More targets, more pathways and more clues for mutant p53.

Mutations in the transcription factor p53 are among the most common genetic alterations in human cancer, and missense p53 mutations in cancer cells can lead to aggressive phenotypes. So far, only few studies investigated transcriptional reprogramming under mutant p53 expression as a means to identify deregulated targets and pathways. A review of the literature was carried out focusing on mutant p53-dependent transcriptome changes with the aims of (i) verifying whether different p53 mutations can be equivalent for their effects, or whether there is a mutation-specific transcriptional reprogramming of target genes, (ii) understanding what is the main mechanism at the basis of upregulation or downregulation of gene expression under the p53 mutant background, (iii) identifying novel candidate target genes of WT and/or mutant p53 and (iv) defining cellular pathways affected by the mutant p53-dependent gene expression reprogramming. Nearly 600 genes were consistently found upregulated or downregulated upon ectopic expression of mutant p53, regardless of the specific p53 mutation studied. Promoter analysis and the use of ChIP-seq data indicate that, for most genes, the expression changes could be ascribed to a loss both of WT p53 transcriptional activation and repressor functions. Pathway analysis indicated changes in the metabolism/catabolism of amino acids such as aspartate, glutamate, arginine and proline. Novel p53 candidate target genes were also identified, including ARID3B, ARNT2, CLMN, FADS1, FTH1, KPNA2, LPHN2, PARD6B, PDE4C, PIAS2, PRPF40A, PYGL and RHOBTB2, involved in the metabolism, xenobiotic responses and cell differentiation.

Introduction

The tumor suppressor p53 is a 393-amino-acid nuclear phosphoprotein that responds to numerous stress stimuli, including DNA damage[1] and hypoxia.[2] Following homotetramerization, it acts as a transcription factor[3] and modulates the expression of a variety of genes, leading to enhanced DNA repair, control of cell cycle and apoptosis, and maintaining cellular homeostasis.[4, 5, 6] The p53 targets are only partially known, with assessments suggesting their number to be nearly 2000 genes.[7]

CDKN1A, MDM2, BAX, GADD45 and BBC3 are paradigmatic examples of upregulated target genes where p53 exerts its activity via evolutionarily conserved cis-response elements (p53RE).[8] The importance of p53 for the biology of cancer is evident by the fact that colorectal, breast and most other human solid tumors show a high frequency of somatic mutations within the TP53 gene (www.iarc.fr/p53). Moreover, germline mutations within TP53 cause the Li–Fraumeni syndrome, a dominantly inherited cancer proneness syndrome with an elevated risk of developing adrenocortical carcinoma, choroid plexus carcinomas, sarcomas and other types of cancer in multiple sites at a young age.[9] So far, nearly 2000 different single amino-acid changes in p53 have been reported in tumors,[10] and their frequencies vary markedly: next to exceedingly rare mutations, strong hotspots are evident.[11, 12] This latter group of mutations affects, in particular, codons 175, 248, 249, 273 or 282. The impact of mutations on p53 functions can vary from a wild-type-like activity, for example, the R337H mutations associated with predisposition to adrenocortical carcinoma,[13] to a partial function or to a suspected complete loss of function (LOF).[12, 14]

According to Resnick et al., different mutant p53s retaining a partial activity (for example, T123A or S215C) show specific effects on the transactivation of target promoters, leading to mutation-specific altered regulation of hundreds of genes (the ‘piano model'), resulting in a variety of biological consequences.[15, 16] Cells can show lack of control of their cell cycle and weakened apoptosis and DNA repair. However, in selected examples, separation of p53 functions was observed, with defective apoptotic control, but wild-type function in cell cycle arrest.[17] Moreover, knockin mouse models showed varied phenotypes, suggesting the occurrence of mutation-specific gene expression reprogramming also in vivo.[18, 19] However, most studies related to mutant p53 activity were performed on hotspot mutations, using reconstituted assays[12, 14, 20] or other in vitro models. Following these experimentations, it was observed that hotspot mutations have the least transactivating activity of common targets and therefore they were suggested to cause a p53 LOF. Hotspot p53 mutations were reported to be associated with more aggressive malignancies and could confer novel phenotypes in vivo, including an increased metastatic capacity and resistance to chemotherapies.[21, 22, 23, 24, 25, 26, 27] The acquired phenotypes of specific mutant p53s are generally referred to as gain-of-function properties,[28] but it is unclear if these features are restricted to or distinct among specific p53 hotspot mutations. Examining the impact of hotspot p53 mutations at a transcriptome level, a large number of genes are downregulated. However, there are also a restricted number of WTp53 targets whose transactivation seems not to be hampered by p53 mutations.[7] Moreover, there are also genes that are upregulated under mutant but not WT p53 expression. It is not clear whether different mutants can lead to similar transcriptional changes or have different impact on it (like an extension of the ‘piano model') and whether the gained phenotypes can be related to specific genes upregulated in the p53 mutant background. Thus, in this work, the focus was placed on cancer-associated p53 hotspot mutations that exhibit a loss of transactivation function in reconstituted assays,[12, 14] and a review of the literature was performed, with the following aims: (1) to verify whether different p53 mutations can be equivalent for their effects, or whether there is a mutation-specific transcriptional reprogramming of target genes, (2) to understand what is the main mechanism at the basis of upregulation or downregulation of gene expression under the p53 mutant background, (3) to identify the novel candidate target genes of WT and/or mutant p53 and (4) to define cellular pathways affected by the mutant p53-dependent gene expression reprogramming.

