Epithelial cell senescence: an adaptive response to pre-carcinogenic stresses?

Abstract

Senescence is a cell state occurring in vitro and in vivo after successive replication cycles and/or upon exposition to various stressors. It is characterized by a strong cell cycle arrest associated with several molecular, metabolic and morphologic changes. The accumulation of senescent cells in tissues and organs with time plays a role in organismal aging and in several age-associated disorders and pathologies. Moreover, several therapeutic interventions are able to prematurely induce senescence. It is, therefore, tremendously important to characterize in-depth, the mechanisms by which senescence is induced, as well as the precise properties of senescent cells. For historical reasons, senescence is often studied with fibroblast models. Other cell types, however, much more relevant regarding the structure and function of vital organs and/or regarding pathologies, are regrettably often neglected. In this article, we will clarify what is known on senescence of epithelial cells and highlight what distinguishes it from, and what makes it like, replicative senescence of fibroblasts taken as a standard.

What is senescence?

Cellular senescence was defined about 50 years ago as the state reached by human diploid fetal lung fibroblasts after a finite number of cell doublings during in vitro culture [1, 2]. The idea that an internal clock, that could be the shortening of telomeres at each round of replication, counting the cell divisions and limiting them to a maximal number emerged in the 90’s [3]. This mechanism was well established and documented in the next years [4, 5] and was termed replicative senescence (RS). However, cellular states identical or similar to RS can be reached by cells subjected to various endogenous or exogenous stressors including reactive oxygen species (ROS), overactivation of some oncoproteins such as RAS, or physical or chemical DNA damaging agents. All types of cells are, therefore, susceptible to enter senescence, even immortalized cancer cells as well as non-dividing cells. These stress-induced types of senescence were named stress-induced premature senescence (SIPS) for the senescence induced by ROS; oncogene-induced senescence (OIS) for the senescence induced by RAS and other proteins of the same pathway; and therapy-induced senescence (TIS) for senescence induced by DNA damaging agents used in anti-cancer therapy.

It is important to understand that under these four acronyms, there are not four sub-types of senescence operating through different pathways, but only the highlighting four physiologic, pathologic, clinical or experimental contexts of senescence. Actually, the four types of senescence share some common mechanisms and/or can superimpose during a common time window: for example, SIPS relies mainly on the DNA damaging effect of ROS which can preferentially affect telomeres, hence accelerating telomere shortening and RS; TIS occurs mainly as a consequence of the oxidative stress induced by the anticancer therapy, hence resembling SIPS; the RAS oncogene acts through both the production of ROS and the induction of a hyper-replication stress which turns into DNA damage, hence combining SIPS and TIS mechanisms; the endogenous oxidative stress increases late in the replicative life span of in vitro proliferating cells, meaning that cells in RS further enter SIPS. This is probably because of these common and/or intercrossing mechanisms that the senescent phenotype displays some recurrent unifying markers. One can agree on a minimal core of 5 major markers defining senescence, knowing that each of them is not ever found or searched in all studies:

• Cell enlargement, associated with an increased spreading on the substrate;

• Cell cycle arrest, mainly but not exclusively in G1; prolonged and/or irreversible; associated with a variable percentage of polynucleated cells;

• Increase in the activity of the lysosomal β-Galactosidase activity, detectable at the non-optimal pH of 6, referred as senescence-associated (SA) β-Gal activity, reflecting an increase in lysosomal mass and/or autophagic activity;

• Specific epigenetic changes including Senescence-associated heterochromatic foci (SAHF);

• Changes in the composition of the secretome, which becomes enriched in pro-inflammatory cytokines and remodelers of extra-cellular matrix, referred as senescence-associated secretory phenotype (SASP).

Excellent reviews on the main mechanisms of senescence were recently published [6–11]. The aim of the present review is to focalize on the peculiarities, if any, of senescence in epithelial cells.

Modalities of senescence in epithelial cells

Although in vivo studies show that senescence can affect a variety of cell types [12–17], recurrently used in vitro cellular models are very few and usually are fibroblasts (for example, IMR90 and WI-38), as in the initial Hayflick’s studies. Therefore, most of our knowledge and concepts on senescence are actually relevant to fibroblasts. Epithelial cells, which, however, play major structural and functional roles in most human organs and are involved in numerous pathologies, especially cancer, were barely considered in senescence studies. This is probably because, the way they senesce apparently differs in several respects from that of fibroblasts undergoing RS taken as a reference.

And indeed, at least in some normal human epithelial cell types, especially mammary epithelial cells and keratinocytes, and at least in some culture conditions, the senescence which spontaneously establishes in cell culture occurs very atypically, in two steps (Fig. 1). The cells undergo a first growth arrest that occurs after only 10–20 PDs. This first plateau was termed, according to the authors, “selection”, “M0”, “stasis”, “telomere-independent senescence”, “early senescence”, “stress-associated senescence” or simply “senescence”. At this plateau, the cells share the main characteristics of senescence: they have an enlarged and flattened morphology, most of them are SA-β-Gal positive, they express an inflammatory secretome, some cells are polynucleated, and they are cell-cycle arrested mainly in G1 [18–26]. But importantly, this cell cycle arrest is not irreversible in all cells. About one cell on 10⁴ spontaneously and systematically escapes from the senescent state, loses the SA-β-Gal activity and reenters cell cycle to give rise to cell clones that undergo a further 5-45 PDs [22, 24, 27–30]. These post-senescence cells then undergo a second growth plateau termed “M1”, “agonescence” or simply “senescence” or “second plateau”. At this plateau, the cells display again the main markers of senescence: they are spread and enlarged, a greater proportion of them are tetraploid, their SA-β-Gal activity is increased and they express a specific secretome [20, 24, 28, 31–34]. Therefore, according to the presence of the main senescence markers, both growth arrest plateaus can be considered as senescence. However, the two plateaus are not strictly identical, at the morphological and molecular levels (see the following paragraphs). One reason for that might be that cells of the two plateaus derive from cells in different initial states: normal cells for the first plateau, and partially transformed post-first plateau cells for the second plateau (see the following paragraphs). Besides, the two plateaus, especially the second one, are associated with a significant level of non-apoptotic cell death [24, 29], explaining why senescence in epithelial cells was sometimes equated to crisis. We will name in this review the two growth plateaus of human keratinocytes and mammary epithelial cells “the first and the second senescence plateaus”. When we will refer to RS of fibroblasts or to other non-epithelial cell types, we will use the term “reference senescence”.

Fig. 1.

Two-step senescence of human keratinocytes and mammary epithelial cells. a Schematic representation of the in vitro growth curves. b Schematic representation of DNA damage and cell cycle arrest pathways involved in the two senescence plateaus. The filled rectangles represent events common to keratinocytes and mammary epithelial cells. The hatched rectangles represent events specific to one cell type. The dashed lines represents non investigated phases

The normal human mammary epithelium consists of two layers, an inner luminal milk-producing one and an outer contractile myoepithelial one. The so-called human mammary epithelial cells (HMECs) express myoepithelial markers such as CD10. Cells of the luminal lineage are called breast primary epithelial cells (BPECs). These two subtypes of mammary epithelial cells undergo the two-step senescence, but only when grown in serum-free culture conditions [22, 26, 35–38]. Regarding keratinocytes, two subtypes were used in senescence studies: normal human epidermal keratinocytes (NHEKs) and normal human oral keratinocytes (NHOKs). As for mammary epithelial cells, the culture modalities were shown to impact on their senescence: culture on feeder layers made of mitomycine C-treated fibroblasts alleviates the first senescence plateau [39].

For other epithelial cell types, the studies are sparse and it is often difficult to determine, according to the data accessible in the articles, whether the in vitro spontaneous senescence occurs in one or two steps, given that the first senescence plateau can be easily missed if incorrectly monitored (our personal expertise). For normal, immortalized or transformed epithelial cells entering senescence following a stress or a pharmacologic or genetic manipulation, senescence is generally acquired rapidly, apparently in only one step (Table 1).

Table 1.

Characteristics of senescence in different types of epithelial cells, normal or cancerous, occurring in vitro or in vivo, either spontaneously or in response to a variety of inducers

