TP53 mutation, epithelial-mesenchymal transition, and stemlike features in breast cancer subtypes.

Altered p53 protein is prevalently associated with the pathologic class of triple-negative breast cancers and loss of p53 function has recently been linked to the induction of an epithelial-mesenchymal transition (EMT) and acquisition of stemness properties. We explored the association between TP53 mutational status and expression of some genes involved in the canonical TGF-β signaling pathway (the most potent EMT inducer) and in two early EMT associated events: loss of cell polarity and acquisition of stemness-associated features. We used a publicly accessible microarray dataset consisting of 251 p53-sequenced primary breast cancers. Statistical analysis indicated that mutant p53 tumors (especially those harboring a severe mutation) were consistent with the aggressive class of triple-negative cancers and that, differently from cell cultures, surgical tumors underexpressed some TGF-β related transcription factors known as involved in EMT (ID1, ID4, SMAD3, SMAD4, SMAD5, ZEB1). These unexpected findings suggest an interesting relationship between p53 mutation, mammary cell dedifferentiation, and the concomitant acquisition of stemlike properties (as indicated by the overexpression of PROM1 and NOTCH1 genes), which improve tumor cells aggressiveness as indicated by the overexpression of genes associated with cell proliferation (CDK4, CDK6, MKI67) and migration (CXCR4, MMP1).

1. Introduction

TP53 tumor suppressor is the most commonly altered gene in human breast cancer where it is mutated in about 30–40% [1]. TP53 gene mutations result in altered and stable p53 proteins that function as dominant negative with gain-of-function properties, including drug resistance, and contribute to malignant progression with detrimental effects on patient's outcome [2]. In particular, clinical evidence indicated that altered p53 proteins are prevalently associated with the pathologic class of triple-negative breast cancers, that is, tumors characterized by the immunohistochemical expression of basal cytokeratins and epidermal growth factor receptor (EGFR), but negative for estrogen (ER), progesterone receptor (PR), and HER2 expression [3-5]. Recently, triple-negative tumors have been also associated with a new less common subtype, known as claudin-low [6].

Wild-type p53 functions as a sequence-specific DNA binding transcription factor that regulates a plethora of target genes involved in DNA repair, cell cycle control, apoptosis, senescence, angiogenesis, and other fundamental biological process [7]. Therefore, it is not surprising that mutations of TP53 gene or inactivations of its signaling pathway are prerequisite for the development of tumors. Most mutations occur within the central DNA binding domain (exons 5–8) and, in particular, at several specific amino acids required for DNA binding. According to the type of mutation (point mutation, deletion, insertion, or stop codon), p53 protein synthesis may be totally inhibited or may generate functionally altered molecules that affect cell homeostasis in a different manner [8]. In fact, it is well known that not all p53 mutations have equal effects: some of them confer loss of function, others have a dominant negative effect and still others are classified as wild-type-like protein and represent mutant forms with a limited biological effect [9, 10].

Recently, some excellent papers have provided experimental evidence linking loss of p53 function to the induction of epithelial-mesenchymal transition (EMT) and acquisition of stemness properties in different tumor cell lines [11-13]. EMT is a key program in embryonic development, the aberrant reactivation of which may induce progression, invasion, dissemination, and finally metastasis in cancer cells. The most evident peculiarities of a cell undergoing EMT process are loss of the epithelial phenotype and acquisition of mesenchymal features and abnormal motility capabilities [14, 15]. The most potent inducer of EMT is transforming growth factor-β (TGF-β) that triggers the activity of several transcription factors (ZEB1/TCF8, ZEB2/SIP1, Snail, Slug, Twist, and Ids), which in turn repress the expression of genes coding for epithelial markers and activate the expression of mesenchymal genes [16]. According to recent acquisition, p53 should prevent EMT by repressing ZEB1 and ZEB2 expressions via miRNAs activity. Consequently, p53 loss-of-function should downregulate miRNAs expression, the transactivation of transcription factors promoting EMT, and the emergence of tumor cells with stemlike properties [17-19].

So far, despite the availability of a huge amount of information on the transcript profile from microarray analysis, the interrelations among p53 mutations and genes involved in EMT have not been specifically assessed. Therefore, to investigate the association between TP53 mutational status and EMT process, we interrogated a publicly accessible microarray dataset consisting of 251 p53 sequenced primary breast cancers [20]. Adopting an unconventional approach, we did not use the whole transcript profile but we selected a priori panel of genes experimentally recognized as involved in the canonical TGF-β signaling pathway and in two early events associated with EMT: loss of epithelial cell polarity and acquisition of stemness-associated features. To delineate a more comprehensive picture of the relationship among p53 mutation, EMT, and tumor aggressiveness, we also considered some genes involved in cell proliferation, apoptosis, and metastatic spread.