Selection of the published literature

In order to identify genes differentially modulated upon the expression of mutant p53, only potentially unbiased transcriptome studies published in the literature were collected. In fact, in transcriptome studies, target genes are analyzed without formulating any a priori hypothesis and, virtually, all the genes are evaluated with the same relevance. An extensive literature search was carried out using PubMed (http://www.ncbi.nlm.nih.gov/pubmed) to collect original papers. Articles were selected by screening title, abstract and full text, and only those reporting the effects of ectopic expression of p53 mutants on the transcriptome were considered further.

Out of over 2000 known p53 mutations reported by the IARC (www.iarc.fr/p53) or UMD TP53 databases,[10, 29] only 12, falling in 11 different codon sites (Figure 1), were studied through global gene expression changes. Those 11 mutated codons lie within the sequence-specific DNA-binding domain, correspond to hotspot mutations in tumors and result in LOF in functional assays. Studies on p53-dependent transcriptomes were few and heterogeneous in their experimental design, with a variable p53 status of the cell lines used, thus limiting the strength of the comparisons. Therefore, conservatively, conclusions on mutant p53-dependent gene deregulation were drawn only when at least three independent p53 mutations showed a coherent effect on the same target. Now on, genes upregulated under the ectopic expression of at least three different mutant p53 genetic backgrounds are defined as UMB, whereas DMB are the genes downregulated under at least three different mutant p53s. The results were obtained relying on the statistical analyses imbedded within each study, and a list of differentially expressed genes was compiled for further analysis (Supplementary Data S1). For each chosen article, in Table 1 the p53 missense mutation studied, and the cell lines used to perform the experiments were reported.

Figure 1

Number of genes (Log10) upregulated (black bars) and downregulated (white bars) following in vitro studies where a mutant form of p53 is overexpressed. Only few codons were assayed and for each mutation it is shown that the number of genes going overexpressed is approximately similar to those downregulated. The missense mutations falling within the codon 248 (R248Q and R248W) were considered as a unique one, in order to empower the study.

Table 1

List of selected article

Article	Cells used	Mutation tested
88	TKS, WTK1	M237I
89	HCT 116	A138P; R175H
90	H1299	R175H
91	HME1	R175H; R273H; R280K; R249S
65	H1299	R175H; R273H; D281G
92	U2OS	R175H; V157F; R248Q
93	LNCaP	G245S; R248W; R273H; R273C
94	H1299	R175H; R248W; R273H; D281G
95	H1299	D281G
96	H1299	R175H; R273H; D281G

For each article, the investigated p53 missense mutations and the cell lines used to perform the experiments are reported.

In silico analyses of promoters and pathways

COMPASSS (COMplex PAttern of Sequence Search Software),[30] a software that allows to perform custom pattern searches in entire genomes, was used to analyze the promoters. Given that most of the deregulated genes are not well-established p53 target genes, the focus was placed on the identification of non-canonical p53REs,[31, 32] particularly a half-site RE motif. In the exploratory search, a conservative approach was used by limiting inspection to 2-kb upstream of annotated transcriptional start sites, not allowing mismatches in the half-site decameric motif, and requiring the presence of a cluster of at least two half-sites within one nucleosome.[32] Hence, the following input were used: RRRCWWGYYY(N0-50)RRRCWWGYYY and NRRCWWGYYN(N0-50)NRRCWWGYYN. Two closely spaced p53 half-sites either in a direct orientation (RRRCWRRRCW) or lacking the CWWG core (WGYYYRRRCW), or having a relaxed motif definition (RRNCNNGNYN) (all sequence features that have been associated with genes repressed by WTp53),[8] were also queried. Thus, COMPASSS was used to analyze the promoters of UMB and DMB genes and to measure the ‘baseline' number of p53REs found in the whole human genome. Then, a binomial distribution-based statistics (approximated as normal distribution) was used, in order to verify whether the promoters were enriched for the input motifs, as compared with the baseline level.