Epithelial cell type	Telomere shortening	p16↑ and/or pRb↓	p53↑ and/or p21↑	DDR	Growth inhibition	Cell morphology	SA-β-Gal	Others	Senescence inducer	Time to senescence	PDs to senescence	References
Amnion
Human amnion epithelial cells						Enlarged flattened	Yes	p19ARF↑ P-p38MAPK↑	Cigarette smoke-induced	3 h	N.D.	[171]
Primary amnion epithelial cells						Nuclear changes	Yes	P-p38MAPK↑ COX2↑	Polybrominatediphenyl ethers (PBDEs)-induced	48–72 h	N.D.	[172]
Primary human amnion epithelial cells				γ-H2AX foci (IF)			Yes	IL6, IL8 secretion↑ P-p38MAPK↑	T-oligos ([TTAGGG]2) that mimic telomere fragments	48 h	N.D.	[173]
Biliary tract (bile ducts)
Mouse biliary epithelial cells (cultured)					Yes	Enlarged	Yes		H2O2-induced	8 days	N.D.	[174]
Human biliary epithelial cells (in vivo)		p16↑	p21↑			Enlarged nucleus	Yes	Multinucleation	Primary biliary cirrhosis	N.A.	N.A.	[174]
Normal mouse intrahepatic biliary epithelial cells		p16↑	p21: no change		BrdU↓		Yes		H2O2-induced	6 days	N.D.	[175]
Biliary epithelial cells	Yes	p16↑	p21↑	γ-H2AX foci					Primary biliary cirrhosis	N.A.	N.A.	[176]
Normal human primary biliary epithelial cells (BECs)					Yes	Enlarged		Multinucleated	Spontaneous	N.D.	1 or 2 passages	[177]
Normal mouse primary biliary epithelial cells (BECs)			p21↑		Yes	Enlarged		Multinucleated	Early chronic liver allograft rejection (CR)	N.A	N.A.	[178]
Biliary tract (gallbladder)
Human primary gallbladder epithelial cells (HGEC)		p16↑	N. D.	γ-H2AX	BrdU↓	Enlarged and flattened	Yes		Lysolecithin	4–8 days	N.D.	[162]
Breast
Normal human mammary epithelial cells (HMECs)					BrdU↓	Enlarged and flattened		HP1γ positive cells (IF)	Spontaneous	60 days	PDs 110	[179]
Normal human mammary epithelial cells (HMECs)		p16↑							Spontaneous	N.D.	3–8 PDs at 1rst plateau 12–35 PDs at 2nd plateau	[61]
Normal human mammary epithelial cells (HMECs)		p16↑	p21: no change		Yes		Yes		Spontaneous	N.D.	15–25 PDs	[22]
Normal human mammary epithelial cells (HMECs)	Yes				BrdU↓	Enlarged	Yes	Polyploidy	Spontaneous	N.D.	10–12 PDs at 1rst plateau 70–75 PDs at 2nd plateau	[24]
Normal human mammary epithelial cells (HMECs)				pS/TQ foci	Growth curve BrdU↓				Overexpression of H-RasV12	N.D.	N.D.	[180]
Breast primary epithelial cells (BPECs)		p16↓			Yes		Yes		Spontaneous	N.D.	13–15 PDs at 1rst plateau 30–50 PDs at 2nd plateau	[26]
Normal mouse mammary epithelial cells					Yes	Flat, enlarged	Yes	IL-1beta↑ p27↑	1,25(OH)2D3	48–96 h	N.D.	[181]
MCF-7 cell line derived from breast adenocarcinoma			p21↑		Yes		Yes		OIS overexpression of oncogenic ERBB2	2–7 days	N.D.	[182]
MCF-7 cell line derived from breast adenocarcinoma	Telomerase activity↓		p53↑ p21↑			Enlarged nuclei	Yes		Adriamycin	4–5 days	N.D.	[183]
MCF-7 cell line derived from breast adenocarcinoma		Ph-Rb↓			BrdU↓		Yes		T-oligo-induced senescence	4–7 days	N.D.	[184]
Colon
Human colorectal cancer cells (HCT 116)					BrdU↓	Enlarged and flattened	Yes	IL8 secretion↑	5-fluorouracil (5-FU)	N.D.	N.D.	[185]
Colorectal tumor derived epithelial cells	Yes				Yes				Unstable primary cultures	N.D.	<28; <13; <8 passages	[186]
Epithelial cancer cell line (colon LS174T)		p16: no change	p53↑ p21↑		Yes	Enlarged and flattened	Yes		Active metabolite of CPT-11 (SN38)	120 h	N.D.	[187]
Human epithelial colon cancer (CaCo2) cells					Yes Cyclin A↓		Yes		H2O2 + peptide from buffalo cheese acid whey	24 h	N.D.	[188]
Human colon cancer cell line HCT116		Ph-Rb↓	p53 Ph-Ser15↑ p21↑		BrdU↓	Enlarged nuclei	Yes		Camptothecin	1–5 days	N.D.	[189]
Human colon cancer cell line HCT116					Yes		Yes		Irinotecan (CPT-11)	48 h	N.D.	[190]
Epidermis
Normal human epidermal keratinocytes (NHEKs)		p16↑ at 1rst and 2nd plateau	p21↑ at 1rst plateau		Yes	Enlarged	Yes			N.D.	15 PDs at 1rst plateau 18 PDs at 2nd plateau	[20]
Normal Human Epidermal Keratinocyte (NHEKs)s					Yes	Enlarged		Multinucleation Vacuoles No increase in apoptosis		N.D.	25 PDs at 1rst plateau 35 PDs at 2nd plateau	[34]
Normal human epidermal keratinocytes (NHEKs)					Yes	Enlarged and flattened Flow cytometry (SSC/FSC) TEM	Yes	Lysosomal activity Autophagy markers	H2O2	4 days	N.D.	[92]
Human epidermal keratinocyte (HEK) cells		p16↑	Ph-p53↑ p21↑		Yes		Yes	ROS production	Engineered fullerene derivatives	48 h	N.D.	[191]
Normal human epidermal keratinocytes					Yes				Spontaneous	140 days	16 PDs	[192]
Normal human epidermal keratinocytes		p16↑	p53: no change p21: no change						Spontaneous		42 PDs	[193]
Normal human epidermal keratinocytes		p16↑							Spontaneous		13 PDs on plastic 81 PDs on feeder layer	[39]
Esophagus
Cell line of human fetal esophageal epithelium (SHEE) established with E6–E7 of HPV type 18	Yes					Enlarged and flattened			Spontaneous	N.D.	Between the 12th to the 16th passage, then proliferation resumed	[194]
Normal human esophageal keratinocytes (EPC2)	Yes	p16↑ P-Rb↓	p21 and p53: no change				Yes	Cyclin D1, D2↑	Spontaneous	≈100 days	40–44 PDs	[70]
Immortalized human esophageal epithelial EPC2-hTERT cells		p16↑ P-Rb↓	p21↑		Inhibition of DNA synthesis		Yes	Cyclin D1↑ p14ARF↓	OIS H-RAS at PD215	13 days	N.D.	[70]
Eye (cornea)
Human corneal epithelial cell (in vivo)		p16↑	p21↑		Yes		Yes	IL6, IL8 secretion	TGF-β1	48–72 h	N.D.	[153]
Human corneal epithelial cell (HCEC)		p16↑ (from 55 to 72 years)	p21↑ (71–72 years)						Old corneal epithelium	N.A.	N.A.	[153]
Eye (lens)
Normal human lens epithelial cells (HLECs)					Growth curve BrdU↓	Enlarged and flattened	Yes		H2O2	2 weeks	N.D.	[195]
Bovine lens epithelial cells (BLECs)	Yes					Flattened increased cell size	Yes		Spontaneous	N.D.	20 PDs	[196]
Immortalized human corneal epithelial cells				γ-H2AX foci (IF)			Yes		Airborne Particulate Matter (PM2.5)	24 h	N.D.	[197]
Eye (retina)
Human retinal pigment epithelial ARPE-19 cell line		p16↑	P-p53 Ser15↑ p21↑	γ-H2AX	Yes		Yes		H2O2	8 days	N.D.	[198]
Normal human retinal pigment epithelial (RPE) cells		p16: no change Rb↓	p21↑			Enlarged and flattened	Yes	Fibronectin, SM22, CTGF, Apo J mRNA expression	H2O2	48 h	N.D.	[199]
Normal human retinal pigment epithelial (RPE)		p16: no change Rb↓	p21↑			Enlarged and flattened	Yes	Fibronectin, SM22, CTGF, Apo J mRNA expression	TGFβ1 TGFβ2	48 h	N.D.	[199]
Normal human retinal pigment epithelial (RPE)							Yes	TGFβ2↑ ECM components↑	Lipoprotein	24 h	N.D.	[200]
Human retinal pigment epithelial ARPE-19 cell line			p53↑		Yes		Yes		Doxorubicin	16–72 h	N.D.	[201]
Normal human retinal pigment epithelial (RPE)						Morphological changes	Yes	Caveolin↑	H2O2	2 h	N.D.	[202]
Human retinal pigment epithelial ARPE-19 cell line		P-Rb↓	p53↑ p21↑		BrdU↓		Yes		Oxidative stress-induced senescence (tBH, H2O2)	72 h	N.D.	[203]
Human retinal pigment epithelial RPE340 cell line	Yes				Yes BrdU↓		Yes		Spontaneous	≈120 days	57 PDs	[204]
Human retinal pigment epithelial RPE340 cell line	Yes				Yes BrdU↓		Yes		Cryopreservation (−10 °C/min)	33 days	29–31 PDs	[204]
Human retinal pigment epithelial RPE340 cell line					BrdU↓		Yes		Hyperoxia-induced premature senescence	N.D.	Between 2 and 12 PDs	[205]
Normal human retinal pigment epithelial (RPE) cells	Yes				BrdU↓	Flattened, enlarged	Yes		Spontaneous	N.D.	58 PDs	[206]
Immortalized human retinal pigment epithelial cells (hTERT)	Yes						Yes		Spontaneous	N.D.	56–60 PDs	[207]
Normal human retinal pigment epithelial (RPE)					Yes				Spontaneous	N.D.	15 PDs	[208]
Kidney
Mouse tubular cells In vivo		p16↑	p21↑	γ-H2AX foci (IF)			Yes		Aged mouse kidneys	N.A.	N.A.	[155]
Human tubular cells In vivo		p16↑	p21↑	γ-H2AX foci (IF)			Yes		Lead acetate in aged mouse kidneys	N.A.	N.A.	[155]
Normal mouse primary tubular cells		p16↑			BrdU↓		Yes	p15↑ p19↑	Spontaneous	6 days	N.D.	[155]
Normal human primary tubular cells		p16: no change	p53↑ p21↑	γ-H2AX foci (IF)	BrdU↓ (IF)		Yes	Cyclin D1↑	γ-Irradiation	10 days	N.D.	[155]
Normal human primary proximal tubular cells		p16↑	p53↑ p21↑		BrdU↓ (IF)		Yes		Spontaneous		Passage 6	[209]
Normal human primary proximal tubular (HPT) cells	Yes	p16↑	p21↑		Yes				Cyclosporine A	24 h	N.D.	[210]
Human kidney epithelial cells		p16 positive cells (IF)							Renal transplants	N.A.	N.A.	[211]
Human kidney epithelial cells	Yes						Yes		Renal transplants	N.A.	N.A.	[212]
Rat kidney epithelial cells	Yes	p16↑	p21↑				Yes		Rat model of chronic renal transplant rejection	N.A.	N.A.	[213]
Human kidney epithelial cells (in vivo)		P16↑	p21↑					P27↑	Chronically rejected renal allografts	N.A.	N.A.	[214, 215]
Renal tubular epithelial cancer cell lines MDCK and WT 9–7			P-p53↑ p21↑			Enlarged and flattened	Yes		Roscovitine	24 h	N.D.	[216]
Normal rat kidney proximal tubular epithelial cells (in vivo)							Yes	Accumulation of mitochondrial DNA (mtDNA)	Cadmium-induced nephropathy	N.A.	N.A.	[217]
Lung
Normal human bronchial epithelial cell line BEAS-2B						Increased cell size	Yes	Autophagy markers	Cigarette smoke	6 h	N.D.	[218]
Human small airway epithelial cells				γ-H2AX foci			Yes	Increased mitophagy	Cigarette smoke	10–15 days	N.D.	[219]
Adenocarcinomic human alveolar basal epithelial A549 cell line					Yes	Enlarged and flattened	Yes	Increased lysosomal content	Pyocyanin, a major P. aeruginosa toxin	3–7 days	N.D.	[220]
Type II-like alveolar epithelial cell line (A549)			p21↑			Flat, enlarged	Yes	Increased lysosomal mass Growth inhibition	Cigarette smoke exposure	36 h	N.D.	[221]
Alveolar epithelial cells (in vivo)	Yes	p16↑	p21↑						Emphysema	N.A.	N.A.	[222]
Airway epithelial cells (in vivo)		p16↑	p21↑				Yes		Lung allograft	N.A.	N.A.	[223]
Normal primary human bronchial epithelial cells (HBECs)			p21↑		Yes		Yes		TGFβ	48 h	N.D.	[224]
Type II-like alveolar epithelial cell line (A549)			p21↑		Yes	Flat, enlarged, increased cell size	Yes	Increased lysosomal mass	Bleomycin	120 h	N.D.	[225]
Primary rat type II alveolar epithelial cells						Flat, enlarged	Yes		Bleomycin	24 h	N.D.	[225]
Human type II epithelial-like (A549) cells	Yes						Yes	Lysosomal mass↑	Lipopolysaccharide	1–7 days	N.D.	[226]
Oral mucosa
Primary oral mucosal keratinocytes		p16↑			BrdU↓				Spontaneous	N.D.	35 PDs at 1rst plateau 40 PDs at 2nd plateau	[25]
Normal human oral keratinocytes (NHOKs)	No		p53↓ p21↓		Yes	Enlarged nuclei	Yes	ROS↑ 8-oxo-dG↑	Spontaneous	30 days	19–20 PDs	[60, 227, 228]
Normal human oral keratinocytes (NHOKs)		p16↑ Ph-Rb↓			Yes	Enlarged	Yes		Spontaneous	30 days	22 PDs	[229]
Normal human oral keratinocytes (NHOKs)		p16↑	p53↓ p21↓				Yes	p15↑ p57↑ TGF-b1 secretion	Spontaneous	30 days	20 PDs	[230]
Normal human oral keratinocytes (NHOKs)		p16↑			Yes		Yes	IL6, IL8↑	Bisphosphonates (BPs)	4 days	N.D.	[231]
Normal human oral keratinocytes (NHOKs)		p16↑ Ph-Rb↓	p53↓ p21↓		Yes Arrest in G2	Enlarged			Spontaneous	40 days	30 PDs	[48]
Pharynx
Normal human pharyngeal epithelial cells (HPECs)		p16↑					Yes		Spontaneous	N.D.	Passage 3	[232]
Prostate
Primary prostate epithelial cells						Flattened, enlarged, increased cell size	Yes		γ-irradiation	48 h	N.D.	[233]
Prostate cancer cell lines DU145 and LNCaP cells						Flattened, enlarged, increased cell size		Polyploidy, multinucleation, increase in side scatter (SSC)	DAC (a methyltransferase inhibitor) Doxorubicin Docetaxel	7 days	N.D.	[234]
Normal primary human prostatic epithelial cells		p16↑					Yes	p27↑ IL-1α↑	Spontaneous	4–8 days	N.D.	[158]
Normal human prostatic epithelial cells						Flattened, enlarged cells	Yes		Spontaneous	N.D.	30 PDs	[157]
Normal human prostate epithelial cells		p16↑ P-Rb↓	P53: no change p21: no change			Flattened, enlarged cells	Yes	Multinucleation	Spontaneous	N.D.	5th passage	[69]
LNCap cells				γ-H2AX foci (IF)	Yes	Enlarged and flattened	Yes	HP1β, IGFBP3 and Cathepsin B production (SASP)	Androgen depletion	16 days	N.D.	[235]
Human prostate epithelial cancer cell line WFU273Ca			p21↑				Yes		1,25-Dihydroxyvitamin D3 and PI3K/AKT inhibitors	24 h	N.D.	[236]
Urinary tract
Normal human uroepithelial cells (HUCs)		p16↑ P-Rb↓	P53: no change p21: no change		Cyclin D↓ p27↓ p19↓	Enlarged			Spontaneous	N.D.	30–40 PDs	[71]
Normal human uroepithelial cells (HUCs)		p16↑					Yes		Spontaneous	N.D.	Passage 3	[237]
Others
Various epithelial cancer cell lines							Yes		Doxorubicin Retinoic acid	2–8 days	N.D.	[238]
Human epithelial cancer cell line HEK293		p16: no change Ph-Rb↓	p21↑ p53/Ser15↑	γ-H2AX↑	Yes	Enlarged and flattened	Yes (5-10 days)	p27↑	Epstein-Barr Virus (EBV) replication and transcription acti- vator (Rta/BRLF1)	3–10 days	N.D.	[239]
Melanoma cells 115			p53↑ p21↑			Enlarged and flattened	Yes	Various growth factors, cytokines and Wnt ligands↑	Doxorubicin	5 days	N.D.	[240]
T84 intestinal epithelial cells		P-Rb↓	p21↑			Enlarged nuclei	Yes	SAHF Cyclin D1↓	Bacterial heat-stable enterotoxin (ST) peptides	10 days	N.D.	[241]

