p21 suppresses inflammation and tumorigenesis on pRB-deficient stratified epithelia

Abstract

The retinoblastoma gene product (pRb) controls proliferation and differentiation processes in stratified epithelia. Importantly, an in contrast to other tissues, Rb deficiency does not lead to spontaneous skin tumor formation. As the cyclin dependent kinase inhibitor p21 regulates proliferation and differentiation in the absence of pRb, we analyzed the consequences of deleting p21 in pRb-ablated stratified epithelia (hereafter pRb^ΔEpi;p21-/-). These mice display an enhancement of the phenotypic abnormalities observed in pRb^ΔEpi animals, indicating that p21 partially compensates pRb absence. Remarkably, pRb^ΔEpi;p21-/- mice show an acute skin inflammatory phenotype and develop spontaneous epithelial tumors, particularly affecting tongue and oral tissues. Biochemical analyses and transcriptome studies reveal changes affecting multiple pathways, including DNA damage and p53-dependent signaling responses. Comparative metagenomic analyses, together with the histopathological profiles, indicate that these mice constitute a faithful model for human head and neck squamous cell carcinomas. Collectively, our findings demonstrate that p21, in conjunction with pRb, plays a central role in regulating multiple epithelial processes and orchestrating specific tumor suppressor functions.

Introduction

The pocket protein family (pRb, p107 and p130) plays unique and overlapping roles in differentiation and cell cycle control (1). Of them, the Rb gene is the predominant family member mutated in human tumors. This reflects the essential role of pRb in the control of the G1 to S phase transition in cell cycle (2–4). Rb encodes a nuclear phosphoprotein that actively represses genes required for the G1-S phase transition through the formation of complexes with members of the E2F transcription factors family (5). During normal cell cycle progression, pRb is functionally inactivated through phosphorylation by cyclin/cdk complexes. This releases the E2Fs allowing the G1 to S phase transition. Consistent with this, forced expression of pRb in cells leads to growth arrest, and restoration of Rb function in pRb-deficient tumors suppresses tumor growth (6). pRb inactivation is further regulated by two families of cdk inhibitors or cki that regulate the cdk kinase activity (7, 8).

The protein p21, the Cdkn1a gene product, belongs to the cki family and is a critical mediator of p53-induced cell-cycle arrest (9–11). The evidence for the tumor suppressor activity of p21 comes from the analysis of p21-deficient mice (12–14), which display increased susceptibility upon carcinogenic treatments (15–18) and also develop spontaneous tumors in aged mice (19). Moreover, p21 deficiency aggravates the phenotype of mice with a variety of genetically engineered oncogenic alterations (20–24). Interestingly, independent from its functions in cell cycle, p21 also modulates differentiation and apoptosis, through the direct association and control of several transcription factors and/or transcription coactivators, in specific cell types such as keratinocytes (25–30).

The epidermis is a unique model to study proliferation and differentiation. Different lines of evidence suggested the involvement of the pRb pathway in epidermal homeostasis (31–35) and in the mouse skin carcinogenesis (8, 24, 36, 37, 38–40). However, due to the early embryonic lethality displayed by Rb-deficient mice (41–43), the role of Rb in mouse skin carcinogenesis has not been established until recently. We and others have used the Cre/loxP technology to inactivate Rb gene in epidermis (44, 45). The pRb-deficiency in epidermis (pRb^ΔEpi mice) promotes increased proliferation and altered differentiation leading to moderate epidermal hyperplasia and hyperkeratosis (45). However, no spontaneous tumors were observed in these mice (45) unless the related gene p107 is also ablated (46, 47). Moreover, the absence of pRb leads to reduced skin tumor development and increased rate of malignant conversion upon chemical carcinogenesis experiments (48, 49). These effects are mainly mediated by acute induction of p53, which creates a selective pressure in the benign tumors leading to premature p53 loss and thus malignant conversion (48). In agreement, we found that pRb and p53 cooperate to suppress mouse epidermal tumors (50) and p107 loss reduces the p53 transcriptional functions in pRb^ΔEpi mice (47). Remarkably, spontaneous tumors arising in pRb^ΔEpi;p53^ΔEpi mice are highly aggressive, display early signs of chromosomal instability and high metastatic behavior (50–52). The genomic profiling of these spontaneous tumors also reveal a significant overlap with human malignancies of multiple tissue origin distinguished by poor prognosis, altered p53 status and high metastasis incidence (53, 54).

Given that i) p21 is a bona fide transcriptional target of p53 (9, 11), ii) the reported functional cooperation between pRb and p21 (20–22), and iii) the involvement of p21 in epidermal homeostasis and carcinogenesis (15–18, 25, 26, 28), we sought to generate mice lacking p21 in the absence of epidermal pRb (Rb^F19/F19;K14cre;Cdkn1a-/-, hereafter pRb^ΔEpi;p21-/-mice). Importantly, and in contrast to the individual gene knockouts, pRb^ΔEpi;p21-/-mice develop an skin inflammatory processes followed by spontaneous tumor development. Based on their histopathology and transcriptome analyses, these tumors resemble human head and neck squamous carcinomas.

Results

Generation and epidermal characteristics of pRb^ΔEpi;p21-/- mice

The absence of pRb in epidermis promotes increased proliferation and altered differentiation, but is insufficient to allow tumor development. This might indicate that other proteins may exert the functions of pRb in its absence to suppress tumorigenesis. This seems to be the case for the other pocket protein p107 but not for p130 (46, 47, 55). Importantly, these overlapping roles require in some cases the upregulation of the compensating protein (56). In genome-wide transcriptome analysis of primary pRb^ΔEpi keratinocytes and epidermis we have previously observed a clear induction of the Cdkn1a gene (46, 55). This is also observed at the protein level (Fig 1A). This induction suggests that an active pathway involving this cki may limit the growth of pRb-deleted in epidermis as previously reported in other systems (20–22).

Figure 1. Epidermal phenotype of pRb^ΔEpi;p21-/- mice.

