Genomic Analysis Revealed New Oncogenic Signatures in TP53-Mutant Hepatocellular Carcinoma

The TP53 gene is the most commonly mutated gene in human cancers and mutations in TP53 have been shown to have either gain-of-function or loss-of-function effects. Using the data generated by The Cancer Genome Atlas, we sought to define the spectrum of TP53 mutations in hepatocellular carcinomas (HCCs) and their association with clinicopathologic features, and to determine the oncogenic and mutational signatures in TP53-mutant HCCs. Compared to other cancer types, HCCs harbored distinctive mutation hotspots at V157 and R249, whereas common mutation hotspots in other cancer types, R175 and R273, were extremely rare in HCCs. In terms of clinicopathologic features, in addition to the associations with chronic viral infection and high Edmondson grade, we found that TP53 somatic mutations were less frequent in HCCs with cholestasis or tumor infiltrating lymphocytes, but were more frequent in HCCs displaying necrotic areas. An analysis of the oncogenic signatures based on the genetic alterations found in genes recurrently altered in HCCs identified four distinct TP53-mutant subsets, three of which were defined by CTNNB1 mutations, 1q amplifications or 8q24 amplifications, respectively, that co-occurred with TP53 mutations. We also found that mutational signature 12, a liver cancer-specific signature characterized by T>C substitutions, was prevalent in HCCs with wild-type TP53 or with missense TP53 mutations, but not in HCCs with deleterious TP53 mutations. Finally, whereas patients with HCCs harboring deleterious TP53 mutations had worse overall and disease-free survival than patients with TP53-wild-type HCCs, patients with HCCs harboring missense TP53 mutations did not have worse prognosis. In conclusion, our results highlight the importance to consider the genetic heterogeneity among TP53-mutant HCCs in studies of biomarkers and molecular characterization of HCCs.

The TP53 gene is the most commonly mutated gene in human cancers and mutations in TP53 have been shown to have either gain-of-function or loss-of-function effects. Using the data generated by The Cancer Genome Atlas, we sought to define the spectrum of TP53 mutations in hepatocellular carcinomas (HCCs) and their association with clinicopathologic features, and to determine the oncogenic and mutational signatures in TP53-mutant HCCs. Compared to other cancer types, HCCs harbored distinctive mutation hotspots at V157 and R249, whereas common mutation hotspots in other cancer types, R175 and R273, were extremely rare in HCCs. In terms of clinicopathologic features, in addition to the associations with chronic viral infection and high Edmondson grade, we found that TP53 somatic mutations were less frequent in HCCs with cholestasis or tumor infiltrating lymphocytes, but were more frequent in HCCs displaying necrotic areas. An analysis of the oncogenic signatures based on the genetic alterations found in genes recurrently altered in HCCs identified four distinct TP53-mutant subsets, three of which were defined by CTNNB1 mutations, 1q amplifications or 8q24 amplifications, respectively, that co-occurred with TP53 mutations. We also found that mutational signature 12, a liver cancer-specific signature characterized by T>C substitutions, was prevalent in HCCs with wild-type TP53 or with missense TP53 mutations, but not in HCCs with deleterious TP53 mutations. Finally, whereas patients with HCCs harboring deleterious TP53 mutations had worse overall and disease-free survival than patients with TP53-wild-type HCCs, patients with HCCs harboring missense TP53 mutations did not have worse prognosis. In conclusion, our results highlight the importance to consider the genetic heterogeneity among TP53-mutant HCCs in studies of biomarkers and molecular characterization of HCCs.
TP53 is the most frequently mutated gene in human cancers (Kandoth et al., 2013). The p53 protein modulates multiple cellular functions, including transcription, DNA synthesis and repair, cell cycle arrest, senescence and apoptosis (Vogelstein et al., 2000). Mutations in TP53 can abrogate these functions, leading to genetic instability and progression to cancer (Vogelstein et al., 2000). Across 12 major cancer types (excluding HCC), 42% of cancers harbored TP53 somatic mutations, with at least 20% mutational rate in 10/12 cancer types and TP53 mutations are associated with inferior prognosis and unfavorable clinicopathologic parameters, such as tumor stage (Kandoth et al., 2013). Furthermore, TP53-mutant tumors are highly enriched among tumors driven by copy number alterations (CNAs), with most remaining TP53-mutant tumors associated with the presence of somatic mutations in the Wnt and/or the RAS-RAF-ERK signaling pathways (Ciriello et al., 2013).
The pattern of TP53 mutations is reminiscent of both an oncogene and a tumor suppressor gene (Vogelstein et al., 2013). The majority (86%) of TP53 mutations are in the DNAbinding domain (Olivier et al., 2010;Kandoth et al., 2013). Most mutations in the DNA-binding domain are missense (88%) and approximately 1/3 of missense mutations affect the hotspot residues R175, G245, R248, R249, R273, and R282 (Olivier et al., 2010). Outside the DNA-binding domain, most mutations (∼60%) are nonsense or frameshift (Olivier et al., 2010). Mutant p53 proteins may lose the tumor suppressive functions and exert dominant-negative activities, but may also gain new oncogenic properties (Olivier et al., 2010;Muller and Vousden, 2014). Indeed, on the immunohistochemical level, p53 is generally detectable to various extents in samples with missense mutations but is undetectable in samples with truncating or frameshift mutations (Hall and McCluggage, 2006;Soussi et al., 2014).
Finally, molecular classification studies have invariably grouped TP53-mutant HCCs under the umbrella of the aggressive subclass, but it is also clear that this subclass is molecularly, biologically and clinically heterogeneous (Boyault et al., 2007;Hoshida et al., 2009;Goossens et al., 2015).
Given the diverse pattern of TP53 mutations, taking advantage of The Cancer Genome Atlas (TCGA) dataset, in this study we sought to determine the pattern of TP53 somatic mutations in HCCs and its association with clinicopathologic features. Additionally, as TP53 mutations are associated with HCC molecular subclasses with poor prognosis, we sought to define the oncogenic and mutational signatures among TP53-mutant HCCs.

