TP53 Mutations in Human Cancer: Database Reassessment and Prospects for the Next Decade

Introduction

“TP53 (MIM #191170) is the most frequently mutated gene in human cancer.” This sentence, found in the introductions of thousands of publications, can be traced back to 1990, 1 year after the description of the first TP53 mutations in lung and colorectal carcinoma [Baker et al., 1989; Takahashi et al., 1989]. Hundreds of novel cancer genes were subsequently identified, but none were capable of stealing this infamous crown from TP53. Even whole-genome sequencing of several thousand cancer genomes has not revealed any new suitors for the throne and to this day TP53 mutations are still frequently found among the five most significantly mutated genes in the most common human cancers [Stratton, 2011; Kandoth et al., 2013].

Nonetheless, the frequency of TP53 mutations is highly variable depending on the type of cancer (Fig. 1). Although nonsynonymous single-nucleotide variants (nsSNVs) are the most common TP53 alterations, it is worth noting that osteosarcoma is one of the few cancers that display a high frequency of TP53 gene deletion, an observation made in 1987 and confirmed by more recent whole-genome analysis, although the basis of this observation remains unknown [Masuda et al., 1987; Barretina et al., 2010]. The frequency of TP53 alterations appears to range from less than 5% in cervical carcinoma to 90% in ovarian carcinoma (Fig. 1), but these numbers must be taken with caution due to several factors.

First, most cancer types are heterogeneous entities comprising several subtypes [Ogino et al., 2012]. Diversities in cancer were initially defined via histological criteria and then confirmed by multiple molecular analyses that linked them to specific landscapes of genetic alteration. For example, two types of lung cancer have been identified: small cell lung cancer (SCLC) and non-SCLC (NSCLC). This latter is in turn divided into three main subtypes: adenocarcinoma (Adc), squamous cell carcinoma (SCC), and large cell carcinoma (LC) [Cooper et al., 2013]. The mutational profiles of the subtypes are different [CLCG Project, 2013]. KRAS mutation is frequent in Adc, but less so in the other subtypes (Supp. Fig. S1A). Similarly, the frequency of TP53 mutation varies across the types and subtypes, ranging from 50% in Adc and LC to 80% in SCLC and SCC (Supp. Fig. S1A). Mutational events are also distinct as the association of each subtype with smoking history is heterogeneous. Breast carcinoma too is a heterogeneous disease with multiple subtypes. Recent molecular profiling studies have identified four major subtypes: luminal A, luminal B, basal like, and HER2 [Weigelt et al., 2010]. The frequency of TP53 alteration in these subtypes ranges from 12% in luminal A to more than 80% in basal like (Supp. Fig. S1B) [Curtis et al., 2012]. Furthermore, for unknown reasons, basal-like tumors display a high frequency of TP53 nonsense and frameshift mutations compared with other breast cancer subtypes [Dumay et al., 2013]. In colon cancer, TP53 mutations occur less frequently in tumors with deficiencies in mismatch repair genes associated with high-microsatellite instability [Pugh et al., 2012].

A second factor influencing the frequency of TP53 mutations is the stage of development of the tumor. In prostate cancer, the frequency of TP53 mutations is low in primary tumors (10%–20%) but can reach 50% in metastatic tumors [Schlomm et al., 2008]. In leukemia and lymphoma, TP53 mutations are found more frequently at relapse, during blastic phase (CML) or in acute transformation (MDS) [Xu-Monette et al., 2012] and Malcikova et al. (2014) in this special issue. Further longitudinal analyses will be needed to assess the chronology of TP53 mutation in various types of cancer and to understand how other genetic and/or epigenetic modifications cooperate with these mutations.

The third factor that strongly modulates TP53 mutation frequency is exogenous features such as viral or bacterial infection [Levine, 2009; Moody and Laimins, 2010; Schetter and Harris, 2012]. Most human viruses impair TP53 activity, thus constraining the cell to proficient viral DNA replication, a mechanism that was, on another note, at the root of the discovery of TP53 [Kress et al., 1979; Lane and Crawford, 1979; Linzer and Levine, 1979]. In cervical cancer, the human papillomavirus E6 protein targets the TP53 protein to the proteasomal degradation pathway [Scheffner et al., 1990]. This interaction is the most significant association between a viral protein and its cellular target. In liver cancer associated with hepatitis B virus (HBV) infection, this interplay between viral proteins and TP53 is less clear as many HBV-positive hepatocellular carcinomas display TP53 mutation [Hussain et al., 2007]. Furthermore, recent evidence suggests that bacterial infection may trigger the TP53 pathway and activate specific TP53 isoforms [Terrier et al., 2013].

Finally, the highly heterogeneous geographical distribution of carcinogen exposure shapes TP53 mutation both qualitatively and quantitatively. Thus, patient cohorts originating from different areas of the globe will have different landscapes of mutation, in TP53 and other regions of the genome [Ozturk, 1991; Hartmann et al., 1997].

The TP53 Network: A Target in Human Cancer

The TP53 protein acts as a central hub that receives, integrates, and transmits multiple signals, generated during various stress events, to ensure cell and tissue homeostasis. Although the most important activity of TP53 is to act as a direct transcription activator for several hundred genes, it is also able to act as a transcription repressor. Moreover, it has several transcription-independent activities that make the investigation of this protein very complex. Discussing all the aspects of the various signaling pathways regulated by TP53 is beyond the scope of this article. However, several recent reviews are available [Aylon and Oren, 2011; Levine et al., 2011; Berkers et al., 2013].

Although gene mutation is the major determinant of TP53 inactivation, it is indisputable that the inactivation of genes closely associated with the TP53 networks could substitute for inactivation of TP53 itself. There should be significant heterogeneity among these related entities for several reasons. First, many of these pathways have significant tissue specificity, as illustrated by the specific requirements of particular types of cancer to target a specific set of driver mutations. The observation of gene deletions in sarcoma, and not TP53 mutations per se, is one of the many arguments that comfort tissue specificity heterogeneity [Masuda et al., 1987; Barretina et al., 2010]. Second, TP53 is associated with multiple networks that control the cellular response to several types of stress, and each one can be the driving force selected to impair TP53 at different steps of neoplastic transformation. TP53 has been shown to be an important player in angiogenesis, DNA damage response, oncogene activation, and aneuploidy propagation prevention to name just a few of its many roles [Vousden and Lane, 2007]. Whether the same set of TP53 activities and partners must be targeted for each function of TP53 is currently unknown. Finally, it is obvious that as a partner becomes more distant from TP53 in the network, its association with TP53 becomes weaker and the effect of its inactivation increasingly independent of TP53.