A) Representative Western blot of primary keratinocyte protein extracts showing the expression of the quoted proteins. Each lane contains a pool of extracts from 3 different mice of each genotype. The number below each lane denotes de densitometric value of each band normalized according the corresponding actin value. B) Aspect of adult mice bearing epidermal-specific Rb loss, p21- null and doubly deficient mice. C, D) Detail of the head (C) and spontaneous wounds (D) in a pRb^ΔEpi;p21-/- mouse. E-H’) Representative histology appearance of skin sections from pRbF/F (E), p21-/- (F), pRb^ΔEpi (G), pRb^ΔEpi;p21-/- (H) mouse by post natal day (pnd) 10 and in pRb^ΔEpi;p21-/- by pnd 30 (H′). Note the epidermal phenotype in pRb^ΔEpi;p21-/- mice characterized by hyperplasia, hyperkeratosis, parakeratosis (denoted by arrows in inset of H’) and dermal cell infiltrates. I-L’) Characterization of epidermal differentiation in pRbF/F (I, I’), p21-/- (J, J’) pRb^ΔEpi (K, K’) and pRb^ΔEpi;p21-/- (L, L’) mouse showing the expression of keratin K5 (in red in I, J, K, L), keratin K6 (in red in I’, J’, K’, L’) and keratin K10 (in green) by pnd 30. Yellow staining indicates co-expression. Bars=100μm.

To confirm this hypothesis, pRb^ΔEpi mice were bred with p21-deficient mice (12) to generate double deficient mice. These mice (pRb^ΔEpi;p21-/-), in an enriched FVB background, were obtained at the expected mendelian ratio. However soon after birth, mutant animals developed a characteristic phenotype that included reduced size, frail appearance, scaliness (hyperkeratosis) and a very sparse hair coat (Fig 1B-D). The hair loss was more evident in the face and snout (Fig 1C). We also noticed the presence of spontaneous wounds (Fig 1D).

Compared to Rb^F/F as control mice (45) (Fig 1E), the epidermis of mice lacking p21 (Fig 1F) was indistinguishable and displayed no significant abnormalities, in agreement with our previous observations (24). In contrast, the epidermis of mice bearing specific Rb ablation displayed a mild hyperplasia (increased number of cell layers) and hyperkeratosis (thickening of the stratum corneum) (Fig 1G, see also (45)). These alterations become clearly aggravated by the simultaneous loss of Cdkn1a gene (Fig 1H). The pRb^ΔEpi; p21-/- mouse epidermis showed severe hyperplasia and hyperkeratosis. By post-natal day (pnd) 30, we also observed inflammatory infiltrates (Fig 1H’), epithelial pustules (microabscesses denoted by an arrow in Supp Fig S1A) and parakeratosis (presence of nucleated cells in the stratum corneum, indicating altered terminal differentiation, denoted by arrows in inset of Fig 1H’). The analysis of several differentiation markers confirmed these alterations, with expanded keratin K5 expression, keratin K10 expression confined to uppermost cell layers and also massive expression of keratin K6 in the interfollicular epidermis (Fig 1L, L’). All these characteristics were dramatically aggravated as compared with pRb^ΔEpi (Fig 1K, K’) or p21 deficient epidermis (Fig 1J, J’), which in turn is undistinguishable from control mice (Fig 1I, I’).

The specific ablation of pRb in epidermis results in increased proliferation in the basal layer and also in aberrant proliferation in suprabasal layers (45). We thus investigated whether the subsequent loss of p21 also aggravates this effect. Analysis of BrdU incorporation in control (Fig 2A, F), p21-/- (Fig 2B, F), pRb^ΔEpi (Fig 2C, F) and pRb^ΔEpi;p21-/- (Fig 2D, F) skin demonstrated significantly increased proliferation in the basal epidermal layer of double deficient mice. Notably, the proliferation in the suprabasal layers, characteristic of pRb^ΔEpi epidermis, was also significantly increased in pRb^ΔEpi;p21-/- mice (Fig 2E, G). Increased proliferation was accompanied by increased E2F activity as measured by luciferase experiments in primary keratinocytes (Fig 2H), likely as a result of the increased E2F1 protein levels observed (Fig 1A). Finally, we also analyzed whether the disengagement between proliferation and differentiation processes, which is characteristic of pRb^ΔEpi mouse keratinocytes, was also aggravated in pRb^ΔEpi;p21-/- cells. The proliferation of primary keratinocytes was analyzed at different time points upon differentiation induction by increasing the calcium concentration in the culture medium (45). In agreement with our previous observations (45), the absence of pRb caused a delay in the cell cycle arrest compared with control or p21-/- cells; however, by 48-72 hours in high calcium, control, p21-/- and pRb^ΔEpi keratinocytes displayed no significant BrdU incorporation, whilst in pRb^ΔEpi;p21-/- a considerable number of cells still proliferate (Fig 2I). Moreover, when terminally differentiated control cells where re-stimulated by decreasing calcium, they were unable to re-enter cell cycle, whereas pRb^ΔEpi and pRb^ΔEpi;p21-/-, and to a minor extent p21-/- cells, displayed rates of proliferation close to those observed in non-differentiated cells (Fig 2I).

Figure 2. Epidermal proliferation in pRb^ΔEpi;p21-/- mice.

A-D) Representative double immunofluorescence images showing BrdU incorporation (in green) in epidermis of pRbF/F (A), p21-/- (B), pRb^ΔEpi (C) and pRb^ΔEpi;p21-/- (D, E) mouse along with K5 (in red in A, B, C, D) or K10 (in red in E). Bars=100μm. Dashed line in (E) denotes the epidermal dermal boundaries. F) Quantitative analysis of BrdU incorporation in K5-positive basal cells in mice of the quoted genotype. G) Quantitative analysis of BrdU incorporation in K10-positive suprabasal cells in mice of the quoted genotype. Data come from three independent experiments scoring five sections of at least three age mated mice of each genotype and are shown as mean ± SEM. p Values were obtained by one-way ANOVA and the corresponding comparison by Bonferroni test. H) Relative luciferase activity of E2F responding elements in primary keratinocytes of the quoted genotypes. Data come from three independent experiments. The luciferase values were normalized to those obtained in control RbF/F keratinocytes and are shown as mean ± SEM. p Values were obtained by one-way ANOVA and the corresponding comparison by Bonferroni test. I) Percentage of BrdU incorporation in primary keratinocytes of the quoted genotypes growing under low (0.05 mM) and high (1.2 mM) Ca²⁺ medium for the indicated times and re-stimulated with low Ca²⁺ medium. Data come from the analysis of three independent experiments scoring at least 1000 cells on each and are shown as mean ± SEM. p Values were obtained by two-way ANOVA and Bonferroni post test.