The complex pattern of gene transcription changes was further analyzed with the tool Database for Annotation, Visualization and Integrated Discovery, in order to detect whether mutations within p53 could affect specific biological pathways.[33] Database for Annotation, Visualization and Integrated Discovery uses all the human genes as background to perform the comparisons, and if a group of genes is enriched within a specific biological process or pathway, the P-value of the modified Fisher's exact test will be lower than the cutoff (0.05). First, the short lists of UMB and DMB genes, either separated or combined, were used as input, but the total number of genes was not large enough to obtain statistically significant results. Thereafter, the analyses were repeated with a broadened input list, that is, the list of genes changing their expression under at least one p53 mutant background (that is, those reported in Supplementary Table S1).

Results and Discussion

Similar deregulation profile of genes by distinct p53 mutation hotspots

By observing UMB and DMB genes, consistent trends emerged (Table 2, see also Supplementary Table S1). A total of 401 genes were found downregulated under the ectopic expression of at least three different mutant p53s, whereas 260 genes were found upregulated (reported in Tables 3 and 4, respectively). Given the heterogeneity among studies, it is likely that consistent findings reveal true p53 target genes. Similar, although less confident, results could be obtained when the expression of genes was evaluated comparing at least two p53 mutants: 446 genes were found to be downregulated, whereas 503 genes were found to be upregulated. Thus, overall, the gene expression reprogramming did not seem to differ in relation to the mutant p53 hotspot analyzed. However, it should be also noticed that a small share of genes (48 out of 1846, 2.6%) was described as behaving discordantly, in relation to the p53 mutant assayed. It should be acknowledged that a systematic comparison of all mutant p53s under the same experimental conditions was not found in the literature, and thus subtle mutation-specific differences cannot be ruled out.

Table 2

Number of genes upregulated and downregulated following in vitro studies where a mutant form of p53 is overexpressed

Number of genes	% Of the total number of genes	Number of mutations on different p53 codons, leading to upregulation	Number of mutations on different p53 codons, leading to downregulation
151	8.2%	0	1
698	37.8%	1	0
446	24.2%	0	>1
503	27.2%	>1	0
401	22%	0	⩾3
260	14%	⩾3	0
Inconsistent
48	2.6%	⩾1	⩾1

Only 48 out of 1846 were described as behaving differently according to the p53 mutated codon assayed. The greatest majority of genes (949) showed a reproducible upregulation or downregulation when various p53 mutations were assayed.

Table 3

List of genes consistently downregulated where at least three independent mutations were assayed