Via the sequencing of several thousand cancer genomes and the systematic identification of significant mutated genes in different types of cancer, the analysis of mutual exclusivity (or association) of different genomic alterations is possible [Ciriello et al., 2012]. Mutually exclusive genomic events identify genes that are active in common biological pathways where alteration of both genes would not add a selective advantage compared with alteration in one. Furthermore, these mutually exclusive events provide strong genetic evidence that the altered genes are functionally linked in a common biological pathway.

Figure 2A presents the most obvious branches of the TP53 network that are impaired in human cancer. The MDM2 and MDM4 proteins are the prevailing targets since they act as direct negative regulators of TP53, downregulating its transcriptional activity (both MDM2 and MDM4) or targeting it to the ubiquitin–proteasome protein degradation pathway (MDM2 only) [Wade et al., 2013]. The MDM2 gene is amplified in several tumors including breast carcinoma, glioblastoma, and sarcoma, whereas MDM4 is amplified in melanoma, retinoblastoma, and breast carcinoma (Fig. 2B) [Oliner et al., 1993; Laurie et al., 2006; Wade et al., 2013]. Also, an abnormal accumulation of the mdm2 protein can occur in the absence of gene amplification, suggesting multiple mechanisms of deregulation and thus warranting different diagnostic procedures. In sarcoma, glioblastoma, or lung Adc, the distribution of TP53, MDM2, and MDM4 alterations is highly exclusive, confirming that they target the same pathway in these types of cancer (Fig. 3A–C). We currently do not know whether the phenotype resulting from downregulation of TP53 via the oncogenic activities of MDM2 in these tumors is the same as that resulting from an oncogenic mutant TP53. The CDNK2A gene encodes the 19/ARF protein that negatively regulates the interaction between TP53 and MDM2 by targeting the latter in the nucleolus upon an oncogenic stress (Fig. 2A) [Li et al., 2011]. Deletion or methylation of this gene is frequently observed in human cancer, particularly in tumors with wild-type TP53 (Fig. 3C). In breast carcinoma, TP53, PIK3CA, GATA3, and FOXA1 mutations appear almost mutually exclusively relative to one another (Figs. 2B and 3D). GATA3 mutations are more frequent in luminal-A type cancers presenting a low frequency of TP53 mutations, suggesting that both genes act on a similar pathway, at least for this specific subtype [Curtis et al., 2012]. The GATA3 transcription factor is an essential component for the differentiation of mammary stem/progenitor cells into luminal cells, but its relationship with the TP53 pathway is still unclear [Zheng and Blobel, 2010]. The inverse correlation between TP53 mutations and PIK3CA is also specific to breast cancer, as it is not observed in lung or colorectal carcinoma (Fig. 3D).

The TP53 network is also under the direct or indirect control of a number of miRNAs having either positive or negative effects on TP53 (Fig. 2A) [He et al., 2007; Wrighton, 2009]. Direct negative regulation can occur via the binding of specific miRNAs to the 3′UTR of the TP53 mRNA (miR-125b, miR-504, and miR33 among others). Conversely, some miRNAs have a net positive effect because they target MDM2 or other negative regulators of TP53. It would be temptingly simple to state that either the upregulation of miRNAs associated with a negative effect or the downregulation of those associated with a positive effect will impair the TP53 pathway. However, such a picture would be oversimplified and would not take into account the fact that most miRNAs have multiple targets and that the deregulation of any one of them may lead to a multitude of effects. In addition to TP53, mir-125b also directly represses other target genes including BBC3 (PUMA), CCNC, CDC25A, or BAK1 [Hermeking, 2012]. Indeed, knowledge is lacking as to whether or not alterations in miRNA networks upstream or downstream of the TP53 pathway can substitute for TP53 mutations and this issue will be complex to assess, as several parameters (such as tissue specificity) are tightly associated with miRNA expression.

Trends in TP53 Reports

Before the release of cancer genome high-throughput sequencing studies, mutation reports were highly stable because modifications to published data were very infrequent. Despite the obvious errors due to sequencing artifacts, incorrect data analysis or badly managed typography, authors almost never changed or retracted their works, and many dubious mutation reports still pollute the scientific literature. In contrast, mutations obtained via NGS are directly stored in large databases after being processed. Raw data from tumor DNA sequencing must go through a pipeline of different algorithms that successively remove background noise associated with the methodology, then neutral variants stored in the dbSNP at the NCI (http://www.ncbi.nlm.nih.gov/SNP/) before finally labeling a modification as a specific tumoral variant [Chin et al., 2011]. Thereafter, the greatest challenge is distinguishing driver mutations from random passenger mutations. The former confer a growth advantage and, by definition, reside in the subset of genes known as cancer genes, whereas the latter are distributed randomly in the genome at positions that do not lead to phenotypic change.

While updating the most recent release of the TP53 mutation database, we noticed many fluctuations in the various datasets of TP53 mutations associated with specific studies. For example, TP53 data from the TCGA study on ovarian cancer have been modified several times since their publication in 2010 [Bell et al., 2011]. Some modifications were due to improvements in the pipeline used to identify significant mutated genes and the use of novel algorithms to predict mutational effects on protein structures and functions [Lawrence et al., 2013]. The significance of other modifications was more obscure. For example, we noticed that all synonymous single-nucleotide variants (sSNVs) had been removed from the cbioportal database (http://www.cbioportal.org/public-portal/) but not from others. Similar observations were made for other sets of data. Although sSNVs are often associated with passenger mutations, they can also have drastic effects on splicing or RNA stability. In the case of TP53, several sSNVs have been clearly identified as detrimental for TP53 splicing (see the section Silent Mutations).

This fluctuation and heterogeneity in the various data sets will make it difficult for curators to carefully update mutation databases and be among the many reasons for the disappearance of locus-specific databases in their current format [Soussi, 2014].

