database of germ-line TP53 mutations to identify patients with multiple TP53 alterations in a single allele that would allow us to study progressive p53 inactivation in tumorigenesis.

These analyses first revealed a significant difference between the TP53 somatic mutation spectra derived from tumors with single versus multiple TP53 mutations, including a 48% reduction in frequency of “hot spot” TP53 mutations. We further adapted a p53 transcriptional assay in yeast to determine the order by which multiple p53 mutations can occur in patients. We validated this method by establishing the mutation sequence of three TP53 mutants in a unique patient with a germ-line p53 mutation (R283H) and astrocytoma progression. This genetic analysis also established the existence and distribution of genetically distinct cell populations in the primary tumor; one represents the remnant of a precursor lesion that we called “preastrocytoma.” We further established the transcriptional activity of these mutants toward the proapoptotic gene BAX and the cell cycle-arrest gene CDKN1a. We propose a molecular model explaining why the germ-line p53 mutant 283H cannot transactivate the BAX gene but can activate CDKN1a and induce growth arrest in human glioblastoma cells. Overall, these findings improve our understanding of the molecular mechanisms at the origin of human tumor initiation and progression.

The tumors reported in the database were split in three groups according to the number of TP53 mutations (single, double, and multiple). The codon distributions in each group were analyzed by pairwise comparison with the asymptotic Kolmogorov-Smirnov two-sample test (SASv8.0 software; SAS Institute, Inc.). Because the codon distributions of mutations in tumors with double and multiple mutations were not significantly different, we combined them for the next analysis. The proportions of nine TP53 mutants were calculated in single versus double/multiple mutant groups with a 95% CI (23).

Patient (41 yrs old, male) anamnesis did not reveal a family history of cancer (patient 3; Ref. 9). His brain was irradiated with 5400 cGy after removal of the first tumor, and tumor recurrence occurred 28 months later. The irradiation did not contribute to TP53 mutations because they were already present in the first tumor.

p53 transcriptional assays in yeast were performed as described previously (19, 24). The yeast strains used were yIG397 (MATa ade2-1 leu2-3,112 trp1-1 his3-11,15 can1-100 ura3-1 URA3 3×RGC::pCYC1::ADE2), YPH-p21 (MATa ura3-52 lys2-801 ade2-101 trp1-Δ63 his3-Δ200 leu2-Δ1 URA3 p21::pCYC1::ADE2), and YPH-bax (MATa ura3-52 lys2-801 ade2-101 trp1-Δ63 his3-Δ200 leu2-Δ1 URA3 bax::pCYC1::ADE2). The pLS76 yeast-expressing vector (CEN6/ARS4, LEU2) containing TP53^WT was used as a control. The cDNA coding for each of the mutants TP53^283H, TP53^267W, TP53^258D, and TP53^267W+283H was cloned by homologous recombination in pSS16. TP53 cDNA rescue from yeast and sequencing were as described previously (9). Six white and pink and 50 red colonies were sequenced/tumor. This assay underestimates the frequency of white (but not red or pink) colonies by ∼3–5% (25). This experimental error is likely because of the introduction of random point mutations in the WT p53 coding sequence during the RT-PCR step, which disrupts p53 transactivation function (24).

The SE of the frequency of each type of yeast colony (corresponding to one TP53 allele) was calculated with a normal approximation confidence for a binomial proportion (see the Fig. 2 C legend).

The heterozygosity status of locus 17p13.1 was analyzed as described previously (26) using genomic DNA extracted from patient lymphocytes and tumors.

A plasmid pc53-SNC expressing p53^wt under a CMV promoter (27) was used for site-directed mutagenesis (QuickChange site-directed mutagenesis kit; Stratagene) using the following primers: (a) 283H-f, 5′-cctgggagagaccggcacacagaggaagagaatc-3′; (b) 283H-r, 5′-gattctcttcctctgtgtgccggtctctcccagg-3′; (c) 267W-f, 5′-ggtaatctactgggatggaacagctttgaggtg-3′; (d) 267W-r, 5′-cacctcaaagctgttccatcccagtagattacc-3′; (e) 258D-f, 5′-catcatcacactggacgactccagtggtaatc-3′; and (f) 258D-r, 5′-gattaccactggagtcgtccagtgtgatgatg-3′. Mutations were verified by sequencing.