These results demonstrate that, whereas the loss of p21 causes minor epidermal defects by itself, it leads to an important aggravation of the phenotype generated by the epidermal specific ablation of Rb gene producing severe gross skin alterations affecting proliferation and differentiation of the epidermal cells.

Inflammatory processes in pRb^ΔEpi;p21-/- mouse skin

As abovementioned, by pnd 30 the epidermis of pRb^ΔEpi;p21-/- mice is characterized by massive inflammatory infiltrates forming abscesses (Supp Fig S1A) that cause epidermal swelling and fragility (Supp Fig S1B) leading to the formation of wounds (Fig 1D, and Supp Fig S1B, C). This may contribute to the severe reduction of mouse viability (Supp Fig S1D). Although we had previously observed a minor inflammation in adult pRb^ΔEpi (52), these observations were unexpected, so we aimed to characterize them in detail. Compared to control (Fig 3A-D), p21-/- (Fig 3A’-D’) or pRb^ΔEpi (Fig 3A”-D”), the skin of pRb^ΔEpi;p21-/- displayed large numbers of lymphocytes, macrophages, mastocytes and γδT cells (Fig 3A’”, B”’, C”’, D”’ and Supp Fig S1E). Of note, such inflammatory events were not detected in newborn skin (Supp Fig S2)

Figure 3. Inflammation in skin of pRb^ΔEpi;p21-/- mice.

A-F”’) Representative immunohistochemistry (A-C”’ and F-F”’) and immunofluorescence (D-D”’, F”’) images showing the presence of T cells (CD3ε positive) (A-A”’, F), macrophages (F480 positive) (B-B”’, F’), mastocytes (C-C”’, F”) and γδT cells (D-D”’, E”’) in pRbF/F (A, B, C, D), p21-/- (A’, B’, C’, D’) pRb^ΔEpi (A”, B”, C”, D”) and pRb^ΔEpi;p21-/- (A”’, B”’, C”’, D”’) mouse skin and pRb^ΔEpi;p21-/- skin grafts (F, F’, F”, F”’). Bars=100μm. E) Representative Western blots showing the expression of the specified proteins in newborn skin extracts of mice of the quoted genotype. Each lane contains a pool of extracts from 3 different mice of each genotype. The corresponding densitometric analysis of the different bands normalized according the corresponding Actin value is provided in Supp Fig S3.

The inflammatory processes in skin are associated with increased NFκB and Stat3 signaling and the production of specific cytokines (57). To monitor whether these pathways are affected in pRb^ΔEpi;p21-/- prior to overt inflammatory events, we studied the expression of key elements of these signaling pathways in newborn skin samples by western blot. The results (Fig 3E, and Supp Fig S3) demonstrated activation of Stat3 and NFκB in double deficient mouse epidermis as indicated by increased phosphorylation of Stat3 and p65, and the reduced expression of IκBα (Fig 3E and Supp Fig S3B, D, H). In addition, with the exception of a moderate increase of p65 in pRb^ΔEpi and pRb^ΔEpi;p21-/- newborn mouse epidermis (Fig 3E and Supp Fig S3C), we observed no significant differences in the expression of different signaling elements of these pathways such as total Stat3, IKKβ, IKKα, and/or IKKγ (Fig 3E and Supp Fig S3A, E, F, G).

Next we analyzed, using an antibody array, the expression of several cytokines. As the loss of p21 is generalized (i.e.: also affects the possible inflammatory cells) and several of these cytokines were also produced by inflammatory cells, we included in the analysis protein extract of epidermis of pnd 30 from all genotypes, new born pRb^ΔEpi;p21-/- mouse epidermis (without inflammatory signs) and from grafts of new born pRb^ΔEpi;p21-/- mouse epidermis on NOD/SCID mice after 3 months of transplantation. Notably, we noticed the increased presence of inflammatory cells in these grafts (Fig 3F-F”’), in spite of the immunodeficiency of the host. This probably indicates that these immune cells were already present in the newborn grafted skin, which did not show evidences of inflammation, and they proliferated, and survived throughout the entire transplant span. Examples of antibody cytokine array are provided in Fig 4A. The quantitative analysis of the experiments (Fig 4B) showed, in comparison with control, p21-/- and pRb^ΔEpi skin extracts, the increased protein levels of GCSF, GMCSF, IL16, CXCL1, CCL3, CXCL2, TREM1, and the decrease of CCL1 and CXCL9 in pRb^ΔEpi; p21-/- extracts. However, their relative protein levels varied among adult mice (all the above commented changes), new born (upregulation of GMCSF, CXCL1 and TREM1 and repression of CXCL9 and CCL1) and grafted skin (upregulation of GCSF, GMCSF, IL16, CXCL1, CXCL2, TREM1 and repression of CXCL9 and CCL1). The comparative analysis of these three situations (overt inflammatory state in pRb^ΔEpi;p21-/- mouse epidermis by pnd 30, no inflammation in newborn pRb^ΔEpi;p21-/- mouse epidermis and skin grafts) indicate that the upregulation of GMCSF, CXCL1 and TREM1 and repression of CXCL9 and CCL1 is probably mediated by the epidermal keratinocytes rather than being exclusively produced by the inflammatory cells recruited to the skin. Regarding the specific changes, we detected that cytokines responsible for the recruitment of: macrophages (GMCSF) and neutrophils (CXCL1) were particularly increased. Remarkably CXCL1 has been shown to promote epidermal wound healing and also squamous tumor development (58). We also detected a significant increase of the inflammatory response enhancer (TREM-1), which is up-regulated in tumor-associated macrophages (TAMs) (59), and the reduction of CXCL9, a chemoattractant of peripheral blood lymphocytes and NK cells, with anti-angiogenic activity (60).

Figure 4. Cytokine production in skin of pRb^ΔEpi;p21-/- mice.

A, B) Representative example of pnd 30 mice skin samples (A) and summary (B) of the cytokine arrays analysis of whole skin protein extracts (n= 3) obtained from mice of the quoted genotypes. In pRb^ΔEpi;p21-/- the cytokine expression of newborn skin and graft skin samples (n=3) were included for comparison. Data come from three independent experiments. The densitometric values were normalized to those obtained in control RbF/F samples and are shown as mean ± SEM. * denotes a p Value<0.05 as obtained by two-way ANOVA and Bonferroni post test.