The trend of TP53 mutation description since 1989 is shown in Figure 4 (see Soussi in this issue for a more in-depth discussion on mutation evolution [Soussi, 2014]). This graph, updated from its original publication in 2011, shows a renewal of trends due to the release of more than 4,000 mutations by the various tumor genome sequencing projects and the widespread use of NGS platforms [Soussi, 2011a]. Before this rebound, the numbers for publications and mutations published each year had culminated at about 200 and 2,400, respectively, before beginning to decline at about the turn of the century and through to 2009 (Fig. 4A and B, blue column). As discussed in the accompanying article, this was largely due to space considerations in scientific journals [Soussi, 2014]. After the nadir of 2009, the results of large studies using NGS started becoming available and led to a powerful rebound of new data in the early 2010s (Fig. 4A and B, red column). Since the TP53 gene is the most frequently mutated gene in human cancer, we can expect that the large sequencing projects still underway will continue to generate great amounts of TP53 mutation data.

The current issue of the TP53 database contains 45,000 mutations with 1,540 variants issued from single-nucleotide substitutions and 2,000 frameshift mutants issued from small insertions or deletions. The trend in the number of variants being published for the first time, TP53 variant novelty, is shown in Figure 4C and D. During the first years, this trend was high because the pool of potential inactive TP53 variants was large. Variant novelty peaked in the years 1993-95 then began decreasing regularly as studies describing TP53 variants became increasingly redundant (Fig. 4C).

It is interesting to note that this trend is highly specific for SNVs as the rate of frameshift mutation novelties has remained stable over the past 20 years and is furthermore similar across conventional Sanger sequencing and NGS studies (Fig. 4C and D). The novel TP53 variants detected between 2010 and 2013 are mostly localized in exons 2, 3, 4, and 10, which were not commonly analyzed in previous studies.

TP53 SNV novelty is reaching a plateau, suggesting that we may now have the full set of variants selected for a loss of tumor-suppresser function (Fig. 4D). With this set of variants in hand, it is now possible to paint an accurate landscape of the TP53 residues that are essential for the tumor-suppressive effect of TP53.

Curation of the TP53 Mutation Database

There is a marked difference in the frequency of the 45,000 somatic mutations included in the current updated issue of the database (http://p53.fr). Although the database is composed mainly of true driver mutations, it also contains a significant number of nondriver mutations; indeed, their identification is essential for providing an accurate database. Several algorithms have been developed to infer the pathogenicity of nsSNVs in human diseases. Some of them, such as Sift or Mut Assessor, can be used for any gene, whereas other are more specific to TP53 and rely on specific features of the protein [Ng and Henikoff, 2006; Reva et al., 2011]. It is beyond the scope of this article to review these procedures, but briefly their sensitivity does not exceed 75%–80% with a bias toward hot-spot mutations. Furthermore, they can have a high false-positive detection rate. For example, the passenger mutation, p.R175C, is identified as highly damaging by most algorithms because it is localized in a highly conserved residue.

For more than 10 years, we have been concerned by the quality of the data included in the TP53 database [Soussi and Beroud, 2001, 2003; Soussi et al., 2005, 2006a, 2006b; Edlund et al., 2012]. To circumvent obviously biased manual curation, we have developed and refined several statistical procedures to provide a highly curated set of TP53 mutations (Supp. Fig. S2).

Mutant diversity in the database can be apprehended from two angles: TP53 variants can be considered irrespectively of the context of the mother publication, or they can be investigated in relation to the other variants described in the same publication.

Heterogeneity of TP53 Variants

The most uncontested and solid information for TP53 variants is their frequency in the database. Five missense variants are found more than 1,000 times, 61 between 100 and 1,000 times, 445 between 10 and 100 times, and 1,147 less than 10 times (Fig. 5). Frequent mutants are undeniably driver mutations selected during the progression of neoplasia. Functional analyses that include both in vitro and in vivo analyses have shown that all TP53 hot-spot mutations displayed a clear loss of activity (Fig. 6A–C) [Ory et al., 1994; Kato et al., 2003].

The situation is less clear for mutants appearing less frequently, and establishing the border between significant and nonsignificant mutants is not an easy task. Infrequent mutants, and more particularly those described only once or twice (630 variants, 2% of the database), may be rare driver mutations with low penetrance, passenger mutations, or sequencing artifacts.

The transcriptional activity of more than 2,000 mutants was quantified in a yeast system using various response elements found in common TP53 target genes such as CDKN1A, BAX, or MDM2 [Kato et al., 2003]. A global analysis of all TP53 mutants found in human cancer indicated that loss of activity is very heterogeneous, ranging from total inactivation to activity greater than that of the wild-type protein (Fig. 6A and B). TP53 mutants can be categorized according to their origin, tumor, cell line, or germline. For the latter two, the detection of TP53 mutations is less prone to sequencing artifacts, as the DNA is of better quality and the mutations are either heterozygote or homozygote, which eases analysis. Another advantage in cell lines is that TP53 status benefits from accruing verification over multiple studies and this redundancy results in a good panel of certified TP53 mutants. For germline mutations, sequencing is performed in clinical genetics laboratories with stringent quality controls that are not mandatory in research laboratories. The pattern of TP53 germline mutations is similar to somatic mutations with a high frequency of transition at CpG dinucleotide (see the review of Kamihara et al. [2014] in this issue). For tumors, sample origin is highly variable, ranging from frozen cells or tissue to formalin-fixed paraffin-embedded (FFPE) specimens obtained after microdissection. These various features explain the differences in the distribution of TP53 mutant activity: as shown in Figure 6A, the great majority of mutant TP53 reported in cell lines or in constitutional DNA are inactive, whereas those observed in tumors display a larger heterogeneity. The updated TP53 database now includes 6,500 mutations (250 variants) obtained via NGS of DNA extracted from frozen tissue. Loss-of-function in this specific set of mutants displays a distribution similar to that observed for cell lines or germ cells, a good indicator of the high quality of these NGS studies (Fig. 6A).