Transcriptional activity of mutant p53s was tested in glioblastoma cell line LN-Z308, which is p53-null, PTEN mutated, and wt for p14^ARF and p16 (25, 28). The cells were transiently cotransfected with each p53 expression vector, a luciferase reporter plasmid driven by p53REs from either the human CDKN1a or BAX genes, and a β-gal expression vector. The transfections were performed in 6-well dishes using 10⁵ cells/well, 5 μl of GenePorter (Gene Therapy Systems), and 1.01 μg of total DNA (10 ng of CMV-p53, 900 ng of p53RE-luciferase, and 100 ng of CMV-LacZ) for each reaction. Subsequently, cells were grown for 48 h at 37°C and lysed in 200 μl of lysis buffer (Tropix) containing 1 mm EDTA, 1 mm EGTA, and 1 mm phenylmethylsulfonyl fluoride. Cell extracts (10 μl/transfection) were tested for luciferase and β-gal activities using a dual light chemiluminescence assay (Dual Light Kit; Tropix). β-gal activity was used to verify transfection efficiency. The results were normalized to lysate protein concentrations. The levels of transfected p53 were determined by Western blot using anti-WT/mutant p53 antibody DO7 diluted 1:1000 (DAKO) and antimouse IgG horseradish peroxidase (Promega). Equal protein loading was verified by membrane staining with red Ponceau and detection of α-actin [goat polyclonal antibody I-19 (Santa Cruz Biotechnology)] with antigoat horseradish peroxidase (Roche 605275). The assays were repeated three times in triplicate.

LN-Z308 and U251MG glioma cells were transfected transiently with expression constructs for wt p53, p53^283H, or p53 ^267W+283H using GenePorter as described above. Cells were harvested 48 h after transfection, and total RNA was prepared using Trizol (Life Technologies, Inc.). The primers for CDKN1a (annealing temperature, 55°C) were 5′-GTTCCTTGTGGAGCCGGAGC-3′ (sense) and 5′-GGTACAAGACAGTGACAGGTC-3′ (antisense). Quantitative real-time RT-PCR for CDKN1a mRNA was performed using the Titan One Tube RT-PCR system (Roche Molecular Biochemicals) and SYBR Green 1 (Molecular Probes) on an iCycler (Bio-Rad). The C_t (threshold cycle) was defined as a fractional cycle when the fluorescence generated by the PCR product and SYBR Green 1 crosses a fixed threshold value in each reaction (horizontal orange line).

The individual mutations in protein and DNA were made in the p53^wt-DNA crystal structure (Protein Databank⁵ file ID, 1TSR; Ref. 29) using Sybyl mutate side chain function (Sybyl 6.5 Tripos, Inc.). A minimization was performed on each mutant to highlight regions of structural instability. Parameters for the minimization included the Tripos force field, Kollman charges, and no water. DNA nucleotide substitutions resulted in the structural instability of C8, A9, and K120 in the p53-BAX structure and A11 and T12 in the p53-CDKN1a structure. The side chain search for possible K120 positions was performed on the mutated yet unminimized structures. However, bp were manually corrected for optimal bp hydrogen bonding and stacking. Using Sybyl systematic search command, the following searches were calculated: (a) K120/R283-CDKN1a; (b) K120/R283-CDKN1a; (c) K120/R283-BAX; and (d) K120/R283-BAX; with atomic radii at 95% or 85% and an angle rotation of 5 degrees for each χ angle of K120. The number of unique conformations for each search was recorded. No difference was seen between side chain searches including electrostatics calculations and those that did not include electrostatics calculations.

LN-Z308 cells were transfected and selected with 800 μg/ml Geneticin for 10 days in six replicates. Resistant clones were counted 7 days later. The test was repeated twice.