ABAT

C15orf41

COBL

EDNRB

GABARAPL1

GTPBP2

KIAA1211

MARVELD2

P9

PSA

SHC4

TBXA2R

VCAN

ABCA12

C16orf5

COL18A1

EFS

GAD1

H19

KIAA1324

MCC

PADI3

PSEN2

SHROOM2

TEAD3

VSNL1

ABCG1

C17orf103

COL2A1

EGR2

GADD45A

HAGH

KIAA1751

MDFIC

PAK6

PSTPIP2

SI

TFAP2E

WDR8

ABHD15

C19orf59

CPE

ENC1

GAGEB1

HES2

KITLG

MDM2

PALMD

PTGES

SLC2A13

TGFBR1

WFDC5

ABHD4

C1orf187

CPN2

EPB49

GALE

HHATL

KLRK1

MEF2A

PARD6G

PTPN22

SLC2A8

TINAGL1

WNT5A

ABHD6

C2orf3

CRISPLD2

EPHB4

GAMT

HIC2

KREMEN2

MGC16703

PARP10

PVRL4

SLC2A9

TINP1

WWP1

ABTB2

C4A

CRYAB

EPPK1

GATA3

HIST1H2AE

KRT78

MGC4248

PARP14

PYHIN1

SLC35D1

TLR3

XLalphas

ACPP

C4orf18

CSNK1G1

ETNK1

GCH1

HLA-DMB

KRTAP2-1

MIB2

PCBP4

RALGDS

SLC35E4

TM7SF3

YPEL3

AHSA2

C5orf4

CST1

ETV7

GCLC

HOXC13

KSR

MICALL2

PCLO

RASAL1

SLC39A8

TMEM144

ZBTB1

AIFM2

C6orf204

CST11

F2RL2

GDAP1L1

HOXD10

LAMP3

MIR

PDE3A

RASGEF1B

SLC4A11

TMEM27

ZMAT3

AK1

C7orf10

CTSH

FAM105A

GDF15

HSDL2

LCE1B

MLF2

PDE4C

RASSF6

SLC4A4

TMEM63B

ZNF197

AKR1B10

C7orf57

CYFIP2

FAM134B

GDF9

HSPA5BP1

LDLRAP1

MME

PDE4D

RDH10

SLC9A1

TMG4

ZNF236

ALDH7A1

C9orf100

CYLC1

FAM13C

GGT1

HTR2A

LHX3

MPZL2

PENK

RET

SLC9A3R1

TMTC3

ZNF385A

AMACR

C9orf98

CYP2C9

FAM43A

GGT6

IDUA

LIN54

MRAS

PERP

retII

SLCO2B1

TNFRSF10B

ZNF441

ANK1

CABC1

CYP2S1

FAM46C

GH2

IFIT3

LOC132671

MRPL44

PF4V1

RGS12

SMARCD3

TNFRSF14

ZNF492

ANO4

CABYR

CYP3A7

FAM69B

GHR

IFITM1

LOC203274

MSX1

PGF

RHOD

SOM

TNP2

ZNF746

APAF1

CACNA2D2

D4S234E

FAM82A1

GJC1

IL24

LOC283585

MT1G

PLA2G2A

RNASE7

SORBS1

TNRC6C

ZNF786

APOBEC3C

CALML3

DDB2

FAM87A

GLS

IL2RB

LOC284837

MTMR6

PLAC8

RNF128

SORL1

TP53I11

ARAP1

CAPN1

DDC

FBXO2

GLS2

INPP1

LOC286434

MYO6

PLAT

RNF144B

SPAG1

TP53I3

ARG1

CARD18

DFNB31

FBXO22

GM2A

INPP5D

LOC348938

MYO7A

PLEK

RNF4

SPN

TP53INP1

ARHGAP6

CASP6

DGKZ

FGF1

GNA11

ISG15

LOC80154

NADSYN1

PLXNB3

RPL36

SSB2

TP53TG1

ARHGEF3

CBFA2T3

DHRS2

FITM2

GNA14

ISYNA1

LRDD

NCRNA00085

PMAIP1

RPS27L

STAR

TP73

ASPA

CCNA1

DISP1

FLJ14312

GNG2

ITFG1

LRP10

NEFL

PODXL

RPS6KA1

STARD5

TRAF4

ATP11B

CCT6B

DKK2

FLJ32065

GPC1

KAT2B

LTB4R

NHLH2

POLH

RRM2B

STAT4

TREM2

BCHE

CDH10

DMRTC1

FLJ36336

GPR126

KCNB1

LUM

NIFU

POMZP3

RTN1

STOX2

TRIM11

BCL11B

CDKN1A

DNAJC18

FLJ40773

GPR137B

KCNC4

MAB21L1

NLRP1

POU3F1

SAA2

SULF2

TRIM2

BDKRB1

CDKN1C

DNAJC21

FLNC

GPR155

KCNJ12

MAD1L1

NOTCH1

PPM1F

SAC3D1

SYK

TRIM22

BDNF

CEACAM1

DPYSL4

FMN1

GPR56

KIAA0247

MAEL

NOTCH3

PPP2R2B

SCN3B

SYNC

TRIM3

BLNK

CES2

DSG3

FREQ

GPR87

KIAA0284

MAFB

NRCAM

PPP2R2C

SCNN1G

SYTL2

TSGA10

BTF3L3

CLCA2

DUOX1

FRMD8

GRAMD2

KIAA1026

MAGEA4

NRP2

PRKX

SEC14L5

TAF3

TSPAN14

BTG2

CLDN19

DUSP13

FXYD2

GREB1

KIAA1052

MAP2K3

OIP106

PRKY

SERPINB5

TAGLN

TSPY1

C13orf31

CNNM4

EDAR

G6PC

GRN

KIAA1199

MAPK13

OTP

PRRX1

SESN1

TAP1

ULBP2

The listed genes are reproducibly deregulated irrespectively on the mutated codon. Note that all the assayed mutations fall within the p53 DNA-binding domain. Bolded genes are in common with the functional assay proposed by Riley et al.[3] Underlined are genes for which a high-confidence p53 occupancy sites had been mapped.