Sample Selection and Histologic Assessment
From TCGA liver hepatocellular carcinoma (LIHC) project (The Cancer Genome Atlas Research Network, 2017), 373 tumors with available somatic mutational data 1 (accessed April 2017) (Gao et al., 2013) were included in the study. Images of diagnostic hematoxylin & eosin (H&E) slides were retrieved from the cbioportal and reviewed by three expert hepatopathologists (SA, MSM and LMT) according to the guidelines by the World Health Organization (Bosman et al., 2010) to define the presence or absence of cholestasis, Mallory bodies, tumor infiltrating lymphocytes (TILs), vessel infiltration and necrotic areas. 4-point scale Edmondson and Steiner system was adopted for tumor grading as previously described (Edmondson and Steiner, 1954;Alexandrov et al., 2013). Clinical information were obtained from the cbioportal (Gao et al., 2013).

Classification of TP53 Somatic Mutations
TP53 somatic non-synonymous and splice region mutations for the 373 HCCs were retrieved from the cbioportal (accessed April 2017) (Gao et al., 2013). TP53 mutations were stratified according to (i) the mutation type as single-nucleotide missense mutations (also encompassing synonymous mutations affecting splice region, Supplementary Methods and Supplementary Table S1) or deleterious mutations (encompassing splice site, nonsense, in-frame, and frameshift mutations); (ii) whether the mutations were within or outside of the DNA-binding domain. For correlative analyses with clinicopathologic parameters, the sample (TCGA-DD-A1EE) with three TP53 mutations (A161S, H193R and C277 * ) was classified as harboring deleterious mutation.