2. Materials and Methods

2.1. Materials

As reported in the original paper [20], gene expression profile was determined by using the Affymetrix Human Genome HG-U133A and -B GeneChip, and microarray dataset while was available at the ArrayExpress website (http://www.ebi.ac.uk/arrayexpress/), with the accession number E-GEOD-3494. Patients and tumors characteristics were provided as supporting information in the original paper [20].

2.2. Gene Selection

According to the aim of the study, we selected 147 genes (Table 1). Specifically, the panel was composed of 27 genes recognized as involved in TGF-β-induced EMT, 57 involved in epithelial cell plasticity, 13 coding for stemlike properties and 31 involved in cell proliferation, apoptosis, and metastatic spread. In addition, to describe breast cancer subtypes, 19 genes coding for luminal and basal markers were also considered. The 147 genes corresponded to 352 Affymetrix probe sets, as verified by GeneAnnot system v2.0 (http://bioinfo2.weizmann.ac.il/geneannot/), that additionally provided information about the quality of each probe set in terms of sensitivity and specificity score [21] (Supplementary Table 1 available at doi: 10.1155/2012/254085).

Table 1

Genes entered in the study.

	Official gene symbol	Gene name	Entrez gene ID	Ensembl genomic location
Apical junctional complex	CDH1	Cadherin 1, type 1, E-cadherin (epithelial)	999	16q22.1
	CDH2	Cadherin 2, type 1, N-cadherin (neuronal)	1000	18q12.1
	CLDN1	Claudin 1	9076	3q28
	CLDN2	Claudin 2	9075	Xq22.3
	CLDN3	Claudin 3	1365	7q11.23
	CLDN4	Claudin 4	1364	7q11.23
	CLDN5	Claudin 5	7122	22q11.21
	CLDN6	Claudin 6	9074	16p13.3
	CLDN7	Claudin 7	1366	17p13.1
	CLDN8	Claudin 8	9073	21q22.11
	CLDN9	Claudin 9	9080	16p13.3
	CLDN10	Claudin 10	9071	13q32.1
	CLDN11	Claudin 11	5010	3q26.2
	CLDN12	Claudin 12	9069	7q21.13
	CLDN14	Claudin 14	23562	21q22.3
	CLDN15	Claudin 15	24146	7q22.1
	CLDN16	Claudin 16	10686	3q28
	CLDN17	Claudin 17	26285	21q22.11
	CLDN18	Claudin 18	51208	3q22.3
	CLDN23	Claudin 23	137075	8p23.1
	CTNNA1	Catenin (cadherin-associated protein), alpha 1	1495	5q31.2
	CTNNB1	Catenin (cadherin-associated protein), beta 1	1499	3p22.1
	CTNND1	Catenin (cadherin-associated protein), delta 1	1500	11q12.1
	JAM2	Junctional adhesion molecule 2	58494	21q21.3
	JAM3	Junctional adhesion molecule 3	83700	11q25
	JUP	Junction plakoglobin	3728	17q21.2
	MAGI1	Membrane associated guanylate kinase, WW and PDZ domain containing 1	9223	3p14.1
	MARVELD2	MARVEL domain containing 2	153562	5q13.2
	OCLN	Occludin	4950	5q13.2
	PVRL1	Poliovirus receptor-related 1 (herpesvirus entry mediator C)	5818	11q23.3
	PVRL2	Poliovirus receptor-related 2 (herpesvirus entry mediator B)	5819	19q13.32
	PVRL3	Poliovirus receptor-related 3	25945	3q13.13
	PVRL4	Poliovirus receptor-related 4	81607	1q23.3
	SYMPK	Symplekin	8189	19q13.32
	TJP1	Tight junction protein 1 (zona occludens 1)	7082	15q13.1
	TJP2	Tight junction protein 2 (zona occludens 2)	9414	9q21.11
	TJP3	Tight junction protein 3 (zona occludens 3)	27134	19p13.3
