Integrated Analysis of TP53 Gene and Pathway Alterations in The Cancer Genome Atlas

SUMMARY

The TP53 tumor suppressor gene is frequently mutated in human cancers. Analysis of five data platforms in 10,225 patient samples from 32 cancers reported by The Cancer Genome Atlas (TCGA) has enabled comprehensive assessment of p53 pathway involvement in these cancers. Over 91% of TP53 mutant cancers exhibit second allele loss by mutation, chromosomal deletion, or copy neutral loss of heterozygosity. TP53 mutations are associated with enhanced chromosomal instability, including increased amplification of oncogenes and deep deletion of tumor suppressor genes. Tumors with TP53 mutations differ from their non-mutated counterparts in RNA, miRNA and protein expression patterns, with mutant TP53 tumors displaying enhanced expression of cell cycle progression genes and proteins. A mutant TP53 RNA expression signature showed significant correlation with reduced survival in 11 cancer types. Thus, TP53 mutation has profound effects on tumor cell genomic structure, expression, and clinical outlook.

INTRODUCTION

The p53 tumor suppressor protein is a transcription factor that inhibits cell division or survival in response to various stresses, thus acting as a key failsafe mechanism of cellular anti-cancer defenses (Kastenhuber and Lowe, 2017; Vousden and Prives, 2009; Macheret and Halazonetis, 2015). Frequent mutation of TP53 in human cancers was first described by Vogelstein and Minna and colleagues (Baker et al., 1989; Nigro et al., 1989; Takahashi et al., 1989) and was further catalogued in TP53 mutation databases (Bouaoun et al., 2016; Caron de Fromentel and Soussi, 1992; Hollstein et al., 1991; Leroy et al., 2014). These compilations established that the pattern of TP53 mutations in several cancers were linked to carcinogen exposure and provided key insights into key p53 functional domains and the role of p53 in cancer etiology (Caron de Fromentel and Soussi, 1992; Hollstein et al., 1991). The majority of TP53 mutations occur in the central DNA binding domain and result in inactivation of transcription factor functions. Some missense mutations have been associated with dominant negative inhibition of wildtype p53 and/or oncogenic gain of function in the absence of wildtype p53 in experimental contexts (Muller and Vousden, 2013, 2014; Soussi and Wiman, 2015). Missense mutations often render p53 protein resistant to proteolytic degradation by E3 ubiquitin ligases like MDM2, ensuring high levels of stable mutant p53 protein (Frum and Grossman, 2014). The most typical TP53 mutation configuration is a single TP53 mutation with loss of the remaining TP53 allele through a large-scale deletion on chromosome 17p (Baker et al., 1989; Baker et al., 1990). Other less frequent configurations include mutation of both TP53 alleles or mutation of one allele and retention of the second wildtype allele. Homozygous TP53 deletion is a rare event likely because of closely linked cell essential genes (e.g. POLR2A) (Liu et al., 2015; Mulligan et al., 1990). Studies in acute myeloid leukemia (AML) have shown that copy neutral loss of heterozygosity can lead to tumors with the same TP53 variant in both alleles (Jasek et al., 2010).

The p53 pathway is affected by numerous upstream regulators and in turn it regulates large numbers of targets transcriptionally and non-transcriptionally (Kastenhuber and Lowe, 2017; Mello and Attardi, 2017). Regulators of p53 include binding proteins that stabilize or destabilize it, as well as enzymes that post-translationally modify p53 and activate, deactivate, or modulate its functions (Dai and Gu, 2010; Nguyen et al., 2014). p53 transcriptionally targets hundreds of genes, target specificity depending on cell type and context as well as the nature and intensity of the cell stressor (el-Deiry et al., 1993; Fischer, 2017; Mello and Attardi, 2017; Vousden and Prives, 2009). p53 upregulates genes that encode cell cycle inhibitors, apoptosis inducers, DNA repair proteins, and metabolic regulators (Vousden and Prives, 2009). The p53 protein also downregulates genes associated with cell cycle progression, though the process involves CDKN1A/p21-dependent indirect mechanisms of gene regulation (Engeland, 2018).

Clinically, TP53 mutations have been linked to a poorer prognosis for some cancers, but this remains controversial (Robles and Harris, 2010). Many variables, such as cancer type, clinical stage, study size, or quality of TP53 mutation detection can affect prognostic determinations. TP53 mutation status is not always indicative of p53 signaling status in a cancer cell (Leroy et al., 2014). For example, p53 signaling can be attenuated through non-mutational mechanisms such as amplification of p53 negative regulators MDM2, MDM4, or PPM1D (Lu et al., 2008; Matheu et al., 2008; Soussi and Kroemer, 2018; Wasylishen and Lozano, 2016). Thus, a more accurate readout of p53 function in human cancers not based strictly on mutation of the TP53 gene might lead to more accurate prognostic predictions.

We have capitalized on the integrated TGCA approach in which large numbers of tumors in many cancer types have been simultaneously examined on five independent data platforms along with extensive clinical annotation to develop a comprehensive picture of the role of TP53 mutation. We broadly show how TP53 mutation confers an array of genomic, transcriptomic, and proteomic changes in cancers that play a major role in regulating clinical outcomes.

RESULTS

TP53 mutation profile in the TCGA dataset is similar to that of current TP53 databases.

We analyzed TP53 mutations in whole exome sequences of 10,225 TCGA patients across 32 different cancer types. We identified 3,786 patients with TP53 mutations. TP53 mutation frequencies by cancer type were variable, with ovarian cancer and uterine carcinosarcoma showing greater than 90% incidence and seven cancer types displaying less than 5% incidence (Fig. 1A). The spectrum of TCGA TP53 mutations obtained by whole exome sequencing was broadly similar to that detected by conventional Sanger sequencing. Notable hotspot mutations at codons 175, 248, and 273 in the TCGA dataset were also the most frequently mutated TP53 codons observed in the Sanger dataset from the UMD_TP53 database (36,350 patients with TP53 mutations, 4300 different variants) (Leroy et al., 2014) (Fig. 1B,C, S1A,B). The majority of TCGA TP53 mutations occurred in the central DNA binding domain encompassing exons 4–8 and aligned with those reported in the Sanger dataset of the UMD_TP53 mutation database. In both groups, missense (~40%), frameshift deletion (~20%), and frameshift insertion (~10%) comprised roughly 70% of all mutations (Fig. S1B). Less frequent were in frame deletions and insertions, synonymous mutations, and splice site mutations (Fig. S1B). In both datasets, mutation types were enriched in the central domain, though the small in frame deletions and insertions were surprisingly dispersed (Fig. S1A). The TCGA mutation dataset also includes infrequent TP53 variants in the two newly discovered alternative TP53 exons, exon beta and gamma (Fig. 1C, S1B). Whether these variants are true somatic pathogenic variants or rare benign polymorphisms is unknown as this region of the TP53 gene has not been thoroughly analyzed.

Figure 1. TP53 mutation profile in the TCGA dataset is similar to that of current TP53 databases.

(A) TP53 mutation frequency varies depending on tumor type. For each cancer type, the relative fraction of non-truncating and truncating mutations are indicated. (B) Distribution of global TP53 mutations in the TCGA dataset are similar to the Sanger sequenced subset of the UMD_TP53 Mutation Database. Major TP53 hotspot mutations are indicated. (C) Exon distribution of TP53 mutations for all variants in the Sanger subset of the UMD_TP53 Mutation Database (top panel) is similar to that observed for TCGA dataset TP53 mutations (middle panel). Previously unreported TP53 single nucleotide variants (SNVs) discovered by TCGA are in bottom panel. (D) Distribution of codon SNV sites along the p53 protein recorded to date. For each codon nine SNVs are possible. Red bars indicate SNVs observed for each codon, while green bars indicate SNVs not yet reported. Unmutated residues at codon 114 and hotspot residues at codon 175, 248, and 273 are shown by asterisks. TAD: transactivation domain; Pro: proline rich domain; OLI: tetramerization domain; I to V: evolutionarily conserved domains. (E) TP53 variants in functional domains of p53. The DNA-binding surface of p53 protein is composed of two large loops (L2 and L3) stabilized by a zinc ion. Sheet S10 and Helix H2, is a component of the LSH (loop Sheet Helix) domain that includes loop L1. These various domains are essential for DNA recognition. Variants are depicted for each structural domain. Sanger: number of TP53 variants found in the Sanger dataset of the UMD_TP53 database; TCGA: number of TP53 variants in the TCGA dataset. Variant frequencies are denoted by bars. The heat map corresponds to loss of function of TP53 variants measured by the functional assays of Kato et al. (1); Kotler et al. (2) and the three functional assays of Giacomelli et al. (3–5). Also see Fig. S1,S2.

We combined TP53 SNVs from the TCGA and UMD_TP53 datasets to assemble a graph showing the fraction of 9 possible codon changes for each of the 393 TP53 triplets encoding the full length p53 (Fig. 1D). The central domain shows at least 6 of the 9 possible codon mutations in virtually every location, whereas sites in the N- and C-terminal domains show an absence of any codon changes. The N-terminal mutation free region correlates closely with the MDM2 binding domain of p53. Because MDM2 is an E3 ubiquitin ligase that mediates p53 degradation, mutation of this binding site should stabilize p53 rather than inactivate it. A mutation resistant C-terminal domain was also noted, but the mechanism for mutation avoidance is unclear. The absence of variants at codon 114 in loop L1 confirms a study demonstrating that TP53 variants at this position have little or no functional effect (Zupnick and Prives, 2006).