Genomic and Transcriptomic Data Analysis
Gene-level copy number ("gistic2_thresholded, " 370/373 samples) and expression ("IlluminaHiSeq, " 367/373 samples) data were retrieved from the UCSC Xena Functional Genomics Browser 2 accessed April 2017). Gene-level copy number data were used to define genomic regions with differential frequencies of copy number alterations between HCCs with missense TP53 mutations, with deleterious TP53 mutations, or with wild-type TP53. Copy number states −2, −1, 0, 1, and 2 were considered homozygous deletion, heterozygous loss, copy number neutral, gain and high-level gain/amplification, respectively.
Transcriptomic data were in the form of gene-level, log-transformed, upper-quartile-normalized RSEM values. Molecular classification was performed according to Hoshida et al. (2009), using the Nearest Template Prediction: http://software.broadinstitute.org/cancer/software/genepattern. The R package limma was used to perform quantile normalization and for differential expression analysis. Multiple correction was performed using the Benjamini-Hochberg method. Genes with adjusted P-value < 0.05 were considered as differentially expressed.
The number of somatic mutations per sample was obtained from the cbioportal (Gao et al., 2013

Mutational Signatures
Decomposition of mutational signatures was performed using deconstructSigs (Rosenthal et al., 2016), based on the set of 30 mutational signatures ("signature.cosmic") (Alexandrov et al., 2013;Nik-Zainal et al., 2016), for the 358 samples with at least 30 somatic mutations. Mutational signatures with >20% weight were considered to have substantial contribution to the overall mutational landscape. For each sample, the mutation signature with the highest weight was considered the dominant mutational signature.

Statistical Analysis
Associations between TP53 mutations and clinical/histologic features were assessed using Mann-Whitney U, Chi-squared or Fisher's exact tests as appropriate. Survival analyses were performed using the Kaplan-Meier method and the log-rank test. Univariate and multivariate analyses for OS and DFS were performed using the Cox proportional-hazards model. Mutual exclusivity and co-occurrence of somatic mutations were defined using the cbioportal (Gao et al., 2013). Statistical analyses comparing copy number profiles and defining genes up-regulated when gained or amplified and genes down-regulated when lost were performed as previously described (Supplementary Methods) (Piscuoglio et al., 2014). All tests were two-sided. P < 0.05 were considered statistically significant. Statistical analyses were performed with R v3.1.2 or SPSS v24 (IBM, Münchenstein, CH).

Clinicopathologic Characterization and Molecular Classification of HCCs
TP53 mutation status was available for 373 HCCs subjected to whole-exome sequencing by TCGA (The Cancer Genome Atlas Research Network, 2017). Analysis of the clinical details of the patients revealed that the median age at diagnosis was 61 (range 16-90) and that 67.5% were male (Supplementary Table S3). Half of the patients were Caucasian (50.8%), with most remaining patients being Asian (43.9%). The most frequent primary risk factor was alcohol consumption (33.1%), followed by HBV (30.0%) and hepatitis C virus (HCV) infection (15.9%). Overall, history of at least one primary risk factor was noted in 74.2% patients (Supplementary Table S3).
We performed a comprehensive histopathologic review of the diagnostic H&E slides for all 373 included cases to assess Edmondson grade, the presence of cholestasis, Mallory bodies, vessel infiltration, necrotic areas, and TILs (Figure 1 and Supplementary Table S3). Most samples were of intermediate grade, with 33.2, 60.6, and 5.4% graded as of Edmondson grades 2, 3, and 4, respectively. No sample was classified as of Edmondson grade 1. Cholestasis, Mallory bodies, vessel infiltration, necrotic areas, and TILs were present in 21.6, 22.0, 34.1, 24.8, and 47.3% of cases, respectively.
Molecular classification was performed for the 367 HCCs for which expression data were available according to Hoshida et al. (2009). 31.3, 21.5, and 47.2% of HCCs were classified as S1, S2 and S3, respectively (Supplementary Table S3).