Table 4

List of genes consistently upregulated where at least three independent mutations were assayed

ABLIM1	CA9	DRG1	HIST1H4C	LPP	MYO5B	ProSAPiP1	SLCO4A1	XRCC5
ACTA2	CARS	E2F3	HMGB2	LRRFIP1	NAP1L1	PRPF40A	SLPI	YARS
ADAMTSL4	CBR4	E2F5	HOMER1	LRRK1	NARS	PRSS7	SMA4	ZBTB45
ADPRTL2	CCNB1IP1	EBAG9	HOMER2	LYN	NBEA	PRSS8	SMTN	ZNF217
AGGF1	CCNB2	EFHA1	HSP90AB1	MAD2L1	NDC80	PVRL3	SNTB2	ZNF238
AKT2	CCNH	EXPH5	HYAL3	MAL2	NDUFA4L2	PYGL	SPAG5	ZNF24
ALDH1A3	CCNL1	F2R	ID1	MAP2K5	NFATC2IP	QARS	SQSTM1	ZNF273
ALDH2	CD14	FADS1	ID3	MAP4	NFKBIA	RAD51C	SRM	ZNF415
ALDH3A1	CD44	FAM169A	IL1RL1	MAPKAPK2	NKTR	RBBP6	SS18	ZNF44
ANGPT1	CDC2	FBXO31	IMPDH2	MAPKAPK3	NLRX1	RBBP8	STAMBP	ZNF579
APS	CDC6	FDFT1	INADL	MAPKAPK4	NMI	RHOBTB2	STAT3	ZNF580
AR	CDH1	FECH	INF2	MARCH_6	NPC1	RHOG	STATH	ZNF652
ARHGEF2	CDKL3	FTH1	INPPL1	MARCKS	NUP153	RLN1	TAF1A
ARID3B	CDS1	FZD3	ITGA6	MAZ	PAQR3	RNF44	TARS
ARNT2	CEBPB	GAGE3	ITPR3	MCL1	PARD6B	RNF6	TFAP2A
ASB13	CKS1B	GAPVD1	JUNB	MCM3	PARP1	RP3-402G11.5	TMC6
ATF3	CLEC18C	GARS	JUP	MCM6	PCGF2	RPGRIP1	TNFRSF9
ATIC	CLMN	GHDC	KIAA0516	ME1	PDE3B	RPS6KA3	TRAF3IP2
BAG2	CPT1B	GINS1	KIF13B	MED13	PDLIM5	RRM1	TREM1
BARD1	CPVL	GLI2	KIF24	MEF2D	PGAP3	SARNP	TRIM29
BCAN	CREB1	GMIP	KIF2C	MELK	PGM1	SEC31A	TROVE2
BMP6	CTPS	GNB2L1	KLF16	MEST	PHKB	SEZ6L2	TUBB
BPTF	CTSF	GOSR1	KPNA2	MFGE8	PI3	SF3A2	TUSC3
BTG3	CUL5	GPR153	KRAS	MINK	PIAS2	SFPQ	TXNIP
BTN3A3	CUL7	GTF3A	KRT16	MMP28	PIK3CA	SGPP2	TXNL4A
C10orf116	DAAM1	GUCY1A3	LARP	MNX1	PLEKHH1	SIRT6	UCN
C13orf1	DCAF4	HAS3	LOC115871	MOCOS	POLA2	SLC16A4	UGT1A10
C16orf45	DHFRL1	HAX1	LOC120450	MRPL46	POLD2	SLC26A2	UGT2B28
C1orf63	DHRS9	HBA2	LOC139376	MRPS6	POLR2E	SLC29A2	UGT2B7
C21orf63	DICER1	HHL	LOC81691	MTAP	PRDM15	SLC4A7	UPF1
CA2	DMXL1	HIBCH	LPHN2	MYC	PRKCI	SLC6A8	VIM

The listed genes are reproducibly deregulated irrespectively on the mutated codon. Note that all the assayed mutations fall within the p53 DNA-binding domain. Bolded genes are in common with the functional assay proposed by Riley et al.[3] Underlined are genes for which a high-confidence p53 occupancy sites had been mapped.