	VCL	Vinculin	7414	10q22.2
Apoptosis	BAX	BCL2-associated X protein	581	19q13.33
	BCL2	B-cell CLL/lymphoma 2	596	18q21.33
	CASP2	Caspase 2, apoptosis-related cysteine peptidase	835	7q34
	ERBB2	V-erb-b2 erythroblastic leukemia viral oncogene homolog 2 (avian)	2064	17q12
	MDM2	Mdm2 p53 binding protein homolog (mouse)	4193	12q15
	TP53	Tumor protein p53	7157	17p13.1
	TP53INP1	Tumor protein p53 inducible nuclear protein 1	94241	8q22.1
	XIAP	X-linked inhibitor of apoptosis	331	Xq25
Basal markers	ACTA1	Actin, alpha 1, skeletal muscle	58	1q42.13
	CD44	CD44 molecule (Indian blood group)	960	11p13
	EGFR	Epidermal growth factor receptor	1956	7p11.2
	KRT5	Keratin 5	3852	12q13.13
	KRT6A	Keratin 6A	3853	12q13.13
	KRT6B	Keratin 6B	3854	12q13.13
	KRT14	Keratin 14	3861	17q21.2
	KRT17	Keratin 17	3872	17q21.2
	TP63	Tumor protein p63	8626	3q28
	VIM	Vimentin	7431	10p13
Cell cycle	CCNB3	Cyclin B3	85417	Xp11.22
	CCND1	Cyclin D1	595	11q13.3
	CCNE2	Cyclin E2	9134	8q22.1
	CDK2	Cyclin-dependent kinase 2	1017	12q13.2
	CDK4	Cyclin-dependent kinase 4	1019	12q14.1
	CDK6	Cyclin-dependent kinase 6	1021	7q21.2
	CDKN1A	Cyclin-dependent kinase inhibitor 1A (p21, Cip1)	1026	6p21.2
	CDKN1B	Cyclin-dependent kinase inhibitor 1B (p27, Kip1)	1027	12p13.1
	CDKN2A	Cyclin-dependent kinase inhibitor 2A (melanoma, p16, CDK4)	1029	9p21.3
	CDKN2B	Cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4)	1030	9p21.3
	CDKN2D	Cyclin-dependent kinase inhibitor 2D (p19, inhibits CDK4)	1032	19p13.2
	GAS1	Growth arrest-specific 1	2619	9q21.33
	MKI67	Antigen identified by monoclonal antibody Ki-67	4288	10q26.2
Cell polarity complexes	CRB1	Crumbs homolog 1(Drosophila)	23418	1q31.3
	CRB3	Crumbs homolog 3 (Drosophila)	92359	19p13.3
	DLG1	Discs, large homolog 1 (Drosophila)	1739	3q29
	DLG2	Discs, large homolog 2 (Drosophila)	1740	11q14.1
	DLG3	Discs, large homolog 3 (Drosophila)	1741	Xq13.1
	DLG4	Discs, large homolog 4 (Drosophila)	1742	17p13.1
	DLG5	Discs, large homolog 5 (Drosophila)	9231	10q22.3
	INADL	InaD-like (Drosophila)	10207	1p31.3
	MPP5	Membrane protein, palmitoylated 5 (MAGUK p55 subfamily member 5)	64398	14q23.3
	LLGL1	Lethal giant larvae homolog 1 (Drosophila)	3996	17p11.2
	LLGL2	Lethal giant larvae homolog 2 (Drosophila)	3993	17q25.1
	PARD3	Par-3 partitioning defective 3 homolog (C. elegans)	56288	10p11.21
	PARD3B	Par-3 partitioning defective 3 homolog B (C. elegans)	117583	2q33.3
	PARD6A	Par-6 partitioning defective 6 homolog alpha (C. elegans)	50855	16q22.1
	PARD6B	Par-6 partitioning defective 6 homolog beta (C. elegans)	84612	20q13.13
	PARD6G	Par-6 partitioning defective 6 homolog gamma (C. elegans)	84552	18q23
	PRKCI	Protein kinase C, iota	5584	3q26.2
	PRKCZ	Protein kinase C, zeta	5590	1p36.33
	SCRIB	Scribbled homolog (Drosophila)	23513	8q24.3
Epithelial-mesenchymal transition	AKT1	V-akt murine thymoma viral oncogene homolog 1	207	14q32.33
	AKT2	V-akt murine thymoma viral oncogene homolog 2	208	19q13.2
	BMP1	Bone morphogenetic protein 1	649	8p21.3
	ID1	Inhibitor of DNA binding 1, dominant negative helix-loop-helix protein	3397	20q11.21