Combining a functional analysis with a determination of the frequency of each mutant in the entire database leads to the striking observation of a clear inverse correlation between the frequency of TP53 mutants and their activity: frequent TP53 mutants are always inactive, whereas approximately half of the mutants reported only once have activity greater than 50% compared with wild-type TP53, further underlining the very limited importance of these mutations during transformation (Fig. 6B).

The Signature of the Mutational Process in Human Cancer: The TP53 Contribution

For many years, the molecular epidemiology of human cancer was restricted to a few genes. Early work on the ha-ras gene showed that mutations had tumor-specific spectrums [Rodenhuis et al., 1988]. Unfortunately, missense mutations that activate the oncogenic activity of ha-ras are restricted to just a few codons (12, 13, and, to a lesser extent, 61 and 146), consequently limiting the impact of studies. In contrast, analyses of TP53 mutations have been successfully used to establish links between exposure to carcinogens and various types of cancer (Supp. Fig. S3). Many reviews detailing these various findings are available [Brash, 1997; Giglia-Mari and Sarasin, 2003; Staib et al., 2003; Toyooka et al., 2003; Slade et al., 2009; Boffetta, 2010; Soussi, 2011b; Schetter and Harris, 2012].

In vitro and animal studies performed during the twentieth century identified specific mutational signatures associated with various mutagens [Soussi, 2011b]. Numerous sequencing studies performed on the TP53 gene made it possible to identify these signatures in human cancers associated with specific exposures. The most remarkable example is that of tandem mutations, which are only observed in skin cancers and specifically induced by ultraviolet radiation (Supp. Fig. S3). The relationship between G->T transversion and lung cancer in smokers is also very noteworthy as is the mutation of codon 249 observed in aflatoxin B1-induced liver cancers (Supp. Fig. S3). The studies leading to these discoveries were possible because TP53 was the only gene that united several specific features used to study the origin of carcinogenesis in humans: (1) mutation in many types of cancer; (2) high mutation frequency; (3) predominantly modified by point mutations; (4) small enough to be relatively easy to analyze; (5) large number of codons that can be targeted by mutations (more than 200); and (6) multiple types of substitution can be identified in a single codon.

With the recent progress in sequencing methodologies, it is now possible to extend this type of analysis to the entire genome, a process that not only confirms the information derived from TP53 studies, but also demonstrates that all types of alterations, passenger and driver, can be used to track mutagen fingerprints (Fig. 7) [Stratton et al., 2009]. Therefore, various biases that could interfere with TP53 studies, such as the low frequency of TP53 mutations in some cancers or the small size of the gene, can now be overcome. The use of passenger mutations for spectrum analysis offers the following advantages: (1) they can be found in large numbers in a single tumor (current analyses indicate that several tens of thousands of passenger mutations can be found per tumor genome), thus enabling single tumor profiling; (2) the large size of the human genome makes the analysis independent of the target size or sequence, allowing a detailed analysis of sequence context that was impossible in small genes; (3) due to their origin, they are true neutral mutations preventing any mutational profile bias that could occur during selection; (4) the lack of selective pressure allows an analysis of each type of cancer, avoiding the problem of specific gene inactivation in particular cancers [Alexandrov et al., 2013a, 2013b; Kandoth et al., 2013; Lawrence et al., 2013].

In lung cancer, the patterns of mutational events are very similar and show a striking similarity to the mutational events observed in the TP53 gene (Fig. 7) [Pleasance et al., 2010]. The predominance of GC->TA transversions is a typical signature of carcinogen exposure associated with smoking. These observations are similar regardless of the strategy used to sequence the cancer genome (exome or whole-genome sequencing). For other types of cancer, such as colorectal cancer or brain tumors, the pattern of mutations found in the whole genome is also similar to that described for TP53 (Fig. 7). They are predominantly GC->AT transitions, thus implicating an endogenous methylation-driven process (most probably deamination of 5-methylcytosine [5-mC]) as a major causative factor. Some upper urinary tract urothelial cell carcinomas (UTC) have been associated with exposure to aristolochic acid (AA), a natural compound found in traditional herbal preparations used as health supplements and remedies. The analysis of TP53 mutations in UTC associated with AA exposure shows a high frequency of AT>TA transversion, a signature specific to the mutagenic effect of AA [Grollman et al., 2007]. Further sequencing of the entire genome of similar tumors confirms this particular landscape (Fig. 7) [Hoang et al., 2013; Poon et al., 2013]. Considering this association between cancer and exposition to carcinogens, the sequencing of genomes from liver cancer associated with aflatoxin B1 exposure will be very interesting, as the signature on the TP53 gene is specific to codon 249. The identification—or lack of identification—of other mutation hotspots will provide more insight on the specificity of DNA adduct formation.

Recent analyses have uncovered other patterns of mutations associated with apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) cytidine deaminases, with a high frequency of C>T or C>G substitutions in the nucleotide context sequence TCW (W: A or T) [Burns et al., 2013; Roberts et al., 2013]. Using the TP53 gene for specific patterns of mutations in various trinucleotide contexts is not really feasible since exonic mutations are shaped by both the nucleotide sequences and the resulting amino acids that must be selected by the mutation. Other specific patterns have also been uncovered but it is still unclear as to whether they have endogenous or exogenous origins; in vitro and in vivo mutagenesis assays will be required to uncover their etiology [Alexandrov et al., 2013a; Lawrence et al., 2013]. Further large-scale analyses of tumor genomes, such as the cancer 10K project focused on sequencing the genomes of 10,000 tumors of each of the most common cancers, may be of great help in increasing specific cancer etiology knowledge highly valuable for the development of preventive medicine.

Distribution of TP53 Mutations

The high concentration of mutations in exons 5–8 encoding the DNA-binding domain of the TP53 protein was already obvious 25 years ago [Hollstein et al., 1991; Caron de Fromentel and Soussi, 1992]. Although bias could occur due to the systematic analysis of these exons, sequencing the full-length gene has just softened the distribution (Fig. 8A and B).

The distribution of mutations in the TP53 protein is unique among all cancer genes, both oncogenes and tumor suppressors. All but five residues of the 393 amino acid long TP53 protein have been the target of at least one mutation in human cancer and each residue in the core domain has been found to be mutated a minimum of five times. This vast scattering of TP53 mutations is due to the high fragility of the core region; indeed, large-scale mutational analysis of each of its residues confirms its extreme sensitivity to substitutions.