The specific markers arguing for the establishment of the senescent phenotype are indicated. The time and the PDs to senescence are also indicated. The cell types are classified by organs and the organs are presented by alphabetical order

N.D. not determined, N. A. not applicable, IHC immunohistochemistry, IF immunofluorescence, WB western blot

Is the DDR pathway involved in epithelial cell senescence?

One mechanism common to RS, SIPS, OIS and TIS is the involvement of the DNA Damage Response (DDR) and the p53/p21 pathways. P21 (CDKN1) is a cyclin-dependent kinase inhibitor (CKI) which arrests the cell cycle progression at the G1/S and G2/M transitions by inhibiting the CDK4,6/cyclinD and CDK2/cyclinE complexes, respectively. P21 is upregulated at senescence by the transcription factor tumor suppressor p53 (TP53), which is itself activated as a result of the activation of the DDR pathway. Despite its name that echoes a pathway activated by all kinds of DNA damages, the DDR pathway is a strictly-defined pathway involving two couples of kinases, ATM/CHK2 and ATR/CHK1, activated by three types of specific DNA damages: shortened and/or deprotected telomeres, DNA double-strand breaks, and stalled or collapsed replication forks [40]. One important feature of the cell cycle arrest associated to reference senescence is its great stability which renders the reference senescent state almost irreversible. This stability relies on the fact that the senescence-associated DNA damages that activate the DDR pathway remain unrepaired. This is not due to a loss in the repair capacities, but to the fact that the damages are located at repetitive sequences, the telomeric ones or other intrachromosomic repetitive sequences, which present a great challenge to the repair machinery and are, therefore, prone to replication fork stalling and collapse [41]. Hence, it is established for a long time that overexpressing telomerase, that ensures telomere length maintenance, or invalidating p21 is sufficient to bypass RS of fibroblasts [42, 43].

Compared to RS of fibroblasts which occurs after a long exponential growth phase of 50–100 population doublings (PDs) that can last several months in in vitro culture, epithelial cells generally senesce slightly prematurely, i.e., after 20–60 PDs (Table 1 and references herein). This suggests that epithelial cells could senesce, for an important part, in response to a stress other than telomere shortening. However, the shortening of telomeres at epithelial cell senescence was generally recorded, but the correlative activation of the DDR pathway was often not clearly demonstrated (Table 1 and references herein).

Specifically regarding keratinocytes and mammary epithelial cells at the first senescence plateau, it was clearly shown that there is no significant telomere shortening and no activation of the DDR and p53/p21 pathways [19, 22, 24, 33, 38, 39, 44–47]. In NHOKs, the expression of p53 even decreases at senescence because of a change in H3 acetylation at the p53 promoter [48]. In accordance, re-expressing hTERT does not increase the lifespan of pre-first plateau HMECs or NHEKs in contrast to presenescent fibroblasts [47, 49]. The first senescence plateau of human keratinocytes and mammary epithelial cells is, therefore, clearly telomere—and DDR-independent (Fig. 1).

For the second senescence plateau, the situation is more confusing and seems to differ between keratinocytes and mammary epithelial cells. In HMECs, cells at the second plateau display shortened telomeres and expression of hTERT into post-first plateau cells results in cells able to bypass the second plateau and proliferate beyond 100 PDs [24, 26, 47, 50]. Similarly, transfecting the human papilloma virus type 16 E6 protein, which induces the degradation of p53 as well as induces telomerase activity [51–53], enables cells to bypass the second plateau [54, 55]. Interestingly, overexpressing c-MYC in post-first plateau HMECs is sufficient to bypass the second plateau, in correlation with an increase in telomerase activity [56, 57]. P53 is activated in HMECs at the second plateau in correlation with an accumulation of DNA damage foci positive for 53BP1 and γH2AX. However, its intended role would not be to arrest the cell cycle, since the expression of p21 and p27 do not increase. It would be rather to promote the survival of the otherwive arrested cells which undergo massive cell death when p53 function is abrogated using the genetic suppressor element GSE22 [33, 58]. Therefore, the second senescence plateau of HMECs resembles, but is not strictly similar, to reference senescence (Fig. 1). In NHEKs, in contrast to HMECs, telomere shortening makes only little progress from the first to the second plateau and 80% of the chromosomes at the second plateau still display telomeres detectable by FISH [20]. In spite of this, the expression of p21 is up-regulated, at least at the messenger level [20] (Fig. 1). Moreover, in NHEKs, the second plateau is not as irreversible as is a RS plateau: some cells can spontaneously escape, at a low frequency (10⁻⁷), even though they maintain p21 expression [20, 28]. Therefore, the nature of the second senescence plateau of human keratinocytes still needs further investigations.

Is the p16/RB pathway involved in epithelial cell senescence?

P53/p21 is recognized as the main inducer of the onset of reference senescence. However, the p16/Rb pathway is also strongly involved in the reference senescent cell cycle arrest, and, actually, is considered as a senescence marker as universal as the SA-β-Gal activity. As discussed by Ben-Porath and Weinberg more than 10 years ago, it is today still not clear whether the p16/Rb and the p53/p21 pathways act in parallel and are both necessary for the onset of reference senescence or if the p16/Rb pathway is dispensable for the establishment of the senescent cell cycle arrest, but is involved in its further stabilization and reinforcement [59]. Like p21, p16 (CDKN2) is a CKI which specifically targets the CDK4,6/cyclinD complexes. The tumor suppressor RB is under the control of these CDK/cyclin complexes. When these complexes are functional, RB is phosphorylated; under this state it cannot interact with the transcription factor E2F which then activates its target genes promoting the G1/S transition. When the CDK/cyclin complexes are inhibited by p16, RB becomes hypophosphorylated, able to interact with and inhibit E2F; the cell cycle is then arrested at the G1/S transition.

In keratinocytes and mammary epithelial cells at the first senescence plateau, p16 is clearly overexpressed and RB hypophosphorylated [19, 20, 22, 25, 26, 39, 44, 60, 61]. This p16/RB activation plays a major role in the cell cycle arrest, since it can be prevented by the human papilloma virus type 16 E7 protein which binds to RB and disrupts its interaction with E2F [18, 47] or by shRNA to p16 [56]. Moreover, overexpressing CDK4 in HMECs is enough to bypass the first senescence plateau [62]. Therefore, the first senescence plateau of human keratinocytes and mammary epithelial cells is clearly p16-dependent (Fig. 1).

Regarding the second plateau, the data differ according to the two cell types. In HMECs and BPECs, p16 is not expressed at the second plateau, because of an epigenetic extinction occurring during the escape from the first plateau [22, 26]. Alternatively, an increased expression and activity of p15 and a loss of CDC25A were documented [33], and an overexpression of CDK4 was shown to be sufficient to bypass the plateau [62]. In NHEKs in contrast, p16 is re-expressed at the second plateau, and even at a level higher than that recorded at the first plateau [20] (Fig. 1). hTERT, but interestingly also hTERT mutants defective for telomere maintenance, can immortalize second plateau NHEKs through the induction of BMI1, a subunit of the polycomb repressive complex 1 involved in the epigenetic down-regulation of p16 [63]. Conversely, BMI1 overexpression can immortalize post-first plateau HMECs by inducing hTERT transcription and increasing telomerase activity [64]. BMI1 was also shown to be able to extend the lifespan of corneal keratinocytes and of keratinocytes derived from embryonic stem cells [65].

Interestingly, HMECs with BRCA1 haploinsufficiency, a protein involved in the DDR pathway and at the origin of inherited breast cancers, enter in the first senescence plateau as wild-type HMECs, confirming that the first plateau is DDR-independent. Wild-type and mutated cells both overcome the first senescence plateau at the same frequency, but BRCA1^mut/+ cells enter the second senescence plateau more rapidly than wild-type cells, without activation of the DDR pathway and of p53. The levels of p16, p15, p18 and p19 are not increased either, however, RB is clearly hypophosphorylated. In NHEKs harboring the same BRCA1 mutation, the second plateau also occurs prematurely and is only dependent on p16 up-regulation and RB hypophosphorylation [66]. These results confirm the prominence of the p16/RB pathway in senescence of HMECs and NHEKs, even for the second senescence plateau in some instance.

Recently, it was shown that OIS occurring in a population of NHEKs mixing stem cells and transient amplifying cells does not rely on the activation of the p16/RB pathway, but on the expression of p21, independently of p53, whose expression significantly decreases. However, when OIS was induced only in transient amplifying keratinocytes, both p53 and p21 decreased, and the cell cycle arrest was mediated by p16 [67]. In HMECs undergoing OIS upon overexpression of an activated form of RAF-1, the levels of p16 and p53 decreased while that of p21 remained unchanged, whereas in fibroblasts similarly undergoing RAF-1-induced OIS, the level of p16 strongly increased [68]. Therefore, the pathways responsible for OIS-associated cell cycle arrest are strongly dependent on the cell type.

Studies performed with other epithelial cell types reporting the up-regulation of p16 and/or the hypophosphorylation of RB at spontaneous senescence are rather sparse (Table 1 and references therein). Interestingly, three studies performed with uroepithelial cells, prostate epithelial cells and esophageal keratinocytes documented the activation of p16/RB but not of p53/p21 [69–71]. The importance of p16 was also emphasized in human oesophageal epithelial cells immortalized by hTERT, in which the sole overexpression of p16 is sufficient to induce a premature senescence [70].