Development of spontaneous tumors in pRb^ΔEpi;p21-/- mice

A progressive weakening and reduced survival in doubly deficient mice was observed. We thus carried out full necropsies in a large cohort of mice of the different genotypes. Upon examination of sections of various tissues, multiple spontaneous tumors were found (Table 1). No spontaneous tumors were observed in control or pRb^ΔEpi mice (n=50 of each) in agreement with our previous data (45). In p21-/- mice, we observed a reduced tumor incidence (4/44 mice), which, as reported (19), only occurred in aged mice (48-52 weeks of age). On the contrary, we found a very high tumor incidence in pRb^ΔEpi;p21-/- mice (39/44 mice, including 5 mice sacrificed by 2-5 weeks showing no macroscopic growth). These tumors were exclusively of stratified epithelia origin and predominantly correspond to squamous cell carcinomas (SCCs) (Table 1). Tumors were first observed by 2 weeks of age and the vast majority of the mice studied displayed tumors by 40 weeks (Fig 5A). This spontaneous tumor development in pRb^ΔEpi;p21-/- mice could reduce the viability of these mice. Of a particular relevance, we found that 21 of the 44 pRb^ΔEpi; p21-/- mice analyzed displayed SCCs in the tongue, detected as white overgrowths affecting primarily the dorsolateral areas (Fig 5B, B’), which also showed massive γδT cell infiltration (Fig 5C). Moreover, and contrary to all the other genotypes (Fig 5D), 3 out of 5 transplants generated with pRb^ΔEpi;p21-/- skin evolved to form outgrowths suggestive of tumor formation (Fig 5E). Histology analyses confirmed that they correspond to papillomatous lesions and well-differentiated SCCs (Fig 5F) characterized by high grade of differentiation and increased proliferation (Fig 5F’).

Table 1. Tumor development in pRb^ΔEpi;p21-/- mice.

Tumor type	Localization	Incidence (%)*	Age**
SCC	Lips	10/44 (22.7%)	2
SCC	Tongue	21/44 (47.7%)	4
SCC	Oral epithelia	3/44 (6.8%)	4
SCC and CIS	Skin	5/44 (11.3%)	4
Adenosquamous CC	Eyelid	1/44 (2.3%)	6
SCC and CIS	Ear skin	4/44 (9.1%)	12

^*Number of mice showing this specific tumor (percentage in mouse cohort)

^**Weeks of age at which the first tumor was observed

SCC: Squamous Cell Carcinoma

CIS: Carcinoma In Situ

AdenosquamousCC: Adenosquamous Cell Carcinoma

Figure 5. Spontaneous epithelial tumors observed in pRb^ΔEpi;p21-/- mice.

A) Kaplan Meier curves showing the spontaneous tumor development in mice of the quoted genotypes. p Value was obtained by log rank test. Macroscopic appearance (B), and histology (B’) of spontaneous tongue tumor in pRb^ΔEpi;p21-/- mice. C) Immunofluorescence image showing the presence of γδT cells in the boundaries of a spontaneous tongue tumor in pRb^ΔEpi;p21-/- mice. Macroscopic appearance (D, E) and histology (F) of skin grafts from pRb^ΔEpi (D) or pRb^ΔEpi;p21-/- (E, F). F’) Immunofluorescence image showing BrdU incorporation (in green) along with K5 expression (in red) in skin grafts from pRb^ΔEpi;p21-/-Bars= 200μm.

In conclusion, the combination of pRb and p21 deficiency results in the spontaneous development of tumors of stratified epithelia origin. This is in contrast with the absence of tumors in mice bearing the epithelial-specific ablation of Rb gene (45). In addition, it is worth mentioning that most of these tumors were not previously reported to occur in p21-null mice, or they appeared at very long latency, thus indicating that pRb and p21 cooperate to suppress tumor development in stratified epithelia.

Deregulated gene expression in pRb^ΔEpi;p21-/- mouse epidermis

To investigate the possible molecular mechanisms underlying the aggravated phenotype, and in particular, the tumor susceptibility displayed by pRb^ΔEpi;p21-/- mice, we carried out microarray analyses from whole skin RNA of quoted genotyped newborn mice (see Supp Materials and Methods). We found that, compared with control mouse skin, the simultaneous loss of Rb1 and Cdkn1a genes leads to the overexpression of 1445 transcripts and underexpression of 1065 transcripts (Supp Table S1, Supp Table S2 and Fig 6A). Enrichment analysis of gene ontology terms (61) indicated that the underexpressed genes were predominantly involved in oxido-reduction and electron chain transport processes, whereas cell adhesion, cytokine-mediated signaling pathway, epithelium development and cell proliferation were the functions predominantly affected by the overexpressed genes, in line with the observed epidermal phenotype of pRb^ΔEpi;p21-/- mice (Fig 6B). Of note, we also observed that multiple genes affecting specific pathways highly relevant to epidermal homeostasis and tumorigenesis, displayed deregulated expression. These included Hedgehog (Shh, Ptch1, Smo, Sfrp1), Notch (Psen1, Hey2, Heyl, Notch1) and Wnt/Sox (Wnt11, Wnt5a, Wnt2, Sox13, Sox2, Sox21, Sox5, Lgr5, Ltbp3, Msx2, Ctnnb1, Myc, Ovol1) signaling pathways. We monitored the expression of some of these genes by qRT-PCR (Supp Fig S4A, B, E). Furthermore, as we found deregulated expression of genes acting as activators and inhibitors of these pathways, we also monitored their relative functionality by using common readout genes. The increased expression of Gli1 (Supp Fig S4C) and Hes1 (Supp Fig S4D) supported the activation of Shh and Notch pathways in the epidermis of pRb^ΔEpi;p21-/- mice. The analysis of Lgr5, Myc, Ovol1 and Axin2 (Supp Fig S4E), and the determination of increased active βcatenin by western blot (Supp Fig S4F) also indicated overactivation of Wnt pathway.

Figure 6. Gene expression of pRb^ΔEpi;p21-/- newborn mouse skin.