It is important to keep in mind that the present analysis focuses on the major isoforms of the TP53 protein. As extensively discussed in an accompanying paper, a substantial number of mutations do not target all TP53 isoforms [Soussi et al., 2014]. Furthermore, beta and gamma exons localized in intron 9 have never been formally analyzed for mutations and only a handful of mutations have been described. A more intensive analysis of this region in human tumors must be undertaken. Any potential discoveries of mutation targeting will shed light on a putative tumor-suppressive activity for its isoforms.

The TP53 protein contains 64 amino acid residues that are conserved in all vertebrates. Two of them, resulting from codons 19 and 23, have never been the target of a single substitution in human cancer. This is not surprising since Phe19 and Trp23 are essential for the binding of MDM2 to TP53, and any substitutions at these positions would be lethal for the cell [Kussie et al., 1996]. Residue Leu26, also important for MDM2 binding, is not fully conserved in every TP53 but similarly to Phe19 and Trp23, it has never been found mutated. Among the other conserved residues, all but three, Lys121, Val97, and Pro98, are frequently substituted in human cancer (between 25 and 2,000 times).

Lysine 121 is localized in the L1 loop of the protein (residues 113–123), a cold spot region for mutation. Neighboring residue 120 is essential for DNA binding as it makes direct contact with a guanine base in the major groove of the DNA [Cho et al., 1994]. Several authors have shown that artificial mutations in the L1 domain changed the specificity of DNA binding and mutants at positions 120, 121, or 123 displayed an increased affinity for some p53RE and increased apoptotic activity [Saller et al., 1999; Zupnick and Prives, 2006]. Although these observations explain why such mutants are not selected during tumorigenesis, the modular role of this L1 loop in distinguishing various TP53 target genes has not yet been identified. The function of the two other cold spot codons at positions 97 and 98 is currently unclear.

Splice Mutations

Eukaryotic gene splicing is highly complex, involving signal sequences localized not only in introns but also in exon [Wang and Cooper, 2007]. Although mutations at donor and acceptor splice sites or branch point sequences localized in introns are well known to impair splicing, it is more difficult to associate splicing defects with exonic mutations. Exonic splicing enhancer (ESE) and exonic splicing silencer (ESS) are DNA sequence motifs (six bases) localized within an exon [Crotti and Horowitz, 2009]. They modulate the splicing of premRNA either positively (ESE) or negatively (ESS). These regulatory sequences were first described more than 15 years ago, but their identification remains difficult, although there are several computer programs that predict ESE and ESS in the TP53 gene. We currently do not know whether mutations at these positions impair splicing in addition to other defects such as DNA binding.

Splice mutations in the TP53 gene are underestimated but they remain nonetheless uncommon. In the current version of the TP53 database, they represent 2% of all mutations but this number rises to 4% if we select only NGS studies that target the entire TP53 gene. These numbers take into account only those mutations that target the two conserved intronic nucleotides GT and AG localized at the exon/intron boundaries (Figs. 9 and 10). We currently do not know whether mutations can be localized at other positions in the splice sites or in the branching point signals, but if it is possible, the event should remain relatively infrequent as there is a strong selection for the tumor to express a mutant TP53.

As discussed in the previous section, exonic mutations localized close to an intron/exon boundary can be detrimental to correct splicing (Fig. 10). These events are easy to spot when the nucleotide is at the third position of a codon as most mutations will lead to an sSNV (Fig. 10). Analysis of the TP53 gene revealed that four introns (3, 4, 6, and 9) cause no codon interruption. This includes codons 125 and 224 described in the previous section as well as codons 261 and 331 that display sSNVs at low frequencies in the database (three and six times, respectively) (Table 1 and Supp. Table S2).

The distribution of splice mutations is similar to that of missense mutations (Fig. 9). However, more research is needed to determine whether this is due to a bias associated with the high number of studies that have focused on these regions or representative of true selection. Although NGS studies encompassing the whole TP53 gene share similar distributions, more studies are needed to ascertain this observation. This landscape of mutations does not include potential alteration of splicing effects in exons beta and gamma localized inside intron 9 as they were blind from analysis.

Transition at CpG Dinucleotides

Across the entire TP53 gene, 425 CpG dinucleotides have been observed, 42 of which are localized in coding exons. The cytosines in these 42 CpG sites are methylated in normal tissue but the exact purpose of this modification for TP53 regulation is currently unknown [Tornaletti and Pfeifer, 1995]. Cytosine methylation also causes genome instability and probably a third of all transition mutations responsible for genetic diseases and cancers in humans [Cooper and Youssoufian, 1988]. These effects are due to the much higher spontaneous deamination of 5-mC as compared with cytosine. Deamination of 5-mC leads to T:G mismatches, whereas deamination of cytosine generates U:G mismatches. Both are recognized by mismatch-specific DNA glycosylases within the base excision repair system. Several lines of evidence suggest that other mechanisms may also be associated with the high frequency of substitution at CpG dinucleotides. Exogenous carcinogens, such as Benzo(a)pyrene or UV sunlight, have greater affinity for methylated CpG dinucleotides than for their unmethylated counterparts [Denissenko et al., 1997; You et al., 2000]. It is also possible that endogenous mutagens, derived from an altered cell metabolism, target methylated CpG dinucleotides resulting in a high rate of transition.

CpG in coding regions can take three forms depending on their location inside a codon: CGN, NCG, or NNC-GNN, referred here as type I, II, and III, respectively (Fig. 11). The symmetrical nature of a CpG dinucleotide induces methylation of the involved cytosines on the two DNA strands. Transition in type I CpG always results in amino acid substitution or the synthesis of a nonsense mutant regardless of whether it is the first or the second base that is changed (Fig. 11A). Specific targeting of the C on the coding strand will lead to a C to T transition, whereas targeting of the C on the noncoding strand will lead to a similar event that will be translated as a G to A transition on the coding strand. Transition in type II CpG will result in mutation only when the C on the coding stand is modified; a mutation of the opposite stand will alter the third base of the codon, which generally does not change the amino acid residue due to degeneration of the genetic code (Fig. 11B). Inversely, transition in type III CpG will lead to an amino acid change only when the C on the noncoding strand is modified (Fig. 11C).