Screening of the R3 (April 1999) version of the IARC p53 mutations database indicated that 7.9% of 7160 tumors harbored more than one mutation. Some tumors (6.1%) had two mutations, either distributed in both alleles or accumulated in a single allele (database does not discriminate). Other tumors (1.8%) had more than two mutations, suggesting that one TP53 allele underwent several mutational events. The spectra of mutations in tumors with double and multiple p53 mutations could not be distinguished (P = 0.17), but both were significantly different from the distribution of single p53 mutations (P = 0.013 and P = 0.002, respectively). This included a decreased prevalence of 48% (95% CI, 42.5–54.6%) for the nine most frequent p53 mutants found in tumors and known to completely abrogate p53 function (Ref. 20; Fig. 1). Consequently, infrequent p53 mutants are more common in tumors with multiple p53 mutations than in those with single mutation. Several studies have suggested that infrequent mutants are rare because they induce only partial inactivation of p53 functions (17, 18, 20). These findings are compatible with the hypothesis that, in alleles with double mutations, the initial mutation might have induced partial loss of function and that during tumor progression a second mutation was necessary to fully inactivate p53.

To dissect the function of sequential accumulation of TP53 mutations in tumor initiation and progression, we screened the IARC database for patients with germ-line mutations (193 patients) to identify those with multiple TP53 mutations and a clinical history of tumor progression. Only one patient suitable for our analysis was found (patient 3; Ref. 9). He was initially operated on for a low-grade astrocytoma (WHO grade II) and reoperated on 28 months later for a glioblastoma (WHO grade IV) in the same location, suggesting tumor progression. The patient carried a germ-line TP53 mutation (R283H), and the primary and recurrent tumors acquired two somatic mutations (R267W and E258D).

We first established by LOH analysis that TP53 alleles had been inactivated exclusively by point mutations and that no allelic deletions had occurred. TP53 alleles present in DNA from patient blood and tumors were analyzed by PCR using primers flanking a polymorphic site in intron 1 of the gene (26). A difference in the number of tandem repeats allowed us to distinguish the size of maternal and paternal alleles by electrophoresis. We found two fragments of equal intensity in all three samples (Fig. 2 A), suggesting that tumor formation did not involve genetic events leading to TP53 allele loss but rather resulted from monoclonal evolution of cells that sequentially accumulated two somatic point mutations.

To understand the role each mutation had played during tumor initiation and early progression, it was important to first establish the order in which they had occurred. This could have proceeded in two different ways (see Fig. 2 B). If the first mutation occurred at codon 267, then we have scenario 1; if it occurred at 258D, then we have scenario 2. Each scenario defines three genetically distinct cell types that we refer to as X (left), Y (middle), and Z (right) and defines the number of alleles contained in each cell type as 2x, 2y, and 2z. Because all three types of mutated TP53 alleles (283H, 267W+283H, and 258D) were found in the grade II astrocytoma (9), we can assume that all three cell types could be present in the primary tumor, albeit in different proportions. The total relative number of alleles in the primary tumor will be 2x + 2y + 2z% = 100%. How can we determine which mutation occurred first? It is important to notice that the number of alleles of each type (wt, 283H, 267W+283H, and 258D) will vary depending on the scenario. Indeed, in scenario 1, two cell types (X and Y) contain wt alleles (in white), whereas only X contains 283H alleles (in pink). In scenario 1, there will be x + y% wt alleles and x% 283H alleles (e.g., more white than pink). In contrast, scenario 2 would result in x% of wt alleles and x + y% of 283H alleles (e.g., more pink than white). Similar predictions can be made for the 258D and 267W+283H alleles (in red). On the basis of these calculations, we wished to establish a method that would allow us to quantify the different percentages of alleles present in the tumors. Such a method would aid in establishing the correct scenario and mutation order.

For this purpose, we adapted a p53 transcriptional assay in yeast (24). In this assay, tumor-derived TP53 mRNA is reverse transcribed, and the resulting cDNA is transformed into an ADE2-deficient yeast strain. The expression of an ectopic ADE2 gene in this yeast is controlled by a p53RE derived from the human RGC (30). Each transformed yeast cell expresses a single cDNA representing the TP53 status of one TP53 allele from the tumor. If the TP53 cDNA encodes p53^wt, the yeast cell will form a white colony on agar plates containing limiting amounts of adenine. If it encodes a mutant p53, adenine insufficiency results in a red yeast colony. The frequency of white and red colonies reflects the proportion of TP53 alleles in the analyzed tissue because the presence of a missense mutation does not alter the expression or stability of TP53 mRNA. Using this assay, we determined wt and mutant TP53 allele proportions in the microdissected primary and recurrent tumors and were able to infer the temporal occurrence of the two somatic mutations.

