SenNet recommendations for detecting senescent cells in different tissues

Abstract

Once considered a tissue culture-specific phenomenon, cellular senescence has now been linked to various biological processes with both beneficial and detrimental roles in humans, rodents and other species. Much of our understanding of senescent cell biology still originates from tissue culture studies, where each cell in the culture is driven to an irreversible cell cycle arrest. By contrast, in tissues, these cells are relatively rare and difficult to characterize, and it is now established that fully differentiated, postmitotic cells can also acquire a senescence phenotype. The SenNet Biomarkers Working Group was formed to provide recommendations for the use of cellular senescence markers to identify and characterize senescent cells in tissues. Here, we provide recommendations for detecting senescent cells in different tissues based on a comprehensive analysis of existing literature reporting senescence markers in 14 tissues in mice and humans. We discuss some of the recent advances in detecting and characterizing cellular senescence, including molecular senescence signatures and morphological features, and the use of circulating markers. We aim for this work to be a valuable resource for both seasoned investigators in senescence-related studies and newcomers to the field.

Introduction

Cellular senescence has been classically described as a stable cell cycle arrest that develops in response to stress or damage, thereby triggering multiple intracellular phenotypic changes. In humans, mice and some other species, cellular senescence has been shown to contribute to various physiological and pathological processes¹.

Senescence is influenced by various intrinsic and extrinsic factors and can be observed as early as during embryogenesis^2,3. Although senescence is an important, physiologically programmed function during tissue repair, wound healing and embryonic development, in other contexts it is typically the result of stochastic stress, either incurred acutely or accumulated chronically⁴. For example, many anticancer therapies have the adverse effect of inducing senescence in healthy, non-cancerous cells, thereby contributing to local and systemic inflammation⁵.

Although the mechanisms driving cellular senescence in tissues are not fully understood, evidence suggests that the accumulation of senescent cells is associated with chronic inflammation, which can be detrimental and result in tissue dysfunction^1–6. The reason for the accumulation of senescent cells with age — whether more senescent cells are generated or fewer are cleared — is not yet well understood. Several lines of evidence suggest that the ability of the immune system to clear senescent cells decreases with age, leading to their accumulation in tissues^7–10.

In recent years we have witnessed a growing interest in unravelling the physiological functions and pathological consequences of cellular senescence and its underlying molecular mechanisms. However, the identification, characterization and isolation of senescent cells in vivo present considerable challenges owing, in part, to technical complexities, resulting in only partial understanding of cellular senescence in mammals. A key challenge is the lack of a specific marker of cellular senescence, as virtually all markers that have been proposed can be absent from senescent cells in specific contexts (for example, cell types, senescence type), or they can be present in non-senescent cells. Moreover, the lack of widely accepted definitions of phenotypes of senescence has further hindered progress in this field.

The main goal of this Expert Recommendation is to provide a tissue-specific overview of senescence markers in the context of ageing and disease. Additionally, we aim to offer guidance on using these markers to identify and characterize diverse senescence phenotypes in both healthy and diseased tissues.

The National Institutes of Health (NIH) Cellular Senescence Network (SenNet) Consortium¹¹ aims to map senescent cells in multiple tissues along the lifespan of both humans and mice using state-of-the-art single-cell technologies. The Biomarkers Working Group within SenNet has observed a recurring issue within the scientific literature regarding cellular senescence in various tissues in the context of ageing and age-related conditions, namely a tendency towards oversimplification of cell-state annotation, exacerbated by the absence of a universally accepted definition of cellular senescence. Importantly, a considerable portion of the literature heavily relies on analyses of homogenized tissue samples, which disregards the unique attributes of distinct cell types.

Recognizing this gap in knowledge, the Biomarkers Working Group has dedicated substantial effort to meticulously reviewing the available literature to comprehensively collate and summarize the current data for senescence-associated markers within specific tissues and cell types in vivo. We focused on data derived from methodologies that offer single-cell resolution and take into consideration cell identity. Additionally, our selection criteria focused on publications that evaluated multiple senescence-associated markers. To our knowledge, the compendium of information presented in this Expert Recommendation represents an unprecedented endeavour in terms of its comprehensive evaluation of existing data. This effort enabled us to formulate precise and targeted recommendations, which we refer to as SenNet guidelines for detecting senescent cells across tissues (SenGuiDe), with the hope of substantially advancing the field.

In this Expert Recommendation, we first introduce the nine hallmarks of cellular senescence. We then explore our current knowledge of senescence markers in 14 tissues in normal ageing and disease, and discuss the challenges in identifying senescent cells in tissues. Finally, we provide recommendations for the use of senescence markers, artificial intelligence and circulating markers and techniques to identify and characterize senescent cells.

Hallmarks of cellular senescence

Senescent cells are characterized by lack of proliferation and the upregulation of anti-apoptotic pathways. They secrete a plethora of chemokines, cytokines, growth factors and proteases referred to as the senescence-associated secretory phenotype (SASP). Cellular senescence is further characterized by changes in chromatin architecture, metabolic adaptations, the accumulation of damaged macromolecules and genomic instability. It is important to note that the presence of individual hallmarks is not sufficient to define the senescent state, as they can also be found in other cell states. Therefore, the presence of multiple hallmarks is required to identify senescent cells. In this section, we introduce the nine known major hallmarks of senescence (Fig. 1).

Fig. 1 |. The hallmarks of cellular senescence.