A) Heatmap showing the expression of differentially expressed genes between RbF/F, Rb^ΔEpi;p21-/-, and pRb^ΔEpi;p21-/- skin samples. The corresponding selected probes (Supp Table S1 and S2) were obtained upon microarray analysis and selection as described in Material and Methods. B) Gene Ontology themes with significant enrichment observed in the pRb^ΔEpi;p21-/- skin samples (red denotes overexpressed and blue underexpressed genes). C, C’) Summary of Chea data showing the presence of p63 (C) and E2Fs (C’) binding in genes upregulated or downregulated in Rb^ΔEpi;p21-/- newborn mouse skin. D) Quantitative RT PCR analysis showing the expression of ΔN and TA p63 isoforms in newborn skin samples from the quoted genotypes. E) Quantitative RT PCR analysis showing the expression of different E2F members in newborn skin samples from the quoted genotypes. The values for each genotype in D and E are represented as mean±SEM. p Values were obtained using one way ANOVA and the specific comparison by Bonferroni post-test.

Next, we studied potential transcription factors involved in the observed gene deregulation. Chromatin Immunoprecipitation Enrichement Analysis (ChEA (62)) revealed a potential involvement of multiple transcription factors in the upregulated and downregulated genes, including E2Fs, Trp53 and Trp63, among others (data not shown). Of note, E2Fs were predicted to bind not only upregulated, but also downregulated genes, whilst p63 appears predominantly associated with the upregulated genes (Fig 6C, C’). The determination of ΔNp63 and TAp63 isoforms by qRT-PCR revealed a significant upregulation of the ΔNp63, but not of the TAp63 isoform in the pRb^ΔEpi;p21-/- mouse epidermis, and to a minor extent in p21-/- mice (Fig 6D). Regarding the E2F transcription factors, we found increased expression of activator E2Fs (E2f1 and E2f2) and also repressor E2Fs (E2f4, E2f5 and E2f7) in the epidermis of pRb^ΔEpi;p21-/- newborn mice (Fig 6E).

Finally, Gene Set Enrichment Analysis (GSEA) (63) was used to analyze whether gene expression differences between control and pRb^ΔEpi;p21-/- newborn skin display enrichment of gene sets collected from various sources (see Supplementary Materials and Methods and Supp Table S3 and S4). Among others, we found a significant enrichment of deregulated genes in pRb^ΔEpi;p21-/- with that observed in the skin of pRb^ΔEpi;p53^ΔEpi, and p53^ΔEpi newborn mice (Table 2 and Supp Table S3) suggesting a major role of p21 in the p53-dependent signaling in epidermis. This is further reinforced by the significant overlap with other gene sets relative to p53-dependent gene expression, DNA damage and DNA repair (Table 2).

Table 2. Summary of GSEA analysis of deregulated genes in pRb^ΔEpi;p21-/- mouse skin.

Gene Set Name (N)1	Description (reference)	Size	NES	FDR q-Val
Martinez_Rb1_And_Tp53_Targets_Up (188)	Genes up-regulated in mice with skin specific double knockout of both RB1 and TP53 by Cre-lox (50).	560	2.60	<0.0001
Martinez_Tp53_Targets_Up (181)	Genes up-regulated in mice with skin specific knockout of TP53 (50).	559	2.54	<0.0001
Mcmurray_Tp53_Hras_Cooperation_Response_Up (10)	Up-regulated ‘cooperation response genes’: responded synergistically to the combination of mutant TP53 and HRAS in YAMC cells (colon) (101)	19	1.94	0.001
Brocke_Apoptosis_Reversed_By_Il6 (33)	Genes changed in INA-6 cells (multiple myeloma) by re-addition of IL6 after its initial withdrawal for 12h. (102)	107	1.70	0.035
Dutta_Apoptosis_Via_Nfkb (10)	NF-kB target genes involved in the regulation of programmed cell death (103)	26	1.64	0.033
Inga_Tp53_Targets (5)	Genes whose promoters contain TP53 response elements (104)	15	1.60	0.045
Kyng_Dna_Damage_Up (20)	Genes with GO annotation and up-regulated after DNA damage in cell lines from young donors. (105)	84	1.55	0.090
Reactome_P53_Independent_Dna_Damage_Response (30)	Genes involved in p53-Independent G1/S DNA damage checkpoint (http://www.reactome.org/)	42	-1.71	0.057
Pyeon_Hpv_Positive_Tumors_Up (22)	Downregulated genes in cervical carcinoma and head and neck tumors positive for human papilloma virus (HPV) compared to those negative for HPV. (93)	47	-1.97	0.027
Martinez_Tp53_Targets_Dn (190)	Genes down-regulated in mice with skin specific knockout of TP53 (50).	520	-2.83	<0.0001
Martinez_Rb1_And_Tp53_Targets_Dn (223)	Genes down-regulated in mice with skin specific double knockout of both RB1 and TP53 by Cre-lox (50).	516	-2.90	<0.0001

¹N: number of enriched genes

NES: normalized enrichment score.

NES >0: enrichment in pRb^ΔEpi;p21-/-

NES<0: enrichment in control

FDR: False discovery rate. Significant < 0.25

SIZE: Number of genes within each gene set

DNA damage in pRb^ΔEpi;p21-/- mouse epidermis

The GSEA revealed a possible activation of the DNA damage response (DDR) in pRb^ΔEpi;p21-/- mouse epidermis (Table 2). To confirm these indications, we monitored DDR in mouse skin by analyzing the expression of phosphorylated histone H2AX (γH2AX). While no significant evidences of γH2AX were observed in control, p21-/- or pRb^ΔEpi mouse skin (Fig 7A, B, C), clear nuclear staining was observed in pRb^ΔEpi;p21-/- mice 30 days after birth (Fig 7D). Moreover, positive γH2AX was also observed in samples by 10 days after birth (Fig 5D’) and in skin transplants (Fig 7D”), predominantly affecting suprabasal keratinocytes, and in parallel with p53 induction (Fig 7E). Similar γH2AX induction was biochemically detected in whole skin protein extracts (Fig 7F). Noteworthy, the γH2AX signal in most cases covered all the nucleus, which is the pattern observed upon exposure to replication stress (RS) generating agents such as hydroxyurea or ATR inhibitors (64). In addition, the possible involvement of RS in the DNA damage of pRb^ΔEpi;p21-/- mouse epidermis is also suggested by the predominant presence of γH2AX signal in suprabasal cells: as the cells committed in the differentiation program do not cease DNA replication due to the simultaneous absence of Rb and p21 (Fig 2E, G, I), they could accumulate damages. Also, in agreement with the damage being of replicative origin, the biochemical analyses showed an activation of the ATR-Chk1 pathway (Fig 7G and Supp FigS5), which coordinates the response against RS (65, 66). The RS leads to the accumulation of single stranded DNA, which is prone to recombine and initiate genomic rearrangements (65, 66). Importantly, the RS-damaged cells are predominantly eliminated through mitotic catastrophe rather than apoptosis due to the presence of unreplicated chromosomes in mitosis‥ We thus analyzed the presence of aberrant mitosis and mitotic catastrophes in epidermal skin sections. We found these aberrancies only in pRb^ΔEpi;p21-/- mouse epidermis (Supp Fig S6). On the contrary, we did not detect significant apoptotic events in the skin or in tumors of pRb^ΔEpi;p21-/- mice, (Fig S6E’, F’, G) despite the increase in p53 levels in the skin or in tumoral lesions (Fig 7G and Supp Fig S6E, F). Thus, and as previously shown for the activation of oncogenes (67), concomitant depletion of Rb and p21 on the skin leads widespread levels of RS, which activate the DDR-p53 axis.