Spectrum of TP53 Somatic Mutations in HCCs
Given that TP53 is one of the most frequently mutated genes in HCCs and its diverse spectrum of mutations in human cancers, we sought to define the spectrum and type of TP53 mutations found in HCCs. A total of 116 somatic non-synonymous TP53 mutations and 2 synonymous TP53 mutations affecting splice regions were identified in 115 (30.8%) cases, including one case with three distinct mutations and one case with two. Missense (including missense and synonymous mutations affecting splice region, Supplementary Methods and Supplementary Table S1) and deleterious (including nonsense, frame-shift, in-frame, splice site) mutations accounted for 73 (62%) and 45 (38%), respectively (Figure 2). Compared to other cancer types characterized by the TCGA, there was no difference between HCC and non-HCC tumor types in terms of the ratio of missense vs deleterious mutations (P = 0.197, Fisher's exact test).
Our results demonstrate that the spectrum of TP53 mutations in HCCs is distinct from that in non-HCC tumors, with HCC-specific recurrent hotspot mutations and a near absence of highly recurrent TP53 mutations found in other cancer types.

TP53 Status Correlates with Specific Histopathologic and Clinical Features of HCCs
Next, we sought to define whether TP53 mutation status correlated with clinicopathologic parameters. TP53 mutations were more frequently found in male patients ( Correlation with histologic features revealed that TP53mutant HCCs were associated with high Edmondson grade, accounting for 12.1, 38.5, and 65.0% of cases classified as Edmondson grades 2, 3, and 4, respectively (P < 0.001, Chi-squared test, Table 1). TP53 mutations were less frequent in HCCs associated with cholestasis (17.5% vs. 38.4%; P = 0.003, Fisher's exact test) and were more frequent in HCCs with necrotic areas (43.5% vs. 26.9%; P = 0.004, Fisher's exact test, Table 1). The presence of TILs was associated with less frequent TP53 mutations (37.4% vs. 62.6%; P = 0.013, Fisher's exact test; Table 1). No association was found between TP53 mutation status and the presence of Mallory Bodies or vessel infiltration.
Correlating TP53 status with molecular classification, (Hoshida et al., 2009) TP53-mutant HCCs were preferentially enriched in the S1 and S2 subclasses (36.5% and 42.5% vs. 21.8% in S3, P = 0.001, Chi-squared test, Table 1). Stratifying TP53-mutant HCCs into those with missense or deleterious mutations did not reveal association between TP53 mutation types and molecular classification (P = 0.459, Chi-squared test, Supplementary Table S4). These results demonstrate that, additional to the wellestablished associations with the male gender, HBV/HCV infection and high Edmondson grade, TP53 mutations were less frequent in HCCs with cholestasis or TILs, but were more frequent in HCCs with necrotic areas.