The assay showed three types of yeast colonies in both tumors: (a) white; (b) red; and (c) pink. Sequencing of the TP53 cDNA found in white colonies showed wt sequence, whereas pink colonies contained the 283H germ-line mutation. The pink color was interpreted as an intermediate phenotype resulting from a partial loss of the capacity to transactivate the RGC p53RE controlling the ADE2 gene (24). The red colonies harbored two different alleles, either TP53^267W+283H or TP53^258D (indicating that mutation 267W occurred in the germ-line TP53^283H allele and mutation 258D ocurred in the wt allele). The red color indicates that these two mutants have totally lost the capacity to transactivate the RGC p53RE. The relative frequencies of the colonies in the grade II astrocytoma were as follows: (a) 15 ± 1.96%, white; (b) 11 ± 1.72%, pink; and (c) 74 ± 2.41%, red (Fig. 2,C). The higher percentages of white as compared with pink colonies suggest that the first scenario is the most probable (Fig. 2,D). Astrocytomas are diffusely infiltrating tumors; therefore, this tumor probably contained some normal cells (astrocytes, vascular cells, and immune cells). This does not affect the difference in the percentages of pink and white colonies because normal cells of this patient will generate equal amounts of each. To further confirm scenario 1, we examined the relative numbers of TP53^267W+283H and TP53^258D alleles in the tumors. The first scenario predicts an excess of TP53^267W+283H [(y + z)%] over TP53^258D alleles (z%) in the primary tumor (Fig. 2,B). Because both mutants give red colonies in this assay, we sequenced TP53 cDNA extracted from 50 red colonies/tumor to derive an estimate of the relative frequency of each. In the primary tumor, the frequency of TP53^267W+283H was 40%, whereas that of TP53^258D was 34%, figures compatible with the first scenario (Fig. 2 D).

With progression to glioblastoma (WHO grade IV), an increased frequency of red colonies was observed: (a) 96 ± 0.66% were red (TP53^267W+283H or TP53^258D); (b) 1.7 ± 0.44%, were pink (TP53^283H); and (c) 2.3 ± 0.51% were white (TP53^wt; Fig. 2,C). Sequencing of the two red alleles in 50 colonies revealed equal frequencies (48%) for each (Fig. 2,C). These data are compatible with both scenarios (Fig. 2 D).

In conclusion, these data suggest that the first mutation occurred at codon 267 of the TP53^283H germ-line allele and that the TP53^wt allele was subsequently hit at codon 258.

Next, we were interested in determining the proportions of the three cell populations with differing TP53 genotypes in primary and recurrent tumors. The predominant population would have likely determined the tumor diagnosis. Consequently, its TP53 genetic status would be indicative of the number of mutations necessary to reach that stage. Minor populations in the tumor might represent remnants of earlier stages that had a competitive disadvantage during clonal evolution, or they could be newly evolved, representing the first stages of progression. It was also of interest to know how many “normal” cells (defined as cells with the constitutive TP53 genotype) might be intermingled with tumor cells in both tumors despite their microdissection.

Knowing the mutation order and the allele distribution in both tumors, establishing the cell proportions was a straightforward task. Indeed, when expressed in percentages, the relative number of each cell type is twice that of its corresponding alleles. In the primary tumor, we have 22% (2 × 11%) TP53^283H/WT cells, 8% (2 × 4%) TP53^267W+283H/WT cells, and ∼70% (2 × ∼35%) TP53^{267W+283H/258D} cells (see Fig. 3,A, grade II astrocytoma). This suggests that the major cell population (70%) in this tumor had the TP53^{267W+283H/258D} genotype (mutation of both TP53 alleles). These cells outgrew an earlier cell population with the TP53^267W+283H/WT genotype from which they had derived. Remnants of this earlier cell population could still be detected in the tumor (8% of the cells; Fig. 3 A, preastrocytoma). We decided to call this intermediate stage preastrocytoma because it preceded the first clinically detectable tumor in this patient. A substantial percentage (22%) of normal cells was present in this tumor, as might be expected for an infiltrating astrocytoma. In the recurrent glioblastoma, 96% (2 × 48%) of the cells had the TP53^{267W+283H/258D} genotype mainly because of a reduction in the presence of normal cells in the tumor (decrease from 22% to 4%) and the disappearance of the precursor cell population (from 8% to 0%). Because most cells in the primary tumor had already mutated both TP53 alleles, progression to glioblastoma likely involved alterations in other genes.