	ID2	Inhibitor of DNA binding 2, dominant negative helix-loop-helix protein	3398	2p25.1
	ID3	Inhibitor of DNA binding 3, dominant negative helix-loop-helix protein	3399	1p36.12
	ID4	Inhibitor of DNA binding 4, dominant negative helix-loop-helix protein	3400	6p22.3
	SMAD2	SMAD family member 2	4087	18q21.1
	SMAD3	SMAD family member 3	4088	15q22.33
	SMAD4	SMAD family member 4	4089	18q21.2
	SMAD5	SMAD family member 5	4090	5q31.1
	SMAD6	SMAD family member 6	4091	15q22.31
	SMAD7	SMAD family member 7	4092	18q21.1
	SMURF1	SMAD specific E3 ubiquitin protein ligase 1	57154	7q22.1
	SMURF2	SMAD specific E3 ubiquitin protein ligase 2	64750	17q24.1
	SNAI1	Snail homolog 1 (Drosophila)	6615	20q13.13
	SNAI2	Snail homolog 2 (Drosophila)	6591	8q11.21
	TCF3	Transcription factor 3 (E2A immunoglobulin enhancer binding factors E12/E47)	6929	19p13.3
	TGFBI	Transforming growth factor, beta-induced, 68 kDa	7045	5q31.1
	TGFBR1	Transforming growth factor, beta receptor 1	7046	9q22.33
	TGFBR2	Transforming growth factor, beta receptor II	7048	3p24.1
	TGFBR3	Transforming growth factor, beta receptor III	7049	1p22.1
	TGIF2	TGFB-induced factor homeobox 2	60436	20q11.23
	TWIST1	Twist homolog 1 (Drosophila)	7291	7p21.1
	TWIST2	Twist homolog 2 (Drosophila)	117581	2q37.3
	ZEB1	Zinc finger E-box binding homeobox 1	6935	10p11.22
	ZEB2	Zinc finger E-box binding homeobox 2	9839	2q22.3
Luminal markers	CD24	CD24 molecule	100133941	6q21
	ESR1	Estrogen receptor 1	2099	6q25.1
	GATA3	GATA binding protein 3	2625	10p14
	KRT7	Keratin 7	3855	12q13.13
	KRT8	Keratin 8	3856	12q13.13
	KRT18	Keratin 18	3875	12q13.13
	KRT19	Keratin 19	3880	17q21.2
	MUC1	Mucin 1, cell surface associated	4582	1q22
	PGR	Progesterone receptor	5241	11q22.1
Metastasis-related genes	CXCR4	Chemokine (C-X-C motif) receptor 4	7852	2q22.1
	CXCR5	Chemokine (C-X-C motif) receptor 5	643	11q23.3
	MMP1	Matrix metallopeptidase 1 (interstitial collagenase)	4312	11q22.2
	MMP2	Matrix metallopeptidase 2 (gelatinase A, 72 kDa gelatinase, 72 kDa type IV collagenase)	4313	16q12.2
	MMP3	Matrix metallopeptidase 3 (stromelysin 1, progelatinase)	4314	11q22.2
	MTA1	Metastasis associated 1	9112	14q32.33
	MTA2	Metastasis associated 1 family, member 2	9219	11q12.3
	MTA3	Metastasis associated 1 family, member 3	57504	2p21
	PAK6	P21 protein (Cdc42/Rac)-activated kinase 6	56924	15q15.1
	TIMP1	TIMP metallopeptidase inhibitor 1	7076	Xp11.23
Stemlike features	ABCG2	ATP-binding cassette, sub-family G (WHITE),member 2	9429	4q22.1
	ALDH1A1	Aldehyde dehydrogenase 1 family, member A1	216	9q21.13
	ALDH1A3	Aldehyde dehydrogenase 1 family, member A3	220	15q26.3
	BMI1	BMI1 polycomb ring finger oncogene	648	10p12.2
	JAG1	Jagged 1	182	20p12.2
	JAG2	Jagged 2	3714	14q32.33
	NANOG	Nanog homeobox	79923	12p13.31
	NOTCH1	Notch 1	4851	9q34.3
	NOTCH2	Notch homolog 2 (Drosophila)	4853	1p12
	NOTCH3	Notch homolog 3 (Drosophila)	4854	19p13.12
	NOTCH4	Notch homolog 4 (Drosophila)	4855	6p21.32
	NUMB	Numb homolg (Drosophila)	8650	14q24.3
	PROM1	Prominin 1	8842	4p15.32