The TCGA effort identified a total of 384 previously unreported variants, the majority of which were frameshift insertions and deletions, though 69 unreported single nucleotide variants (SNVs) were noted (Fig. S1C–E). These SNVs were more frequently observed in exons 2–4 and 10–11 (Fig. 1C, bottom panel), likely because historical TP53 mutation studies generally sequenced only exons 5–8 or 5–9.

The availability of functional data for most missense TP53 variants is an unusual feature among cancer genes. In one study Kato et al. measured the transcriptional activity of TP53 variants in yeast assays using 8 different TP53 response elements (Kato et al., 2003). A clear correlation between TP53 variant loss of activity and their frequency in human tumors has been shown (Soussi et al., 2005). Recently, two large-scale analyses of TP53 variants were performed in mammalian cells using cellular assays measuring variant effects on cell proliferation as a quantitative functional readout (Giacomelli et al., 2018; Kotler et al., 2018). Using the Kato data, we confirm the clear correlation between TP53 loss of activity and the high frequency TP53 variants in cancer (Fig. S2A,B). TP53 variants from the TCGA dataset are also clearly associated with a loss of activity for the 8 different TP53 response elements (see Fig. S1B for the CDKN1A/p21/WAF1 and AIP1 promoter results). Similarly, we showed that functional analysis data of Kolter et al. (2108) and Giacomelli et al. (2018) correlate strongly with the frequency of TP53 variants in human cancer and that TP53 variants from the TCGA dataset are generally associated with deleterious TP53 variants (Fig. S2B, lower graphs). The previously unreported TCGA TP53 variants showed a relatively high residual p53 activity in all functional assays (Fig. S2A, and red boxes in Fig. S2B).

With respect to p53 three dimensional structure, deleterious TP53 variants were clustered in specific regions of the p53 protein such as loop L2 and L3, and part of a Helix Loop Helix including sheet S10 and helix H2 (Fig. 1E). These domains are highly enriched for mutations in the UMD_TP53 Mutation Database and TCGA dataset and variants exhibit major functional consequences at virtually every codon (Fig. 1E). Loop L3 in particular is critical for DNA binding and exhibits the highest number of mutations.

Inactivation of both alleles occurs in over 90% of TCGA cancers with TP53 mutations.

Tumor suppressors are considered to be recessive at the genetic level; inactivation of both tumor suppressor alleles is generally required for an oncogenic phenotype (Knudson, 1989). While TP53 is a tumor suppressor, there has been evidence that missense mutations in TP53 result in expression of a stabilized protein of altered conformation that exhibits both dominant negative activity toward wildtype p53 protein as well as gain of function oncogenic activity (Muller and Vousden, 2013; Soussi and Wiman, 2015). Therefore, mutation of only one TP53 allele could potentially result in a significant oncogenic phenotype.

To assess TP53 allele status in the TCGA dataset, we performed an integrated analysis of TP53 mutant tumors using two data platforms, exome sequence data and copy number data. First, we examined the 10% of TP53 mutant tumors (381 of 3786 total tumors with TP53 mutations) with two distinct TP53 mutations. Functional analysis of the TP53 mutations revealed that both mutations were usually inactivating (Fig. 2A). To determine whether the two TP53 mutations were on the same allele (cis) or on different alleles (trans), we examined individual DNA sequencing reads that were in the same exon and 6–75 nucleotides apart. Those tumors with individual reads consistently displaying both mutations were considered cis for TP53 mutations and those tumors consistently exhibiting only one of the two mutations were considered trans. Representative read files for cis and trans tumors are shown in Fig. 2B. For those 35 two TP53 mutation tumors with tightly linked mutations, 28 (80%) were trans and 7 (20%) were cis (Fig. 2C). Assessment of these 35 tumors for copy number status revealed that 6 of the 7 cis tumors exhibited TP53 copy number loss while 27 of the 28 trans tumors retained a diploid TP53 status (Fig. 2C). Thus, there is selection for wildtype TP53 allele loss in the cis tumors, whereas trans mutant TP53 alleles show no selection for allele loss. When we looked at functional effects of individual TP53 mutations in the cis and trans tumors, 5 of the 7 cis tumors contained at least one TP53 mutation with partial WT function, whereas almost all of the trans TP53 mutations resulted in two TP53 alleles without any WT function (Fig. S3A).

Figure 2. Inactivation of both alleles occurs in most TCGA cancers with TP53 mutations.

(A) p53 functional analyses in some TCGA tumors with two TP53 mutations. Heat maps show relative transcriptional activity of TP53 variants compared with wildtype TP53 based on data in Kato et al. (2003). Each column shows a p53 transcriptional target and each row shows a TP53 variant. W - WAF1 (CDKN1A); M – MDM2; B – BAX; 14 – 14-3-3 sigma (SFN); A – AIP (TP53AIP1); G – GADD45A; N – NOXA (PMAIP1); P – p53R2 (RRM2B). Database variant frequency is shown as a blue bar. (B) Individual DNA sequence reads from tumors with two closely linked TP53 mutations show trans (left) and cis (right) mutation configurations. Mutations are at the top. Gray boxes represent individual nucleotides. Gray lines show individual sequence reads with colored segments indicating mutations. (C) Most tumors with two closely linked TP53 mutations have trans mutations and retain diploid copy number (NO LOH), while a minority have cis mutations but show wildtype TP53 allele loss (LOH). (D) Tumors with one TP53 mutation exhibit TP53 copy number loss while tumors with 0 or 2 TP53 mutations are largely diploid. Copy number values at the TP53 locus for tumors with 0 (top panel), 1 (middle panel), or 2–3 (lower panel) TP53 mutations are shown. On the X axis TP53 copy number values are binned in 0.1 value increments where 0 represents diploidy and values of −0.4 to −0.6 are roughly equivalent to a haploid copy number. Significant differences between each category are indicated. ****(p < 1E-50); ***(p < 1E-25). (E,F) Median variant allele fractions (VAF) in tumors with one TP53 mutation approximate 1.0, indicating frequent loss of both wildtype TP53 alleles. Uterine corpus endometrial carcinoma (UCEC) (E) and head and neck squamous cell carcinoma (HNSC) (F) were stratified by mutation number and copy number. A copy number (CN) of 0 is considered diploid and CN of −1 is considered haploid. “Mult Mut” indicates tumors with 2+ TP53 mutations. VAF distributions are shown and median values are indicated by the central bar in the box and whiskers plots. Statistical significance was indicated by t tests. See also Fig. S3.

We also examined TP53 copy number across all TCGA cancers stratified by 0, 1, or 2+ TP53 mutation categories. While tumors with no TP53 mutations were generally diploid at the TP53 locus, tumors with a single TP53 mutation showed significant skewing towards copy number values indicating loss of a single TP53 allele (Fig. 2D). Tumors with two or more TP53 mutations largely retained a diploid copy number.

A closer look at TP53 DNA copy number status in the single TP53 mutation tumors showed that 66% of these tumors displayed TP53 copy number loss, while 34% were apparently diploid in TP53 copy number. Analysis of individual DNA sequencing reads within multiple cancer types with a single TP53 mutation revealed that the variant (mutant) allele fraction (VAF) of the TP53 reads averaged close to 1.0 in tumors with TP53 allele copy number loss (Fig. 2E,F, Fig. S3B,C). Most tumors with a single TP53 mutation in a diploid context also displayed a variant allele fraction close to 1.0. This suggests that the mutant TP53 allele is frequently duplicated through mitotic recombination or another gene duplication mechanism. Consistent with our earlier result (Fig. 2C), tumors with two or more TP53 mutations averaged a variant allele fraction near 0.5 (Fig. 2E,F, Fig. S2A). In one tumor type (UCEC), tumors with a TP53 silent mutation showed a variant allele fraction approximating 0.5, indicating little selection for TP53 allele loss or for mutant allele duplication (Fig. 2E). Among tumor types with frequent TP53 mutations, all showed these patterns (Fig. 2E,F, Fig. S3B,C). Among six tumor types with frequent TP53 mutation, 91.3% exhibited variant allele fractions (VAFs) consistent with loss of both wildtype alleles, while only 8.7% displayed a likely retention of the wildtype TP53 allele (Fig. S3D). This pattern is repeated among all other cancer types with statistically sufficient numbers of TP53 mutations (Fig. S3D). Thus, the data indicates a strong selection for loss of the second TP53 allele after mutation of the first allele.

P53 RNA and protein expression are highly variable and dependent on mutation type.

In the previous section over 90% of mutant TP53 tumors displayed loss of the wildtype TP53 allele. To confirm this phenomenon at the RNA expression level we examined p53 variant allele fractions in the p53 RNAseq data. After adjusting RNA VAFs for tumor purity, we calculated that over 92% of 799 TCGA tumors with TP53 missense mutations had p53 VAF values near 1.0 (Fig. S3E). Thus, TP53 generally behaves as a recessive tumor suppressor both at the DNA and RNA level.

Analysis of p53 RNA expression levels in the TCGA tumors revealed wide intertumoral variation within both wildtype and mutant TP53 tumors (Fig. S4A). Across all cancer types, p53 RNA in tumors with missense (or in-frame deletion/insertion) TP53 mutation was modestly increased relative to that in WT TP53 tumors (Fig. S4B,C). However, p53 RNA expression in tumors with truncating TP53 mutations (nonsense, frameshift insertion/deletion, splice site) was reduced compared to either WT or missense MUT TP53 tumors (Fig. S4B,C). This is likely due to nonsense-mediated mRNA decay processes.