To summarize, the main mechanisms inducing the cell cycle arrest of epithelial cells undergoing spontaneous senescence are not fundamentally different from those of reference senescence, but the hierarchy and the kinetics of the events would be almost opposite. In reference senescence, the DDR-p53/p21 pathway seems to be first activated, suggesting it is responsible for the onset of senescence; the cell cycle arrest is then rapidly reinforced by the activation of the p16/RB pathway, giving rise to only one visible and very stable senescence plateau. In epithelial cells undergoing a two-step senescence, the p16/RB pathway is activated first and alone. It induces an early growth arrest, rather instable, from which a few cells evade and undergo, several population doublings later, a second growth plateau which involves a DDR-like pathway or again the p16/RB pathway, according to the cell type (Fig. 1). Very interestingly, a transcriptional profiling followed by principal component analysis clearly distinguished HMECs at the first plateau from HMECs at the second plateau, and also clearly distinguished both of them from isogenic mammary fibroblasts at reference senescence. Genes functionally associated with “response to stress” were up-regulated at both epithelial senescence plateaus [38]. In epithelial cell types other than keratinocytes and mammary cells, the p16/RB pathway could also be the main driver of the establishment of senescence, associated or not with telomere shortening. The activation of the p53/p21 pathway could be dispensable, although the studies are too few to be definitely conclusive. In situations where epithelial cells are submitted to various senescence-inducing stresses, both pathways seem to be activated, within a few hours or days (Table 1 and references therein). Alternative pathways are also sometimes involved. For example, it was shown that HT29 colorectal cancerous cells are able to undergo OIS, despite the inactivation of p16 and p53 in this cell line. The cell cycle arrest results in this context from a p53-independent up-regulation of p21 and a down-regulation of CDC25A [72].

Epithelial cell senescence is associated with high autophagic activity

Several studies have searched for positive or negative regulators of epithelial cell senescence, other than p53/p21 or p16/RB. These studies were performed with a variety of epithelial cell types, normal, immortalized or cancerous, and the senescent state was obtained in vitro or in vivo, spontaneously, in normal or pathologic contexts, after drug treatment or after modulation of the expression of a specific gene. The stock of the results we could collect gives a list of 70 different positive or negative regulators of epithelial cell senescence (Table 2 and references herein). To highlight some pathways potentially linking these regulators, we performed for this review article a bioinformatics analysis using STRING, a database and web resource of known and predictive protein–protein interactions (http://string-db.org/). We entered in the analysis the list of the 70 regulators of Table 2, increased by the senescent cell cycle regulators p53, p21, p16, RB, p15, p27, cyclin D1 and cyclin D2. Sixty four out of the seventy five regulators revealed to be interconnected. These 64 interacting proteins were then entered into CYTOSCAPE (http://www.cytoscape.org/) to better visualize the network. The results are given in Fig. 2. The interaction network highlights 10 major nodes having more than 10 connections. Amongst them, there are p53 (TP53, 29 connections), p21 (CDKN1A, 14 connections) and p16 (CDKN2A, 11 connections), pointing out once more, the importance of these pathways in senescence.

Table 2.

Molecular actors and pathways (others than p53/p21 and p16/RB) that positively or negatively mediate the establishment of the senescent phenotype in epithelial cells exclusively (normal, immortalized or transformed)

Regulators/pathways involved in epithelial senescence	Epithelial cell type	Senescence inducer	Senescence markers used	References
Positive regulators
ABCC3 (ATP binding cassette subfamily c member 3 = MRP3)	Immortalized human epithelial cells (HECs)	ABCC3 overexpression	Cell morphological changes SA-β-Gal IL8 accumulation	[242]
AKT1 (AKT serine/Threonine kinase1)	Normal human primary esophageal epithelial cells	AKT overexpression	Cell morphological changes SA-β-Gal p16, p21	[243]
ALOX15B (arachidonate15-lipoxygenase, type B = 15-LOX-2)	Normal human primary Prostatic epithelial cells	15-LOX2 overexpression	Cell morphological changes SA-β-Gal BrdU incorporation	[244]
ALOX15B (arachidonate15-lipoxygenase, type B = 15-LOX-2)	Human cancer prostate epithelial PC3 cell line	15-LOX2 overexpression	Cell morphological changes SA-β-Gal	[244]
ATF4 (activating transcription factor 4)	Mouse primary renal tubular epithelial cells	ATF4 overexpression	Cell morphological changes SA-β-Gal SAHF p16	[245]
ATG5 (autophagy related 5)	Bronchial epithelial cells	Cigarette smoke	SA-β-Gal p21 γ-H2AX foci	[246]
ATG5 (autophagy related 5)	Primary tubular epithelial cells	OIS (HRas) γ-irradiation	p16 γ-H2AX Autophagy markers BrdU incorporation	[89]
ATM (ATM serine/threonine kinase)	Mouse intrahepatic biliary epithelial cells	Oxidative stress-induced cellular senescence	SA-β-Gal BrdU incorporation p21 (IF), P-p53 (IF)	[128]
ATM (ATM serine/threonine kinase)	Mouse cancerous keratinocytes PDVC57	shATM	SA-β-Gal P-p53 P21 HP1γ	[247]
ATRAID (all-trans retinoic acid induced differentiation factor = Apr3)	Human retinal pigment epithelial cells and ARPE–19 cells	Replicative oxidative stress (H2O2 and tert-BHP)	SA-β-Gal Growth inhibition p53 p21	[248]
Autophagy	Human biliary epithelial cells (BEC)	Oxidative stress DNA damage	Cell morphological changes SA-β-Gal CCL2, CX3CL1 secretion	[90]
Autophagy	Normal human epidermal keratinocytes (NHEKs)	Spontaneous	Cell morphological changes Flow cytometry (SSC/FSC) Corpses	[29]
BMP4 (bone morphogenetic protein-4)	Human retinal pigment epithelial ARPE-19 cell line	BMP4 overexpression	SA-β-Gal p53 p21 Phospho-Rb BrdU	[203]
CAV1 (caveolin 1)	Normal mouse lung epithelial cells	Bleomycin-induced cellular senescence	SA-β-Gal p16 Phospho-Rb MacroHistone 2A	[130]
DKK1 (dickkopf-1 WNT signaling pathway inhibitor 1)	Squamous esophageal epithelial cell line EPC2	Reflux esophagitis	SA-β-Gal Loss of PhosphoRb p21 γ-H2AX	[249]
DUSP6 (dual specificity phosphatase 6 = MKP-3)	Normal rat epithelial cells	Etoposide	Increased cell size SA-β-Gal p53 p21	[250]
Constitutive fragment p95 of ERBB2 (Erb-B2 receptor tyrosine kinase 2)	MCF7	Expression of the HER2 fragment	γ-H2AX foci 53BP1 foci p53-Ser15 p21 Phospho-Rb Matrix metalloprotease 1 (MMP1), angiopoietin-like 4 (ANGPTL4), IL-11, and IL-6 secretion	[251]
ER stress	Normal mouse proximal tubular epithelial cells	Premature through RAGE overexpression	SA-β-Gal SAHF Growth arrest p21	[81]
FOXP3 (forkhead Box P3)	MCF7 and HCT116 cells	Etoposide- or doxorubicin-induced senescence	SA-β-Gal Growth inhibition p53	[252]
GDF15 (growth differentiation factor 15)	Human tracheobronchial epithelial (hTBE) cells	Cigarette smoke	SA-β-Gal Growth inhibition p21 p16	[253]
HOPX (hOP homeobox)	Human bronchial epithelial cells (HBECs) and various non-small cell lung cancer cell lines	HOPX knockdown and/or overexpression	SA-β-Gal p53 p21 γ-H2AX foci SAHF	[254]
IGFBP7 (insulin growth factor-like binding protein 7 = IGFBP-rP1)	Human mammary epithelial cells MCF-7 (breast cancer cell line)	Overexpression of IGFBP7	Enlarged, flattened morphology SA-β-Gal Growth inhibition BrdU incorporation	[255]
IFI16 (interferon gamma inducible protein 16)	Human cancer prostate epithelial PC3 cell line	IFI16 overexpression	Cell morphological changes (enlarged flattened, increased granularity) SA-β-Gal Growth arrest p21	[256]
MAP2K3 (mitogen-activated protein kinase kinase 3)	Human mammary epithelial MCF10A breast cancer cell line	MAP2K3 overexpression	SA-β-Gal Growth inhibition p53	[257]
MAPK14 (mitogen-activated protein kinase 14 = p38MAPK)	Human retinal pigment epithelial ARPE-19 cell line	Oxidative stress-induced senescence (tBH, H2O2)	SA-β-Gal Growth arrest p53 p21 Phospho-Rb Apo J mRNA expression	[203]
MST1R (macrophage stimulating 1 receptor = RON)	Madin-Darby canine kidney epithelial cells (MDCKs)	RON overexpression	Cell morphological changes SA-β-Gal Growth inhibition	[258]
MTOR (mammalian target of rapamycin)	Human retinal pigment epithelial ARPE-19 cell line (immortalized)	Rapamycin	Cell morphological changes SA-β-Gal p16	[259]
PRDX1 (peroxiredoxin 1)	mammary epithelial cells in vivo	H-RasV12	SA-β-Gal	[260]
PRDX1 (peroxiredoxin 1)	MCF10A and MCF7 breast cancer cell lines	H2O2	SA-β-Gal	[260]
PTTG1 (pituitary tumor-transforming 1)	MCF10A and MCF7 breast cancer cell lines	PTTG1 overexpression	SA-β-Gal p53 p21 p16 IL-8, GROα, TNF-α, and Serpin E1 secretion	[261]
Raf-1 (Raf-1 proto-oncogene, serine/threonine kinase)	Human mammary epithelial cells (HMECs) not expressing p16INK4a	OIS (Raf-1)	Enlarged, flattened, vacuolated No increase in SA-β-Gal activity Growth arrest	[68]
RAD9A (RAD9 checkpoint clamp component A)	Lung cancer (A549 and H1299) and breast cancer (MCF7 and MDA-MB 231) cell lines	RAD9A overexpression	Cell morphological changes SA-β-Gal Growth inhibition p21	[262]
REL (REL proto-oncogene, NF-KB subunit NF-kB)	Normal human epidermal keratinocytes (NHEKs)	cRel overexpression	Enlarged and flattened morphology SA-β-Gal Polynucleation Growth inhibition Resistance to apoptosis	[96]
SMAD3 (SMAD family member 3)	Normal primary mouse skin keratinocytes	OIS v-rasHA	SA-β-Gal p15INK4b Growth arrest	[263]
SMURF2 (SMAD specific E3 ubiquitin protein ligase 2)	Normal human mammary epithelial cells (HMECs)	SMURF2 overexpression	Enlarged and flattened morphology SA-β-Gal	[264]
SOD2 (superoxide dismutase 2, mitochondrial)	Normal human epidermal keratinocytes (NHEKs)	SOD2 overexpression	Enlarged morphology Growth curves Corpses Autophagy markers	[92]
SQSTM1 (sequestosome 1)	Normal human biliary epithelial cells	Etoposide or starvation-induced senescence	SA-β-Gal	[265]
STIM1 (stromal interaction molecule 1)	DU145 and PC3 cells	STIM1 overexpression	SA-β-Gal	[266]
TFAP4 (transcription factor AP-4)	Immortalized human retinal pigment epithelial cell	AP-4 overexpression	Enlarged and flattened morphology SA-β-Gal Growth inhibition	[267]
TP53 (tumor protein 53)	Human urinary bladder epithelial cell line EJ	p53 overexpression	Enlarged and flattened morphology SA-β-Gal Growth inhibition granularity (TEM)	[268]
TP63 (tumor protein 63) ∆Np63α isoform)	Normal bronchial epithelial cells	∆Np63α overexpression	Cell morphological changes SA-β-Gal	[269]
TP63 (tumor protein 63)	Normal mouse primary keratinocytes	p63 ablation	Flattened and enlarged morphology SA-β-Gal	[270]
TGFB1 (transforming growth factor beta1)	Immortalized human mammary epithelial cells (HMEC-hTERT)	OIS H-Ras-V12	Enlarged cell shape SA-β-Gal Growth inhibition	[271]
TGFB1 (transforming growth factor beta1)	Normal human retinal pigment epithelial (RPE) cells	H2O2 induced senescence	SA-β-Gal Fibronectin, SM22, CTGF, Apo J mRNA expression	[199]
WNT5A (Wnt family member 5A)	Normal primary human ovarian surface epithelial (HOSE) cells	Wnt5 overexpression	Enlarged cell shape SA-β-Gal	[272]
Negative regulators
ATG5	Normal primary human bronchial epithelial cells (HBECs)	Cigarette smoke	SA-β-Gal p21	[91]
Autophagy	Human mammary epithelial cells	Autopahgy inhibition	Cell morphological changes SA-β-Gal	[273]
BMI1 (BMI1 proto-oncogene, polycomb ring finger)	Normal mouse intrahepatic biliary epithelial cells (cultured)	BMI1 knockdown	SA-β-Gal p16 BrdU	[175]
CARM1 (coactivator-associated arginine methyltransferase)	Primary alveolar epithelial type II (ATII) mouse lung cells	CARM1 knockdown and/or knock out	SA-β-Gal p21 p16	[274]
CDC25C (cell division cycle 25C)	Normal human mammary epithelial cells	Spontaneous	p21 p27 Cyclin A, D1, E Rb	[33]
CCND1 (cyclin D1) CDK4/6 (cyclin dependent kinase 4/6)	Human mammary epithelial cells	CCND1 inhibition	Cell morphological changes SA-β-Gal Autophagy markers	[273]
CKB (creatine kinase B)	Normal primary human bronchial epithelial cells (HBECs)	Cigarette smoke	SA-β-Gal p21 Cell cycle arrest IL8	[275]
DAXX (death domain associated protein)	Mouse ovarian surface epithelium	DAXX inhibition	SA-β-Gal p53 p21 p27	[276]
EGF (epidermal growth factor)	human mammary epithelial cells	EGF medium depletion	SA-β-Gal p21 Phospho-Rb	[277]
ESR1 (estrogen receptor 1)	Immortalized human mammary epithelial cells (hTERT) and MCF10A cell line	OIS H-rasV12	SA-β-Gal	[278]
FOXM1 (forkhead box M1)	NHEK	FOXM1 knockdown	SA-β-Gal Growth inhibition p16 Phospho-Rb	[279]
GLIS2 (GLIS family zinc finger 2)	Wild-type and Kif3a null kidney epithelial tubular cells	GLIS2 knockdown	SA-β-Gal Growth inhibition p53 p16 γ-H2AX foci H3K9me3	[280]
GLO1 (glyoxalase 1)	Kidney (in vivo)	Old rat kidney overexpressing GLO1	SA-β-Gal p53 p16	[281]
Glucose metabolism	Immortalized human mammary epithelial cells	Glucose metabolism inhibition by 2DG or G6PC3	SA-β-Gal Growth inhibition DEC1 mRNA	[282]
HSPA1A (heat shock protein family A (Hsp70) member 1A = Hsp72)	Human mammary epithelial MCF10A breast cancer cell line	OIS oncogenic PI3K	SA-β-Gal p53 p21	[283]
HSPA1A (heat shock protein family A (Hsp70) member 1A = Hsp72)	Human mammary epithelial MCF10A breast cancer cell line	OIS H-Ras	Enlarged, flattened, highly vacuolized cells SA-β-Gal	[283]
HSPB1 (heat shock protein family B (small) member 1 = Hsp27)	Human mammary epithelial MCF10A breast cancer cell line overexpressing Hsp27	Doxorubicin-induced senescence	SA-β-Gal p53 p21	[284]
HSP90	EBV-positive NPC cell line C666-1	HSP90 inhibition by AT13387	SA-β-Gal SAHF p21 p27	[285]
ID1 (inhibitor of DNA binding 1, HLH protein)	Immortalized prostate epithelial cell line, NPTX cells	ID1 inactivation	SA-β-Gal Growth inhibition Cyclin B1↓ p21 mRNA p27 mRNA	[286]
KDM5B (lysine demethylase 5B = JARID1B)	CRC cell lines (Colo201, DLD1 and HCT116)	JARID1B depletion	SA-β-Gal p16 Growth inhibition	[287]
KRAS G12D (proto-oncogene, GTPase)	Primary pancreatic duct epithelial cells (PDECs)	KRAS G12D overexpression	SA-β-Gal Growth curves p16	[288]
LC3B	Normal primary human bronchial epithelial cells (HBECs)	Cigarette smoke	SA-β-Gal p21	[91]
MAD2 (mitotic arrest deficiency protein 2)	Human mammary epithelial MCF-7 (breast cancer cell line)	MAD2 inhibition	Enlarged, flattened, highly vacuolized cells SA-β-Gal p21 Growth inhibition IL6, IL8 production	[289]
MAP3K7 (mitogen-activated protein kinase kinase kinase 7 = TAK1)	Human retinal pigment epithelial ARPE-19 cell line	H2O2	SA-β-Gal Growth inhibition p53 MMP9 activity	[290]
MMP7 (matrix metallopeptidase 7)	Normal primary human mammary epithelial cells (HMECs)	MMP7 inhibition	SA-β-Gal Cell cycle distribution	[32]
NR2E1 (nuclear receptor subfamily 2 group E member 1 = TLX)	Prostate epithelial cancer cell lines LNCaP, DU145	NR2E1 knockdown	SA-β-Gal Growth arrest p53 p21	[291]
PARK2 (Parkin RBR E3 ubiquitin protein ligase)	Primary human bronchial epithelial cells	Cigarette smoke	SA-β-Gal Autophagy markers p21	[292]
PRKCH (protein kinase C eta)	Human mammary epithelial MCF-7 (breast cancer cell line)	Etoposide H2O2	SA-β-Gal p21 p27 IL6 secretion	[138]
PRKCI (protein kinase C iota)	Human mammary epithelial MCF-7 (breast cancer cell line)	PRKCI inhibition	Enlarged, flattened morphology SA-β-Gal BrdU γ-H2AX	[293]
PRMT1 (protein arginine methyltransferase 1)	MDA-MB-231 cells	PRMT1 knockdown	Enlarged, flattened morphology SA-β-Gal Growth inhibition p21	[294]
RRM2 (ribonucleotide reductase regulatory subunit M2)	Epithelial ovarian cancer (SKOV3) cell line	RRM2 knockdown	Enlarged, flattened morphology SA-β-Gal BrdU γ-H2AX foci and WB 53BP1 foci	[295]
SIRT6 (sirtuin 6)	Normal primary human bronchial epithelial cells	TGF-β-induced senescence	Enlarged, flattened morphology SA-β-Gal p21 IL1β	[224]
SIRT6 (Sirtuin 6)	Normal primary human bronchial epithelial cells	Cigarette smoke	SA-β-Gal p21 γ-H2AX foci	[246]
SIRT6 (Sirtuin 6)	Primary human keratinocytes	Stress-induced senescence	SA-β-Gal	[296]
SNAI1 (snail family transcriptional repressor 1)	Human prostate cancer cell line LNCaP	SNAI1 inhibition	SA-β-Gal	[297]
TERF2 (telomeric repeat binding factor 2)	Normal human epithelial keratinocytes (NHEKs)	DN TRF2 overexpression	SA-β-Gal 53BP1 foci Phospho-p53Ser15 p21 BrdU incorporation	[298]
TFF1 (trefoil factor 1)	Pancreatic and prostate cancer cells PC3 cells, HS766T cells	Knockdown of TFF1	Enlarged, flattened morphology SA-β-Gal Growth inhibition	[299]
TWIST1 (twist family BHLH transcription factor 1)	H460 NSCLC cancer cells	TWIST inhibition	SA-β-Gal p21 p27	[300]
TWIST1 (twist family BHLH transcription factor 1)	Immortalized human prostate epithelial cell lines	Cisplatine H2O2	SA-β-Gal p53 p21 γ-H2AX foci	[301]
TWIST2 (twist family BHLH transcription factor 2)	SiHa and HeLa cells	Knockdown of Twist2	SA-β-Gal Growth inhibition CBX3 positive cells cyclinD1 mRNA expression Cdk4 mRNA expression	[302]
RNF114 (ring finger protein 114 = ZNF313)	MCF7, HCT116, PC3	ZNF313 knockdown or adriamycin-induced senescence	SA-β-Gal Cell growth p21	[303]