2.3. Statistical Analysis

As some genes are recognized by more than a single probe set, each of which characterized by an individual specificity and sensitivity that differently contribute to gene expression value, a gene expression mean value was calculated after weighting each probe set for its own sensitivity and specificity score. Specifically, each expression value (already log2 transformed in the original dataset) was multiplied for the semisum of sensitivity and specificity scores of the corresponding probe set.

Prediction Analysis for Microarray (PAM) analysis was used to identify genes associated with the TP53 mutation status. PAM methodology minimizes the classification error using cross validation. For the selected genes, shrunken centroids across the different mutation groups were plotted. The FDR level was estimated through a permutation method.

To identify the tumors characterized by a similar intrinsic phenotype (tumor subtypes), unsupervised hierarchical cluster analysis was performed using the subset of genes coding for luminal and basal markers, HER-2, and claudins. The choice of the number of clusters to be used was supported by mean silhouette values [22]. PAM methodology was used to detail differential gene expression among clusters [23].

Principal Components Analysis (PCA) and PCA-based biplots were used to assess gene expression among clusters [24]. Moreover, for evaluating the associations among genes, specific subsets, not used to build PCA, were passively projected over the PCA-based biplots of intrinsic phenotypes.

All analyses were performed using open source software R 2.14.1 packages stats, cluster, and HDMD (http://www.R-project.org/).

3. Results and Discussion

As described in the original paper [20], the cases series was composed of 251 tumors, 58 of which characterized by a TP53 mutation. Of the 58 mutant tumors, 37 had point mutations and 21 had “severe” mutations, that is, insertions (n = 3), deletions (n = 11), and stop codons (n = 7), that result in frame shift and truncations with deleterious functional consequences.

To identify genes differentially expressed between mutant and wild-type p53 tumors, we first applied a PAM analysis on the overall cases series (Figure 1). According to the expected loss-of-function, mutant p53 tumors were characterized by the underexpression of genes under p53 control (i.e., CDKN1A, coding for p21, and TP53INP, coding for the p53-inducible nuclear protein 1). In addition, they were showed as underxpressed several genes coding for epithelial cell polarity complex (DLG4, DLG5, INADL, PARD3B, PARD6B) and apical junctional components (CLDN11, JAM2, JAM3, MAGI1, OCLN, PVRL2, TJP1, TJP3). Conversely, mutant p53 tumors overexpressed CDH2, the gene coding for N-cadherin, and some genes involved in cell proliferation (CCNE2, CDK4, CDK6, CDKN2A, CDKN2D, MKI67) and metastatic spread (CXCR4, MMP1). An overall similar pattern of expression was found when we interrogated another publicly available microarray dataset where p53 mutational status was known [3] using the same panel of 147 genes. Despite the different microarray platform used (42K cDNA microarrays instead of Affymetrix GeneChip), mutant p53 tumors were associated with a dramatic underexpression of genes coding for apical junctional components (INADL, JAM2, JAM3, MAGI1, PARD6B, PVRL2) coupled with the overexpression of genes coding for N-cadherin (CDH2) or involved in cell proliferation and metastatic spread (CCNE2, CDK4, CDK6, MKI67, MTA1) (Supplementary Figure 1). Notably, in both datasets, p53 mutant tumors were associated with the overexpression of PROM1, supporting the experimental evidence indicating the relation among p53 mutation and the reacquisition of some stemlike properties according to an EMT-like process [12, 13].

Figure 1

Shrunken centroids for wild-type TP53 and mutant TP53 tumors. Left-sided bars indicate lower expression in the subgroups relative to overall centroid; right-sided bars indicate higher expression in subgroups relative to overall centroid.

As regards the genes related to the canonical TGF-β signaling pathway, mutant p53 tumors showed the downregulation of many genes coding for pivotal elements of this pathway (ID1, ID4, SMAD3, SMAD4, SMAD5, TGFBR2, TGFBR3, ZEB1) coupled with the overexpression of SMURF2, coding for a SMAD-specific E3 ubiquitin protein ligase, and TGIF2, coding for a transcriptional repressor interacting with TGF-β activated SMAD proteins (Figure 2). A similar pattern of expression (i.e., downregulation of SMAD2, SMAD3, SMAD5, SMAD7, TGFBR2, ZEB1, and overexpression of SMURF2 and TGIF2) was found in Langerod dataset [3] (Supplementary Figure 2). This unexpected finding could be explained taking into account that TGF-β is a multifunctional cytokine and a powerful tumor suppressor that governs many aspects of mammary epithelial cells physiology and homeostasis [25]. Consistent with the notion that estrogen receptor and TGF-β signaling pathways are major regulators during mammary gland development [26], it is not surprising that p53 mutant tumors concomitantly underexpressed ESR1 (coding for ERα), TGFBR2, TGFBR3 (coding for TGF-β receptors) and ID1, ID4, SMAD3, SMAD4, SMAD5, ZEB1 (coding for key elements of the pathway).