The nine hallmarks of senescence, which provide a foundation for the identification of senescent cells within tissues. 1, Cell cycle arrest: upregulation of p53, p16^INK4A (encoded by CDKN2A) and p21^CIP1 (encoded by CDKN1A) results in persistent activation of RB family proteins, inhibition of E2F transactivation (not shown) and cell cycle arrest. 2, Activation of the DNA damage response, including phosphorylation of the histone variant H2AX (γH2AX) and focal assembly of checkpoint and DNA repair factors (for example, T53BP1) at sites of DNA damage across the genome (orange circles). 3, Nuclear changes, which include global loss and local gain of heterochromatin, known as senescence-associated heterochromatic foci (SAHF) and characterized by gain of histone H3 trimethylated at Lys9 (H3K9me3) and H3K27me3, activation of transposable elements such as long interspersed element-1 (LINE-1), senescence-associated distension of satellites (SADS), telomere-associated foci (TAF) formation, and loss of lamin B1 (LMNB1) and lamina-associated domains. 4, Upregulation of anti-apoptotic factors such as ABL1, EFNB1, BCL2L1, EFNB3, SERPINB2 and PI3K. 5, The senescence-associated secretory phenotype (SASP) refers to the release of cytokines, chemokines and growth factors, together comprising an inflammatory response that influences the cellular microenvironment. Some protein secretion happens through extracellular vesicles. A subset of mitochondria in senescent cells show mitochondrial outer membrane permeability, and release of mitochondrial DNA (mtDNA) into the cytoplasm, which contributes to the SASP. 6, Increased lysosomal content and elevated activity of β-galactosidase (β-gal). 7, Metabolic adaptations are caused by changes in the function of mitochondria, which increase the production of reactive oxygen species (ROS). 8, Cell surface markers: elevated expression of senescence-associated protein and glycan receptors. 9, Changes in morphology: senescent cells become flat (in cell culture), enlarged and vacuolized (not shown). They have enlarged nuclei and undergo nuclear blebbing (not shown), leading to the formation of cytoplasmic chromatin fragments (CCFs), which contribute to the SASP. DDR, DNA damage response; ER, endoplasmic reticulum.

Cell cycle arrest

Many senescent cells accumulate the CDK2 inhibitor CDKN1A (also known as p21^CIP1) and the CDK4/6 inhibitor CDKN2A (also known as p16^INK4A), resulting in the persistent hypo-phosphorylation of RB family proteins, inhibition of E2F transcription factors, downregulation of the proliferation marker MKI67 and, subsequently, cell cycle arrest. This arrest is enforced by various factors, including by chromatin changes at E2F target genes¹² and ribosome biogenesis defects¹³. However, there is currently no specific marker exclusively indicative of the senescent cell cycle arrest. Activation of RB and p53 (encoded by TP53 in humans, Trp53 in mice) also occurs in other forms of cell cycle arrest¹⁴, and CDKN2A is expressed in certain non-senescent cells¹⁵ and is not expressed in all senescent cells¹⁶. Therefore, detecting senescence-associated cell cycle arrest requires quantifying multiple factors and features¹⁷.

The SASP

Senescent cells secrete a wide array of pro-inflammatory cytokines, chemokines and interleukins such as IL-6, IL-1α and IL-1β, along with cell-growth modulators, proteases and their inhibitors, angiogenic factors and insoluble factors such as fibronectin, collagens and laminin, and other inflammatory molecules such as GDF15, TGFβ1 and IFNγ, collectively termed the SASP. Evidence of protein secretion through exosomes is also emerging^18–20. The composition and strength of the SASP can vary depending on the duration of senescence, the origin of the senescence stimulus and specific cell types^17,21. Single-cell RNA sequencing (scRNA-seq) reveals considerable cell to cell variability in the expression of SASP factors²². In later stages of senescence, an interferon type 1 response is triggered, partly facilitated by the de-repression of long interspersed element-1 (LINE-1) retrotransposons²³. Senescence can be transmitted to neighbouring cells, a phenomenon commonly known as secondary senescence. Two mechanisms of transmission have been described: paracrine senescence is mediated by the SASP²⁴, whereas juxtracrine senescence, involving direct cell to cell contact, is mediated by NOTCH1–JAG1 signalling^25,26.

Nuclear changes

Nuclei of senescent cells undergo substantial changes with crucial functional implications. Post-translational modifications of histones and DNA methylation changes contribute to a global loss of heterochromatin and to focal gains of heterochromatin during senescence^27,28, the activation of LINE-1 (ref. 28) and the appearance of senescence-associated distension of satellites (SADS)²⁹. Lamin B1 (LMNB1) depletion leads to loss of lamina-associated domains and changes in gene expression³⁰. Oncogene-induced senescence is characterized by the formation of senescence-associated heterochromatic foci (SAHF), which are visible DAPI-dense foci, which are organized into a core of chromatin modified by histone H3 trimethylated at Lys9 (H3K9me3) and a shell of chromatin modified by H3K27me3 (ref. 31). This phenotype can be observed in other forms of senescence in cell culture such as replicative senescence, although to a lesser degree. PML nuclear bodies have been reported to mediate senescence through the regulation of E2F factors³². HMGB1, a protein belonging to the high mobility group family of proteins, which are involved in the regulation of chromatin structure and gene expression, is released from the nucleus into the extracellular environment during cellular senescence. Its extracellular presence serves as a signal of cellular stress or damage for immune cells³³.

Cell surface markers

Several proteins have been investigated as potential markers for identifying senescent cells both in cell culture and in tissues using an inducible CDKN2A or CDKN1A expression system (DEP1, NTAL, EBP50, STX4, VAMP3, ARMCX3, B2MG, LANCL1, PLD3 and VPS26A)³⁴. Additionally, DPP4, TNFRSF10D (also known as CD264), NOTCH1, NOTCH3, CD36, oxidized vimentin, ICAM1 and uPAR have also been used as markers of senescence³⁵, as well as CD274 (refs. 10,36), PDL2 and MHC class I molecules^37,38.