To gain an understanding of the progressive alteration in p53 function mediated by these p53 mutants during tumor initiation and progression, we expressed them in yeast and in the p53-null glioma cell line LN-Z308 (28). We examined their ability to transactivate reporter genes (ADE2 or luciferase, respectively) under the control of p53RE from the CDKN1A and BAX promoters.

The transcriptional assay in yeast indicated that p53^283H had WT transactivating ability toward the CDKN1a promoter (white colonies) but strongly reduced activity toward the BAX promoter (mostly red colonies; Fig. 4 A). p53^267W+283H and p53^258D failed to induce either promoter (red colonies; data not shown). p53^267W alone could transactivate the CDKN1A but not the BAX promoter (data not shown).

Transcriptional activity in glioblastoma cells (Fig. 4,B) showed that p53^283H transactivated the CDKN1A promoter but had a reduced activity on the BAX promoter (100% and 42% of wt activity; P = 0.0009). p53^267W+283H and p53^258D completely lost the capacity to induce CDKN1A [5% (P < 0.0001) and 1% (P < 0.0001) of wt activity, respectively] and BAX [1.2% (P < 0.0001) and 0.7% (P < 0.0001) of wt activity, respectively]. p53^267W had substantially reduced the capacity to induce transcription from the BAX promoter (23% of wt; P = 0.0016) and had a markedly reduced capacity to activate the CDKN1a promoter (22% of wt activity; P < 0.0001). The control p53^273H was inactive on both promoters. Western blot analysis indicated that differences in transcriptional activity were not because of a deficiency of exogenous p53 protein to be expressed within the cells (Fig. 4,B). Separate immunocytochemistry experiments also showed that these mutants are expressed predominantly in the nucleus (data not shown). To confirm that transcriptional activation also occurred on the endogenous CDKN1A promoter, we transfected glioma cells with the different p53 cDNA expression vectors and measured the activation of the endogenous CDKN1A gene by quantitative real-time RT-PCR (Fig. 4 C). The C_ts for amplification were identical for p53^WT and p53^283H, indicating equal ability to activate transcription. The C_ts for p53^267W+283H and mock-transfected cells were identical, confirming that the second mutation abrogated p53 transactivation ability.

Taken together, these results suggest that p53^283H had a strongly reduced ability to induce transcription from the BAX promoter but maintained transcription of the CDKN1A promoter. Accumulation of a second mutation at R267W abrogated this activity. p53^258D was functionally similar to hot spot mutant p53^273H because it eliminated p53 transactivation on both promoters.

The above data suggest that the various mutants identified differ in their ability to bind specific p53REs. To examine the structural basis for these differences, we used computer visualization softwares to perform three-dimensional modeling of the p53-DNA interactions based on the published crystal structure of the p53^wt-DNA complex (Protein Databank⁵ file ID, 1TSR; Ref. 29). Localization within the p53 tertiary structure of the three mutated amino acids revealed that only R283 is located at the protein-DNA interface (Fig. 5A). Therefore, substitution R-H at position 283 could directly modify p53 binding to DNA. This residue has two important roles in efficient and specific binding to DNA. First, it stabilizes the p53-DNA complex through non-sequence-specific interactions with the phosphatidic backbone of DNA (green arrow, Fig. 5,B). Second, it influences sequence-specific binding of p53 to the DNA through its interaction with K120 (red arrow, Fig. 5,B). K120 is one of the three p53 residues involved in sequence-specific interactions with the nitrogen bases of DNA (yellow arrow; Ref. 29). The 4.25 Å distance between K120 and R283 creates a Van der Waals interaction that stabilizes K120 in the correct orientation for sequence-specific interaction with DNA. Substitution R283H allows incorrect conformations of K120, which in turn loses the direct protein-to-base interactions with DNA (Fig. 5,C). The model predicts that the p53RE of the BAX promoter (and likely other promoter with a p53RE element with 5′ Py substitutions) will bind p53 with a lower affinity than the p53RE of the CDKN1A promoter. The nucleotide interacting with K120 is cytosine in the BAX p53RE (second nucleotide 5′ of the p53RE), whereas it is adenine in the CDKN1A p53RE. The substitution of a Pu with a Py at this site reduces the length of the hydrogen bond with K120 from 2.3 to 1.68 Å, resulting in Van der Waals overlap (Fig. 5,D, compare cytosine in green with adenine in white). The interaction of K120 with the BAX p53RE is unfavored further by the presence of a dThd instead of a guanine in the first 5′ nucleotide. Indeed, the methyl group of the dThd is very close to the β-carbon of K120 (3.27 Å instead of 6.08 Å), resulting again in Van der Waals overlap (Fig. 5 D, compare dThd in green with guanine in white). This predicts that the combination of a p53 R283H mutation and PuPu-PyPy substitution at the 5′ end of the p53RE will induce K120 to preferentially assume a conformation that prevents its binding to DNA.