The expression of p53 protein in the wildtype and mutant TP53 tumors as measured by RPPA displayed intertumoral variability as well as some variability based on cancer type (Fig. S5A). Cancer types with high fractions of mutant TP53 tumors generally showed significantly higher levels of p53 protein expression relative to tumors with wildtype p53, though there were a few exceptions (Fig. S5B). When the TP53 mutations in a tumor type were stratified by truncating and non-truncating mutations, both wildtype TP53 and truncating mutant TP53 tumors exhibited low p53 protein levels while the non-truncating mutant TP53 tumors showed significantly higher p53 expression (Fig. S5C,D). This result may be due to nonsense-mediated decay of p53 truncating mutant RNA and that non-truncating TP53 mutations often lead to a p53 conformational change with resistance to degradation.

TP53 mutation is significantly correlated with increased chromosomal instability.

P53 has been called the “guardian of the genome”, based on accumulated evidence that it plays a major role in preserving genomic stability (Lane, 1992; Smith and Fornace, 1995; Tarapore and Fukasawa, 2002). We examined copy number data for 25,000+ loci in each of the 10,225 TCGA tumors and stratified this data by TP53 status. We analyzed major copy deviations from diploidy (copy number values of 0). Loci greater than four-fold that of normal copy number (copy number values of greater than 2) were considered indicative of amplification and were counted. Loci with copy number values of less than −1 were considered indicative of deep deletions and these were tallied. The fraction of all TCGA wildtype and mutant TP53 tumors with amplification and deep deletion were each plotted across the entire genome (Fig. 3A). The global genomic profile for deep deletions (CN < −1) shows peaks of enhanced deletions at major tumor suppressors such as CDKN2A and RB1. Generally, the mutant TP53 tumors displayed a much greater fraction of tumors with deletions at each frequent deletion site than their wildtype counterparts (Fig. 3A, upper panel). Likewise, peak regions of amplification (e.g. MYC, CCND1, CCNE1) showed a higher fraction of amplifications in mutant TP53 tumors relative to wildtype TP53 tumors (Fig. 3A, middle panel). These divergences in amplification frequencies between wildtype and mutant TP53 were highly significant (Fig. 3A, bottom panel). Across the entire genome of all TCGA cancers, mutant TP53 tumors exhibit roughly 2.5-fold the total number of deletions and amplifications as do wildtype TP53 tumors (p = 0) (Fig. 3B). When the relative TP53-dependent genomic instability of individual tumor types was compared mutant TP53 tumors displayed significantly more chromosomal instability in 19 of 23 tumor types with sufficient TP53 mutations for comparison (Fig. 3C).

Figure 3. TP53 mutation is correlated with increased chromosomal instability.

(A) Genomic profile aggregated from copy number data of all TCGA tumors shows that frequent amplifications and deep deletions occur significantly more in MUT TP53 than in WT TP53 tumors. Fraction of copy number losses (CN < −1) (top panel) and gains (CN > 2) (middle panel) for wildtype (blue line) and mutant (orange line) tumors are shown. Peaks correspond to frequent regions of deep deletion (top panel) and amplification (middle panel) and are labeled with the tumor suppressor gene or oncogene at the epicenter of each peak. Statistical significance in relative frequency of deletion or amplification between WT and MUT TP53 tumors at each gene are indicated in the bottom panel. (B) Mutant TP53 tumor genomes display roughly 2.5 fold higher rates of amplification and deep deletion relative to their wildtype TP53 counterparts. Gene loci with copy number values greater than 2 (amplification) and less than −1 (deep deletion) were totaled for all wildtype and all mutant TP53 tumors and divided by total loci number in each TP53 category. (C) Most cancer types show significantly increased rates of amplification and deep deletion in the MUT TP53 tumors (Mut TP53 Sig Inc) compared to wildtype TP53 tumors (WT TP53 Sig Inc). The fraction of loci with deep deletion or amplification in MUT TP53 tumors was divided by that in WT TP53 tumors to give a ratio. Significance was determined by t test. (D) Of the six most frequent deep deletions, five occur significantly more frequently in the mutant compared to the wildtype TP53 group. (E) Of the six most frequent amplifications, five occur significantly more frequently in the mutant compared to the wildtype TP53 group. (F) Three frequently amplified negative regulators of p53 (MDM2, MDM4, and PPM1D) are significantly more amplified in WT relative to MUT TP53 tumors. (G) Nucleotide level mutation rates are increased in MUT TP53 tumors. Median mutation numbers per tumor for wildtype and mutant TP53 tumors are shown in the box and whiskers plots. An unpaired t test showed that the mutant TP53 tumors have significantly more total mutations per tumor compared to their wildtype counterparts. See also Table S1.

When frequent regions of amplification and deletion were examined more closely, major oncogenes and tumor suppressor genes were usually located at the epicenter of each. Again, most of these amplicons and deletions were significantly more frequent in TP53 mutant tumors (Fig. 3D,E, Table S1A,B). A notable exception were three frequent amplification regions in which MDM2, MDM4, and PPM1D were located at the epicenter, and which were significantly more amplified in wildtype TP53 tumors (Fig. 3F). All three of these genes encode negative regulators of p53 function, illustrating a non-mutational mechanism of inactivating p53 in tumors.

We also assessed the effects of TP53 mutation on genomic instability at the individual nucleotide level by comparing whole exome mutation rates in individual wildtype and mutant TP53 tumors. Across all TCGA tumors we found a moderately increased rate of whole exome nucleotide level mutations in the individual mutant TP53 tumors. The median number of whole exome mutations across all wildtype TP53 tumors was 68 compared to 150 in mutant TP53 tumors (Fig. 3G). There did not appear to be significant differences in the types or patterns of mutations based on TP53 status. Despite the apparent increase in nucleotide level mutation rates in mutant TP53 tumors, it is difficult to determine whether TP53 mutation plays a causative role in nucleotide level instability or whether increased TP53 mutations are correlated with other processes that globally affect cellular mutation rates and patterns. Consistent with the latter idea, a recent analysis of cancer mutational signatures indicated that TP53 mutation patterns were often dependent on mutational signatures commonly found in the relevant tissue of origin (Giacomelli et al., 2018).

Comparison of global RNA, microRNA, and protein expression reveal p53-dependent pathways in cancer.

Because p53 is a transcriptional regulator, we compared gene expression patterns in the wildtype and mutant TP53 tumors for each cancer type (Fig. 4A–D, Table S2). We directly compared RNA expression levels for each gene and ranked by t test the most significant up- and downregulated genes in wildtype versus mutant TP53 tumors. Of the top 20 most upregulated genes in wildtype TP53 tumors, 14 were established p53 upregulated target genes (Fig. 4A, Table S2, S3). P53 target genes EDA2R, RPS27L, and SPATA18 were each significantly upregulated in 21 different wildtype TP53 cancer types (Fig. 4A, Table S3). To perform pathway analyses based on differences in individual gene expression, we identified the top 500 most consistently upregulated genes in wildtype TP53 tumors relative to mutant TP53 tumors and performed Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) on them (Table S4A). Not surprisingly, as shown in Fig. 4B, most of the significantly enhanced pathways in wildtype TP53 cancers involved p53-related signaling pathways. Even those enriched pathways not directly related to p53 signaling are functionally associated with p53, and included a number of apoptosis pathways.

Figure 4. Comparison of global RNA expression reveals p53-dependent pathways in cancer.

(A) The most significantly upregulated gene RNAs in wildtype TP53 cancers are mostly known p53 target genes and the number of cancer types in which they were upregulated is indicated. Direct p53 target genes are indicated by red bars. See also Table S2 and S3. (B) Pathway analyses based on genes expressed at significantly higher rates in wildtype TP53 cancers shows that p53-related pathways (red bars) are highly enriched. See also Table S4A. (C) The most significantly upregulated RNAs in mutant TP53 cancers are cell division promoters. For each cancer the top 100 and 500 gene RNAs most highly expressed in mutant relative to wildtype TP53 tumors were identified and the top 20 upregulated genes across all mutant TP53 cancers are shown. Roles for cell division, G2/M checkpoint control, E2F target genes, and documented repression by p53 are indicated by blue boxes. Significant upregulation in individual cancer types are indicated by red and pink boxes. Red arrows indicate four genes comprising the mutant p53 signature discussed later (Fig. 7). (D) Pathway GSEA analysis based on mutant TP53 upregulated genes in TCGA cancers confirms importance of cell cycle regulation (blue bars). Also see Table S4B.

Using similar approaches as described above, we also identified the most consistently upregulated genes in mutant TP53 cancers. All but one of the 20 most consistently upregulated genes were directly related to cell division regulation (Fig. 4C). Many of these top 20 genes also played roles in cell cycle promotion (particularly G2/M checkpoint control), were E2F targets (E2F is a major S phase/G2 promoting transcription factor), and were documented to be repressed by wildtype p53 protein (Fig. 4C). GSEA performed on the top 500 most upregulated genes in mutant TP53 tumors revealed highly significant enrichment of pathways directly involved in promoting cell cycle progression (Fig. 4D, Table S4B).

Differential microRNA expression was also compared, as p53 is known to transcriptionally target a number of specific miRNAs (Hermeking, 2012; Lujambio and Lowe, 2012). Of the top 20 microRNAs consistently upregulated in wildtype TP53 cancers, 6 of these had previously been shown to be direct transcriptional targets of p53 (Fig. 5A, Table S5A). A literature search of these top 20 miRNAs indicated that 16 of them had previously been associated with tumor suppressor activity in other experimental contexts. Among the 20 miRNAs consistently upregulated in mutant TP53 cancers, 9 had been experimentally associated with oncogenic function (Fig. 5B, Table S5B). An examination of the gene targets of the miRNAs upregulated in mutant and wildtype TP53 cancers suggested that in WT TP53 cancers there appeared to be enrichment for miRNAs that enhance apoptosis, p53 signaling, and hypoxia as well as suppressing cell cycle progression and NOTCH signaling (Fig. 5C). In contrast, miRNAs enriched in mutant TP53 cancers appeared to promote cell cycle progression and reduce apoptosis (Fig. 5C).