Hypothesized mechanisms at the basis of DMB and UMB phenotypes

In order to better understand the possible mechanisms related to the changes of expression caused by mutations within p53, DMB genes were first compared with the information from Riley's list,[3] who reported 126 experimentally validated p53 target genes. These genes were crossed with those reported in Table 3 and 26 in common were found (bolded in Table 3). Almost all of them, 25, were genes normally activated by the WTp53. Then, COMPASSS was used (the detailed statistics are reported for each chromosome and for each p53RE motif in Supplementary Table 2) and it was observed that DMB genes were enriched for p53RE motifs typically found in genes transactivated by WTp53. This was expected and was consistent with the comparison made with Riley's data. Thus, it is conceivable that, for most of the DMB genes, the lack of expression is related to the LOF of p53.

When Riley's list was compared with the UMB genes (Table 4), only three were in common (bolded), preventing to draw any conclusion. According to COMPASSS, UMB genes were specifically enriched for a pair of the p53RE variant motif (RRNCNNGNYN) that was previously related to WTp53-dependent gene repression.[8] Out of 260 UMB genes, 242 contained a putative repressor element. It is, however, important to note that the p53RE variant pattern search may retrieve false-positive results. This motif was also enriched over the baseline for the DMB genes (this because it represents a more degenerated version of the canonical p53RE) confirming the difficulty in separating p53-upregulated and p53-downregulated genes purely on the basis of the cis-regulatory elements.[3, 8, 32] The fact that WTp53 could bind p53RE within specific UMB genes is reinforced by studies of chromatin immunoprecipitation followed by DNA sequencing (ChiP-seq) coupled to transcriptome analysis.[34] In fact, high-confidence p53 occupancy sites have been mapped not only for 57 DMB genes but also for 23 UMB genes (underlined in Tables 3 and 4). In summary, also UMB genes could be explained with the loss of activity, that is, a loss of transcriptional repression, toward specific targets. Actually, it was shown recently that p53 bears a repressor activity for genes such as CHEK1, BCL2, ARF and FOS,[8] MDR1 and Lasp1.[8, 35] Moreover, experiments in the yeast assays showed that mutant p53s lose the transactivating capability towards several target REs from target genes.[12, 14, 20]

Three generally accepted mechanisms of direct p53-mediated transcriptional repression are known: (1) steric interference by masking overlapping transcription factor binding sites,[36, 37] (2) sequestration of transcription activators[38] and (3) recruitment of histone deacetylases.[3, 39] Moreover, other indirect mechanisms were suggested, such as the transcriptional activation of microRNAs, known inhibitors of mRNA translation and stability.[39, 40] Thus, a ‘full-loss-of-function hypothesis' could explain both the UMB and the DMB genes. However, alternatives are discussed further in final remarks section.

Novel targets for WTp53

Previous analyses were also useful to detect novel putative direct p53 targets. In fact, a short list of highly likely candidate p53 targets was obtained applying the in silico analysis of p53REs within the DMB and UMB genes, crossed with the results from a ChIP-seq study.[34] In Supplementary Table S3, all the UMB and DMB genes positive for a p53REs within the promoter (through COMPASSS) were listed. Following the cross with the ChIP-seq study, known p53 targets were found (including ATF3, BTG2, BTG3, MYC, CDKN1A, ENC1, TP53I3 and TP53INP1). However, interestingly, a restricted number of novel potential p53 targets were also suggested. These are: ARID3B, ARNT2, CLMN, FADS1, FTH1, KPNA2, LPHN2, PARD6B, PDE4C, PIAS2, PRPF40A, PYGL and RHOBTB2. Intriguingly, some of them, belonging to the UMB category, were shown to be in causal relationship with features of the malignant phenotype and their upregulation in tumor correlates with a worsening of the prognosis. For example, an overexpression of ARID3B in human neuroblastoma cell lines is more common in stage IV neuroblastoma than in stages I–III, indicating its role in the progression of malignant neuroblastoma.[41] The upregulation of ARNT2 is also common in neuronal-derived tumors. ARNT2 forms complexes with HIF-1a (Hypoxia-inducible factor 1-alpha), and it allows for initiating hypoxia/nutrient deprivation-induced vascular endothelial growth factor expression, therefore permitting tumor angiogenesis.[42] FTH1 was found to be overexpressed in tumorspheres, and its upregulation has an important anti-apoptotic role.[43, 44] Moreover, FHIT overexpression was shown to have a role in increasing the multidrug resistance of cancer cells,[45] whereas its silencing caused an increased sensitivity.[46] KPNA2 was found to be highly expressed in different types of cancer, and its aberrant expression is often linked to a poor prognosis.[47] Finally, PARD6B was found amplified and overexpressed in a high number of breast cancer cell lines. The encoded protein, PAR6B, has a central role in tight junction assembly, maintenance of cell polarity, all features important for tumor progression and invasion.[48] Although the precise mechanism at the basis of the upregulation is not established (a loss of transcriptional repression is likely, as stated before), the increase in gene expression could, at least in part, explain some of the novel phenotypes gained by cancer cells, including angiogenesis, drug resistance and altered cell–matrix and cell–cell interactions.