Only those articles that show a functional role of these regulators by the use of inhibition (genetic or pharmacological) or overexpression (vectors, agonists) techniques were taken into account. The context inducing senescence and the markers used in the studies are indicated

Fig. 2.

Protein–protein interactions between the known regulators of epithelial cell senescence. The 70 regulators of epithelial cell senescence listed in Table 2 plus the cell cycle regulators p53, p21, p16, RB, p15, p27, cyclin D1 and cyclin D2 were analyzed using STRING for predictive protein-protein interactions and then entered into CYTOSCAPE to visualize the network. The diameters of the circles are proportional to the number of interactions. Red circles represent positive regulators of epithelial cell senescence, blue circles negative regulators, and purple circles regulators reported either as positive or negative regulator according to the study

Apart from these expected nodes, this analysis highlighted other nodal proteins that could, therefore, play a central role in epithelial cell senescence. One very central is the ATP-dependent molecular chaperone HSP90, with 22 connections. There are four isoforms of HSP90, two in the cytosol (HSP90α and β), one in the mitochondrion (TRAP1) and one in the endoplasmic reticulum (GRP94). HSP90 is needed for the correct folding of newly synthesized proteins and for the renaturation of denatured or misfolded proteins. Its client proteins include p53, CDK4, CDK6, hTERT, some MMPs, and kinases of the RAS pathway, i.e., some important inducers or components of the senescent phenotype. However, its clients are much more numerous than that: about 1–2% of the cellular proteins and this level rises to 4–6% in stressed cells. AKT is also at a node with 15 connections in epithelial cell senescence regulation. AKT controls, amongst others, the mTOR pathway (for review see [73]), one major regulator of nutrients availability and protein synthesis. The central place of these two proteins in the regulatory network of epithelial cell senescence suggests that epithelial cell senescence could be associated with an alteration of cell proteostasis.

Interestingly, when the ATPase activity of HSP90 is pharmacologically inhibited, its client proteins become degraded by the proteasome [74]. However, there is a general decline in the proteasome activity at senescence (for reviews see [75–78]). Therefore, one expects that senescence in epithelial cells could be associated with an accumulation of abnormal proteins and an induction of the unfolded protein response (UPR) pathway (Fig. 3). And indeed, an involvement of the UPR pathway in senescence was documented (for review see [79]), but again only very few studies were performed in epithelial cells (for specific references on UPR in senescent epithelial cells see [80, 81]).

Fig. 3.