Figure 5. Comparison of global microRNA, and protein expression reveals p53-dependent pathways in cancer.

(A) Tumors with wildtype TP53 show upregulation of a subset of miRNAs relative to tumors with mutant TP53. For each tumor type those miRNAs most significantly upregulated for expression in wildtype relative to mutant TP53 cancers were determined. The top 20 most significantly upregulated miRNAs are shown. Blue rectangles indicate whether each miRNA is a direct p53 transcriptional target or has been shown to exhibit tumor suppressor activity. Pink rectangles indicate the miRNA is significantly upregulated in specific cancers. See also Table S5A. (B) Tumors with mutant TP53 show significant upregulation of a subset of miRNAs relative to tumors with wildtype TP53. For each tumor type, the top 20 miRNAs most significantly upregulated for expression in mutant TP53 cancers are shown. Green rectangles indicate miRNA oncogenic activity and blue rectangles indicate significantly upregulation in specific cancers. Also see Table S5B. (C) Schematic diagram outlining miRNAs significantly upregulated in wildtype TP53 cancers (left) and mutant TP53 cancers (right), their key target genes and proposed pathway impacts. (D) GSEA analysis of RPPA data indicates that proteins most upregulated in TCGA mutant TP53 tumors are enriched for cell cycle progression (green bars) and DNA damage response (blue bars). See Table S6B.

We also compared the TCGA RPPA data to determine whether TP53 mutation status affected protein expression (Table S6A,B). In TP53 mutant tumors, we observed significant upregulation of proteins associated with cell cycle progression, such as cyclin B1, cyclin E1, FOXM1, and CDK1 (Fig. 5D, Table S6B). As expected, mutant TP53 cancers contained enhanced expression of p53, presumably the mutant form which is known to be stabilized following a non-truncating mutation. Proteins associated with DNA damage response were also shown to upregulated in mutant TP53 cancers.

Genomic alterations correlated with TP53 mutation status vary by cancer type.

Driver oncogenic aberrations in the same pathway are rare since only one alteration is generally necessary to induce pathway signaling alteration. Thus, frequent mutually exclusive alterations within a cancer type may reveal multiple members of one signaling pathway, whereas frequent co-occurring mutations may be indicative of distinct but cooperating pathways. To catalogue the aberrations that share similar oncogenic endpoints with TP53 mutations, we identified the genomic alterations that are mutually exclusive with TP53. For this purpose, we used the Mutex algorithm, which performs a greedy search to identify modules of genetic alterations with maximized mutual exclusivity score (see Methods) (Fig. 6A). We have performed the analysis in 32 cancer types using both mutation and copy number data and detected all the mutual exclusivity modules that involve TP53. Searches in the Pathway Commons database identified known pathway interactions between TP53 and a fraction of the mutually exclusive genes (Fig. 6B). MDM2, KRAS and ARID1A are mutually exclusive with TP53 in more than one cancer type and their functional relation to TP53 has been characterized. In addition, a group of genes demonstrated mutual exclusivity with TP53 in at least one cancer type but searches in signaling databases did not capture a known functional relationship to TP53. Those genes are grouped without a connection to TP53 in the pathway diagram on Figure 6B. Among those, NUP107, FRS2 and CPM were mutually exclusive with TP53 in multiple cancer types. It is not certain whether these genes are oncogenic drivers and have a true functional relation with TP53 as they co-locate with MDM2 on the chromosome 12q, suggesting co-amplification. Mutations and copy number alterations showing mutual exclusivity with TP53 are shown in Fig. 6C. In most cases, TP53 was mutually exclusive with known driver genes. In GBM, TP53 mutations were mutually exclusive with alterations in MDM2, CDKN2A and EGFR. In breast cancers, we observed mutually exclusivity between alterations in TP53 and GATA3 in two separate modules with contributions from PI3KCA in module and MAPK3 and CDH1 in the other. Interestingly, TP53 mutations were mutually exclusive with mutations in KRAS in LUAD and with HRAS in HNSC. ARID1A mutations were mutually exclusive with TP53 mutations in UCEC and STAD, while ARID1B mutations were detected in COAD. MDM2 amplifications were mutually exclusive with TP53 in three cancer types (GBM, BLCA, SARC).

Figure 6. Genomic alterations mutually exclusive with TP53 mutations.

(A) Schematic overview of Mutex algorithm for determining gene mutual exclusivity (see Methods). (B) Pathway representation of mutually exclusive partners of TP53 across all cancer types. The color code represents the cancer type in which mutual exclusivity is observed. White indicates mutual exclusivity in more than one cancer. Genes are grouped based on their topology in the network. The disconnected gene groups on the right do not have connections to TP53 in the Pathway Commons database. (C) The oncoprint representation of TP53 mutation mutual exclusivity genomic modules. Note the partners in SARC are all amplified on the same samples, which is a chromosomal event at chromosome 12q. The p values represent significance of the observed mutual exclusivity between each gene and TP53 as calculated by a Fisher’s Exact Test. See also Fig. S6.

With respect to TP53 co-occurring mutations, we noted that IDH1 and ATRX were highly significant in LGG and GBM (Fig. S6A), as has been previously reported (Kannan et al., 2012; Liu et al., 2012). In UCEC, TP53 and PPP2R1A, a negative growth regulatory subunit of protein phosphatase 2, have highly significant co-occurring mutations, as previously noted (McConechy et al., 2012). TP53 mutations also significantly occur with certain copy number alterations, particularly those that regulate cell cycle progression and cell division (Fig. S6B). These include amplifications of CCNE1 in UCEC and STAD, CCND1 in HNSC, CCND2 in LGG, and MYC in PAAD, UCEC, and BRCA, as well as TERT in LUAD and BRCA.

Development of a prognostically useful mutant TP53 signature.

There are numerous studies examining whether TP53 mutation affects survival and in some cases TP53 mutation has been associated with poorer prognosis (Robles and Harris, 2010). In our initial stratification of the TCGA cancers into wildtype and mutant TP53 categories, we examined the effect of TP53 mutation on overall survival. For most TCGA cancer types, no statistically significant difference in overall survival was observed for mutant versus wildtype TP53 cancers (Table S7A), but this may be due to the limited clinical followup for some of the TCGA patient population. We showed that TP53 mutations across all TCGA cancers were significantly more frequent in patients with less than one year survival post-diagnosis (Fig. S7A). We assessed age at initial diagnosis for each cancer type and found significant differences only in a few cancer types (Fig. S7B). Notably, TP53 mutations in CESC associated with later diagnosis, suggesting human papillomavirus infection may be more carcinogenic than TP53 mutation (since these two events tend to be mutually exclusive) (Crook et al., 1992).

Using TP53 mutation as a prognostic marker, while useful in some contexts, may not be useful in others. For example, p53 protein function can be inactivated not only by mutation but by overexpression of key p53 regulatory proteins such as MDM2, MDM4, or PPM1D. Thus, we sought to develop an expression signature correlated with TP53 mutation that might be more prognostically predictive. We tested a mutant p53 expression signature based on the expression of four cell cycle regulatory genes consistently upregulated in mutant TP53 cancers, CDC20, CENPA, KIF2C, and PLK4 (Fig. 4D, arrows). We ranked all TCGA tumors of each type for relative RNA expression of these four signature genes and averaged their rankings for a final signature score. High signature scores correlated well with TP53 mutation status. For each cancer type with sufficient TP53 mutation numbers we then compared overall survival of patients displaying signature scores within the bottom quartile to those within the top quartile of mutant TP53 signature expression. We found that 11 of 24 cancer types with high p53 signature resulted in significantly poorer overall survival relative to their low signature counterparts (Fig. 7A–D, S7C–F, Table S7B). For 8 of these 11 cancer types, the mutant TP53 signature was distinctly better at predicting poorer outcomes than TP53 mutation status (Fig. 7B,D, S7C–F). Two representative cancers, LGG and SKCM, showed strong correlation of TP53 mutation and p53 signature (Fig. 7A,C), but p53 signature was much more prognostic than TP53 mutation status (Fig. 7B,D). We found similar prognostic results for the four gene signature in five tested non-TCGA cancer RNA expression datasets. We also applied the Web-based software program Kaplan-Meier Plotter (Nagy et al., 2018) on 16 TCGA cancer types using the four gene signature and found strong correlations between our prognosis calculations and the Kaplan-Meier Plotter-based calculations (Table S7B).

Figure 7. Development of a prognostic mutant TP53 signature.

(A) Lower grade glioma (LGG) stratified by mutant p53 expression signature correlates with a number of clinical and molecular parameters. For all LGG, RNA expression of the four genes comprising the mutant p53 signature (CDC20, PLK1, CENPA, and KIF2C) were ranked from low to high expression and combined (“Mut p53 Signature”). The bottom and top signature expression quartiles are demarcated by black vertical lines. Also indicated are tumor grade and mortality. Below the signature is shown TP53 mutation status and copy number and mutation status of a number of cancer driver genes relevant to LGG development. p values to the right indicate the significance of the difference of the low and high signature quartiles for each parameter. (B) p53 mutant signature status is more prognostically predictive than TP53 mutation status for overall survival in LGG. Log rank analysis was performed on the LGG overall survival data based on TP53 mutation status (top panel) or on mutant p53 signature status (bottom panel) in a top versus bottom quartile analysis. (C) Skin cutaneous melanoma (SKCM) stratified by p53 expression signature shows correlates with TP53 mutation and mortality. The heat maps shown here are similar to those described for panel (A), though driver genes relevant to SKCM are shown. (D) p53 mutant signature status is more prognostically predictive than TP53 mutation status for overall survival in SKCM. Overall survival in SKCM is compared based on TP53 mutation status (top panel) or on mutant p53 signature status (bottom panel) as in panel B. See also Fig. S7 and Table S7.