Fig 7. DNA damage response in pRb^ΔEpi;p21-/- mice.

A-D”) Representative images of the immunohistochemistry detection of histone γH2AX in the epidermis of pRbF/F (A), p21-/- (B), pRb^ΔEpi (C) and pRb^ΔEpi;p21-/- (D-D”) mice by pnd 30 (A, B, C, D) pnd 10 (D’) and in pRb^ΔEpi;p21-/- skin grafts (D”). Bar=50μm. E) Double immunofluorescence showing the co-expression of p53 (red) and histone γH2AX (green) in the epidermis of pRb^ΔEpi;p21-/- mice. Bar=10μm. Dashed line denotes the epidermal-dermal boundaries. Inset denotes a high magnification of double labeled nuclei. F) Western blot analysis showing histone γH2AX expression in skin extracts of mice of the quoted genotype. Histone H2AX was used as loading control. The number below each lane denotes de densitometric value of each band normalized according the corresponding H2AX value. G) Western blot analysis of the quoted DNA damage proteins in the epidermis of mice of the quoted genotypes. In F and G, each lane comes from an independent sample. The corresponding densitometric analysis of the different bands normalized according the corresponding Actin value is provided in Supp Fig S5.

pRb^ΔEpi;p21-/- mice represent a model of human Head and Neck Squamous Cell Carcinoma (HNSCC)

The SCC development observed in oral tissues and tongue of pRb^ΔEpi;p21-/- mice might indicate that these genetically engineered mice would represent a model of HNSCC. Human HNSCC is the sixth most common human cancer worldwide and, in spite of recent progress in therapeutic management of this disease, the improvement of overall survival in HNSCC patients is still low (68). Therefore the development of animal models recapitulating the characteristics of human HNSCC would allow advances of new molecularly targeted therapies (68–70). Moreover, our GSEA analysis also showed a significant overlap with previous transcriptome studies of human HNSCC (Table 2, Supp Table S3 and Supp Table S4). Consequently we aimed to validate pRb^ΔEpi;p21-/- mice as a potentially suitable model of human HNSCC.

To this, we compared the genes overexpressed in pRb^ΔEpi;p21-/- mice with human HNSCC transcriptome studies using the Oncomine database (71, 72). The comparison revealed a significant overlap between overexpressed mRNAs in pRb^ΔEpi;p21-/- newborn mouse epidermis with homologous mRNAs overexpressed in multiple HNSCC expression datasets (Supp Table S5). This overlapping allowed us to obtain a signature of commonly overexpressed gene both in mouse epidermis and human HNSCC studies selected (Fig 8A). Of note, we also observed a significant overlap between the upregulated mouse genes and those downregulated in human HNSCC present in Oncomine database and characterized by increased stage, grade and metastatic likelihood, poor clinical outcome, and also in metastasis compared with primary tumors (Supp Table S6). These findings would suggest that the pRb^ΔEpi;p21-/- mice might represent a faithful model of good prognosis HNSCC. To test this aspect, we monitored the mouse-human signature in a gene expression dataset which contains HNSCC primary tumors with follow-up information of metastatic events (73), and showing a significant overlap by GSEA (Fig 8B, B’, C). Unsupervised sample clustering using this common gene signature identified two patient groups (Fig 8D), with statistically significant differences in metastasis-free survival (Fig 8E). Importantly, samples with poor clinical outcome display underexpression of the pRb^ΔEpi;p21-/- signature genes, reinforcing the hypothesis that the pRb^ΔEpi;p21-/- mice might represent a potential model of human HNSCC characterized by good prognosis. Importantly, human HNSCC is characterized by intense immunosuppression (74) probably mediated by myeloid derived suppressor cell populations (MDSCs). These cells are a heterogeneous population of early myeloid cells, which expand as a consequence of chronic inflammation in response to proinflammatory mediators. The MDSCs are mainly found in peripheral blood and tumor tissue and to a lesser extend in lymph nodes. Phenotypically, in mouse-bearing tumors, these cells are identified by the expression of Gr-1+CD11b+ surface markers with subsets expressing CD49d representing the most suppressive ones (75). We observed massive enlargement of submaxilar and inguinal lymph nodes in pRb^ΔEpi;p21-/- mice (Supp Fig S7A) and increased levels of Gr-1+CD11b+ myeloid derived suppressor cells (MDSCs), which also express CD49d, in peripheral blood and lymph nodes (Supp Fig S7B-C). These observations parallel with the mechanism of immune evasion observed in patients with head and neck cancers (76). Finally, our comparative study in Oncomine database revealed that upregulated genes are also present in HNSCC cell lines sensitive or resistant to specific chemotherapeutic drugs (Supp Table S7), indicating that the pRb^ΔEpi;p21-/- mice could also be used as tools to test the specificity of these compounds in tumor therapy or chemoprevention.

Fig 8. Mouse-Human comparative gene expression.

A) Common gene expression signature between pRb^ΔEpi;p21-/- mouse and at least 4 out of 7 different human HNSCC studies selected From Supplementary Table S5. Only the most significant comparison from Pyeon study was included. A threshold median p-Val<0.00001 between the possible comparisons was used. B, B’, C) Examples of GSEA endplots of the comparison between genes deregulated in pRb^ΔEpi;p21-/- newborn mouse skin and different clusters of the Ricjkman dataset (73). D) Unsupervised sample clustering using common genes was done in Rickman dataset (73), which contains HNSCC primary tumors with follow-up information of metastatic events. Clustering gave rise to 2 sample groups (D), with statistically significant differences in metastasis-free survival (E). Importantly, samples with poor clinical outcome display underexpression of the pRb^ΔEpi;p21-/- signature genes.