Changes in cell morphology

In cell culture, senescent cells undergo distinct morphological changes: they become flat, enlarged and vacuolized, and may display multiple or enlarged nuclei. Senescent cells establish cytoplasmic bridges that allow them to signal to neighbouring cells through direct intercellular protein transfer³⁹. Senescent cells in culture have been observed to display enlarged nuclei, nuclear blebbing and the formation of cytoplasmic chromatin fragments (CCFs). CCFs are generated by blebbing of chromatin through the nuclear membrane mediated by the compromised integrity of the nuclear lamina⁴⁰, and trigger an interferon response via the cGAS–STING pathway⁴¹.

Increased lysosomal content

In senescent cells, lysosomes increase in both number and size. This increase primarily comprises lysosomes containing indigestible autofluorescent materials^42,43, such as lipofuscin, which is an aggregate of oxidized proteins, lipids and metals that accumulates and can be detected by the histochemical stain Sudan Black B⁴³. The increased lysosomal mass has been linked to senescence-associated β-galactosidase (SA-β-gal) activity.

Metabolic adaptations

Senescent cells are characterized by heightened energy-producing oxidative metabolism surpassing that of proliferating cells, which sustains essential ATP-consuming proteo-stress pathways that are vital for their survival^44–46. Many drivers of mitochondrial dysfunction can also result in cellular senescence. These drivers include mitochondrial DNA (mtDNA) depletion and mutations, inhibitors of the electron transport chain, loss of HSPA9, loss of SIRT3 and SIRT5, and disruptions in complex I assembly⁴⁷. Mitochondrial dysfunction-associated senescence is primarily driven by the accumulation of cytosolic NADH, leading to ATP depletion and AMPK activation and cell cycle arrest²¹.

Senescent cells are characterized also by increased production of reactive oxygen species (ROS), primarily produced by dysfunctional mitochondria⁴⁸. Both intracellular and extracellular ROS can contribute to the induction of senescence and to stabilizing cell cycle arrest⁴⁹. A subset of mitochondria in senescent cells exhibit mitochondrial outer membrane permeability, by which mtDNA is released into the cytoplasm, which contributes to the SASP⁵⁰.

The DNA damage response

In response to DNA damage, cells activate the DNA damage response (DDR). The DDR triggers cellular senescence through ATM-mediated and ATR-mediated stabilization of p53, which activates CDKN1A to halt cell cycle progression. A persistent DDR also triggers the SASP⁵¹. Phosphorylation of Ser139 of histone H2AX (known as γH2AX) at sites of DNA damage facilitates the assembly of foci of checkpoint factors and DNA repair factors, including TP53BP1, MDC1, NBS1 and phosphorylated SMC1; phosphorylation of RAD17 can also be observed in senescent fibroblasts⁵². During senescence, telomeric DNA damage and attrition have been shown to result in the formation of DDR foci (for example, γH2AX and TP53BP1) that co-localize with telomeres, referred to as telomere-associated foci (TAF)^53,54.

Upregulation of anti-apoptotic pathways

Senescent cells typically exhibit the upregulation of proteins involved in anti-apoptotic pathways^50,55–58. The targeting of survival pathway factors (ABL1, EFNB1, BCL2L1, EFNB3, SERPINB2 and PI3K) by RNAi was shown to reduce the viability of senescent cells to a greater extent than non-senescent human cells⁵⁵.

Senescence markers in tissues

In this section, we discuss our current understanding of senescence markers in mouse and human tissues in the context of both normal ageing and ageing-related diseases. We focus on studies that have leveraged multiple hallmarks to support the presence of cellular senescence. A database of markers categorized by tissue can be found in Supplementary Table 1. The database contains information from 224 peer-reviewed papers and 328 markers, covering a total of 14 mouse and human tissues, along with circulating markers. Because a recurring question is how to prioritize markers in senescence studies, we attempt to provide recommendations based on the level of evidence in the collective body of work comprising our database.

Table 1 features senescence markers documented in our database in three or more tissues. Collectively, these markers represent six of the nine senescence hallmarks, with the SASP being most represented. Within the SASP, IL-6, TNF, IL-1α, IL-1β, SERPINE1 (also known as PAI-1) and TGFβ1 have been reported in more than ten tissues. Notably, CDKN2A and CDKN1A emerge as the predominantly reported markers for cell cycle arrest. H2AX is the most utilized marker for assessing the DDR across tissues, followed by telomere length and TAF. SA-β-gal staining marks increased lysosomal content in 11 tissues, and nuclear changes evidenced by the loss of LMNB1 and nuclear HMGB1 have been documented in 5 and 7 tissues, respectively. Upregulation of the anti-apoptotic factor BCL2 was reported in four tissues.

Table 1 |.