Summary of the known cellular and molecular events involved in epithelial cell senescence. Oxidative stress is at the center of the events involved in epithelial cell senescence. It is responsible, at least in part, of the DNA damages that induce the cell cycle arrest and the SASP, as well as of the induction of housekeeping machineries. It is maintained by positive loops involving some components of the SASP, the mitochondrion and the p38MAPK

In addition, one can assume that the potential disruption of proteostasis at epithelial cell senescence and the decline in proteasome activity could result in an increase in autophagy. Chaperone-mediated autophagy, macroautophagy and microautophagy are three processes of cellular component degradation involving lysosomes. The three processes differ in the type of cargos they degrade and in the way those cargos are captured by the lysosome (for reviews on autophagy see [82–85]). In brief, macroautophagy cargos are various, including native molecules, protein aggregates, intact or altered organelles. The cargos are first isolated inside a double membrane vesicle, the autophagosome, which then fuses with a lysosome forming an autolysosome in which the sequestered material is degraded. Chaperone-mediated autophagy (CMA) targets only altered proteins exposing a specific KFERQ motif. These proteins are taken over by chaperones and directly internalized by the lysosome through the LAMP2 receptor. The process of microautophagy is less well known. It involves a direct internalization of the cargo by the lysosome by a mechanism of lysosomal membrane invagination. A general decrease in CMA with aging was described [77, 78, 86], but direct evidence of such a decrease in senescent cells, especially epithelial ones, is lacking. In contrast, an increase in macroautophagy at senescence was very well described, including for epithelial cells [7, 29, 87, 88] (Fig. 3). However, the roles and the consequences of this increased macroautophagic activity are not clear. A first possibility is that macroautophagy would be a necessary part of the senescent phenotype which would help senescent cells to maintain their homeostasis and survive at long term. A second possibility is that the macroautophagic activity would increase at senescence because altered proteins and other cell components accumulate, but would never reach a level high enough to ensure a good clearance, thereby inducing senescence. A third possibility would be in contrast that the macroautophagic activity would reach a very high level at senescence, leading to the degradation of too numerous vital cell components and finally to cell death. All three scenarios were independently shown to occur in association with epithelial cell senescence. The first scenario was described during biliary epithelial cell senescence and for murine kidney tubular epithelial cells undergoing TIS, OIS or spontaneous senescence [89, 90]. Macroautophagy was proven to be part of the senescence process because once inhibited, the senescence markers, including components of the secretome, declined [90]. The second scenario was demonstrated for senescent bronchial epithelial cells in the context of chronic obstructive pulmonary disease. Senescence induced by cigarette smoke extracts was shown to be accompanied by an increased macroautophagy. However, the macroautophagy level revealed to be insufficient for efficiently eliminating the ubiquinated proteins which continue to accumulate [7, 91]. We have demonstrated the last scenario for NHEKs at the first senescence plateau. A high macroautophagic activity is induced in senescent NHEKs because of their high level of oxidative stress affecting mitochondria and the nucleus. This high macroautophagic activity is associated with an increase in Beclin1 and a decrease in BCL2 expressions. It leads to a specific pathway of cell death associated with plasma membrane permeabilization without activation of the apoptotic pathway [29, 30, 92]. Similarly, human ovarian surface epithelial cells undergoing OIS end up dead within a few days by a caspase-independent mechanism with features of macroautophagy [93]. As in senescent NHEKs, Beclin1 is overexpressed, and its silencing increases cell survival [29, 93].

Role of oxidative stress in epithelial cell senescence

Oxidative stress, the result of an imbalance between the rates of ROS production and degradation, is known for decades to play a major role in senescence [94, 95]. Several lines of evidence point it as the main inducer of epithelial cell senescence.

Normal human epidermal keratinocytes at the first senescence plateau have an intracellular concentration in ROS (mostly H₂O₂) that is about 30-fold higher than during exponential growth phase [30]. Continuously treating NHEKs with catalase, an anti-oxidant enzyme that degrades H₂O₂, or with N-acetyl-cysteine, a general antioxidant, delays the occurrence of the first senescence plateau [19, 96] and also decreases the level of autophagic cell death [92].

The pathways activated by ROS in senescent epithelial cells could be numerous. In our STRING/CYTOSCAPE analysis of epithelial cell senescence regulators, p38MAPK (MAPK14) stands out as a major node with 15 connections (Fig. 2). P38MAPK is a mitogen-activated protein kinase which belongs with JNK to the group of stress kinases known to be activated following oxidative stress, DNA damage and other stresses [97–99]. P38MAPK is well-known to be involved in SIPS of human and murine fibroblasts [100]. It was demonstrated to be the main mediator between oxidative stress and the up-regulation of p16 [59]. P38MAPK is also able to activate p21 in a context of OIS occurring in colorectal cell lines in which p16 and p53 are inactivated [72]. Caveolin1, a scaffolding protein of caveolae, is also at a node in the network of epithelial cell senescence regulators with 14 connections (Fig. 2). It is both a marker and an actor of oxidative stress: as p16, it is up-regulated by oxidative stress through the activation of p38MAPK; in turn, it directly or indirectly inhibits anti-oxidant pathways by inhibiting the anti-oxidant enzyme thioredoxin reductase 1 (TrxR1) and the redox-sensitive transcription factor NRF2, respectively, thereby increasing the level of oxidative stress [101]. Another node in the network of epithelial cell senescence regulation is SMAD3, with 16 connections (Fig. 2). In lens epithelial cells, low levels of oxidative stress lead to the nuclear translocation of SMAD3 and to the up-regulation of transglutaminase 2, an enzyme which catalyzes protein aggregation [102]. In mammary cancer cells, the activation of SMAD3 upon oxidative stress induces its interaction with BRCA1 [103]. Interestingly, senescence was shown to occur in prostate cancer cells (PC3) upon androgen receptor expression and activation. This senescence occurs without activation of the DDR pathway, but with activation of p21 and RB hypophosphorylation, independently of p16 up-regulation but dependently on an increase in ROS [104]. Taken together, these data suggest that senescence of epithelial cells could to be strongly dependent on oxidative stress, with p38MAPK as a frequent relay in the pathways.

Although this oxidative stress could induce senescence through damages to several types of macromolecules and organelles, DNA seems to be the preferential target for an induction of the cell cycle arrest. One of the most common oxidative DNA damage is single-strand breaks (SSBs) [105, 106]. SSBs are defined as a breakdown of the DNA phosphate–sugar backbone only on one DNA strand, often accompanied by one nucleotide loss creating altered 5′ and 3′ ends. The SSB repair (SSBR) pathway first involves a step of detection of the break by a poly(ADP)ribose polymerase (PARP), mainly PARP1. Hence activated, PARP1 synthetizes poly(ADP)ribose (PARs) which accumulate at the breaks making foci detectable by fluorescence microscopy. PARs favor in turn the recruitment of the scaffold protein XRCC1. The next steps are common with the Base Excision Repair (BER) pathway. The XRCC1 protein becomes phosphorylated by the casein kinase 2α (CK2α), what favors the recruitment, stabilization and activation of the polynucleotide kinase phosphatase (PNKP) and Aprataxin (APTX), the enzymes involved in the restoration of the 3′ and 5′ ends of DNA, a step necessary to the final action of the polymerases β, δ and ε, and of the ligases I and IIIα [105–109]. In a recent study [19], we have shown that NHEKs and HMECs at the first senescence plateau specifically accumulate abnormal XRCC1 foci: they are 5-fold bigger than normal; XRCC1 is non-phosphorylated; PNKP, ligases I and IIIα are absent and no DNA polymerase activity can be detected by BrdU incorporation. In consequence, the SSBs remain unrepaired as evidenced by comet assays. This dysfunctioning of the SSBR pathway results from a down-regulation of the expression of PARP1 at the messenger level, leading to a drastic decrease in protein expression and activity. The few PARs synthetized by the remnant activity are, however, enough to recruit XRCC1, but insufficient to recruit CK2α, leaving XRCC1 unphosphorylated. That is the reason why the XRCC1 foci remain unresolved and why XRCC1 continues to accumulate at the foci which hence become abnormally large. Re-expressing PARP1 restores the recruitment of CK2α, the phosphorylation of XRCC1, the recruitment of PNKP and the resolution of the foci.

Interestingly, the down-regulation of PARP1 at the first senescence plateau of NHEKs was shown to be dependent on the senescence-associated oxidative stress [19]. The details of this regulation are not yet established. Whatever the point, it is interesting to note that the mechanism of accumulation of SSBs upon oxidative stress is double: ROS create the DNA breaks directly as chemical compounds able to attack the DNA backbone, and ROS are responsible for letting them unrepaired and accumulating by indirectly inhibiting the transcriptional expression of the SSBR-initiating enzyme.

In addition, it is the formation of the abnormal non-phosphorylated XRCC1 foci that could activate the p16/RB pathway and induce senescence. Indeed, invalidating XRCC1 using RNA interference in NHEKs already at the first senescence plateau suppresses the up-regulation of p16 and restores the phosphorylation of RB [19]. The paradigm is that XRCC1 would recruit different proteins according to its phosphorylation status: when phosphorylated, it would regularly recruit repair proteins, when unphosphorylated, it would rather recruit protein(s) of a cell cycle arrest pathway. One of them is p38MAPK: it is activated in NHEKs at the first senescence plateau and this activation is abrogated upon XRCC1 invalidation [19], suggesting it is activated by the abnormal unphosphorylated XRCC1 foci. However, whether p38MAPK is directly or indirectly recruited and activated at these foci remains to be documented. The other partners in the pathway leading from p38MAPK to p16 up-regulation have also to be characterized. P16 expression is known to be controlled during senescence by several mechanisms that were reviewed in 2012 [110], including involvement of the polycomb repressor complexes PRC1 and PRC2, and AP1, ETS and Id1 transcription factors. However, again, most of the studies on this subject were performed in fibroblasts. More recently, 22 senescence-associated miRs were identified in HMECs, as well as in mammary fibroblasts. Four of them, 26b, 181a, 201 and 424, were shown to coordinately repress expression of the CBX7, EED, EZH2 and SUZ12 components of PRC2, thereby leading to p16 up-regulation [111]. The TGFβ/SMAD3 pathway was also implicated in the induction of p16 in mouse keratinocytes undergoing OIS [112].

Therefore, the first senescence plateau of keratinocytes and mammary epithelial cells would be a stress response to endogenous oxidative stress, whose level is too low to induce double-strand breaks and the DDR/p53-p21 pathway, but high enough to induce single-strand breaks, whose accumulation is enforced by the down-regulation of PARP1, and whose aberrant signalization activates a p38MAPK/P16-RB cell cycle arrest pathway (Fig. 1).

The origin of the oxidative stress associated with the senescence of epithelial cells is still today unclear. Mitochondria are obviously dysfunctional during NHEK first senescence plateau, displaying oxidative damages and being the target of macroautophagic elimination [29, 30, 92]. This mitochondrial dysfunction results, at least in part, from an up-regulation of the redox enzyme MnSOD (SOD2). The function of MnSOD is to dismutate the toxic ${\text{O}}_2^{.-}$ into H₂O₂, which in turn is detoxified by catalase and a series of peroxidases. The uncoordinated up-regulation of MnSOD at NHEK senescence induces an imbalance between the production and degradation of H₂O₂, leading to an H₂O₂ accumulation, which contributes to the onset of the first senescence plateau of NHEKs [19, 20, 96]. The up-regulation of MnSOD was also shown to induce senescence in breast and prostate epithelial cell lines [113]. The up-regulation of MnSOD at senescence results from the activation of transcription factors of the NF-κB family [96, 114]. The ability of NF-κB transcription factors to induce or reinforce senescence was clearly shown in several cell types, including normal, immortalized or cancerous epithelial cells [115]. However, why NF-κB activity increases is not known and can just be extrapolated from data obtained in other cell types. In hepatocarcinoma cells and melanocytes, it was shown that NF-κB activation follows the accumulation of DNA damage and mainly involves ATM and PARP1 [116, 117]. In partial contradiction, it was shown in fibroblasts that NF-κB activation is independent of the DDR pathway but dependent of p38MAPK [118]. These data suggest that NF-κB could be induced at senescence by oxidative stress and oxidative DNA damage. A positive loop between oxidative stress and NF-κB activation might, therefore, exist.

Escape from epithelial cell senescence

Reference senescence is associated with a very stable cell cycle arrest which relies on the permanent activation of the DDR pathway [41]. It is assumed that to become cancerous a normal cell has to bypass or overcome this cell cycle arrest, hence becoming immortal, a property supposed to be a prerequisite to cancerogenesis. So, globally, reference senescence is assumed to be a cell-autonomous tumor suppressor state.