To facilitate the use of the mutant p53 four gene signature as a potential prognostic tool in a clinical setting, for each cancer type we normalized RNA expression values of each of the signature genes to that of a linked control gene, resulting in a four gene normalization set. Each normalization gene was chosen by (a) having a median expression virtually identical to that of its linked signature gene, (b) having an extremely low average deviation from median expression across all tumors of that type, and (c) having no correlation with expression of the signature gene across all tumors. Thus, ratios of total expression of the four signature genes to the four normalization genes would be dependent only on expression of the signature genes (Fig. S7G,H). Increasing ratios of signature to control gene expression above 1.0 would represent tumors with predicted poorer prognosis, while ratios below 1.0 would predict better prognosis (Fig. S7H). Testing of this normalization approach on all TCGA cancer types that were prognostically dependent on the four gene signature (Table S7B) showed virtually identical overall survival results between the two methodological approaches (Table S7C).

DISCUSSION

The TCGA-sponsored large scale analyses of 32 different cancer types using five high throughput data platforms integrated with relevant clinical data has provided a rich data set from which to derive important insights (Cancer Genome Atlas Research et al., 2013). Despite over 90,000 papers that have been published on p53, previous studies have generally utilized tumor samples of one cancer type and perhaps one or two experimental methodologies to examine p53 effects. The TCGA large scale integrated multi-data platform approach facilitates extraction of statistically significant patterns that might be obscured by background noise associated with smaller, less diverse data sets.

The integration of two data platforms, exome sequencing and DNA copy number, was particularly useful in the analysis of individual TP53 alleles in both wildtype and mutant TP53 tumors. First, we found that roughly 10% of tumors with TP53 mutations had two distinct TP53 mutations, and both alleles were affected in 80% of cases. Moreover, about two-thirds of tumors with a single TP53 mutation exhibited loss of the wildtype TP53 allele and a high fraction of the remaining third with diploid TP53 copy number exhibited copy neutral loss of heterozygosity. This type of wildtype TP53 loss has been previously reported (Jasek et al., 2010; Parikh et al., 2014; Saeki et al., 2011; Svobodova et al., 2016) and is likely the result of mitotic recombination or gene conversion events in the emerging tumor (Kumar et al., 2015; Stewart et al., 2012). Overall, by our assay methods, over 91% of tumors with TP53 mutations had structural loss of both TP53 alleles, which was further corroborated by analyses of p53 RNAseq data showing that over 92% of single mutation tumors exhibited RNA variant allele fractions close to 1.0. While cell culture studies indicate the mutant p53 protein can behave in a dominant negative fashion to inactivate wildtype p53 protein (Muller and Vousden, 2013)(Giacomelli et al., 2018; Muller and Vousden, 2014; Soussi and Wiman, 2015), the data presented here argue that there is still a strong selection for inactivation of the wildtype TP53 allele in tumors with a single TP53 mutation. Thus, TP53 usually behaves like a classic recessive tumor suppressor in the requirement for inactivation of both alleles (Knudson, 1996).

Genomic instability is a central characteristic of most cancers (Negrini et al., 2010). TP53, since its designation as “guardian of the genome” by David Lane in 1992 (Lane, 1992) has been known to prevent this instability. Studies in cell culture, experimental animal models, and human cancers have shown that mutations in the TP53 gene are associated with enhanced chromosomal instability due largely through loss of cell cycle checkpoint control (Donehower, 1997; Negrini et al., 2010; Smith and Fornace, 1995; Tainsky et al., 1995; Tomasini et al., 2008). However, few large scale systematic studies on genomic instability across multiple cancers have been published. Our examination of TCGA cancer types (those with sufficient numbers of tumors with TP53 mutations) revealed that 19 of 23 examined cancer types had significantly enhanced copy number instability in the MUT TP53 cohort relative to their WT TP53 counterparts. This global copy number instability was closely associated with increased amplification of known oncogenes (e.g. CCND1, CCNE1, ERBB2, MYC) and deep deletion of known tumor suppressors (RB1, PTEN WWOX). An exception to this trend was the enhanced amplification of MDM2, MDM4, and PPM1D loci in WT TP53 tumors. These three genes all encode negative regulators of p53 and thus might undergo selection in nascent WT TP53 cancer cells (Lu et al., 2008; Oliner et al., 2016; Wasylishen and Lozano, 2016). Our genomic instability results are consistent with recent experimental data indicating that p53 is directly involved in suppression of aneuploidy by engaging a ploidy sensing checkpoint that blocks proliferation of tetraploid and aneuploid cells (Dalton et al., 2010; Ganem et al., 2007; Hanel and Moll, 2012; Talos and Moll, 2010). Moreover, amplifications have been shown experimentally to be enhanced by mutant TP53 due in part to defective double-strand break repair and absence of p53-mediated apoptosis in response to proliferation of cells with double-strand DNA breaks (Hanel and Moll, 2012; Lengauer et al., 1998; Livingstone et al., 1992).

Analysis of the MUT TP53 cancers across the RNA, miRNA and protein expression data platforms consistently showed strong enhancement of pathways regulating cell cycle progression. MUT TP53 cancers showed enhanced expression of cell cycle progression genes and S phase promoting E2F target genes. The E2F results are consistent with the findings that wildtype p53 may indirectly suppress E2F1/2 through CDKN1A (p21^CIP1) activation which in turn results in suppression of RB1 phosphorylation and cell cycle inhibition (Polager and Ginsberg, 2009). P53 may also have suppressive effects on a number of E2F target genes including FOXM1, one of the highest differentially upregulated proteins in the MUT TP53 RPPA data set (Barsotti and Prives, 2009).

Studies on TP53 mutations as a prognostic marker have been historically mixed and many variables may contribute to this (Robles and Harris, 2010). One problem in these clinical studies is that there are mutationally independent mechanisms to inactivate the p53 signaling pathway. To circumvent this we searched for downstream transcriptional signatures based on RNA expression data of four genes highly and significantly upregulated across virtually all MUT TP53 TCGA tumors relative to WT TP53 tumors. In addition, these genes (CDC20, CENPA, KIF2C, and PLK1) were cell cycle promoting genes and established p53 repression targets. Eleven of 24 TCGA cancers showed significantly poorer survival with the high p53 signature and significance values were usually more robust than those observed in the previous TP53 mutation prognostic tests. Moreover, no cancer types showed poorer survival with a low p53 signature. Finally, we developed a normalization approach for each of 11 cancer types that would facilitate prognostic predictions on samples from individual patients entering the clinic. Thus, we believe this four gene RNA expression signature could serve as an improved prognostic marker and a better indicator of absence of p53 functionality in some cancer types.

In conclusion, the large scale multi-data platform approach pioneered by the TCGA effort has provided an unparalleled opportunity to better understand structural mechanisms of p53 pathway inactivation and the resulting impact on the genetics and biology of many of the cancers examined. We believe that this paper may facilitate the development of diagnostic and therapeutic tools based on a more robust knowledge of the p53 signaling pathway in cancer.

STAR METHODS

Data Sources.

All analyzes carried out in this paper are derived from The Cancer Genome Atlas Research (TCGA) Network effort. All data files have been deposited in the TCGA Pan-CanAtlas Data portal in the Synapse.org Website. All data used in the analysis reported here is from Data Freeze 1.3.1. Exome sequencing data (syn4924181) from this portal was used to identify TP53 and other relevant gene mutations. For individual TP53 DNAseq reads and RNAseq reads we accessed CGHub BAM files (https://cghub.ucsc.edu/). RNA expression data was obtained from RNAseq files (syn4557678.9 and syn4874822.6). Copy number data was downloaded from the GISTIC copy number files (Syn5049514.1). MicroRNA expression data was obtained from two files (syn45577894.9 and syn4557787.8). RPPA data (syn4557674.9) and clinical data (syn4983466.1) was also downloaded. All publicly available TCGA tumor data complies with U.S. law protecting patient confidentiality and other ethical standards.

For comparison purposes, we also utilized the UMD_TP53 mutation database. Version 2017_R1 was used for all studies. This release includes 80,406 TP53 mutations identified in tumours, cell lines (somatic mutations) or in patients with hereditary cancer (germline mutations) (database freeze Oct 2017). These mutations can be grouped into 6,874 different TP53 variants and are from studies using conventional Sanger sequencing, NGS or both. For comparison with TP53 variants from the TCGA dataset, a specific dataset (Sanger_dataset) was created by selecting only studies using Sanger sequencing. The Sanger dataset includes 37,299 TP53 mutations (4,299 variants). The database also includes functional data for most missense mutations. Residual transactivating activity for WAF (CDKN1A), MDM2, BAX, 14-3-3-sigma, AIP, GADD45A, NOXA and p53R2 promoters was originally published by Kato et al. and later used to assess TP53 variant deleteriousness (Soussi et al., 2005). The residual transcriptional activity of mutant TP53 was assayed in yeast and always compared to wild-type p53 for the same promoter. Two large-scale analysis of the functional activities of TP53 variants in mammalian cells have been released recently and are now included in the UMD TP53 database (Giacomelli et al., 2018; Kotler et al., 2018). Kotler et al. (2018) have defined the growth arrest potential of TP53 variants in H1299, a TP53 null cell line whereas Giacomelli et al. have analyzed the growth suppressive effective of TP53 in three different cellular settings including cells either WT TP53 or TP53KO. Data from both studies have been included in the UMD_TP53 database. For the purpose of all analysis, all datasets have been normalized from 0 to 1 with the lowest value corresponding to the most detrimental activity for TP53. Only missense variants issued from single nucleotide variations have been used for all comparison.