Cellular senescence markers described in three or more tissues

Marker	Full name	HGNC ID	Tissue	Senescence hallmark	Identification technique(s)
CDKN2A (p16INK4A)	Cyclin-dependent kinase inhibitor 2A	HGNC:1787	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, immune system, kidneys, liver, lungs, mammary gland, ovary, pancreas, placenta, skeleton, skin	Cell cycle arrest	scSeq, immunoblot, mouse model, RT–qPCR, scRNA-seq, CyTOF, WB, IHC, IF, snRNA- seq, FISH, RNA-ISH, immunostaining, gene knockout, RNA-seq, bulk RNA-seq
CDKN1A (p21CIP1)	Cyclin-dependent kinase inhibitor 1A	HGNC:1784	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, immune system, kidneys, liver, lungs, ovary, pancreas, placenta, skeleton, skin	Cell cycle arrest	scSeq, immunoblot, RT–qPCR, scRNA-seq, CyTOF, WB, IHC, IF, snRNA-seq, immunoassay (Luminex), bulk RNA-seq, knockout mice functional studies
IL-6	Interleukin-6	HGNC:6018	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, liver, lungs, mammary gland, ovary, pancreas, placenta, skeleton, skin	SASP	Immunoblot, RT–qPCR, scRNA-seq, CyTOF, WB, LC-MS/MS, microarray, Olink PEA, immunoassay (ELISA, Luminex), IF, multiplex immunoassay, IHC, Luminex mouse 23 cytokine array, mouse cytokine array (antibody-based), bulk RNA-seq, antibody arrays
TNF	Tumour necrosis factor	HGNC:11892	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, liver, lungs, pancreas, placenta, skeleton, skin	SASP	Immunoassay, RT–qPCR, CyTOF, scRNA-seq, cytokine array, immunoassay (ELISA, Luminex), RNAscope-ISH, microarrays, flow cytometry, EIA, mouse cytokine array (antibody-based), bulk RNA-seq, antibody arrays
H2AX	H2A.X variant histone	HGNC:4739	Adipose tissue, cardiovascular system, central nervous system, kidneys, liver, lungs, ovary, pancreas, placenta, skeleton, skin	DDR	IF, IHC, WB, RT–qPCR
IL-1α	Interleukin-1α	HGNC:5991	Bone marrow, cardiovascular system, central nervous system, circulating markers, liver, lungs, ovary, pancreas, placenta, skeleton, skin	SASP	RT–qPCR, CyTOF, snRNA-seq, scRNA-seq, cytokine array, immunoassay (ELISA, Luminex), flow cytometry, mouse cytokine array (antibody-based), bulk RNA-seq, antibody arrays
IL-1β	Interleukin-1β	HGNC:5992	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, lungs, ovary, placenta, skeleton, skin	SASP	RT–qPCR, CyTOF, microarray, WB, scRNA-seq, cytokine array, bulk RNA-seq, immunoassay (ELISA), IHC, IF, multiplex immunoassay, antibody arrays
SA-β-gal	Senescence-associated β-galactosidase		Adipose tissue, bone marrow, cardiovascular system, central nervous system, immune system, kidneys, liver, ovary, pancreas, placenta, skin	Increased lysosome content	Colorimetric and fluorescent staining and microscopy, IHC, IF, X-Gal crystal detection via electron microscopy, flow cytometry, SA-β-gal staining, X-gal precipitation
SERPINE (PAI-1)	Serpin family E member 1	HGNC:8583	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, liver, lungs, ovary, pancreas, skeleton	SASP	RT–qPCR, scRNA-seq, Olink PEA, aptamer- based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), IHC, IF, immunoblotting, bulk RNA-seq, CyTOF
TGFβ1	Transforming growth factor-β1	HGNC:11766	Bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, liver, lungs, ovary, pancreas, skeleton	SASP	RT–qPCR, scRNA-seq, WB, microarray, aptamer-based proteomics (SOMAscan assay), IF, RNAscope-ISH, RNA-seq, flow cytometry
p53	Tumour protein p53	HGNC:11998	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, liver, lungs, pancreas, placenta, skeleton	Cell cycle arrest	CyTOF, RNA-seq, WB, scRNA-seq, IF, aptamer- based proteomics (SOMAscan assay), microarray, immunoblots, RT–qPCR, knockout mice functional studies, IHC
CCL2	C–C motif chemokine ligand 2	HGNC:10618	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, liver, ovary, skeleton	SASP	Immunoassay, RT–qPCR, RNA-seq, WB, microarray, snRNA-seq, scRNA-seq, bulk RNA-seq, immunoassay (ELISA, Luminex)
CCL5	C–C motif chemokine ligand 5	HGNC:10632	Bone marrow, central nervous system, circulating markers, kidneys, liver, mammary gland, ovary, pancreas, skeleton	SASP	scRNA-seq, immunoassay (ELISA, Luminex), IHC, RT–qPCR, flow cytometry, bulk RNA-seq
CXCL1	C–X–C motif chemokine ligand 1	HGNC:4602	Bone marrow, cardiovascular system, central nervous system, circulating markers, liver, mammary gland, ovary, pancreas, skeleton	SASP	CyTOF, snRNA-seq, cytokine array, RNA-ISH, immunoassay (Luminex), RT–qPCR, flow cytometry, bulk RNA-seq, scRNA-seq