However, in epithelial cells undergoing two-step senescence, the first senescence plateau is not associated with the activation of the DDR pathway, and therefore, is not all that stable. Both keratinocytes and mammary epithelial cells were shown to be able to spontaneously overcome the first senescence plateau [19, 20, 24, 26, 30]. In NHEKs at the first plateau, cell cycle arrested cells have two possible outcomes: either they die because of an overactivation of macroautophagy or they re-enter into cell cycle. The outcome depends on the level of macroautophagy which in turn depends on the level of oxidative stress: cells with the highest levels of oxidative stress and macroautophagy undergo cell death; those in which these levels are just below, but still much higher than in proliferating pre-first plateau cells, escape cell death and clear a portion of their altered components, so that a few of them re-enter cell cycle [19, 20, 29, 30, 92]. The frequency of cell cycle re-entry is in between 10⁻⁴ to 10⁻⁵ in both HMECs and NHEKs [20, 24]. Spontaneous escape from senescence associated with features of cell death was also shown in amniocytes [119], in irradiated HeLa cells [120], and in cancer cells undergoing OIS or TIS [72, 121, 122]. Colorectal cancer cells escaping from OIS or TIS overexpress the anti-apoptotic proteins MCL1 and BCL-XL [72, 122]. Depletion of MCL1 induces the death of the parental senescent cells and prevents post-senescence emergence [122].

The molecular mechanisms enabling the cell cycle re-entry of senescent epithelial cells are only partially established. Post-first plateau HMECs display an epigenetic extinction by hypermethylation of the p16 promoter [24], suggesting that it is this down-regulation that permits the escape. However, other molecular changes could contribute to post-first senescence plateau evasion since about 2% of all promoters display changes in their methylation status in PSNE HMECs compared to proliferating cells [123]. Post-first plateau NHEKs also display a down-regulation of p16. However, p16 is again reactivated at a high level when cells reach the second plateau [20]. In HMECs, it was proposed that variant cells with hypermethylated p16 promoter pre-exist in the population before the entry in the first senescence plateau [124]. In NHEKs in contrast, it was shown based on monoclonal cultures that all cells of the initial culture are able to enter in the first senescence plateau and then generate post-senescent cells [20]. Besides, the escape from OIS of colorectal cancer cells was associated with p21 down-regulation and CDC25A re-expression [72, 121]. Therefore, the mechanisms promoting senescence escape have to be more deeply reassessed.

The fact that senescent cells that are very large cells would be able to undergo mitosis is rather challenging. We have proven that post-first plateau NHEKs come from the division of bona-fide large senescent progenitors by 2 means. First, we have sorted from a population of NHEKs at the first plateau the largest and most granular cells, i.e., the cells with the more pronounced senescent phenotype. The sorted cells were then plated at low density, stained with the filiation tracers CFDA SE or Dil, and monitored for the appearance of clones of small emerging cells surrounding a senescent one. In the clones, the small emerging cells appeared fluorescently stained, proving they are born from the division of the senescent mother cell. Second, we made an immunofluorescence against cytokeratine 14 and demonstrated by confocal imaging that some small cells were still attached by a cytokeratine 14-positive pedicle to the senescent mother cell [20]. Actually, fully senescent cells do not undergo a classical symmetric mitosis giving rise to two similar daughter cells, but an atypical budding mitosis, also called neosis. In videomicroscopy recordings, the senescent mother cell is most often polynucleated, containing up to 5 nuclei. Almost every nucleus undergoes mitosis (karyokinesis) in a time window of a few hours. For each pair of produced nuclei, one remains in the mother senescent cell and the other buds outwards, taking with it some surrounding cytoplasm (cytokinesis). It thereby gives rise to a new small mononucleated cell which then secondarily undergoes classical mitosis to generate a clone. This process was demonstrated for NHEKs at the first senescence plateau, for senescent amniocytes, and for several cancer cells lines undergoing OIS or TIS [20, 119–122, 125, 126].

Escape from the second senescence plateau of NHEKs was reported, but the mechanisms were not investigated [20, 28]. Although spontaneous escape from the second growth plateau of HMECs was not yet reported, an escape can be induced by overexpressing ZNF217 and c-MYC in post-first plateau cells. The escaped cells display several hundred supernumerary changes in methylation compared to post-first plateau cells [123].

The cells that escape from epithelial cell senescence are (pre)tumorigenic

One important point regarding these spontaneous escapes from epithelial cell senescence is to distinguish them from situations in which cells avoid senescence. Escape from senescence means that cells do enter senescence, remain senescent a few days and then re-renter cell cycle and undergo division to generate a progeny. Cells that avoid, bypass or escape senescence never undergo senescence; that is the case, for example, for fibroblasts overexpressing hTERT. This distinction is tremendously important, because cells that have bypassed senescence have became immortal but have kept most of the properties of normal cells, whereas cells that have escaped from senescence can have acquired during their senescent period DNA damages, mutations or other modifications that could have given them new properties. Therefore, escaping from senescence is not a simple reversion to the anterior proliferation state but is an evolution towards a new status. As will be argued below, this new status is (pre)tumorigenic.

Post-first plateau NHEKs were shown to display some transformation traits and are able to form small tumors. That is the reason why we have called them post-senescence neoplastic emergent (PSNE) cells. The transcriptomic signature of PSNE NHEKs share some genes in common with the transcriptomic signature of psoriasis, an hyperproliferation disease of the epidermis, as well as with transcriptomic signatures of different types of cancers, including non-melanoma skin cancers [27]. Moreover, PSNE NHEKs are slightly engaged in an epithelium to mesenchyme transition (EMT), which is dramatically amplified when these cells are submitted to the secretome of senescent fibroblasts. This EMT is marked by an overexpression of the EMT master genes twist and slug, a down-regulation of the MET receptor, an up-regulation of the PAR1 (F2R) receptor, an up-regulation of ADAM-10, a down-regulation of E-cadherin and an up-regulation of vimentin. PSNE NHEKs submitted to a SASP display fibroblastoid morphology and increased migration capacity [19, 28]. Importantly, despite PSNE NHEKs (as well as PSNE HMECs) are not immortalized since they enter a second senescence plateau [20, 24], they are able to form some small skin hyperplasias and non-melanoma carcinomas when injected in nude mice. Of note, these skin lesions developed only after a delay of about 8 months, i.e., in a context of an already aging microenvironment [20]. Colorectal cancer cells escaping from OIS or TIS also display an EMT, grow in low adhesion conditions, and are as able as parental cancer cells to form tumors in immunocompromised mice [72, 122]. Similarly, an escape from senescence of irradiated HeLa cells gives rise to daughter cells able of anchorage-independent growth [120]. PSNE cells are mutated, as suggested by HPRT assays [19]. Notwithstanding, chromosomal abnormalities visible by cytogenetic analyses are not yet present; they accumulate only when the cells are about to reach the second senescence plateau [20, 24, 26].

The second senescence plateau seems to be more stable than the first one. In HMECs, no spontaneous emergence from the second plateau has yet been reported. In contrast, in NHEKs, a spontaneous second emergence from the second plateau was observed, although at a frequency of about 10⁻⁷, i.e., much lower than the first one. The second emergent cells have still long telomeres, although the telomerase is not reactivated, and massively display chromosomal abnormalities, the majority of whom being aneuploidies and translocations [20, 28]. Colorectal cancer cells escaping OIS or TIS display more DNA damage, probably resulting from replicative stress, than their parental senescent cells, are more resistant to chemotherapeutic drugs, and display chromosome gains and losses [72, 126].

Therefore, the cells that escape from senescence are genetically instable, a well-established characteristics of cancer cells. What event(s) in the progenitor senescent cells initiate(s) this genetic instability? Our recent study on NHEKs is, to our knowledge, the only one tackling this issue. We have shown that oxidative stress and the consecutive SSB generation and down-regulation of PARP1 are necessary and sufficient for causing PSNE, suggesting that the mutagenicity of accumulated unrepaired SSBs is the motor of the process [19].

The secretome of senescent epithelial cells

One major characteristic of the senescent phenotype is its secretome, the SASP, which becomes enriched in pro-inflammatory interleukines and chemokines, in growth factors and extracellular matrix-remodeling proteases. Senescent cells thereby modify their microenvironment, making it a stimulating support for tumor development from premalignant cells. The SASP is also referred as the senescence messaging secretome (SMS) because it reinforces the senescent phenotype in an autocrine manner. It is also sufficient to induce senescence of non-senescent neighbor cells in a paracrine manner. The SASP temporally develops after the installation of the senescent cell cycle arrest, but as for the cell cycle arrest, mainly in response to the senescence-associated DNA damages. The transcriptional expressions of the SASP components are mainly under the control of NF-κB transcription factors which are themselves activated by the DDR pathway, by p38MAPK, and by PARP1. The composition, the mechanisms of induction and the main properties of the SASP are globally shared by all cell types including epithelial cells whatever they undergo RS, SIPS, OIS or TIS (for recent review see [127] and for references specifically on epithelial cells see [128–136]). However, some subtle specificity regarding the epithelial SASP seems to exist.

Coppe et al. compared the composition of the SASP of human normal fibroblasts and prostate epithelial cells induced in senescence by X-ray irradiation [137]. The results indicate a 66% overlap of the two SASPs, with in common cytokines and growth factors such as IL-6, GRO, and GM-CSF. However, the epithelial SASP also presents significant differences compared to the fibroblastic one. There are some cytokines which are up-secreted by senescent prostate epithelial cells compared to senescent fibroblasts such as ENA-78 (CXCL5), MIP-3α (CCL20), VEGF, and IL-1α. There are also some proteins which are less present in the SASP of prostate epithelial cells than in that of fibroblasts, including IL-8, HGF and IL-7. In mammary cancer cells undergoing SIPS and TIS, it was shown that IL-6 expression is enhanced by PKCη, whereas the expression of IL-8 is specifically repressed [138]. Interestingly, there are even a few proteins which are down-regulated in senescent epithelial prostate cells, whereas enriched in the SASP of fibroblasts, notably IGFBP6, known to inhibit the growth of cancer cells [139]. Therefore, the SASP of epithelial cells could be even more pro-inflammatory and pro-tumoral than that of fibroblasts. Interestingly, the same study of Coppe et al. has shown that p53 restrains the SASP of fibroblasts. When p53 is inactivated in fibroblasts, the SASP is amplified, and enriched with several new cytokines, notably ENA-78. Similarly, p53-deficient prostate cancer cells lines develop a SASP after X-ray-induced senescence which is more developed than that of primary prostate cells, with an oversecretion of a large set of cytokines and growth factors [137]. Therefore, the SASP of epithelial cells could be much enriched in proinflammatory and protumoral cytokines because it occurs essentially independently of p53.

The SMS function of the SASP of epithelial cells was little studied. However, a few studies interestingly point on IGFBP7 (IGFBP-rP1/MAC25). IGFBP7 is a low affinity insulin-like growth factor binding protein that is suggested to be a tumor suppressor protein (for review see [140]). IGFBP7 is overexpressed by HMECs and primary prostate epithelial cells [141, 142]. It is able to induce in a paracrine manner senescence of triple negative breast cancer cells (MDA-MB231) through a strong activation of the p38MAPK, an activation of p53 and an up-regulation of p21 [143].

P16-positive senescent epithelial cells accumulate with aging and in hyperplastic and pre-neoplastic lesions

The most innovative results of these last years on senescence concern the occurrence and the functions of senescence in living organisms. Unexpectedly, senescence was shown to occur in vivo at all the stages of life. It occurs during embryonic development to participate in the regulation of tissue partening. It occurs at adult age in a controlled manner in response to inflammation to regulate the repair processes. It also occurs at sites of some pathology, presumably in response to an acute or chronic stress, including some therapeutic interventions. And it occurs in most tissues and organs, at an increasing rate as age progresses, probably in response to chronic age-associated stresses (for reviews see [17, 144]). During development, different cell types in different organs were shown to undergo senescence. However, this senescence appears rather atypical since it is not associated with DNA damage and seems to involve quasi-exclusively p21, without p53 involvement, without p16 involvement, and without the production of a SASP [144]. During the repair process, senescence concerns mostly fibroblasts or myofibroblasts and endothelial cells in the skin or stellate cells in the liver and would act to restrict fibrosis [145–147]. In these two contexts, senescence is globally beneficial to the organism and is only transient, very likely because senescent cells are cleared by the immune system. In contrast, senescent cells accumulate with aging probably because their phenotype is stable, in line with the concepts drawn from in vitro studies, and probably because they are not properly cleared by the immune system. This senescent cell accumulation with aging is globally detrimental to the organism: it restricts lifespan and contributes to several age-associated disorders and pathologies [150]. Therefore, we will discuss here only the studies reporting data relative to senescent epithelial cells accumulating with aging.