Discussion

The retinoblastoma gene plays multiple functions in cell cycle, differentiation and apoptosis. However, in contrast with the widely reported alterations in the so called pRb-pathway, mutations of Rb gene are restricted to few human tumors. In agreement with these observations, mouse models reveal that the ablation of Rb does not produce the development of spontaneous tumors in a large number of tissues. This has been attributed in several instances to the functional compensation by other genes, such as the Rb family members p107 and p130 (77, 78). In mouse epidermis, pRb loss generates a characteristic phenotype but not the development of spontaneous tumors (45). Here we show that the phenotype due to pRb loss is further aggravated by the absence of p21, affecting differentiation and proliferation and allowing spontaneous tumor development. These findings are similar to the reported overlapping functions between pRb and p107 in this tissue (45, 46, 56), which interestingly also led to oral and perioral spontaneous tumor development (47). However, we also observed alterations, such as epidermal parakeratosis and an exacerbated inflammatory response, not observed in pRb^ΔEpi;p107- mice, suggesting that the compensating functions of p21 and p107 over pRb deficiency are different. These observations, together with the recently reported spontaneous tumor development mediated by E2F1 loss in pRb^ΔEpi mice (79) reveal complex and intricate roles of pRb and the retinoblastoma pathway components in stratified epithelia homeostasis through the interaction with other signaling pathways. In this regard, we have also observed deregulation of Wnt, Shh and Notch pathways in pRb^ΔEpi;p21-/- mouse skin.

The p21 protein participates in the pRb-pathway through the negative modulation of cdk activity (25). However, through the interaction with several proteins, such as transcription factors and/or transcription coactivators, it also has many functions independent of its roles in cell cycle modulation (25–28, 30). Interestingly, in certain situations, p21 can control DNA replication even in the absence of pRb, in part through its ability to interact with E2F transcription factors (22, 80). In this regard, we have observed increased E2F activity and deregulated expression of specific E2F family members, which may partially account for the observed deregulated gene expression in pRb^ΔEpi;p21-/- mouse skin. Collectively, it is thus conceivable that p21 can compensate the absence of pRb, which can explain its roles as tumor suppressor. Indeed, p21 deficient mice are prone to develop tumors at advanced age (19, 81), and its deficiency collaborates with other genetic mouse models to increase tumorigenesis (15, 24, 82). In addition, p21 is a transcriptional target of p53, which is also able to cooperate with pRb to prevent spontaneous tumor development (50). Of note, tumors generated in pRb^ΔEpi;p53^ΔEpi display chromosomal instability, slight inflammation (52), high metastatic behavior (51), and remarkable transcriptome similarities with human tumors characterized by increased malignant behavior and poor prognosis (53, 54). The similarities in the transcriptome between pRb^ΔEpi;p53^ΔEpi and pRb^ΔEpi;p21-/- mouse skin, and the presence of aberrant mitosis points to a major role of p21 in mediating p53-dependent signaling in this tissue, at least in the context of pRb deficiency. However, there are important differences between pRb^ΔEpi;p53^ΔEpi and pRb^ΔEpi;p21-/- mouse skin, as we did not observe aggravation of the pRb^ΔEpi phenotype in pRb^ΔEpi;p53^ΔEpi (50)., and we did not detect overt metastasis in pRb^ΔEpi;p21-/- mice.

The inflammatory response found in pRb^ΔEpi;p21-/- mouse skin is accompanied by abundant lymphocytes, macrophages, mastocytes and γδT cells. This inflammatory phenotype is concomitant to the activation of NFκB and Stat3 signaling, two processes in which p21 has a potential role (29), and increased production of inflammatory cytokines. This inflammatory milieu could favor tumor development creating an environment prone to malignancy. In this regard, the pattern of cytokines observed in pRb^ΔEpi;p21-/- mouse is similar to those observed in different types of tumors (58–60, 83). For instance, human HNSCC development requires ancillary signals from inflammatory and “tumor educated” stromal cells that contribute to the survival, proliferation and maintenance of the undifferentiated state of transformed keratinocytes through increased expression of several cytokines (84, 85). Rising evidence indicates that the targeting of pro-tumorigenic autocrine/paracrine loops might represent a valuable therapeutic avenue for HNSCC treatment, in part due to the immunosuppressive characteristics of this type of tumors (74). Unfortunately, the key signaling elements of these inflammatory responses remain largely unknown.

The pRb^ΔEpi;p21-/- mouse epidermis is also characterized by extensive DNA damage and the subsequent activation of the DNA damage response (DDR) (86). Given the roles of pRb and p21 in restricting S-phase entry, this is most likely due to replication stress (RS), which stands for the accumulation of large patches of single stranded DNA that are toxic due to their propensity to recombine and initiate genomic rearrangements (65, 66). This kind of damage is also the same as the one initiated by oncogenes, which also promote a promiscuous S-phase entry (67). In addition to restricting replicative damage, extensive DNA damage causes p21 degradation and apoptosis induction, thereby providing an independent link between DNA damage and this cki. Consistent with the known role of the DDR kinases in activating p53, mutant skin presented increased levels of this tumor suppressor, yet no evidences of apoptosis were found. Noteworthy, this could again be due to the origin of the damage being RS, which eliminates cells through mitotic catastrophe rather than apoptosis.