CXCL8	C–X–C motif chemokine ligand 8	HGNC:6025	Adipose tissue, central nervous system, circulating markers, lungs, ovary, placenta, skeleton, skin	SASP	Immunoassay, Olink PEA, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), multiplex immunoassay, antibody array, bulk RNA-seq, scRNA-seq, RT–qPCR
MMP12	Matrix metallopeptidase 12	HGNC:7158	Adipose tissue, bone marrow, cardiovascular system, central nervous system, circulating markers, kidneys, lungs, skeleton	SASP	Immunoassay, RT–qPCR, scRNA-seq, bulk RNA-seq, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA), RNA-seq, immunoblotting
CCL3	C–C motif chemokine ligand 3	HGNC:10627	Bone marrow, central nervous system, circulating markers, kidneys, ovary, pancreas, skeleton	SASP	scRNA-seq, aptamer-based proteomics (SOMAscan assay), RT–qPCR, flow cytometry, bulk RNA-seq
HMGB1	High mobility group box 1	HGNC:4983	Bone marrow, central nervous system, ovary, pancreas, placenta, skeleton, skin	Nuclear changes	RT–qPCR, scRNA-seq, IHC, bulk RNA-seq, IF, WB, ELISA
MMP2	Matrix metallopeptidase 2	HGNC:7166	Cardiovascular system, circulating markers, liver, pancreas, placenta, skeleton, skin	SASP	WB, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), RT–qPCR, IHC, bulk RNA-seq, scRNA-seq
CCL8	C–C motif chemokine ligand 8	HGNC:10635	Bone marrow, cardiovascular system, circulating markers, kidneys, ovary, skeleton	SASP	RT–qPCR, RNA-seq, microarray, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
CXCL10	C–X–C motif chemokine ligand 10	HGNC:10637	Cardiovascular system, circulating markers, ovary, pancreas, skeleton, skin	SASP	LC-MS/MS, immunoassay (ELISA, Luminex), RT–qPCR, flow cytometry, bulk RNA-seq, scRNA-seq
CXCL2	C–X–C motif chemokine ligand 2	HGNC:4603	Bone marrow, cardiovascular system, liver, mammary gland, ovary, skeleton	SASP	RT–qPCR, bulk RNA-seq, scRNA-seq
GDF15	Growth differentiation factor 15	HGNC:30142	Adipose tissue, cardiovascular system, circulating markers, lungs, placenta, skeleton	SASP	RNA-seq, RT–qPCR, RNA-seq, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), knockout studies, proteomic analyses, IF, IHC, bulk RNA-seq, scRNA-seq
IGFBP3	Insulin-like growth factor binding protein 3	HGNC:5472	Bone marrow, circulating markers, kidneys, ovary, pancreas, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), RNA-seq, IF, WB, Luminex assay, bulk RNA-seq, scRNA-seq
MMP3	Matrix metallopeptidase 3	HGNC:7173	Adipose tissue, cardiovascular system, central nervous system, liver, skeleton, skin	SASP	RT–qPCR, immunoassay, microarray, RNA-seq, bulk RNA-seq, scRNA-seq, antibody arrays, ELISA, IHC
MMP9	Matrix metallopeptidase 9	HGNC:7176	Bone marrow, cardiovascular system, circulating markers, placenta, skeleton, skin	SASP	RT–qPCR, RNA-seq, immunoassay (ELISA, Luminex), antibody array, bulk RNA-seq, scRNA-seq, antibody arrays
CSF2	Colony-stimulating factor 2	HGNC:2434	Bone marrow, cardiovascular system, circulating markers, placenta, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA), antibody array, bulk RNA-seq, scRNA-seq
IGFBP7	Insulin-like growth factor binding protein 7	HGNC:5476	Bone marrow, cardiovascular system, circulating markers, ovary, skeleton	SASP	RT–qPCR, aptamer-based plasma profiling, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), bulk RNA-seq, scRNA-seq
LMNB1	Lamin B1	HGNC:6637	Adipose tissue, bone marrow, central nervous system, placenta, skin	Nuclear changes	IF, RT–qPCR, snRNA-seq, WB
MMP1	Matrix metallopeptidase 1	HGNC:7155	Circulating markers, kidneys, liver, skeleton, skin	SASP	Aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), RT–qPCR, bulk RNA-seq, scRNA-seq, ANTIBODY ARRAYS
BCL2	BCL2 apoptosis regulator	HGNC:990	Bone marrow, central nervous system, ovary, placenta	Upregulation of anti-apoptotic pathways	CyTOF, RT–qPCR, snRNA-seq, bulk RNA-seq, WB, IHC
CCL4	C–C motif chemokine ligand 4	HGNC:10630	Central nervous system, circulating markers, pancreas, skeleton	SASP	scRNA-seq, immunoassay (Luminex), RT–qPCR, flow cytometry, bulk RNA-seq
CCL7	C–C motif chemokine ligand 7	HGNC:10634	Bone marrow, cardiovascular system, circulating markers, skeleton	SASP	RT–qPCR, RNA-seq, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
CDKN2B (p15INK4B)	Cyclin-dependent kinase inhibitor 2B	HGNC:1788	Cardiovascular system, central nervous system, pancreas, skeleton	Cell cycle arrest	RT–qPCR, FISH, IHC, SABiosciences Cellular Senescence RT2 Profiler PCR Array
CSF1	Colony-stimulating factor 1	HGNC:2432	Bone marrow, central nervous system, mammary gland, skeleton	SASP	RT–qPCR, scRNA-seq, snRNA-seq, bulk RNA-seq