Detecting senescence in vivo is more challenging than in vitro: the increase in cell size and spreading of senescent cells was not clearly evidenced in vivo; measuring the SA-β-Gal activity can not be performed on formalin-fixed paraffin-embedded specimens; enumerating senescent cells through their expression of cell cycle regulators or proteins of the DNA damage signalization is strongly dependent on the specificity and affinity of the existing antibodies. These limitations could be overcome by the recent design of a very sensitive derivative of Sudan Black B that detects lipofuscin, an aggregate of oxidized proteins and lipids that accumulates in senescent cells [148].

Sharpless and collaborators have estimated by quantitative real-time PCR, the expression of several CKIs including p16, Arf, p15, p18, p19, p21 and p27 in twenty seven organs of young versus old mice and rats. The results clearly show that p16 is by far the most up-regulated CKI at advanced age with an average increase of about tenfold, compared to an only 1.4-fold increase of p21. Inside the 27 examined organs, p16 up-regulation was detected in a variety of cell types including epithelial cells, stromal cells, lymphocytes and β-cells [149]. A dramatic increase of about 30-fold in p16 expression in the aged mouse kidney was confirmed by Baker et al.; in comparison, p21 was induced by only twofold, whereas in lung and colon, p16 and p21 were up-regulated at approximately the same level [150]. In these studies, the increased p16 expression was correlated with an increase in SA-β-Gal activity and/or expression of some senescent secretome components, indicating that p16 up-regulation is an in vivo marker of senescence [149, 150]. Importantly, the selective elimination of p16-positive cells using a suicide transgene markedly reduced glomerulosclerosis, a hallmark of kidney aging [150]. In the human kidney as in the rodent one, p16 expression was shown to increase about tenfold with age, especially in epithelial tubular cells of the cortex, whereas the expression of p53 and p21 remain unchanged [151]. In kidney from patients with Immunoglobulin A nephropathy, a common disease which shares similar characteristics with kidney aging, the percentage of tubular epithelial cells positive for p16 increases with the grade of the pathology up to 60%. These cells are also positive for SA-β-Gal, and in this situation for p21. Moreover, they oversecrete collagen III [152]. In the human corneal epithelium, both p16 and p21 expression were shown to increase, but interestingly the increase in p16 expression appeared more precocious than that of p21 [153].

von Zglinicki and collaborators performed a study using γH2AX foci as a marker of DDR-induced senescence in mice. In contrast with the Sharpless study showing that p16 increases with age in almost all organs, no significant increase in γH2AX foci was observed in heart and skeletal muscle, eye lens, testis or kidney [154]. In partial disagreement, Berkenkamp et al. did find an accumulation of γH2AX-positive tubular cells in aged mouse kidney, but in only about 1.5% of the cells [155]. In the skin of aged mice, Wang et al. observed a significant increase in γH2AX-positive fibroblasts in the dermis, but again, only in a small percentage of cells (about 5%); no increase in γH2AX-positive keratinocytes was recorded in the epidermis [154]. We obtained similar results for human skin. We used 53BP1 as a DDR marker and observed an accumulation with age of a few 53BP1 foci in about 20% of the fibroblasts, whereas in the epidermis we did not detect any significant increase in DDR-positive cells. We used, in addition XRCC1 to detect cells whose senescence could be dependent on an accumulation of SSBs and on the activation of the SSBR pathway. We recorded an accumulation of XRCC1 foci in about 40% of epidermal cells in correlation with a strong decrease in PARP1 expression, whereas PARP1 remains constantly expressed in the dermis, without any XRCC1 foci accumulation. Epidermal cells also showed signs of oxidative stress revealed by the overexpression of MnSOD, whereas fibroblasts of the dermis did not [19]. Of note, both human epidermis and dermis were shown to be enriched in SA-β-Gal-positive cells at old age, however, without any precise quantification [156].

Therefore, there is clearly an accumulation with age of epithelial cells positive for p16, SA-β-Gal, secretome components and/or DNA single-strand breaks, i.e., senescent. Among those cells, only a minority, if any, seem to display DDR markers. This suggests that in vivo senescence of epithelial cells does not rely on the accumulation of DNA damages able to activate the DDR pathway as for in vitro reference senescence, but rather on the accumulation of other damages, notably oxidative DNA single-strand breaks, responsible for the activation of the p16/RB pathway. This also suggests that senescent epithelial cells accumulated with aging are closer to cells at the first senescence plateau than to cells at the second senescence plateau. Of note, since these cells accumulate with aging, they may not undergo cell death in vivo as they do in vitro. They also may escape immune surveillance for reasons that have to be established.

An accumulation of senescent cells was also reported in several hyperplastic disorders and in benign tumors. The senescent cells accumulating in these contexts are of different types, including epithelial ones. Benign Prostatic Hyperplasia is a very common disease of old men characterized by an increased proliferation of both epithelial and stromal cells. Although debated, it could be a premalignant condition which may predispose to prostate cancer development. In an initial study, Choi et al. reported that 40% of Benign Prostatic Hyperplasia specimens were positive for SA-β-Gal activity. In these tissues, the SA-β-Gal activity was detected only in epithelial cells and correlated with a high prostate weight greater than 55 g [157]. An increased expression of GM-CSF, IL-1α, IL-4 and IL-8 was also recorded in these hyperplasias [132, 158, 159]. These cytokines were shown to increase also in senescent prostatic epithelial cells in vitro and could mediate the hyperproliferation of both prostatic non senescent epithelial cells and fibroblasts [132, 159]. Interestingly, a higher level of oxidized DNA bases was found in Benign Prostatic Hyperplastic tissues compared to normal surrounding ones [160], suggesting that the senescence of epithelial prostatic cells in Benign Prostatic Hyperplasia could be induced by oxidative stress.

Bile duct adenomas and cholangiocellular carcinomas are, respectively, intrahepatic lesions and tumors originating from bile ductular cells. Bile duct adenomas extensively express p16 but this expression is considerably lower in cholangiocellular carcinomas. Conversely, EZH2, a component of the PRC2 complex that represses p16 expression is highly expressed in cholangiocellular carcinomas but not in bile duct adenomas [161]. A similar exclusive expression of p16 and EZH2 was found in the couple papillary hyperplasia versus carcinoma of the gallbladder [162].

In atypical hyperplasia of the breast, which is a benign breast disease that could represent a precursor stage to breast carcinoma, p16 expression was found increased. This increase correlates with increasing age at diagnosis but not with the risk of breast cancer [163]. P16 expression was also found increased in ductal breast carcinoma in situ. In this context of more advanced carcinogenesis, the high p16 expression correlated with an increased risk of subsequent tumor lesion only when associated with a high Ki67 expression. The high p16/high Ki67 cells also display low levels of RB and high levels of E2F3, cyclin E and D1, indicating a deregulated p16/RB pathway.

Several pre-cancerous and cancerous human lesions are positive for an active form of an oncogene of the ras family, either H-ras, K-ras or B-raf which were shown in vitro to induce OIS. Mice expressing a constitutively active K-rasV12 allele develop multiple pre-malignant lung adenomas and pancreatic intraductal neoplasias containing numerous senescent cells positive for p16, p15, Dec1, DcR2, HP1-γ and SA-β-Gal. In contrast, the rare malignant lung or pancreatic ductal adenocarcinomas that developed were negative for the above senescent markers [164]. When p16 or RB were conditionally deleted in combination with K-RAS expression, the pancreatic intraductal neoplasias appeared earlier and rapidly progressed to invasive and metastatic cancers [165, 166].

Taken together, these results suggest that premalignant lesions such as hyperplasias and adenomas are made of spontaneous or RAS-induced senescent epithelial cells arrested in the cell cycle by p16 up-regulation. These lesions are prone to evolve in carcinomas by escaping senescence through an epigenetic inhibition of p16 and/or deregulation of other actors of the p16/RB pathway.

The significance of the accumulation of p16-positive cells, i.e., mainly but not exclusively senescent epithelial cells, with aging was highlighted by the studies of Ian van Deursen using a mouse model in which p16-positive cells can be selectively killed upon administration of a drug activating a suicide transgene. This clearance in mice from 1-year-old attenuated several age-related disorders in several organs at 18 months, delayed tumorigenesis and increased lifespan [150].

Conclusions and perspectives

Senescence in epithelial cells displays most of the characteristics of an adaptive response to stress, mainly oxidative and/or oncogenic stresses. It involves primarily the activation of the p16/RB cell cycle arrest pathway, often independently of the activation of the DDR pathway, and is associated with the activation of stress-response pathways such as autophagy and unfolded protein response (Fig. 3). It differs from the reference senescence by its stability and by its links with tumorigenesis. Compared to DDR-dependent reference senescence which is very stable and tumor-suppressor in a cell-autonomous manner, epithelial cell senescence is less stable and permissive to the generation of neoplastic post-senescent cells, thereby behaving as a mechanism of cancer initiation. The frailty of the cell cycle arrest might rely on the mechanisms of p16 regulation, whose diversity could statistically permit some rare cells to down-regulate p16 and reenter into cell cycle. The senescence-inducing stress, especially the oxidative one, acts to generate discreet DNA damages which represent a source of mutations and genetic instability for the daughter cells. Besides, epithelial cell senescence is also tumor promoter through its SASP. This is true also for reference senescence, but the tumor promoting effect of the epithelial SASP could be higher. Therefore, epithelial cell senescence would be tumor-promoter by two mechanisms and at two stages of tumor development: as a reservoir of damaged tumor initiating cells, and, later, as a producer of a microenvironment favoring tumor progression. This could explain why in humans the most frequent cancers arise from epithelial cells, and why their incidence increases with age.

Therefore, eliminating senescent epithelial cells or minimizing their accumulation in tissues and organ with time could represent therapeutic avenues to prevent or delay cancer incidence, as well as to decrease several other age-related disorders and to increase life span, as demonstrated in the Van Deursen mouse model [150]. To be easily medically applicable, senotherapy, as it is called now, has to be drug-delivered. The presently identified senolytics, i.e., drugs that target some pro-survival pathways and induce cell death preferentially in senescent cells, include the kinase inhibitors Dasatinib and Quercetin, and inhibitors of anti-apoptotic members of the BCL2 family [167–170]. However, until now, the specific expression of the targets of these drugs in senescent epithelial cells has not been checked, nor the efficacy of the drugs in specifically killing senescent cells including epithelial ones, nor their power of delaying or decreasing cancer incidence.

Abbreviations

BER

Base excision repair

BPEC

Breast primary epithelial cell

CKI

Cyclin-dependent kinase inhibitor

DDR

DNA damage response

DSB

Double-strand break

EMT

Epithelium-to-mesenchyme transition

HMEC

Human mammary epithelial cell

NHEK

Normal human epidermal keratinocyte

NHOK

Normal human oral keratinocyte

OIS

Oncogene-induced senescence

PARP1

Poly(ADP)ribose polymerase 1

PD

Population doubling

PSNE

Post-senescence neoplastic emergence

ROS

Reactive oxygen species

RS

Replicative senescence

SA-β-Gal

Senescence-associated-β-galactosidase activity

SAHF

Senescence-associated heterochromatin foci

SASP

Senescence-associated secretory phenotype

SIPS

Stress-induced premature senescence

SMS

Senescence messaging secretome

SSB

Single-strand break

SSBR

Single-strand break repair

TIS

Therapy-induced senescence

UPR

Unfolded protein response

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.