Genomic Instability Is Not Associated with TP53 Mutation Type
Next, we compared the number of somatic genetic alterations between TP53-wild-type and mutant cases. Mutational burden was higher in TP53-mutant HCCs, HCCs with missense TP53 mutations and HCCs with deleterious TP53 mutations than TP53-wild-type cases (P < 0.001, P < 0.001 and P = 0.004, respectively, Mann-Whitney U tests), but no difference was observed between cases with missense or deleterious mutations (P = 0.799, Mann-Whitney U test, Supplementary Figure S1A). Similarly, TP53-mutant HCCs, HCCs with missense TP53 mutations and HCCs with deleterious TP53 mutations all harbored higher number of genes affected by CNAs compared with TP53-wild-type cases (P < 0.001, P < 0.001 and P = 0.001, respectively, Mann-Whitney U tests, Supplementary Figure S1B), with no difference between cases with missense or deleterious TP53 mutations (P = 0.352, Mann-Whitney U test, Supplementary Figure S1B). Consistent with their increased chromosomal instability, TP53-mutant HCCs displayed more frequent gains of chromosomes 1p, 3, 10p and 19p and losses of half the genome, notably of chromosomes 4, 5, 10q, 14, 17p, 18 and 19 (Supplementary Figures S2A-C). The CNA landscapes between HCCs with TP53 missense or deleterious mutations were remarkably similar (Supplementary Figure S2D).
To identify potential CNA drivers associated with TP53 mutations, we interrogated the genes overexpressed when gained and genes downregulated when lost in the regions that showed differential CNA frequencies between TP53-mutant and TP53wild-type cases (Supplementary Figure S2A). Pathway analysis of the copy number-regulated genes revealed that TP53-mutant cases displayed deregulation in pathways associated with EIF2 signaling, protein ubiquitination pathway, RNA polymerase-II complex and DNA repair pathways, and in molecular and cellular functions related to cell death and survival, cell cycle, DNA replication, recombination and repair (Supplementary Figure S3).
To define the oncogenic signatures in TP53-mutant HCCs, we performed unsupervised partitioning of the samples into classes with distinct patterns of likely 'driver' genetic alterations (or 'selected functional elements, ' SFEs), (Ciriello et al., 2013) including mutations in 29 significantly mutated genes, amplifications in 27 recurrently amplified regions, and homozygous deletions in 34 recurrently deleted regions (see Materials and Methods). Among the 144 TP53-mutant HCCs with mutational and CNA data, we found median of 2 mutational (range 0-11) and 2.5 CNA (range 0-13) SFEs in each case and identified four robust oncogenic signature classes (OSCs, Figures 3B-E and Supplementary Figure S4A). HCCs with missense or deleterious TP53 mutations did not cluster separately (P = 0.305, Chi-squared test, Figure 3B), nor HCCs of distinct transcriptomic subclasses (Supplementary Figure S4B).
These observations are concordant with the observation that tumors are primarily driven by either somatic mutations or CNAs but rarely both (Ciriello et al., 2013) (Figure 3E and Supplementary Figures S4C,D). Furthermore, we identified subclasses of TP53-mutant HCCs likely driven by co-occurring CTNNB1 mutations, 8q24.21 (MYC) amplification or 1q amplification in a mutually exclusive manner.

Mutational Signatures in TP53-Mutant HCCs
The somatic mutational landscapes are shaped by endogenous and/or environmental biological and chemical processes (Alexandrov et al., 2013). More than 10 mutational signatures have been identified in liver cancers, including two liver cancerspecific signatures 12 and 16 of unknown etiology, both of which are characterized by frequent T>C substitutions but with different sequence contexts (Alexandrov et al., 2013).
Taken together, our results suggest that the different types of TP53 mutations were associated with distinct mutational processes. Specifically, signature 12 was rarely found in HCCs with deleterious TP53 mutations.

Distinct Types of TP53 Mutations Are Associated with Different Prognoses
Previous studies found that associations between the types of TP53 mutations and prognoses in breast, and head and neck cancers (Olivier et al., 2006;Ozcelik et al., 2007;Vegran et al., 2013;Lapke et al., 2016). Here we hypothesized that patients with HCCs harboring TP53 missense or deleterious mutations may display different prognoses. Considering the patients with available data on OS (n = 372) or DFS (n = 321), we found that patients with TP53-mutant HCCs displayed a more aggressive behavior including shorter OS and DFS than TP53-wild-type patients (P = 0.018 and P = 0.005, respectively, log-rank tests, Figure 5). Patients with missense or deleterious TP53 mutations did not differ in OS or DFS (P = 0.129 and P = 0.148, respectively, log-rank tests, Figure 5). Importantly, while patients with deleterious TP53 mutations had worse OS and DFS than TP53-wild-type patients (P = 0.004 and P = 0.001, respectively, log-rank tests, Figure 5), there was no difference in OS or DFS between patients with missense TP53 mutations and those wildtype for TP53 (P = 0.192 and P = 0.084, respectively, log-rank tests, Figure 5). As an exploratory analysis, we asked whether OSCs or mutational signatures of TP53-mutant HCCs were prognostic. Compared to OSC1 (28 months), OSC2 (26 months) and OSC3 (median not reached), OSC4 was associated with the shortest median OS of 14 months, although the difference was not statistically significant (P = 0.366, log-rank test; Supplementary Figure S4E). Univariate Cox regression analyses revealed that the aflatoxin-associated signature 24 (HR 3.275,, P = 0.013), HBV infection status and the presence of necrotic areas were associated with poor prognosis (Supplementary Table S5). However, in a multivariate analysis, mutational signature 24 was not an independent prognostic indicator (P = 0.242; Supplementary  Table S5).
Taken together, our results showed only patients with deleterious TP53 mutations but not missense TP53 mutations were associated with significantly worse OS and DFS in this cohort.
FIGURE 5 | TP53 mutation status is associated with worse overall and disease-free survival. Overall (A) and disease-free survival (B) of HCC patients with and without TP53 somatic mutations using the Kaplan-Meier method. Median survival for each group is indicated in parentheses. Statistical comparisons were performed using log-rank tests. P < 0.05 was considered statistically significant.