All but one of the 42 “coding” CpG dinucleotides in the TP53 gene have been found to be mutated at least once in human cancer, corresponding to 25% of all the mutations in the database, but there is a high degree of heterogeneity in their frequency (Fig. 11 and Supp. Fig. S4). The inverse relationship between this frequency and the loss-of-activity of the resulting variant is striking and demonstrates that many mutants resulting from these substitutions are not truly selected during neoplasia because they are still active. In contrast, the most frequent mutants issued from transition at CpG dinucleotides correspond to the TP53 hotspot of mutations and display a clear loss-of-function (Supp. Fig. S4).

Examination of the frequency of transition at the two hotspot codons, 248 and 273 (type I CpG, more than 1,000 hits), shows a roughly equal frequency of C>T and G>A transitions (Fig. 11A), which is not surprising since the deamination of C residues in both strands of a symmetrical CpG dinucleotide will occur and be repaired at a similar rate. Therefore, the two TP53 variants resulting from these substitutions should occur at the same frequency if they both act as driver mutations with similar properties. The four TP53 variants issued from transitions at codon 248 or 273 (p.R248Q, p.R248W, p.R273H, and p.R273C) have been extensively analyzed and their loss-of-function has been confirmed both in vitro and in vivo.

In contrast, codon 175, also including a type I CpG, shows a marked disequilibrium in the distribution of mutations, with 1889 G>A (p.R175H) and only 45 C>T (p.R175C) (Fig. 11A).

We reported the disequilibrium for codon 175 as early as 1994 as it was already obvious in a mutation database composed of only a few hundred entries [Ory et al., 1994]. To analyze the role of this codon in TP53 function and/or loss-of-function, we constructed a library of 15 different p53 mutations at position 175 [Ory et al., 1994]. We found that the G>A transition led to variant p.R175H, the most frequent mutant in the database. The biochemical and biological function of p.R175H mutant (c.524G>A) are impaired entirely, and it is furthermore associated with various gain-of-function activities [Blandino et al., 1999]. On the other hand, the p.R175C mutant (c.523C>T) is not impaired for any TP53 function [Ory et al., 1994]. Considering this body of evidence, we hypothesize that p.R175C may be a passenger mutation coselected during neoplastic transformation. This hypothesis is supported by the following observations: (1) 30% of tumors with the p.R175C variant also have other TP53 mutations; (2) the variant is highly uncommon in cell lines; and (3) it has never been described as a germline mutation.

Disequilibrium has also been observed at codons 196 (432 p.R196* vs. 11 p.R196Q), 213 (611 p.R213* vs. 84 p.R213Q), 267 (71 p.R1267W vs. 28 p.R267Q), 282 (881 p.R282W vs. 35 p.R282Q), 306 (302 p.R306* vs. 3 p.R306Q), and 342 (173 p.R342* vs. 4 p.R342Q) with ratios ranging from 2.5 (for codon 213) to 100 (for codon 306) (Fig. 11A). It is worth noting that an arginine to glutamine substitution is observed among all the variants found at low frequencies. This substitution manifests as the tolerated transformation of the positively charged polar amino acid arginine to another polar residue, glutamine, suggesting that these mutations have only a mild effect on TP53 activity.

Three of the above-mentioned variants are found at very low frequencies (p.R196Q, p.R306Q, and p.R342Q) and are obviously passenger mutations, whereas two others, p.R213Q and p.R267Q, are found with significant frequency, suggesting that they are low-penetrant mutants. It is also noteworthy that four of the five frequent variants are nonsense mutations but whether this is due to coincidence or a greater aptitude for selection during transformation for these variants is an open question.

For type II and III CpG, only two mutants, p.P152L (type II, found 138 times) and p.G245S (type III, found 689 times), are obviously driver mutations associated with impaired transcriptional activity (Fig. 11B and C). sSNVs localized at codon 125 are discussed in previous sections.

Other transitions, some resulting in sSNVs others in nsSNVs, occur from one to 24 times in the database and their status as true mutations is questionable (Supp. Fig. S4). Most of the nsSNVs retain a significant residual activity and several of them are found in tumors with multiple TP53 mutations.

Posttranslational Modification

Like many eukaryotic transcription factors, the TP53 protein is the target of multiple posttranslational modifications (PTMs) that modulate its activity [Meek and Anderson, 2009], (see also the review of Nguyen et al. [2014] in the present issue). Several proteins coded from driver genes such as CCND1 or CTNNB1 are impaired by mutations that target residues modified by phosphorylation. Analyzing the impact of mutations on PTM is rather tricky as these protein sites include several continuous residues around the modified amino acid, and mutations changing the net charge or the hydrophobicity of these neighboring residues can also be detrimental for modifications.

Twelve different types of modification targeting 62 out of the 393 residues of the TP53 protein have been uncovered, the most frequent being phosphorylation and ubiquitination [Meek and Anderson, 2009] and Nguyen et al. (2014) in the present issue. All but six of the modification-susceptible residues have been found to be mutated in various cancers, corresponding to 5% of the mutants in the database (Fig. 12). The frequency of these mutations is highly heterogeneous but as discussed later in this review, assessing the impact on PTM for residues in the core region of TP53 is difficult to conceptualize, as these mutations will also impair DNA binding and transcriptional activity (Fig. 12). In contrast, most PTM sites localized outside the core domain are not mutated at a high frequency and those mutations that do occur do not impair TP53 transcriptional activity significantly, suggesting that they are either passenger mutations or sequencing errors (Fig. 12). The only exception is residue 337 localized in the oligomerization domain and discussed in the next section. It should be noted that PTMs on a specific residue are modulated by the neighboring residues, which are important for recognition by modifying enzymes. Therefore, mutations close to PTM sites could be also deleterious. Analysis of the distribution of TP53 mutations close to the various PTM sites does not reveal any specific clustering.