CXCL12	C–X–C motif chemokine ligand 12	HGNC:10672	Cardiovascular system, circulating markers, pancreas, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
FAS	Fas cell surface death receptor	HGNC:11920	Bone marrow, circulating markers, pancreas, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), immunoassay (Luminex), bulk RNA-seq, scRNA-seq
ICAM1	Intercellular adhesion molecule 1	HGNC:5344	Bone marrow, circulating markers, pancreas, skeleton	SASP	RT–qPCR, immunoassay (ELISA, Luminex), bulk RNA-seq, scRNA-seq
IFNγ	Interferon-γ	HGNC:5438	Bone marrow, circulating markers, placenta, skeleton	SASP	RT–qPCR, immunoassay (Luminex), ELISA
IGF1	Insulin-like growth factor 1	HGNC:5464	Cardiovascular system, central nervous system, circulating markers, skeleton	SASP	RT–qPCR, scRNA-seq, snRNA-seq, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq
IGFBP4	Insulin-like growth factor binding protein 4	HGNC:5473	Bone marrow, circulating markers, ovary, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
IL-17A	Interleukin-17A	HGNC:5981	Bone marrow, circulating markers, placenta, skeleton	SASP	scRNA-seq, immunoassay (Luminex), bulk RNA-seq, qPCR
MAPK14	Mitogen-activated protein kinase 14	HGNC:6876	Immune system, kidneys, pancreas, placenta	SASP	Microarray, WB, RT–qPCR, IF, bulk RNA-seq
MMP13	Matrix metallopeptidase 13	HGNC:7159	Bone marrow, cardiovascular system, kidneys, skeleton	SASP	RT–qPCR, RNA-seq, bulk RNA-seq, scRNA-seq
NFKB1	Nuclear factor-κB subunit 1	HGNC:7794	Bone marrow, pancreas, placenta, skeleton	SASP	CyTOF, RT–qPCR, IHC
Telomere length	Telomere length		Cardiovascular system, kidneys, placenta, skin	DDR	FISH, TRAP assay, RT–qPCR
TIMP2	TIMP metallopeptidase inhibitor 2	HGNC:11821	Central nervous system, circulating markers, ovary, skeleton	SASP	snRNA-seq, scRNA-seq, aptamer-based proteomics (SOMAscan assay), immunoassay (Luminex), bulk RNA-seq
TNFRSF1B	TNF receptor superfamily member 1B	HGNC:11917	Circulating markers, lungs, pancreas, skeleton	SASP	Immunoassay (ELISA, Luminex), RT–qPCR, bulk RNA-seq, scRNA-seq
VEGFA	Vascular endothelial growth factor A	HGNC:12680	Bone marrow, circulating markers, lungs, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), cytometric bead array, bulk RNA-seq, scRNA-seq
CSF3	Colony-stimulating factor 3	HGNC:2438	Bone marrow, mammary gland, skeleton	SASP	RT–qPCR, scRNA-seq
CXCL16	C–X–C motif chemokine ligand 16	HGNC:16642	Circulating markers, mammary gland, skeleton	SASP	Aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq, RT–qPCR
HGF	Hepatocyte growth factor	HGNC:4893	Cardiovascular system, circulating markers, skeleton	SASP	RT–qPCR, microarray, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
ICAM3	Intercellular adhesion molecule 3	HGNC:5346	Circulating markers, pancreas, skeleton	SASP	Aptamer-based proteomics (SOMAscan assay), RT–qPCR, bulk RNA-seq, scRNA-seq
IGFBP2	Insulin-like growth factor binding protein 2	HGNC:5471	Circulating markers, ovary, skeleton	SASP	Aptamer-based proteomics (SOMAscan assay), immunoassay (ELISA, Luminex), bulk RNA-seq, scRNA-seq
IGFBP5	Insulin-like growth factor binding protein 5	HGNC:5474	Cardiovascular system, circulating markers, skeleton	SASP	RT–qPCR, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
IL-7	Interleukin-7	HGNC:6023	Cardiovascular system, circulating markers, skeleton	SASP	RT–qPCR, immunoassay (Luminex), bulk RNA-seq, scRNA-seq
INHBA	Inhibin subunit-βA	HGNC:6066	Bone marrow, circulating markers, skeleton	SASP	RT–qPCR, immunoassay (ELISA, Luminex)
Lipofuscin	Lipofuscin		Ovary, placenta, skin	Increased lysosomal content	Sudan Black B staining
MKI67	Marker of proliferation Ki-67	HGNC:7107	Bone marrow, cardiovascular system, skin	Cell cycle arrest	CyTOF, IF
MMP10	Matrix metallopeptidase 10	HGNC:7156	Cardiovascular system, circulating markers, skeleton	SASP	RNA-seq, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq, scRNA-seq
PLAUR	Plasminogen activator, urokinase receptor	HGNC:9053	Central nervous system, circulating markers, skeleton	SASP	scRNA-seq, snRNA-seq, aptamer-based proteomics (SOMAscan assay), bulk RNA-seq
SPP1	Secreted phosphoprotein 1	HGNC:11255	Central nervous system, circulating markers, skeleton	SASP	scRNA-seq, bulk RNA-seq, immunoassay (ELISA)
TAF	Telomere-associated foci		Cardiovascular system, liver, skin	DDR	ImmunoFISH
TNFRSF1A	TNF receptor superfamily member 1A	HGNC:11916	Circulating markers, pancreas, skeleton	SASP	Immunoassay (ELISA, Luminex), RT–qPCR, bulk RNA-seq, scRNA-seq
TP53BP1	Tumour protein p53 binding protein 1	HGNC:11999	Cardiovascular system, kidneys, skin	DDR	IF, IHC, IF foci