Amino acids E258 and R267 do not interact with DNA (Fig. 5,A). Modeling of the p53 crystal structure indicated that substitutions E258D and R267W induce strong stability changes of the tertiary structure of p53 (Fig. 5, E and F). These changes are likely to globally impair p53-DNA binding irrespective of sequence specificity. This interpretation is compatible with our observation of loss of transactivation for both mutants on the CDKN1A, BAX, and RGC promoters. Furthermore, the predicted physical distance between amino acids 267 and 283 does not support the existence of any trivial cooperation between mutations 283H and 267W that would lead to synergistic inactivation of p53 in the double mutant. The p53 molecule with R267W and R283H mutations is likely to adopt a grossly misfolded conformation, leading to reduced or abrogated DNA binding.

p53 acts as a suppressor of neoplastic growth by inducing cell growth arrest and/or apoptosis. To functionally characterize the in vitro growth-inhibitory properties of the mutant proteins found in the patient, we have tested their capacity to inhibit clonal cell growth in monolayer cultures using a colony formation assay. LN-Z308 human glioblastoma cells (p53-null) were chosen because expression of wt p53 in these cells is known to induce growth arrest but not apoptosis (31, 32). Cells were transfected with expression vectors for each of the p53 mutants and neomycin resistance (p53^wt, p53^273H, and the empty vector served as controls). Fig. 6 shows that p53^283H inhibits clonal cell growth of LN-Z308 cells similarly to p53^wt. However, the presence of both mutations (267W+283H) in the same p53 molecule abrogated this function and yielded the same amount of colonies as p53^258D, p53^273H, or the control vector.

A significant number of colonies (10%) was formed after transfection of the TP53^wt and TP53^283H cDNA constructs. It has been previously found for TP53^wt cDNA transfectants that these colonies lose or rearrange exogenous TP53 sequences (27). To evaluate this issue in the case of TP53^283H, we isolated these independent colonies, expanded them into viable clones, and examined p53 expression by Western blot and immunocytochemistry. As controls, we also expanded colonies from p53^258D, p53^267W+283H, p53^267W, p53^273H, and p53^wt. Most of the colonies (95%) obtained with p53^wt or p53^283H were not viable because they could not be expanded, and those few (5%) that did expand failed to express p53 (data not shown).

The initiation steps of human gliomagenesis are poorly understood. The timing and sequence of the progressive loss of single tumor suppressor functions by accumulation of point mutations have not yet been investigated in this process.

We found a general decrease in the frequency of hot spot TP53 mutations in human tumors containing multiple TP53 mutations and, consequently, an increase of rare TP53 mutants in this group. The presence of multiple TP53 mutations in tumors might be due in part to mismatch repair defects leading to alleles with multiple mutations, to artifacts of PCR amplification, and/or to sample contaminations. The latter two would favor the presence of hot spot mutations because these are a more frequent source for sample contamination, and polymerase mutation frequency in vitro is higher on these codons (33). Although we recognize the inherent limits of the p53 database, the finding that tumors with multiple mutations have more infrequent mutations is compatible with the idea that progressive loss of p53 functions may have occurred in these tumors. Infrequent mutants can lead to only partial loss of p53 function (17, 18, 19, 20).