This lack of selection for PTM sites as driver mutations is in line with observations made in murine models [Olsson et al., 2007]. For example, mice expressing single TP53 mutants at various phosphorylation sites displayed only very mild—if any—specific phenotypes and were not highly prone to cancer ([Donehower, 2014; Nguyen et al., 2014] in this special issue). Similarly, mice with triple mutations at the three acetylation sites localized in the DNA binding domain did not display early onset of tumors, suggesting that acetylation at these sites is not essential for the tumor-suppression activity of TP53 [Li et al., 2012].

Oligomerization Domain

Tetramerization is an important feature of TP53 and required for its DNA-binding activity. An unstressed cell contains predominantly dimers of TP53 that oligomerize quickly after DNA damage [Rajagopalan et al., 2011; Gaglia et al., 2013]. The oligomerization domain (residues 326–355) consists of a beta-sheet (residues 326–333), followed by an alpha-helix (residues 335–356) [Clore et al., 1994]. This domain is encoded by exon 9 and 10 and the splitting occurs between codons 331 and 332. As TP53 isoforms beta and gamma do not include the coding region of exon 10, mutations in this region will target predominantly TP53 proteins that express full-length carboxy terminus. The formation of TP53 tetramers follows a particular kinetic with the quick formation of dimeric molecules followed by a slow association of these dimers into tetramers. In tumor cells expressing both wild-type and mutant TP53, poised dimers containing a wild and a mutant subunit have not been observed and consequently it appears as if only wt2/Mut2 tetramers are formed [Natan et al., 2009]. The oligomerization domain also includes a nuclear export signal (residue 340–351) exposed only when the protein is dimeric, suggesting a link between TP53 transport in the nucleus and DNA damage [Stommel et al., 1999].

At first glance, the importance of this domain for TP53 activity should lead to the observation of a high frequency of TP53 mutation. In fact, the opposite trend is observed: only a few mutations have been observed with the exception of codon 337 (Fig. 13). The oligomerization domain has been extensively analyzed and multiple functional data, such as transactivation or oligomeric state defined via multiple criteria or by biological activity, are available for many substitutions whether or not they were found in human cancer (Fig. 13) [Kawaguchi et al., 2005; Kamada et al., 2011]. The heat map summarizing these data shows that the most frequent mutants display a significant loss of transcriptional activities, a dysfunction in tetramerization, and impairment for apoptosis (Fig. 13). The only exceptions are mutants at codon 331 (significantly mutated in human cancer) that do not display any particular loss-of-function. An explanation for this may be that the codon is at the extremity of exon 9 and any mutation could impair splicing (see Pleiotropic Effects of TP53 SNVs in Human Cancer).

Arginine 337 forms an intermolecular salt bridge with aspartic acid at position 352 in TP53 dimers and is involved in the stabilization of the dimer [Clore et al., 1995]. Several variants at this position (p.R337C, p.R337L, or p.R337G) display impaired oligomerization and biological activities, and are frequently mutated in human cancer (Fig. 13). This position also includes p.R337H, found as a germline variant associated with pediatric adrenocortical carcinomas in Brazil, where this low-penetrant mutation is thought to be carried by thousands of people [Ribeiro et al., 2001; Custodio et al., 2013]. Haplotype analysis indicates a founder effect for p.R337H [Letouze et al., 2012]. As a somatic event, it is quite infrequent in the database (10 occurrences). Although the transcriptional activity of this mutant is close to that of the wild-type protein, it is very sensitive to pH change due to the protonation of the histidine residue [DiGiammarino et al., 2002].

Taken together, these data suggest that the oligomerization domain is not a primary target of the selection forces that drive neoplasia and inactivate TP53. Further structural and in vivo analyses of residue 337, which is found as a fascinating founder germline mutation and targeted by several somatic events, will provide valuable information on the relation between oligomerization and TP53 function in normal and tumor cells.

Pleiotropic Effects of TP53 SNVs in Human Cancer

As most TP53 mutations are localized in the core region of the protein that contains the DNA-binding domain (residues 100–300), a binary relationship between TP53 mutations and inactivation of the transcriptional activity of the protein has been established. This relation is supported by multiple in vitro and in vivo studies and it is highly likely that the tumor-suppression function of TP53 is linked to a complex transcriptional program. Recent studies in mouse models have suggested that growth arrest, apoptosis, and senescence were not essential for tumor-suppression function [Brady et al., 2011; Li et al., 2012; Valente et al., 2013]. Consequently, novel target genes and novel pathways must be identified to obtain a more accurate understanding of the heterogeneous behavior of the diverse TP53 variants.

Sequencing of the coding regions of cancer genes has led to a protein-centric view of the consequences of the various SNVs included in the TP53 database. The concept of a eukaryotic gene and its various products must evolve to take into account and fully understand the multiple consequences of SNVs. Although the protein is the final product after translation, a single SNV can have multiple effects on posttranscriptional modifications, translation efficiency, or PTMs, leading to the synthesis of RNA and/or protein variants that carry a combination of multiple defects.

We identified at least five different effects that can be caused by SNVs on the TP53 gene or protein, such as splicing, DNA binding/transactivation, differential targeting of the various isoforms, isoform translation, and PTMs (Fig. 14). Most substitutions can impair DNA binding and transcriptional activity and additional effects can pile up. The TP53 gene is transcribed into eight different mRNAs and translated into 12 protein isoforms [Khoury and Bourdon, 2010]. Several isoforms lack the amino terminus and translation starts at codons 40, 133, or 160. Others do not contain the carboxy terminus due to alternative splicing with two newly discovered exons localized in intron 9. Depending on the position of the mutation, not all isoforms will be impaired. Indeed, mutants at codons 110, 127, or 342 are found more than 100 times in the database despite the fact that they do not target all TP53 isoforms (Fig. 14). To what extent these mutations contribute to the inactivation of the TP53 pathway is currently unknown, as is the significance of the expression of the remaining wild-type isoforms.

The ATG codons 133 and 160 assign amino acids to the DNA-binding site of the TP53 protein but they also act as start codons for isoforms Delta133 and Delta160. Mutations at these positions not only impair the transcriptional activity of the full-length protein, they also prevent the translation of the isoforms (Fig. 14). Translation efficiency is modulated by an overall sequence context, the Kozack sequence, that includes nucleotides adjacent to the start codon and therefore mutations in contiguous codons can increase or decrease the level of the deltaN isoforms.