This table was obtained by selecting a subset of markers from the senescence markers database (Supplemental Table 1) that were described in three or more tissues. Markers are ranked by the number of tissues in which they were reported. The columns correspond to the marker acronym or symbol, its full name or preferred term, the marker’s HGNC ID, the tissues in which the marker was described, the senescence hallmark to which the marker corresponds and the techniques used to detect the marker. CyTOF, cytometry by time of flight; DDR, DNA damage response; EIA, enzyme immunoassay; ELISA, enzyme-linked immunosorbent assay; FISH, fluorescence in situ hybridization; IF, immunofluorescence; IHC, immunohistochemistry; LC–MS/MS, liquid chromatography with tandem mass spectrometry; Olink PEA, Olink proximity extension assay; RNA-ISH, RNA in situ hybridization; RNAscope-ISH, RNAscope in situ hybridization; RNA-seq, RNA sequencing; RT–qPCR, quantitative PCR with reverse transcription; SASP, senescence-associated secretory phenotype; scRNA-seq, single-cell RNA sequencing; scSeq, single-cell sequencing; snRNA-seq, single-nucleus RNA sequencing; TRAP, translating ribosome affinity purification; WB, western blotting.

Beyond marker prioritization, we do not recommend the use of individual markers to determine the senescence state of cells because ample evidence exists of so-called false-positive senescence⁵⁹ (Box 1). Our recommendation is to require the presence of multiple hallmarks of senescence in the same cell, as described in Box 2. We further discuss the main findings reported on the presence of senescence markers in each tissue in ageing and disease.

Box 1. Key challenges in identifying and characterizing senescent cells in tissues in vivo.

Here we discuss the major challenges in identifying senescent cells in tissues, characterizing their phenotypes and distinguishing cellular senescence from other cell states.

• False-positive senescence detection. A challenge in using markers to differentiate senescent from non-senescent cells is that ample evidence exists for the presence of senescence markers in non-senescent cells. For example, polarization of macrophages towards M2 phenotypes results in the expression of CDKN2A (also known as p16^INK4A) and senescence-associated β-galactosidase (SA-β-gal)⁷⁸, two of the most commonly used senescence markers. To avoid this pitfall, it is crucial to use multi-pronged detection methodologies (Box 2).

• Immune-cell senescence. Cellular senescence has now been described in nearly all tissues, including the immune system. Nonetheless, the long-standing use of the term ‘immunosenescence’ to describe a state of cell exhaustion and immune dysregulation observed with ageing, as opposed to cellular senescence involving cell cycle arrest, has created some confusion. Although these two processes are at some levels related and both appear with ageing, at other levels they are distinct. Distinguishing immune-cell markers from senescence markers is challenging, requiring further studies to identify differential features. Such studies will be essential for the validation and quantification of senescence burden in immune cells.

• Senescence in postmitotic cells. Numerous senescence markers have been identified in postmitotic cells such as neurons⁹⁹, cardiomyocytes⁶⁰, osteocytes⁹³, adipocytes²⁷⁰ and others, and their levels in these cells have been observed to increase with age. A long-standing debate in the field revolves around whether these cells can genuinely be considered senescent. Evidence suggests that inducing senescence in vivo can result in phenotypes similar to those seen in the induction of senescence in proliferation-competent cells. Moreover, the removal of postmitotic cells bearing senescence markers has been associated with improved health outcomes in mice^102,206. We argue that, given the consistent evidence of age-dependent increases in cell cycle inhibitors (for example, CDKN1A and CDKN2A), DNA damage foci, senescence-associated secretory phenotype (SASP) factors and other senescence markers in postmitotic cells²⁷¹, postmitotic cells should be considered as senescent cells.

Box 2. Recommendations for the use of senescence markers.

Many cell senescence-associated markers have been described, primarily in cell culture studies, but individually they lack specificity and sensitivity to identify senescent cells in tissues. Specific senescence hallmarks (Fig. 1) can translate into the appearance of different sets of markers in different cell types or forms of senescence. Many senescence-associated markers are also present in non-senescent cells (Box 1). We propose a multi-pronged methodology to identify senescent cells in tissues, which requires evidence of multiple senescence hallmarks in the same cell. We recommend that at least three hallmarks be probed in tissues to identify senescent cells. Drawing from the reviewed literature, we suggest the following hallmarks and provide a list of common markers that could be given priority (markers denoted with an asterisk are the most frequently reported, according to our literature review). We anticipate that this methodology will be especially beneficial when employing low-plex techniques that impose restrictions on the number of markers that can be simultaneously assayed in individual cells. Nevertheless, to comprehensively capture the heterogeneity of senescent cells, especially of the senescence-associated secretory phenotype (SASP), high-plex methods with hundreds to thousands of genes should be used (Box 4).

• Cell cycle inhibition

  • Upregulation of CDKN2A*
  • Upregulation of CDKN1A*
  • Downregulation of MKI67
• The DNA damage response (DDR)

  • Formation of γH2AX nuclear foci*
  • Formation of TP53BP1 nuclear foci
  • Formation of telomere-associated foci (TAF)
• The SASP

  • IL-6*
  • IL-1α
  • IL-1β
  • SERPINE1

Note: SASP composition is tissue-dependent and cell type-dependent. The absence of expression of the most common SASP genes in a cell does not preclude that another, cell type-specific SASP could be expressed. Hence, a lack of these factors should not be used as evidence against a cell being senescent. To truly capture the SASP comprehensively, high-throughput whole transcriptome methodologies such as single-cell transcriptomics and proteomics should be utilized.

• Increased lysosomal content

  • Senescence-associated β-galactosidase (SA-β-gal) activity*
• Nuclear reorganization

  • Nuclear exclusion and subsequent release of HMGB1*
  • Loss of lamin B1 (LMNB1)
  • Senescence-associated distension of satellites (SADS)
• Upregulation of anti-apoptotic pathways

  • Upregulation of BCL2*
  • Upregulation of other BCL2 family proteins

The cardiovascular system

Multiple studies have shown the accumulation of senescence-associated markers in various cardiovascular cell lineages, such as cardiomyocytes, cardiac mesenchymal stromal cells, cardiac fibroblasts and cardiac progenitor cells, both in ageing and in cardiovascular disease contexts.