The Genome Aggregation Database (gnomAD) is a repository of SNP data from 125,748 exome sequences and 15,708 whole-genome sequences from unrelated individuals sequenced as part of various population genetic studies. It contains predominately frequent and rare non-pathogenic SNP from the normal population although a few pathogenic germline variants have been identified (Lek et al., 2016; Soussi et al., 2019).

TP53 Mutation Analyses.

All TP53 mutations were downloaded from the TCGA PanCanAltlas portal in Synapse (synapse.org). Exome sequencing data from Freeze 1.3.1 (syn4924181) was utilized for these analyses. The following types of TP53 mutations were scored as bona fide TP53 mutations for analysis purposes: non-synonymous missense mutations, indels (in frame insertions and deletions and out of frame insertions and deletions, nonsense mutations, and splice-site mutations. Synonymous missense mutations were not scored as bona fide mutations, except for NM_000546.5:c.375G>A (NP_000537.3:p.(Thr125=) and NM_000546.5:c.672G>A (NP_000537.3:p.(Glu224=) two synonymous variants which are known to be a frequent TP53 splice site mutation. T125T is known to be a frequent TP53 splice site mutation. Other mutations not considered bona fide mutations were non-exonic mutations in the 5’ UTR, the 3’ UTR, and introns.

In using the UMD_TP53 database for comparison purposes, minimal genomic information such as genomic coordinates and genetic events were extracted from each dataset to define a correct annotation using HGVS recommendations. In a second step the variant annotation were validated using the Name Checker tool developed by Mutalyzer (https://mutalyzer.nl/) (Wildeman et al., 2008). Mutalyzer handles all types of variations that can target the TP53 gene, such as substitutions, insertions, duplications, deletions, or more complex insertion/deletion. The current version of Mutalyzer (Mutalyzer 2.0.26) uses the stable NCBI sequence NG_017013.2 as a reference for TP53.

Assessment of TP53 Allele Status through integration of TP53 mutation and copy number data.

For six cancer types with high numbers of TP53 mutations (UCEC, LGG, OV, LUAD, LUSC, HNSC) we downloaded all TP53 mutations for each TP53 locus and stratified each tumor into one of two categories: (a) tumors with two or more distinct TP53 mutations, or (b) tumors with one TP53 mutation (tumors with no TP53 mutations were not further analyzed). TP53 copy number data for each tumor with TP53 mutations was then downloaded. Copy number values for the TP53 alleles downloaded from the TCGA PanCanAtlas portal in each tumor were categorized by GISTIC scores (0 = diploid, −1 = haploid, −2 = nullizygous) obtained from the Memorial Sloan-Kettering Cancer Center cBioPortal for Cancer Genomics (http://www.cbioportal.org/public-portal) and each GISTIC score aligned with its TP53 mutation status. As indicated in Fig. 2C, virtually all tumors with two or more TP53 mutations coincided with diploid copy number values. However, two thirds of tumors with one TP53 mutation showed copy number loss and had GISTIC scores of −1, while about one third displayed diploid 0 copy number scores. To determine the variant allele fraction (VAF) in each of these one TP53 mutation tumors, we obtained the total number of wildtype TP53 allele reads and mutant TP53 reads in each tumor from the exome sequencing data. We then adjusted the number of wildtype TP53 allele reads by multiplying this number by the purity fraction of that tumor. The VAF was then determined by dividing the total number of mutant alleles by the purity-adjusted total wildtype allele number. Those diploid (0) one mutation TP53 tumors with VAF less than 0.75 were considered to retain a wildtype TP53 allele (no LOH), while those with VAF greater than 0.75 were considered to be copy neutral LOH tumors (CN LOH). All haploid (−1) one mutation TP53 tumors were considered to have lost the second wildtype TP53 allele (LOH). VAF for each TP53 category (2 TP53 mutations, 1 TP53 mutation – diploid, and 1 TP53 mutation – haploid) were compared by box and whisker plots (GraphPad Prism 7) for all six cancer types as shown in Fig. 2E,F and Fig. S2A. Statistical differences in VAF for tumors with 2 TP53 mutations relative to those with one TP53 mutation were determined by two sided t test.

Analysis of p53 RNAseq Data to Determine Relative Expression of Mutant and Wildtype p53 in Tumors with TP53 Missense Mutations.

To determine relative mutant and wildtype p53 RNA expression in individual TCGA tumors, BAM files containing individual p53 reads were downloaded from the GDC portal. Individual mapped reads were obtained using the SAMtools mpileup utility. For each tumor with a documented non-synonymous TP53 missense mutation, total numbers of p53 wildtype and mutant RNA sequence reads were quantified. For each tumor, tumor purity fractions were determined by downloading tumor purity data from the TCGA Pan-CanAtlas Data portal in the Synapse.org Website. Purity adjusted p53 variant allele fractions (VAFs) were then determined for each tumor. Tumors that retain one expressed copy of WT TP53 and one equally expressed copy of mutant TP53 would be expected to display a VAF value of 0.5. Tumors that lose their wildtype TP53 allele through copy number loss or copy neutral loss should have values approximating 1.0 (or greater if tumors express p53 RNA at higher levels than adjacent normal cells in the tumor sample).

TP53 Status and Global Copy Number Alterations.

GISTIC copy number data for 25,129 individual genetic loci in each of 10,225 tumors was downloaded from the Synapse TCGA PanCanAtlas data portal. For each tumor, the copy number data was stratified by cancer type and by TP53 mutation status (either wildtype or mutant for TP53). Then, every locus was categorized by the following GISTIC copy number filters: (a) CN > +2.0 was considered indicative of amplification; (b) CN > +1.0 was considered copy number gain; (c) CN < +1.0 and > −0.5 was considered diploidy/near diploidy; (d) CN < −0.5 was considered copy number loss; and (e) CN < −1.0 was considered deep deletion. For Fig. 3A the fraction of those tumors at each locus that were in the amplification (CN > 2) and deep deletion (CN <−1) categories was determined. The fraction of copy number gains at each locus for each TP53 category were then graphed across the entire genome (Fig. 3A). The fractions of all loci in all wildtype and mutant TP53 tumors with a GISTIC copy number less than −1 or greater than +2 are shown in Fig. 3B. Wildtype and mutant TP53 tumor differences in amplification/deletion frequencies were statistically determined to be significant by chi-square test.

TP53 Status and RNA Expression-based Pathway Analysis.

Level 3 normalized global RNA expression data files for each tumor were downloaded from the TCGA PanCancerAtlas SYNAPSE portal. For each cancer type, tumors were stratified by TP53 mutation status and the RNA expression of each gene averaged for the wildtype and mutant TP53 tumor cohorts. Then, for each gene 2-sided t tests comparing RNA expression of wildtype versus mutant TP53 tumors were used to determine whether each gene was differentially expressed to a significant degree. Several cancer types had less than five patients with TP53 mutations and these were excluded from analysis. The 500 most differentially upregulated genes (determined by lowest t test p values) in wildtype and mutant TP53 tumors of each cancer type were then ranked (Table S3). Genes highly upregulated in wildtype TP53 tumors were enriched for known p53 target genes (Fig. 4A, Table S2, S3). These were identified from a manually curated list of known p53 target genes from the literature (see Table S3) This p53 target gene list contained genes identified from individual gene experimental studies as well as genes from global expression genomic studies shown to be consistently upregulated in multiple datasets (Fischer, 2017). Pathway analysis was performed on both the 500 most upregulated genes in wildtype and mutant TP53 tumors of each cancer type by Gene Set Enrichment Analysis (GSEA) (Mootha et al., 2003; Subramanian et al., 2005). Uploading of the differentially regulated 500 gene sets into the “canonical pathways” search function of GSEA resulted in ranked lists of the most significantly enriched pathways for each cancer type. The number of cancer types with wildtype TP53 showing enrichment for a particular pathway are shown in Fig. 4B and Table S4A. In the case of mutant TP53 tumors, the 500 most frequently observed upregulated genes across all mutant TP53 cancers were used to search the GSEA “canonical pathways” function (Fig. 4C,D, Table S4B).

TP53 Status and Protein Expression-based Pathway Analysis.

RPPA data was downloaded from the TCGA PanCancerAtlas SYNAPSE portal (syn4557674.9). Similar to methods used for RNA expression data, RPPA values for each protein was stratified according to the TP53 mutation status of the corresponding patient tumor. Mean protein expression levels were determined for proteins in the wildtype and mutant TP53 tumor cohorts (in cancer types with sufficient numbers of TP53 mutations) and significant upregulation of expression in wildtype and mutant TP53 cancers were statistically analyzed by two sided t tests. Wildtype and mutant TP53 tumors were compared in individual cancer types and all cancer types combined. P values for statistical differences in proteins upregulated the combined mutant TP53 cancer types are shown in Fig. 5D and Table S6B. P values for statistical differences in proteins upregulated in wildtype TP53 cancer types are shown in Table S6A.

TP53 Status and miRNA Expression-based Pathway Analysis.

MicroRNA data was downloaded from the TCGA PanCancerAtlas SYNAPSE portal (syn4557787.8). Similar to methods used for RNA expression data, expression values for each miRNA was stratified according to the TP53 mutation status of the corresponding patient tumor. Mean expression levels were determined for miRNAs in the wildtype and mutant TP53 tumor cohorts (in cancer types with sufficient numbers of TP53 mutations) and significant upregulation of expression in wildtype and mutant TP53 cancers were statistically analyzed by two sided t tests in each cancer type. Those miRNAs significantly upregulated in wildtype or mutant TP53 tumors were noted (Table S5, Fig. 5A,B). The most consistently upregulated miRNAs in wildtype and mutant TP53 tumors are indicated in Fig. 5A and 5B, respectively. For each of the top 20 most consistently upregulated genes in wildtype TP53 cancers, literature searches were used to identify those that have been shown to be direct transcriptional targets of p53 and exhibit tumor suppressor functions in experimental contexts (Fig. 5C, Table S5A). For each of the top 20 most consistently upregulated genes in mutant TP53 cancers, literature searches were used to identify those that have been shown to exhibit oncogenic or cell growth promoting functions in experimental contexts (Fig. 5C, Table S5B).