Pathway analysis of deregulated genes using Database for Annotation, Visualization and Integrated Discovery

The possible pathways and biological functions modulated by mutant p53s were evaluated in silico using the tool Database for Annotation, Visualization and Integrated Discovery and in Table 5 the main results (with a KEGG-pathways based analysis) are reported. As expected, the p53 signaling pathway (P=8.6 × 10−8), and pathways related to the control of the cell cycle (P=1.3 × 10−3), is among the most significant semantic terms. Moreover, an over-representation of genes encoding for enzymes in the metabolism of xenobiotics (P=1.3 × 10−4) was found, where, in general, the cytochrome p450 genes are overexpressed. This might be related to the known resistance to chemotherapeutic drugs associated with p53 mutation status of patients' cancer cells.[49] Intriguingly, an enrichment of deregulated genes in pathways devoted to the catabolism of amino acids was also found (for example, ARG1, arginase; PRODH, proline oxidase; GLS2, glutaminase; GAD1, glutamate decarboxylase 1, all downregulated). The amino-acid catabolism leads to the formation of α-ketoglutarate, one of the key substrate for the tricarboxylic acid cycle, which in turn results in enhanced mitochondrial respiration and ATP generation. It is worth to stress here that p53 was shown to have a role not only in the regulation of cell cycle, apoptosis, differentiation, senescence, angiogenesis,[50] antioxidant response[51] and glutaminolysis[52, 53] but also in the modulation of both glycolysis[54, 55, 56] and mitochondrial respiration.[57, 58, 59, 60] Metabolic enzymes including glucose transporters (such as GLUT1 and GLUT4), glycolytic enzymes (such as PGM5 and HK2) and tricarboxylic acid cycle enzymes are downstream targets of p53 (ref. 61), and WTp53 was shown to slow the glycolysis. The inhibition of glycolysis can also be achieved by p53-dependent transcriptional activation of synthesis of cytochrome C oxidative 2, resulting in enhanced mitochondrial respiration.[62] Thus, mutations within p53 could lead to an increase of the glycolysis, characteristic of cancer cells.[63] One of the most important genes linking the energy metabolism with p53 was proposed to be GLS2, encoding for glutaminase. It was shown that GLS2 is transactivated by WTp53, and it regulates the cellular energy metabolism by increasing the production of α-ketoglutarate, the mithochondrial respiration and the ATP generation.[52] It is noteworthy that GLS2 expression was shown to be decreased in hepatocellular carcinomas, whereas its overexpression reduced tumor cell colony formation in an in vitro assay. GLS2 downregulation in cancer could be obtained by LOF mutations within p53, and this is consistent with the fact that GLS2 was found among the DMB genes. In addition, PRODH (proline dehydrogenase) was found downregulated in various transcriptome studies (Supplementary Table S1). Interestingly, PRODH functions as a tumor suppressor, and it suppresses hypoxia-inducible factor signaling by increasing α-ketoglutarate.[64] Thus, overall, p53 mutations could lead to increasing levels of glycolysis and, in parallel, to reduced mitochondrial respiration. This suggests a role of mutant p53s in the Warburg effect. The modulation of α-ketoglutarate production, through the alteration of amino-acid catabolism, could be one possible mechanism to be considered in a putative p53-dependent metabolic shift.