DISCUSSION
In this study, we performed a detailed analysis of TP53 somatic mutational spectrum in HCCs, with nearly all missense mutations (98%) and most deleterious mutations (73%) affecting the DNA-binding domain. Notably, we found that the residues mutated in HCCs differed from those in other cancer types. Hotspot mutations R249S and V157F were common in HCCs but extremely rare in other cancers, while mutations affecting R175 and R273, two of the most frequently mutated residues in other cancers, were nearly absent in HCCs. This latter observation also applies to other HCC datasets (Ahn et al., 2014;Schulze et al., 2015), suggesting that TP53 mutational spectrum in HCC is distinct from that in other cancers.
To determine the genotype-phenotype correlation between TP53 mutation status and clinicopathologic parameters, we performed a detailed assessment of histologic features using H&E slides. We confirmed the established associations with the male gender, HBV/HCV infection and high Edmondson grade. Additionally, TP53 mutations were associated with the presence of necrotic areas, and accordingly, with the absence of cholestasis, a feature more frequently observed in well-differentiated HCCs. Finally, we observed that the presence of TILs was associated with less frequent TP53 mutations, in line with the favorable prognosis associated with tumors with high TILs in other tumor types (Mahmoud et al., 2011).
Analysis of the mutational signatures revealed that signatures 16 of unknown etiology and the age-associated signature 5  were the most prevalent in HCCs. We also found that signature 12 of unknown etiology, characterized by frequent T>C substitutions, was prevalent in TP53-wildtype and HCCs with missense TP53 mutations but were largely absent in those with deleterious TP53 mutation. A previous study reported that the W3 signature, which was highly similar to signature 12 (Fujimoto et al., 2012), was associated with the age of patients. Here we found no difference in the age of patients when we considered tumors with strictly missense or deleterious TP53 mutations (i.e., excluding one patient with both types). The basis of signature 12 is thus unclear and further studies are required to elucidate its biological significance.
Adopting the algorithm of "oncosign" (Ciriello et al., 2013), we identified four robust subclasses of TP53-mutant HCCs with distinct oncogenic signatures. Of these classes, one subclass was likely driven by co-occurring CTNNB1 mutations, while two subclasses were likely driven by amplicon drivers on 1q and 8q. 1q21 amplification has been linked to hepatocarcinogenesis, with ALC1 (CHD1L) overexpression in HCC cells shown to promote G1/S phase transition and to inhibit apoptosis (Ma et al., 2008). The authors further suggested that the oncogenic function of ALC1 might be associated with its role in promoting cell proliferation by down-regulating p53 expression (Ma et al., 2008). The 1q21 amplicon also contains HORMAD1, a gene that has been shown to drive chromosomal instability in breast cancer (Watkins et al., 2015). As for 8q24, in addition to the well-known oncogenic role of MYC, previous studies have also shown that MYC amplification is an indicator of malignant potential and poor prognosis in HCC (Lin et al., 2010), and that the co-occurrence of MYC amplification and p53 alteration may contribute to HCC progression (Kawate et al., 1999). The remaining subclass did not have highly recurrent genetic alterations. Interestingly, this subclass was numerically, though not statistically, associated with the most favorable OS among the four classes. One may speculate that TP53-mutant HCCs lacking additional drivers may constitute a less aggressive subclass. Of note, the features that characterized the four OSCs were largely mutually exclusive, suggesting that distinct oncogenic processes are operative in non-overlapping subsets of TP53mutant HCCs.