As discussed in the previous section, several PTMs such as ubiquitination or neddylation contribute to the stability of the protein. Frequent mutations localized in the core region of the protein (residues Lys132 or Cys141) are modified by PTM and it is difficult to assess whether loss of DNA binding or PTM or both impair TP53 function (Fig. 14). For codon 337, the situation is even more complex as the residue, localized in the oligomerization domain and translated via exon 10, is not present in eight out of 12 isoforms and has furthermore been shown to be methylated. Therefore, the net consequence of any change at this residue will be complex with multiple potential consequences targeting only a subset of TP53 proteins. This is not a trivial point; as discussed previously, it is estimated that thousands of individuals carry the low-penetrant p.R337H TP53 mutation in Brazil. Structural and functional analyses have suggested that DNA binding and transactivation are moderately impaired, but most of the experiments were performed in vitro using the full-length isoform. More in vivo analyses are needed to fully understand how this mutation impairs the TP53 pathway.

Most of the information needed for correct splicing is not at the splice site itself and exonic sequences can have a profound impact on splicing efficiency. For a gene such as TP53 that expresses eight differentially spliced mRNA translated into 12 isoforms with sometimes opposing effects, a delicate equilibrium of the ratio of each protein is required. We can expect that several exonic mutations will be deleterious for both RNA splicing and protein function (Fig. 14). RNAseq technology will be of great aid in addressing this question as it will enable the quantitative and qualitative assessment of the status of all TP53 mRNAs in a tumor.

It is possible that these sidecar effects carry little weight compared with mutations affecting the main activity of TP53 directly, or that the loss of TP53 function is indeed the cumulative result of multiple defects. Exploring the contributions of the diverse variants will be very laborious but also surely mandatory if we hope to understand how each mutant ultimately affects TP53 function.

Conclusion

The analysis of TP53 mutations in human cancer has been at the forefront of molecular epidemiology and led to the identification of specific patterns of mutation associated with carcinogen exposure [Schetter and Harris, 2012]. Whole-genome sequencing of tumors has confirmed that the global mutation landscape is similar to that previously described in the TP53 gene [Alexandrov et al., 2013a; Kandoth et al., 2013]. Thanks to the large size of the human genome, novel original mutation signatures have been uncovered in specific tumor types and will contribute greatly to our understanding of cancer etiology. Novel patterns of mutation specific to TP53 are also emerging and in turn raising interesting questions. For example, the observation that basal-like breast cancers and ovarian cancer share a high frequency of TP53 mutations (more than 80%) with a high number of frameshift mutations needs to be confirmed and investigated. Analyzing whether this is a consequence of deficient DNA repair or specific selection for null TP53 status will be of interest.

The end of the twentieth century was dominated by the concept that single gene alterations could be used as efficient clinical biomarkers. As a result, much time was invested looking for the “Holy Grail” gene. TP53 was one of the genes at the forefront of these studies, resulting in thousands of publications, and yet only a handful of biomarkers are currently in clinical testing [Hutchinson and DeVita, 2009]. Great expectations were put on the use of TP53 mutations as clinical biomarkers for predicting prognosis or response to treatments, hopes supported by the central function of TP53 in the cellular response to DNA damage induced by chemotherapeutic drugs. After 25 years spent scrutinizing TP53 status in all types of cancer and after thousands of analyses and meta-analyses, we now have a much clearer vision of TP53 and can thus nuance our expectations and focus on specific niches where TP53 mutations do indeed have clinical value. Monitoring TP53 mutations in CML or in MDS before relapse or transformation is an example of a direct and clinically pertinent application.

The analysis of germline mutations in various cancer-prone families is also an essential clinical aspect of TP53 and novel observations suggest that its value goes beyond Li–Fraumeni syndrome [Kamihara et al., 2014]. Indeed, TP53 germline mutations have also been observed in families at high risk of breast cancer, albeit at very low frequency, suggesting that the spectrum of tumors associated with these mutations may be shaped by some currently unknown genetic factor.

For many other cancer types, the value of TP53 mutations per se has not been powerful enough to warrant its use in the clinic. Before all else, we need to improve the diagnosis of TP53 alterations qualitatively, taking into account the entire sequence of the gene, including the newly discovered exons. Furthermore, we can acquire a more limpid picture of the landscape of TP53 inactivation in various types of cancer by looking at the whole TP53 pathway, including the miRNA network that regulates TP53 function. One of the predominant issues that has emerged from the cancer genome projects is the identification of specific landscapes of genetic and epigenetic alterations targeting specific pathways, beyond the question of tissue specificity. In this information-rich context, it should be possible to define specific niches where TP53 alterations can act as strong biomarkers. Finally, if we wish to define the true clinical value of the molecule and its mutations, we must identify the exact nature of the tumor-suppressive function(s) of TP53 that are impaired in human cancer.

New methodologies offering increased sensitivity for the detection of mutations in an excess of normal DNA have made feasible the use of circulating tumor DNA as a source of material for genetic analyses [Diehl et al., 2005; Dawson et al., 2013]. Thanks to their high frequency, TP53 mutations will be among the most useful targets in this novel setting.

TP53-based therapy has been in the pipeline of numerous academic and pharmaceutical laboratories [Lane et al., 2010; Selivanova, 2010; Hong et al., 2014]. In the wake of the disappointment surrounding gene and viral therapies, a new generation of chemical drugs targeting mutant TP53 is under development and several candidates are currently in clinical trials [Lehmann et al., 2012]. Drugs to inhibit the interaction between wild-type TP53 and MDM2, thus increasing TP53 activity, are also in development. In particular, these therapies hold the promise of offering efficacy in all types of tumors, whether they express wild-type or mutant TP53 [Vassilev, 2004; Zhao et al., 2010].

For all of the various settings and perspectives described above, the TP53 databases have been invaluable resources. They have played—and continue to play—key roles in linking clinical and basic research and developing numerous working hypotheses. Disabling the TP53 pathway is ubiquitous in human cancer and perhaps even mandatory. Understanding how the various mutations shape tumors and deciphering the heterogeneity of the various targets in different tissues will be precious not only for treating cancer but also for understanding the roles of TP53 in normal cells.