Figure 2

Boxplots of the genes associated with EMT in wild-type (WT) or mutant (mut) TP53 tumors with respect to overall case series (all).

When severe and missense mutations were considered separately, PAM analysis provided evidence that severe TP53 mutations were responsible for the differential gene expression observed in mutant with respect to wild-type p53 tumors, even though some important alterations were already present in missense TP53 mutations, as, for example, the downregulation of some apical junctional components or the overexpression of genes related to cell proliferation and invasion. As shown in Figure 3, in addition to the expected decrease in the expression of TP53, CDKN1A, and TP53INP1, tumors harboring severe mutations were characterized by a dramatic overexpression of genes associated with proliferation (CDK4, CDK6, MKI67) and metastatic spread (CXCR4, MMP1), and by underexpression of several genes involved in epithelial cell identity (DLG5, INADL, JAM2, JAM3, MAGI1, OCLN, PARD3B, PARD6B, PVRL2, SCRIB, TJP1 and TJP3).

Figure 3

Shrunken centroids for wild-type TP53 an mutant (missense or severe mutation) TP53 tumors. Left-sided bars indicate lower expression in the subgroups relative to overall centroid; right-sided bars indicate higher expression in subgroups relative to overall centroid.

As regards the association between missense or severe TP53 mutation and EMT-related genes, Figure 4 indicates that, with respect to tumors harboring a missense mutation, those with severe mutations were characterized by the overexpression of SMURF2, SNAI1, and TGIF2 genes. Of particular interest is the overexpression of SNAI1 gene because of the concomitant overexpression of NOTCH1 pointed out by PAM analysis in tumors with severe TP53 mutation (Figure 3). Indeed, Notch signalling pathway, which is implicated as an important contributor to EMT in tumorigenesis, has been recently suggested to play a direct role on the expression of the Snail transcription factor [27].

Figure 4

Boxplots of the genes associated with EMT in wild-type (WT), missense (MS), or severe (SV) TP53 mutation with respect to overall case series (all).

With respect to missense TP53 mutations, severe ones were also characterized by an increased expression of PROM1, the gene encoding for prominin, a pentaspan transmembrane glycoprotein (CD133) often overexpressed on cancer cells, where it is thought to function in maintaining stem cell properties by suppressing differentiation. This finding is in agreement with the hypothesis that basal cancers, which have been proposed to have a stem cell origin, are virtually all TP53 mutants and express high levels of PROM1 transcript and protein [28]. Unfortunately, since in Langerod dataset [3] severe mutations accounted for only three cases, we were unable to verify all these observations in an independent dataset.

One of the aims of the study was to explore the relationship among p53 mutation, EMT, and tumor aggressiveness, a peculiar characteristic of certain breast cancer subtypes (especially basal-like phenotype). To this specific aim, we performed an unsupervised hierarchical cluster analysis, using the subset of genes coding for luminal and basal markers, HER-2, and claudins, and we looked at the distribution of p53 mutations according to tumor subtype. The analysis indicated that mutant p53 tumors distributed into three main clusters (Figure 5). Of the 58 mutant p53 tumors, 23 were included in Cluster 1, 17 in Cluster 2, and 18 in Cluster3. However, looking at the relative percentage, we found that, on the total number of tumors in each cluster, only 17% (23/133) of Cluster 1 and 19% (18/95) of Cluster 3 tumors had p53 mutations, whereas 74% (17/23) of Cluster 2 tumors did have. Notably, 10 of these 17 mutations were severe mutations.

Figure 5

Tumor dendrogram from clustering 251 tumors and 30 genes associated with intrinsic breast cancer phenotypes (luminal, basal, ERBB2, and claudins). Black bars denote tumors with a wild-type p53; orange bars, tumors with a mutant p53 (missense mutation); red bars, tumors with a mutant p53 (severe mutation).