Table 5

The tool DAVID groups cluster of genes into biological pathways

Category	Term	Count	%	P-value	Benjamini
KEGG_PATHWAY	p53 signaling pathway	26	1.9	8.6 × 10−8	1.6 × 10−5
KEGG_PATHWAY	Metabolism of xenobiotics by cytochrome P450	19	1.4	1.3 × 10−4	7.9 × 10−3
KEGG_PATHWAY	Progesterone-mediated oocyte maturation	23	1.7	3.2 × 10−4	0.015
KEGG_PATHWAY	Ascorbate and aldarate metabolism	9	0.7	3.5 × 10−4	0.013
KEGG_PATHWAY	Alanine, aspartate and glutamate metabolism	12	0.9	5.5 × 10−4	0.017
KEGG_PATHWAY	Cell cycle	28	2	1.3 × 10−3	0.029
KEGG_PATHWAY	Beta-alanine metabolism	9	0.7	2.7 × 10−3	0.052
KEGG_PATHWAY	Arginine and proline metabolism	15	1.1	2.8 × 10−3	0.05

The p53 signaling pathway and pathways related to the control of the cell cycle are deregulated, as expected (analysis carried out including all 1846 genes collected among all the published studies).

Final remarks

As stated before, collected data seem to be in favor of a general lack of activity at the basis of UMB (LOF) and DMB (loss of transcriptional repression) genes. Given that all the p53 mutants taken into consideration here were classified as LOF in in vitro reporter assays, their expected impact corresponds to the lack of the hand in the ‘piano model' analogy[15] and, therefore, the consequence of this gene expression reprogramming could result as a ‘sound of silence'. However, several other aspects deserve discussion and open to the possibility of other mechanisms at play. In fact, the majority of p53 mutations encountered within tumors are of missense type. Moreover, tumors commonly retain and overexpress the full-length mutant p53 (ref. 65) and mutations whose effect is a true ablation of the gene sequence, such as large deletions, nonsense substitutions or in/del frame-shifts, account for only about 16% of the cases (www.iarc.fr/p53). This is in striking contrast to the majority of tumor suppressors (for example, RB1, APC, NF1, NF2 and VHL), where the primary mutations are deletion or nonsense, leading to little or no expression of the respective proteins. Dominance or dominant-negative potential of mutant p53s when heterozygous with the WT allele has been considered as an underlying reason for the high preponderance of p53 missense mutations in cancer. However, one should wonder whether the classification of the hotspot mutations as ‘inactivating' based on in vitro assays (commonly performed on yeast systems) is completely correct. The fact that missense mutations are preferred to the abrogation of the locus suggests that these two possibilities are not equivalent. p53 knockin mutant mouse models produced an altered tumor spectrum as compared with the knockout models, with more metastatic tumors.[66, 67] Similarly, in an analysis of Li–Fraumeni patients, germline missense mutations in TP53 have been shown to be associated with an earlier age of onset (9 years) when compared with germline deletions, suggesting a gain-of-function effect of missense p53 mutants in human tumors.[68] In addition, tumors with mutant p53 proteins may be more aggressive (for example, conferring a poor prognosis) as compared with tumors where p53 is lost.[18, 69, 70, 71] Thus, it could be hypothesized that these mutations alter profoundly, but not completely abrogate, some of the p53 functions. Various authors suggested a gain-of-function at molecular level for mutant p53s.[65, 72, 73, 74, 75, 76] Mutant p53s can directly bind promoters of various targets such as it is for miR-130b,[77] miR-128-2,[78] Axl[79] or NF-kB2.[80] It was also shown that they form aberrant protein complexes with interacting partners, such as NF-Y, Sp1, Ets-1 or VDR, perturbing their activity.[81] More interactions of this type were reported in a recent review.[82] Among them, it is noteworthy to underline that, although WTp53 does not form heterotetramers with p63,[83] mutant p53s have been shown to bind and sequester TAp63 away from its target genes, hampering its anti-metastatic capacity.[84, 85, 86, 87] Furthermore, other studies showed that mutant p53s bind to p63 and use it as chaperone for the transactivation of novel targets.[7] In this regard, it should be considered that the enrichment of the degenerated motif detected with COMPASSS within the promoters of UMB genes could be also a marker of the presence of p63-responsive elements, given the sequence similarities with the p53REs. Thus, we cannot rule out that some of the UMB genes are directly/indirectly transactivated by mutant p53s.

In summary, the elaboration of data already present in the literature allowed to gain novel insights in the biology of p53 and to define novel targets. Further studies are warranted in order to better define the extent of differences in the transactivating activities of different mutant p53s and to validate the novel targets suggested here.