TP53 mutation status predicts worse OS and DFS in HCC patients (Yano et al., 2007;Woo et al., 2011;Cleary et al., 2013). However, we found that patients with deleterious mutations, but not those with missense mutations, were associated with worse OS and DFS compared to patients wild-type for TP53. This is in line with other tumor types, in which different types of TP53 mutations have been associated with different prognoses (Olivier et al., 2006;Ozcelik et al., 2007;Vegran et al., 2013;Lapke et al., 2016). In fact, the risk of death or relapse for patients harboring deleterious mutation is 2.3 times (HR = 2.36 and 2.063, respectively) higher than TP53-wild-type patients. The prognosis for patients with missense mutations appears to sit between those with wild-type TP53 or deleterious TP53 mutations, albeit not statistically different from either group. It is conceivable that the prognostic significance of the type of TP53 mutations may be confirmed in a larger cohort with extensive follow-up.
It has been suggested that TP53 missense mutations have varying capacity to transactivate p53 target genes and to alter the responsiveness to chemotherapeutic agents in breast cancer (Jordan et al., 2010). A differential expression analysis using the HCC TCGA dataset comparing HCCs with TP53 missense mutations and those with TP53 deleterious mutations identified TP53 itself as up-regulated but did not identify significantly altered genes (data not shown). Furthermore, HCCs harboring the missense mutations functionally shown to lack the ability to transactivate genes with p53 response elements (Jordan et al., 2010) did not differ from HCCs with other missense mutations on the transcriptomic level (data not shown). It is thus unclear precisely how the various TP53 mutations may differentially alter the transcriptomic landscape of HCCs. Further functional studies may be required to elucidate how the types of TP53 mutations may affect its biological functions.
In HCC molecular characterization studies to date, HCCs are typically classified as TP53-wild-type or TP53-mutant, where all TP53 mutations were treated as equal (Fujimoto et al., 2012;The Cancer Genome Atlas Research Network, 2017). However, many studies have demonstrated that TP53 can be affected by either (or both) gain-of-function or loss-of-function mutations, with missense mutations preferentially displaying gain-offunction or neomorphic properties (Muller and Vousden, 2014). Our study has demonstrated that HCCs with missense or deleterious TP53 mutations display similar clinicopathologic features, mutational/CNA burden and oncogenic signatures, but are associated with distinct mutational signatures. Clinically, while patients with tumors harboring deleterious TP53 mutations had worse prognosis compared to those wild-type for TP53, there was no statistically significant difference between those with missense mutations and those wild-type for TP53. Our study highlights the importance to consider the type of TP53 mutations in studies of biomarkers and molecular characterization of HCCs.
Our study has limitations. Despite TCGA being the largest genomic study of HCC, it is by no means the only largescale study. However, as one of our aims was to define clinicopathologic correlates, we chose TCGA as it is the only study with publicly available H&E slides for pathology review. Secondly, the power of the OS and DFS analyses was limited due to the cohort size. Further studies may reveal whether prognosis is related to the type of TP53 mutations, as has been shown in other cancers. Thirdly, our analyses did not consider the noncoding genome due to the nature of the sequencing performed by the TCGA. Given the frequent mutations in non-coding regions such as TERT promoter, MALAT1 and NEAT1 (Fujimoto et al., 2012;Schulze et al., 2015), it is conceivable that additional oncogenic signatures within TP53-mutant HCCs may emerge.

CONCLUSION
Our study highlights the genetic heterogeneity among TP53mutant HCCs and that patients with HCCs harboring different types of TP53 mutations may be associated with distinct prognoses. Future work will be required to elucidate whether the co-occurring genetic alterations act synergistically with TP53 mutations to promote carcinogenesis in HCCs.

ACKNOWLEDGMENTS
The authors acknowledge that the manuscript has been accepted and presented as e-poster at the International Liver Cancer Association 2018. However, the work has been expanded and only a small part overlaps with the one presented at the mentioned congress.