Analyses of mutually exclusive and co-occurring genomic alterations.

To understand if TP53 genomic alterations exhibit mutually exclusivity with other gene alterations in cancer patients, we used the Mutex algorithm (Babur et al., 2015) on 33 TCGA studies iteratively. For each study, we compiled an alteration matrix from detected gene mutations and copy number alterations that are also confirmed by altered gene expression, as described in the original Mutex manuscript. Mutex finds groups of genes with less overlap in gene alterations than expected by random, and provides a score for the group based on the worst fitting gene, a significant score making sure every member of the group contributes significantly to the pattern. We did not restrict the search space using pathways, used 0.01 as Mutex score cutoff, and limited maximum group size to 5. Mutex identified 46 distinct groups in 10 cancer studies containing TP53 and other 43 genes. We used ChiBE (Babur et al., 2014b; Babur et al., 2010) to retrieve known pathway relations in Pathway Commons (Cerami et al., 2011) between the genes in the result groups and filtered-in the relations of TP53. For the query, we selected relation types “controls-state-change-of”, “controls-expression-of”, “in-complex-with” and “interacts-with” and merged the last two types (Fig. 6). These binary interactions were derived from detailed processes in Pathway Commons resources using a pattern detection algorithm (Babur et al., 2014a).

To identify co-occurring genomic alterations with TP53 mutation, we utilized the cBioPortal for Cancer Genomics “Enrichments” function (www.cbioportal.org) (Cerami et al., 2012; Gao et al., 2013). By entering “TP53” and “Mutations” for each TCGA PanCanAtlas tumor type in the “Query” box, and then entering “Enrichments”, statistically prioritized lists of altered genes significantly co-occurring with TP53 mutation could be obtained, using both the “Mutations” and “Copy Number” subfunctions. For those cancer types exhibiting highly significant co-occurring gene alterations, these were entered along with TP53 into the “Oncoprint” function to obtain the oncoprints shown in Figure S6. Statistical significance of each co-occurring alteration was accompanied by q values derived from the Benjamini-Hochberg procedure.

TP53 Status and Clinical Parameter Correlations.

Clinical data (overall survival, age at first diagnosis) were downloaded from the patient data section of the TCGA PanCancerAtlas SYNAPSE portal (syn4983466.1). Overall survival data of all cancer types (with sufficient numbers of TP53 mutations) was combined and stratified into three groups: (a) dead within one year post diagnosis; (b) still alive at four or more years post-diagnosis; and (c) dead after one year post-diagnosis or still alive less than four years post-diagnosis. These three groups were then sub-stratified by type of TP53 mutation (Fig. S7A). The significance of the difference in TP53 mutation frequency patterns in the three groups was determined by chi square test. Finally, age at first diagnosis was also compared according to TP53 mutation status for each cancer type and mean values for age at first diagnosis for each cancer type and TP53 mutation category is shown in Figure S7B. For each cancer type, two sided t tests were used to determine whether TP53 mutation status had a significant effect on age of first diagnosis (Fig. S7B, Table S7A). Overall survival (still incomplete due to limited years of followup for a number of patients) was stratified by TP53 mutation status for each of the cancer types with sufficient numbers of TP53 mutations (Fig. S7C). Overall survival for patients with wildtype TP53 tumors was compared to patients with mutant TP53 tumors for each cancer type by log rank test using GraphPad Prism 7 software. Results for each of these comparisons is indicated in Table S7A.

Development and Testing of a Mutant p53 RNA Expression Signature.

The mutant p53 RNA expression signature was based on the aggregated expression of four genes, CDC20, PLK1, CENPA, and KIF2C, in the TCGA PanCancer dataset. These four genes were almost invariably significantly overexpressed in all cancer types with mutant TP53. Moreover, these genes have been established as (a) targets of wildtype p53 repression, (b) promoters of cell cycle progression, (c) components of the G2/M checkpoint, and (d) established E2F targets (with the exception of CENPA). For each individual cancer type (with sufficient numbers of TP53 mutations) each of the four signature genes was ranked by RNA expression levels. These rankings were then added together to give a combined ranking of the four genes across all tumors of a given cancer type to give relative mutant p53 signature values. Those patient tumors in the lowest quartile for mutant p53 signature expression were then compared to those patients in the highest quartile for signature expression for each cancer type. First, each cancer type was tested for how well the signature correlated with TP53 mutation status. For most cancers, tumors with TP53 mutation correlated significantly with a high mutant p53 expression signature. For each cancer type, tumors of the low and high mutant p53 signature quartiles were then compared for overall survival by log rank test. The resulting survival curves are shown in Fig. 7B,D and Fig. S7C–F and detailed in Table S7B.

Many of the TCGA cancer types were further analyzed for overall survival based on mutant p53 signature by the Web-based tool Kaplan-Meier plotter (29907753) by both top and bottom quartile splits and by median splits (comparison of tumors with signature values above and below the median signature values) (Table S7B, last column). Values from these analyses correlated well with our earlier analyses. Likewise, we performed Kaplan-Meier plotter analyses on non-TCGA tumor RNA expression datasets, including breast, ovarian, lung, gastric, and two liver carcinoma datasets and found that these survival analyses closely matched our results with TCGA RNA expression datasets.

The mutant p53 signature studies were developed from TCGA datasets subjected to fairly uniform data collection methods for each cancer type. So, application of the signature analysis to any newly accrued patient cancers might be subject to data collection methods and conditions quite different from the original TCGA data collection and analysis methods. Thus, absolute signature values derived from newly collected patient data might not correlate well with the values obtained from the original TCGA-derived analyses. To internally normalize signature values within each patient tumor sample we matched expression of each of the four mutant p53 signature genes (CDC20, PLK1, KIF2C, and CENPA) with expression of a normal control gene in that individual tumor. To qualify as a matched control gene, the gene had to meet the following criteria: (1) it had to have a nearly identical median expression as the matched signature gene; (2) it had to exhibit a very low average deviation across all tumors within a cancer type; and (3) it had to display no evidence of correlated expression with the matched signature gene (see example in Fig. S7G). For each of the 11 TCGA cancer types that showed prognostic dependence on the mutant p53 signature, we chose four normalization genes based on the above criteria and these are indicated in Table S7C. For each tumor of a given type, we then totaled the expression of the four signature genes and the four normalization control genes and determined ratios of total signature expression/total control expression for all of the tumors. Fig. S7H shows a typical plot of these ratios for TCGA lung adenocarcinomas. When we compared the quartile of tumors with the highest signature/control ratios to the lowest for overall survival by log-rank test, we saw significantly poorer survival for the high quartiles versus the low quartiles (Table S7C), correlating well with the original mutant p53 signature analyses reported in Table S7B. For most cancers, we found that signature/control ratios above 1.8 put the tumor sample in the high risk quartile, while ratios below 0.5 put the sample in the low risk quartile.

To determine whether this approach could be used for a non-TCGA dataset, we examined relative expression of the four LIHC signature and normal genes in an RNA expression dataset of 243 hepatocellular carcinomas from China and obtained very similar results to those of the 366 TCGA LIHC. Nevertheless, if the appropriate tumor RNA expression data is available, we believe it is preferable to develop normalization genes from a locally available dataset using the three criteria described above for choosing normalization genes.

Supplementary Material

Supplementary Figures

Figure S1. TP53 mutation profile in TCGA dataset (related to Fig. 1). (A) Distribution of TP53 mutations in the TCGA dataset according to specific mutation type is similar to the Sanger sequenced subset of the UMD_TP53 Mutation Database. Bar heights indicate mutation frequency at individual codons across the protein. Major TP53 hotspot mutations are indicated. (B) Similar overall fractions of mutation types are observed in the TCGA (left) and UMD_TP53 Mutation (right) databases. Infrequently observed mutation types are magnified at the top. (C,D) Venn diagrams showing the overlap between TP53 variants from the Sanger subset of the UMD_TP53 database (blue) and the TCGA dataset (green) for all TP53 variants (C) or for Single Nucleotide Variants (D). Most TP53 variants unique to the TCGA dataset are nucleotide indels (315 out of 384). (E) Unique Single Nucleotide Variants (SNVs) TP53 variants in the TCGA dataset (69) are dispersed in different parts of the TP53 gene.

Figure S2. P53 Functionality in tumors with TP53 mutations (related to Fig. 1). (A) Analysis of previously unreported TCGA-specific TP53 variant loss of function for the 24 SNVs localized in coding exons. Residual transactivating activity based on TP53 variant functional assays reported by Kato et al. (2003) for WAF (CDKN1A), MDM2, BAX, 14-3-3-σ, AIP, GADD45A, NOXA and P53R2 promoters ranges from 0 (red) to 100% (green) as indicated in the legend. The remaining transcriptional activity for four TP53 hotspot variants is shown at the bottom of the figure. (B) Analysis of functionality of all UMD_TP53 Mutation Database variants (orange box and whiskers plots), eight classes of mutant TP53 alleles based on their frequency of occurrence (indicated by frequency numbers on the X axis) in the UMD_TP53 mutation database (blue box plots), all TCGA TP53 variants (green box plots), as well as previously unreported identified TCGA variants (red box plots). Upper graphs are based on functional data analysis of transcriptional activity of mutant p53 proteins on a CDKN1A (WAF1) p53 response element (left) or an AIP1 p53 response element (right) in yeast-based assays from Kato et al. (2003). The lower left graph is based on functional data of growth suppressive activity of p53 variants in cell-based assays performed by Kotler et al. (2018). The lower right graph is based on tests for dominant-negative activity of variants in cellular functional assays performed by Giacomelli et al. (2018). The Y axis shows mean variant functionality relative to wildtype p53 based on the parameters of each of the functional assays employed.