Tumor initiation and progression might require abrogation of several p53 functions that could be lost at once by a “hot spot” mutation or progressively within the same tumor through multiple partial loss of function mutations. To test this hypothesis, we established the chronological order by which three TP53 mutations occurred during human astrocytoma development and progression in a germ-line TP53^283H patient. We showed that the primary tumor contained three cell populations differing in their TP53 genotype: (a) normal cells (TP53^283H/WT; 22% of cells); (b) grade II tumor cells (TP53^{267W+283H/258D}; 70% of cells); and (c) cells with an intermediate genotype (TP53^{267W +283H/WT}; 8% of cells) from which the tumor cells derived (Fig. 3,A). This provides the first evidence for monoclonal evolution in a WHO grade II astrocytoma and suggests that this tumor derived from a precursor lesion (called preastrocytoma in Fig. 3 A) that was not detected clinically. Complete inactivation of a single TP53 allele by the germ-line and somatic 267 mutation (and potentially other unknown events) is sufficient to create this lesion. Additional studies will have to establish whether this is a preneoplastic or early neoplastic lesion in the brain, analogous to the adenoma stages found in colonic transformation. Tumor progression to glioblastoma likely involved other genetic events typical of malignant astrocytoma (2). The increase in TP53^{267W+283H/258D} cells (from 70% to 96%) that accompanied progression appears due to the loss of normal and preastrocytoma cells, which are likely at a competitive disadvantage as compared with more transformed cells. Most astrocytomas lose p53 function as the result of point mutation in one allele and loss of the second allele (11). The genetics of the present case suggest that subtle alterations in DNA repair enzymes and/or DNA damage-sensing and -signaling proteins may have contributed to the accumulation of multiple mutations in two alleles. We did not examine whether a mismatch repair deficiency contributed to the accumulation of TP53 mutations in these tumors, but this is rare in brain tumors and is usually associated with Turcot syndrome (34, 35). Previous investigations have shown that acquisition of a tumor suppressor gene mutation in a low-grade tumor can lead to tumor progression (5, 6, 7, 8, 9). Our study now demonstrates that a primary tumor can contain several clones of cells with different levels of p53 inactivation because of the sequential accumulation of TP53 mutations. We also demonstrate for the first time the order by which these multiple mutation events occurred during tumor initiation. The cells with both p53 alleles inactivated composed 70% of the primary tumor, suggesting that these cells had an early selective advantage during the initiation of this tumor rather than during its progression. This is in agreement with our prior finding that, on average, 70% of the cells in primary astrocytoma already contain TP53 mutations based on a series of 14 grade II gliomas with TP53 mutations and recurrence (9). Malignant tumor progression most likely involved additional genetic events.

To understand the function of these p53 mutants in tumor initiation and progression, we characterized their transcriptional activity and ability to suppress colony formation. The single mutant p53^283H had wt transcriptional activity toward the CDKN1a promoter but could not induce transcription of the BAX promoter in yeast and human cells. p53^258D and p53^267W+283H had lost the ability to transactivate both promoters. These results confirm in human brain tissue that a subset of rare p53 mutants retains transcriptional activity on some p53 target genes (15, 16, 17, 18, 19, 20, 21). To analyze the structural basis for this selective target gene activation, we modeled the impact of these mutants on the p53-DNA structure using published crystallographic data (29). Previous works have shown that different p53 mutants have distinct effects on the wt p53 protein structure, leading to variable functional effects (36). Our analysis suggests that R283 might stabilize K120 in a configuration that enables it to specifically bind to the second nucleotide (Pu) of p53RE. The mutation R283H would allow K120 to assume other configurations that prevent its binding to DNA. The substitution of two Pu with two Py at the 5′ end of the BAX p53RE creates a Van der Waals overlap with K120. This overlap, combined with the mutation R283H, is predicted to induce K120 to preferentially assume a configuration that prevents binding to the BAX p53RE. Previous work had suggested that K120 established sequence-specific DNA interactions because a K-R substitution prevented BAX but not CDKN1A gene transactivation (19, 20). Here, we provide a putative structural explanation for this observation by showing that K120 orientation is critical to maintain binding to 5′ nucleotides in the BAX p53RE, and we further suggest that it might require stabilization by its structurally adjacent 283 amino acid. Although not specifically tested in this study, our data predict that p53^283H might also have reduced the ability to transactivate PUMA and p53AIP1, two other major p53-induced mediators of apoptosis (37, 38), because the p53REs of both genes have 5′ Pys. Clearly, additional experiments will be needed to validate this model of how the p53-p53RE interaction is structurally altered upon p53 mutation and depending on target p53REs.