PCA-based biplots, drawn using the same subset of genes of hierarchical cluster analysis (Figure 6), showed that Cluster 2 tumors were positively associated with genes related to basal phenotype (KRT5, KRT6A, KRT6B, KRT14, KRT17, EGFR) and with a panel of claudin-coding genes (CLDN1, CLDN6, CLDN10), whereas they were negatively associated with the majority of genes related to luminal phenotype. Conversely, Cluster 3 tumors were positively associated with genes related to luminal phenotype (ESR1, GATA3, MUC1, PGR, KRT18) and with a different panel of claudin-coding genes (CLDN3, CLDN4, CLDN7) and negatively associated with genes related to basal phenotype. Cluster 1 tumors showed a less clear-cut phenotype according to the more heterogeneous nature of this cluster, even though they appeared prevalently, associated with genes related to basal phenotype (KRT5, KRT6B, KRT14, KRT17, TP63). Remarkably, Cluster 2 tumors also showed the concomitant underexpression of ERBB2 gene providing evidence that these tumors had a gene expression profile consistent with the pathologic class of triple-negative tumors (Supplementary Figure 3), which are characterized by the expression of basal cytokeratins (mainly Krt5) and EGFR, but do not express estrogen and progesterone receptors, and HER2.

Figure 6

Active PCA-based biplots of intrinsic breast cancer phenotype-related genes. Samples are colored according to the mutational p53 status (wild-type in green and mutant in blue) and are labeled according to the corresponding cluster (1, 2, or 3). Loading for basal genes are represented by red arrows, luminal genes by pink arrows, claudin genes by orange arrows, and ERBB2 gene by yellow arrow.

When we looked at the expression of EMT-associated genes according to clusters (Figure 7), we found that Cluster 2, consistent with the pathologic class of triple-negative cancers, showed a gene expression profile similar to that of tumors harboring a severe p53 mutation. Conversely, Cluster 3, consistent with the luminal-like phenotype, had a pattern of expression similar to that of wild-type p53 tumors whereas the phenotypically heterogenous Cluster 1 looks like the group of tumors with a missense p53 mutation. In particular, Cluster 2 (consistent with the pathologic class of triple-negative cancers and akin to severe p53 mutated tumors) was characterized by the underexpression of SMAD2, SMAD5, ZEB1, and TGFBR3 and the overexpression of SMURF2, TGIF2, and SNAI1, in agreement with the gene profile observed in tumors harboring a severe TP53 mutation (Figure 4).

Figure 7

Boxplots of the genes associated with EMT in the three main clusters identified by unsupervised hierarchical cluster analysis using the subset of genes coding for luminal and basal markers, HER-2 and claudins.

Notably, when the subset of genes related to stemness properties was passively projected over the PCA-based biplots provided in Figure 6, Cluster 2 tumors were positively associated with PROM1 and NOTCH1, and negatively associated with ALDH1A1, BMI1, NUMB (Figure 8). Similar to the latter but opposite in the sign, was the pattern of association shown by Cluster 3 tumors.

Figure 8

Passive projectionof stemlike-related genes in the space of the first two principal components (PC), (a), and in the space of first and third PCs, (b), of intrinsic phenotype gene set. Loading for EMT or stemlike genes are represented in black.

The imbalance in Numb/Notch pathway observed in Cluster 2 tumors, associated with the overexpression of SNAI1, is of particular interest because the involvement of this pathway in differentiation program and epithelial cancer progression and metastasis. Numb is an evolutionary conserved protein that plays a critical role in cell-fate determination, including control of asymmetric cell division, endocytosis, cell adhesion, cell migration, and ubiquitination of specific substrates as p53. Loss of Numb causes increased activity of the oncogene Notch1 and for this reason, low expression of Numb and high levels of Notch1 have been associated with tumor progression and used as markers of tumor aggressiveness, especially in basal-like breast cancer [29]. The aggressiveness of this group of tumors was corroborated by the observation that Cluster 2 tumors were prevalently poorly differentiated (17/23 tumors were Grade III) with respect to Cluster 1 and Cluster 3 tumors, and by the positive association with genes promoting cell proliferation and metastatic spread. In this context, it can be viewed the overexpression of MMP1 and CXCR4, and the downregulation of TIMP1. Indeed, MMP1 encodes for a matrix metalloproteinase family member (specifically, a collagenase) involved in the breakdown of extracellular matrix whereas TIMP1 encodes for a specific tissue inhibitor of metalloproteases, including MMP-1. Because MMP1 is a target gene for wild-type p53 activity, the functional inactivation of the protein results in a gene overexpression that allows tumor cell migration after degradation of basement membrane and cell detachment [30, 31]. The concomitant overexpression of CXCR4 due to a gain-of-function mutant p53 [32, 33], further contributes to enhance tumor cell migration and metastatic spread [34]. In fact, CXCR4 encodes a C-X-C motif chemokine receptor specific for stromal cell-derived factor-1 (SDF-1/CXCL12), a member of the family of chemoattractant molecules, physiologically involved in the migration of immune cells. The CXCL12/CXCR4 signaling axis is also known to be important for tumor cell migration: CXCR4 expressed on tumor cells, provides a means of homing for metastatic cells to target organs [35]. Due to its implication with tumor dissemination, CXCR4 overexpression has been linked to a poor prognosis in breast cancer patients [35].