Figure S3. Structural and functional status of both TP53 alleles in TCGA cancers with mutant TP53 (related to Fig. 2) (A) TCGA tumors with two TP53 mutations located within 75 nucleotides on the same exon were determined to be either on the same allele (Cis) or on separate alleles (Trans) by analysis of individual DNA sequence reads. Both TP53 mutations and frequencies of occurrence of each variant in the UMD p53 database in each tumor are indicated to the left of the two heat maps using p53 protein (NP_000537.3) as a reference. While each horizontal bar of the heat map corresponds to a single tumor, the columns represent transcriptional function assays for each mutant p53 on known p53-responsive gene promoters (indicated above each column) as described in the legend of Fig. S2. The legend at the bottom indicates relative transcriptional activity of each mutant p53 in the transcriptional assay. Red boxes with diagonal bars indicate TP53 truncating mutations that are assumed to have less than 10% functional activity. (B,C) Median variant allele fractions (VAF) in lung adenocarcarcinomas (B) and low grade gliomas (C) with one TP53 mutation are near 1.0, indicating frequent loss of both wildtype TP53 alleles. Tumors with TP53 mutations were stratified by mutation number and copy number. A copy number (CN) of 0 is considered diploid and copy number of −1 is considered haploid. Individual DNA sequence reads for wild and variant (mutant) TP53 were totaled, adjusted for tumor purity fraction, and variant allele fraction (VAF) determined. VAF distributions for each mutation category are shown and median values are indicated by the central bar in the box and whiskers plots. Statistical significance by t tests compare tumors with two or more (MULT) TP53 mutations versus tumors with a single TP53 mutation that are either diploid (CN 0) or haploid (CN −1) for TP53. (D) Most TCGA tumors display loss of both wildtype TP53 alleles. Based on TP53 VAF and copy number analyses described for Fig. 2E,F and Fig. S2B,C, bar graphs were generated showing fractions of various types of TP53 mutant configurations for all cancer types. TP53 mutations in the “MUT (CN 0)” category were classified as CN LOH (copy neutral TP53 mutation) if they display VAF values greater than 0.75. Those MUT TP53 tumors with VAF values less than 0.75 were classified as “NO LOH” (red segments of bar graph) and are the only mutant TP53 category not showing loss of both TP53 alleles. (E) The vast majority of TCGA tumors with missense TP53 mutations show little or no residual expression of WT p53 RNA. The individual tumor purity-adjusted TP53 variant allele fractions based on RNAseq were quantified for each of 799 TCGA tumors with non-synonymous TP53 mutations. VAF values above 0.75 were considered “high” and VAF values below 0.75 were considered “low”.

Figure S4. p53 RNA expression differs in tumors depending on TP53 mutation status (related to Fig. 4). (A) Individual tumor p53 RNA expression shows dramatic reduction of p53 RNA in those tumors with truncating TP53 mutations. RNA expression values in all breast cancers are stratified according to TP53 mutation status (wildtype, non-truncating mutation, or missense mutation) and TP53 copy number status (diploid or haploid). Mean expression levels for each TP53 category are indicated by horizontal red lines. The significant p53 RNA expression differences between tumors with non-truncating and truncating TP53 mutations (by t test) is indicated. (B) Mean p53 RNA expression in each of the three major TP53 mutation categories for each tumor type with sufficient numbers of TP53 mutations is shown. Asterisks indicate the relative significance of the difference in p53 RNA expression in tumors with truncating and non-truncating TP53 mutations. (C) Violin plots showing p53 mRNA levels in all TCGA cancers were aggregated and stratified into three groups: (a) WT TP53 tumors (blue); (b) MUT TP53 tumors containing missense/in frame TP53 mutations (yellow); and (c) MUT TP53 tumors with nonsense/frameshift/splice site TP53 mutations (black).

Figure S5. P53 protein expression varies in tumors depending on TP53 mutation status (related to Fig. 5). (A) Range of relative p53 protein expression in each of the TCGA tumor types. (B) Correlation of TP53 mutation fraction and median p53 protein expression levels in individual tumors. Red data points indicate cancers in which p53 protein levels are higher in the mutant TP53 tumors of a given type are higher than their WT TP53 counterparts. (C) TCGA breast cancers with non-truncating TP53 mutations express significantly more p53 protein than those with no TP53 mutations or truncating TP53 mutations. Mean protein expression levels for each TP53 category are indicated by horizontal red lines. An unpaired t test was used to show significance of the differences in p53 protein expression between tumors with truncating and non-truncating mutations in TP53. (D) Violin plots showing p53 protein levels in all TCGA cancers (RPPA data) were aggregated and stratified into three groups: WT TP53 tumors (blue); MUT TP53 tumors containing missense/in frame TP53 mutations (yellow); and MUT TP53 tumors with nonsense/frameshift/splice site TP53 mutations (black).

Figure S6. Genomic alterations that significantly co-occur with TP53 mutations in individual cancer types (related to Fig. 6). (A) The cBioPortal for Cancer Genomics algorithms were used to generate oncoprints showing gene mutations that significantly co-occur with TP53 mutations across all samples in six TCGA cancer types. Types of mutations arising in TP53 and co-occurring genes are designated in the legend at the bottom. For missense mutations and in frame mutations the darker squares indicate likely driver mutations and the lighter squares indicate mutations of unknown driver potential. Significance of the co-occurrences are indicated by q values to the right. (B) Deep deletions and amplifications that significantly co-occur with TP53 mutations in eight cancer types. Types of gene copy number alterations are designated in the legend. Significance of the co-occurrences are indicated by q values to the right.

Figure S7. Effects of TP53 mutation status and mutant p53 signature on overall survival (related to Fig. 7). (A) The clinical data for 20 TCGA tumor types with sufficient fractions of TP53 mutations were aggregated and classified into three survival categories: (i) patients who died within one year of initial diagnosis, (ii) patients who were alive 4 years or more after initial diagnosis, and (iii) all other patients. The fraction of each patient survival category that showed TP53 mutation was then determined. Types of TP53 mutation for each patient category were also determined and these are indicated by different colors in each bar (legend at the bottom of panel A). (B) The TCGA clinical data was also examined for all cancer types with sufficient TP53 mutation numbers and the age at initial diagnosis of cancer determined. This data was then stratified by TP53 mutation status. Significant TP53 mutation-dependent differences in age at initial diagnosis for a particular cancer type are noted by asterisks (legend). (C-F) For some TCGA cancers, stratification by mutant p53 signature is more prognostically predictive of overall survival than stratification by TP53 mutation status. For each cancer, overall survival is compared by log rank analysis using stratification by TP53 mutation status (left) or by mutant p53 signature status (right). Log rank p values are given for each comparison. (C) KIRC (kidney renal clear cell carcinoma), (D) KIRP (kidney renal papillary carcinoma), (E) LIHC (liver hepatocellular carcinoma), (F) PAAD (pancreatic adenocarcinoma). (G) Normalization of the mutant p53 signature values to expression of four normal control genes within each tumor facilitates evaluation of individual patient tumors. Total RNA expression values in each of 511 lung adenocarcinomas for the four genes of the mutant p53 signature (CDC20, PLK1, KIF2C, CENPA, blue line) are compared to expression of the normalization genes (UBIAD1, ZMYM6, RPP30, ZNF17, brown line). (H) Ratios of expression in lung adenocarcinomas of four mutant p53 signature genes and four normalization genes from panel G facilitate stratification of high prognosis tumors and low prognosis tumors. Log rank survival analysis of high and low quartiles in this graph gave p values <0.0001 with a high/low hazard ratio of 2.372. See also Table S7C.

Supplementary Tables

Table S1. Genes most frequently deleted (A) and amplified (B) in the PanCan dataset. Related to Fig. 1.

Table S2. Genes showing most differential RNA expression in WT and MUT TP53 cancers. Related to Fig. 4.

Table S3. Ranking of p53 target gene expression in TCGA WT TP53 cancers. Related to Fig. 4.

Table S4. Pathway enrichment in WT (A) and MUT (B) TP53 cancers based on RNA expression. Related to Fig. 4.

Table S5. Analysis of major upregulated miRNAs in WT (A) and MUT (B) TP53 cancers. Related to Fig. 5.

Table S6. Proteins upregulated in WT (A) and MUT (B) TP53 cancers based on RPPA. Related to Fig. 5.

Table S7. Comparison of clinical attributes and outcomes in WT and MUT TP53 cancers (A) and in high and low mutant p53 signatures prior to (B) and after (C) normalization to control genes. Related to Fig. 7.

SIGNIFICANCE.

The TCGA-sponsored multi-data platform analyses of 32 different cancer types has provided a unique opportunity to examine cancer signaling pathways. One key pathway often dysfunctional in human cancers is the p53 signaling pathway. We integrated results from five of the TCGA data platforms in conjunction with clinical data to compare cancers with and without TP53 mutation. Loss of both functional TP53 alleles occurs in over 90% of cancers with TP53 mutations. TP53 mutation was highly correlated with DNA copy number instability, particularly increases in amplification and deep deletion events. In addition, cancers with TP53 mutations displayed distinctive gene expression patterns and such patterns were used to produce an expression signature that is prognostically predictive in at least 11 cancer types.