In contrast, the p53-DNA structure shows that R267 and E258 do not interact with the DNA. Their mutation more likely leads to modification of the p53 tertiary structure, a global effect that is expected to abolish DNA binding independently of DNA sequence. Indeed, we found that these mutants abolish transcription from the BAX, CDKN1A, and RGC promoters.

To verify that the ability of p53^283H to transactivate CDKN1A has functional relevance, we examined its capacity to inhibit colony formation from transfected human glioma cells. p53^283H inhibited colony formation, consistent with its ability to activate the p21 growth arrest pathway. In contrast, p53^258D and p53^267W+283H had totally lost this capacity, leading to the formation of a number of clones similar to the ones obtained with empty vector or p53^273H.

Taken together, these data are compatible with the model illustrated in Fig. 3, B–D. The germ-line mutation R283H in one TP53 allele may result in a reduction of p53-dependent apoptosis because we show that this mutant does not efficiently transactivate BAX. This partial loss of function phenotype, initiated by a 50% reduction in TP53^wt gene dosage, is theoretically expected to reduce p53^wt tetramers by ∼94% in the cells, if one assumes equal transcription, translation, stability, and ability to form tetramers for the mutant and wt p53 proteins. Increased stability of p53 mutants, as is frequently observed, might reduce even further the ratio of p53^WT tetramers in the cells. The activity of the 87.5% p53^WT/mutant heteromers is unknown and will likely vary according to the ratio of wt and mutant p53 molecules/tetramer (Fig. 3,C). Because clonal expansion of cells with two additional p53 mutations occurred during the tumorigenic process, it is likely that there was a selective advantage of eliminating residual wt activity in these heteromers. p53 is known to induce apoptosis in cells sustaining DNA damage, thus a decreased overall proapoptotic activity of p53 tetramers in cells containing damaged DNA might increase their viability and survival. Acquisition of mutation 267W on the TP53^283H allele and subsequent acquisition of mutation 258D on the remaining TP53^wt allele progressively disrupted the capacity of p53 tetramers to induce growth arrest through the p21 pathway. This would have conferred an increasing proliferative advantage and led the cells to clonal expansion. It is likely that the increased genetic instability associated with p53 loss leads to the accumulation of additional genetic defects and ultimately malignant progression (Fig. 3 B).

Although this model hinges on data from a single patient, it allows us to make several interesting predictions. First, the early stages of tumorigenesis and progression can occur through progressive decreases in wt p53 activity. p53 activity is subject to multiple pathways of regulation, and the importance of progressive loss of p53 functions in tumor formation is emerging (16). Second, a decrease in expression of p53-regulated apoptosis mediators (such as BAX, PUMA, and/or p53AIP1) can precede loss of p53 cell cycle-regulatory function (p21) and promote early tumor formation. BAX has been demonstrated previously to be involved in tumor initiation (39, 40) but not in tumor progression because for the latter the cells need to acquire a proliferative advantage (41). Understanding that loss of apoptosis induction might precede loss of cell cycle control during tumorigenesis in a subset of patients has significance beyond the mechanisms that start tumor formation. It could also help define new stage-specific targets for the development of novel treatments and define the most appropriate currently available therapy. Clinical therapies that induce DNA damage and cell death by p53-dependent apoptotic response might not be appropriate for such patients. Third, clonal analysis of tumor evolution will likely uncover subtle steps that have not been described in the current histological classification of tumors such as the preastrocytoma lesion we found within the grade II astrocytoma of this patient. Finally, studying tumor suppressor genes with multiple mutations in single tumors could be particularly informative to understand the progressive loss of function of their products during tumorigenesis. These predictions need to be further validated experimentally when other such cases become available, and it remains to be established whether they will be generally applicable to tumors losing p53 function by a combination of mutation in one allele and loss of the other.

We thank C. Sapienza, S. Pittard, M. F. Hansen, D. Brat, D. Post, and P. Vertino for advice or reading the manuscript, S. N. Devi for technical assistance, M. Olivier for providing us the IARC p53 database, R. Iggo for the p53 yeast assay kit, and N. de Tribolet and M. F. Hamou for patient samples. The statistical analyses on the IARC p53 database were performed as a paid service by H. Chakaraborty of the Emory Biostatistics Core Facility. DNA sequencing and primer synthesis were performed at the Microchemical and Sequencing Core Facilities.