Surprisingly, on the contrary, it should be the negative association, pointed out by PCA-based biplots, between Cluster 2 tumors and BMI1 expression. That, because the role of BMI1 gene in self-renewal of stem cells and as an oncogene in many human cancers where it induces EMT. Although Bmi1 overexpression has been correlated with poor prognosis in several tumor types, a recent study has indicated that, in breast cancer, high Bmi1 expression is limited to the luminal subtype and that it is associated with a good outcome [36]. Under this light, the positive association that we observed between BMI1 expression and Cluster 3 tumors, consistent with the luminal-like phenotype, seems to provide a transcriptomic support to this clinical evidence.

Cluster 1 tumors, which are prevalently p53 wild-type, are more difficult to categorize. Dissimilarly from basal-like and luminal-like, these tumors had an indefinite phenotype characterized by the coexpression of luminal and basal cytokeratins. In addition, the overexpression of several transcription factors (ID2, ID4, SNAI2, TWIST1, ZEB2), known to be under TGF-β control, and the concomitant overexpression of some genes coding for stemlike properties (ABCG2, JAG1, JAG2, NANOG, NOTCH4) makes it difficult to have a correct interpretation of the results. Indeed, it is not easy to establish whether such a phenotypical heterogeneity represents an intermediate step of an EMT-like process, in which tumor cells gain characteristics of mesenchymal cells but have not completely lost epithelial characteristics, or it is simply due to the individual heterogeneity of the tumors forming the cluster.

4. Conclusions

Aim of this in silico study was to investigate the association between TP53 mutational status and expression of a panel of genes related to TGF-β induced EMT and stemlike features, using a publicly accessible microarray dataset consisting of 251 p53-sequenced primary breast cancers. According to recent experimental evidence linking loss of p53 function, induction of EMT and acquisition of stemness properties in different tumor cell lines [11-13], we expected an evident positive association between EMT-related genes and p53 mutations, in particular with severe p53 mutations. In addition, since clinical evidence indicates that p53 mutations are prevalently associated with the pathologic class of triple-negative breast cancers, we expected an overexpression of EMT-related genes in this specific subset of tumors. Our analysis supports the notion that mutant p53 tumors (especially those harboring a severe p53 mutation) were consistent with the aggressive clinic class of triple-negative cancers, but it clearly indicates that, differently from cell cultures [11-13], surgical tumors did not overexpress TGF-β-related transcription factors. Taking into account the physiological role of TGF-β in mammary gland differentiation [25, 26], these unexpected findings seem to suggest an interesting relationship between p53 mutation, mammary cell dedifferentiation, and the concomitant acquisition of stemlike properties which improve tumor cells aggressiveness.

Supplementary Table: lists the 352 Affymetrix probe sets corresponding to the 147 genes entered in the study. For each probe set sensitivity and specificity score are also provided.

Supplementary Figure 1: shows shrunken centroids for wild-type TP53 and mutant TP53 tumors in Langerod dataset (for details see Statistical Analysis description).

Supplementary Figure 2: shows the boxplots of the genes associated with EMT in wild-type (WT) or mutant (mut) TP53 tumors with respect to overall case series (all), in Langerod dataset (for details see Statistical Analysis description).

Supplementary Figure 3: shows the boxplots of the ESR1, PGR, ERBB2, TP53 and TP53INP genes in the three main clusters identified by unsupervised hierarchical cluster analysis using the subset of genes coding for luminal and basal markers, ERBB2 and claudins (for details see Statistical Analysis description).

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.