The tumor spectrum observed in pRb^ΔEpi;p21-/- mice together with the transcriptome studies support the hypothesis of a functional role for pRb and p21 in oral tumorigenesis and in HNSCC development. Importantly, the landscape of mutations in this type of cancer has been recently reported (87–90) and no mutations in RB1 or CDKN1A genes were found. However, indirect evidences may sustain this hypothesis. First, pRb is functionally inactivated in most human tumors, including HNSCC (91), probably through the amplification of CCND1 and inactivation of CDKN2A (87–89). In addition, p53 is frequently mutated in HNSCC (87–89), leading to altered p21 expression (91, 92). Nonetheless, an important part of human HNSCC also displays mutations leading to impaired Notch signaling and alterations in ΔNp63 expression (87–89). These findings have been associated to impaired terminal differentiation in the tumors (87–89). We observed, increased expression of ΔNp63 and deregulated expression of Notch pathway genes in pRb^ΔEpi;p21-/- mouse epidermis. However, the increased expression of Hes1 would indicate overall activation of the Notch pathway rather than inhibition. In this regard, it is worth considering that the tumors observed in these mice were predominantly well differentiated. Moreover, part of the tumor suppressor functions of Notch in HNSCC are associated to a cell proliferation arrest mediated by p21 induction (89) a process obviously impaired in pRb^ΔEpi;p21-/- mice. The possible relevance of these deregulated signaling pathways in mice would deserve future analysis in clinical samples.

In addition, our transcriptome comparative studies also demonstrate significant similarities between deregulated genes in pRb^ΔEpi;p21-/- mice and human HNSCC characterized by HPV infection and/or good prognosis. HPV infection may induce the development of specific HNSCC subtypes associated with better clinical outcome and characterized by specific gene expression patterns (93). These characteristics have been attributed to the expression of the two main viral oncogenes E7 and E6 (93). Importantly, the expression of viral E7 oncoprotein interferes with pRb and p21, supporting an active role of this two tumor suppressors in preventing epithelial tumors (91, 94, 95), and HPV E6 oncogene targets p53 tumor suppressor to degradation, thus disturbing p21 expression. Nonetheless, we recently reported that, in artificial human skin prepared using primary keratinocytes engineered to express the E7 protein and engrafted onto nude mice, the expression of E7 oncoprotein produces upregulation of p21 gene while maintaining significant overlap in mRNA and miRNA expression with human HPV associated pretumoral and tumoral lesions. Given the relevance of HPV in HNSCC development, these observations deserve future research aimed to identify the possible similarities and potential differences between pRb^ΔEpi;p21-/- mouse tumors and human HPV-positive HNSCC as previously reported (96).

Altogether, the present study reveals relevant and cooperative roles of pRb and p21 in epidermis and other stratified epithelia, affecting in a cell autonomous manner proliferation and differentiation and in a non-cell autonomous way other processes, such as inflammation. The results presented here support that p21 plays a key role in limiting cancer onset in the absence of a pRb pathway, and provide a valuable platform for the study of human epithelial malignancies.

Materials and Methods

Mice and histological procedures

All mice husbandry and experimental procedures were approved by the Animal Ethical Committee (CEEA) and conducted in compliance with Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas (CIEMAT) guidelines. Mice were generated and identified by PCR as described (12, 45). Samples were fixed in 4 % buffered formalin or 70% ethanol, embedded in paraffin wax and sectioned (5μm). Mice were injected with bromodeoxyuridine (BrdU; 0.1mg/g weight in 0.9% NaCl) 1 hour before sacrifice. Immunohistochemistry and immunofluorescence were done using standard protocols (50, 52, 97) on deparaffinized sections using antibodies against K5, K6 (1/500 Covance), K10 (1/50 DE-K10 DakoCytomation), γH2AX Ser 139 (Upstate), Cleaved Caspase 3 (1/50, Cell Signaling), γδ T-Cell Receptor-FITC (γδ TCR-FITC) (1/50, BD Pharmingen), Biotin-CD45 (1/50, Pharmingen), F480 (against macrophages, 1/50 Abcam, Cl: A3-1), CD3 ε (against T cells and NK cells, 1/6 DakoCytomation). Mast cells detection was done using Toluidine blue stain, taking advantage of metachromasia properties of these cells (Certistain). BrdU incorporation was monitored as described (98, 99). Secondary antibodies were purchased from Jackson Immunoresearch Laboratory. Control slides were obtained by replacing the primary antibodies with PBS (data not shown). Immunofluorescences were counterstained with DAPI (1/100 Roche).

Cell culture and Luciferase assays

Primary keratinocytes cultures and E2F luciferase assays were performed as previously reported (98, 99) using pSV40-Renilla (Promega) and pE2F-Luc (Dr.X. Lu, Ludwig Institute, London). Two independent transfection experiments (performed in triplicate) were carried out and luciferase activity was normalized to the values obtained with control (RbF/F) cells.

Western blot analysis

Skin and primary keratinocyte extracts were obtained as reported (98, 99). Total protein (50μg) was used for SDS-PAGE, transferred to nitrocelullose (Amersham) and probed with primary antibodies (diluted 1/500) against pRb, p107, p130, E2F1 (Santa Cruz), p21 (Abcam), p53 (Novocastra), phospho Tyr-705 Stat3, Stat3, phospho Ser 536 p65 (Cell Signaling), ATR (Serotec), phospho Ser345-Chk1, Caspase 3, Parp (Cell Signaling) (diluted 1/200); active βcatenin (Millipore) (diluted 1/500); p65, IκB-α, Ikk-γ (Santa Cruz); (diluted 1/2000) Ikk-α, and (1/1000) IKK-β (Imgenex), γH2AX Ser 139 (Upstate), Chk1 (Novocastra) and (1/5000) H2AX (Abcam). Actin (1/500, Sta. Cruz) was used to normalize the loading. Chemiluminescence was performed using manufacturer’s recommendations (Pierce).

The cytokines panel was analyzed following manufacturer’s recommendations (Proteome Profiler™ Mouse Cytokine R&D Systems ARY006). Membranes were incubated with 1mg of pnd 30 protein extracts from skin of pRb^ΔEpi, p21-/-, pRb^ΔEpi;p21-/- or Rb^F/F (as a control) and from new born pRb^ΔEpi;p21-/- mouse epidermis after three months upon grafting onto NOD/SCID mice. This array screens for relative levels of 40 different cytokines and chemokines. Quantification of the relative expression was determined by QuantityOne software.

Mouse skin grafts

Skin grafting was performed following previous protocols (46, 100). Briefly, dorsal full thickness skin pieces of 2 cm² were obtained from pRb^ΔEpi, p21-/-, pRb^ΔEpi;p21-/- or Rb^F/F (as a control) newborn mice (2 days after birth). Donor skin pieces were grafted onto a wound created by removing a similar-sized piece of full thickness back skin from female immunodeficient NOD/SCID recipient mice as described (100). Graft recipient animals were routinely monitored for tumors and sacrificed 12 weeks after grafting.