During normal ageing, mouse cardiomyocytes accumulate the following senescence-associated markers: TAF, cyclin-dependent kinase inhibitors (CDKN1a, CDKN2B (also known as p15^INK4b) and CDKN2a), SA-β-gal activity, SADS and a pro-fibrotic SASP⁶⁰. In mice, aged cardiac mesenchymal stromal cells identified as PTPRC⁻PECAM1⁻ATXN1⁺PDGFRA⁺ (also knownas CD45⁻CD31⁻SCA1⁺PDGFRα⁺) experience a decline in clonogenic and proliferative capacity linked to increase mRNA expression of Cdkn1a, Cdkn2b and Cdkn2a and SASP factors⁶¹. In humans, cardiac progenitor cells defined as KIT⁺PECAM1⁻PTPRC⁻ (also known as c-kit⁺CD31⁻CD45⁻) showed increased levels of CDKN2A, SA-β-gal activity, γH2AX foci, reduced BrdU incorporation into DNA (S phase marker; reduced incorporation indicates cell cycle arrest) and induction of the SASP during ageing⁶². Vascular ageing is characterized by arterial thickening and stiffness and endothelial dysfunction. Multiple senescence-associated markers, including cell cycle inhibitors (Cdkn2a, Cdkn1a and Trp53), DNA damage foci and SASP factors, are upregulated in the aorta of ageing mice^63–65. Both endothelial cells and vascular smooth muscle cells, which are the primary cell types of the aorta, exhibit senescence-associated markers during mouse ageing^63,66.

Myocardial infarction in mice was associated with increased expression of senescence-associated markers such as CDKN2A and CDKN1A, SA-β-gal activity, TAF and increased expression of SASP components in mouse cardiomyocytes⁶⁷ and cardiac fibroblasts⁶⁸. Senescence-associated markers have also been identified in the context of vascular diseases such as atherosclerosis and abdominal aortic aneurysm and thoracic aortic aneurysm in both human and mouse models^69–72. In mouse models of atherosclerosis (Ldlr^–/–), advanced atherosclerotic lesions exhibit SA-β-gal expression and upregulation of Cdkn2a and canonical SASP components⁷¹. Moreover, in humans, vascular smooth muscle cells isolated from atherosclerotic plaques also express senescence-associated markers, including CDKN1A, CDKN2A and SA-β-gal activity⁷². Senescence-associated markers, including CDKN1A, p53, total γH2AX and γH2AX associated with telomeres (TAF), and SASP factors have also been observed in both human and mouse models of thoracic aortic aneurysm and abdominal aortic aneurysm^69,70.

The immune system

The immunosenescence noted in advanced age leads to increased susceptibility to infections and reduced responsiveness to vaccination⁷³. This phenotype may involve elements of cellular senescence, particularly in T cells⁷⁴, although different mechanisms may operate in specialized postmitotic immune cells. Studies have shown that immune cells, including monocytes and tissue-resident macrophages, are the predominant producers of SASP factors such as IL-1α, IL-1β, IL-6, TNF and HMGB1 (refs. 75–78). As features of the SASP overlap with those of immune cells, it is imperative to identify marker(s) that can distinguish between activated resident immune cells and bona fide senescent cells. This unmet need remains a significant gap in the field, posing a challenge to the study of senescence in the immune system.

Tissue-resident immune cells such as macrophages, type 2 innate lymphoid cells, natural killer cells, B cells and effector memory T cells depend on proliferation to maintain themselves⁷⁹. Each of these cell subsets shows age-related dysfunction and produces pro-inflammatory cytokines that overlap with SASP factors⁷⁹, thereby providing evidence for the impact of age and senescence on immune cells. Recent studies in platelets of older people have found increased levels of SELP (CD62P), CD40LG and CD63, suggesting the existence of an age-associated increase in platelet activation, which may be relevant in senescence⁸⁰. Subsets of CD8A⁺ T cells with high SA-β-gal activity, which also exhibit features of telomere dysfunction-mediated and CDKN2A-mediated senescence⁷⁴, and CD8A⁺ effector memory T cells display a SASP regulated by p38 MAPK⁸¹.

Transcription and translation measurements using bulk peripheral mononuclear blood cells have identified various cell cycle regulators, including CDKN2A, p53, CDKN1A, CDKN1B and p38 MAPK, and changes in some of these markers were linked to ageing and/or various chronic diseases^82,83. Analysis of the SenMayo gene panel (Box 3) indicates that approximately 50% of monocytes in the bone marrow of older subjects could be senescent⁸⁴. CD3⁺ T cells isolated from the blood of aged mice and of mice deficient in ERCC1, which is a crucial DNA repair protein, exhibit increased levels of multiple senescent markers, including Ccl2, Tnf, Il1b, Inhba, Spp1, Cdkn2a and Cdkn1a (refs. 85,86). By contrast, vaccine-activated mouse CD8A⁺ T cells that are transferred to new mice can retain functionality and proliferation capacity beyond the lifespan of the original mouse, suggesting that determinants of senescence did not transfer with the T cell and may derive from the new host⁸⁷. It is also important to note that as CDKN2A-expressing peripheral mononuclear blood cells do not appear to survive cryopreservation, quantitative PCR (qPCR) assays that quantify CDKN2A mRNA levels in peripheral T cells have mainly used CD3⁺ cells isolated from fresh blood to ensure representation of the senescent subset⁸⁸. Recently, it was reported that circulating senescent myeloid cells drive neurodegeneration and brain inflammation⁸⁹.

Box 3. Recommendations for the use of cellular senescence signatures and artificial intelligence to identify senescent cells.

Novel computational tools take advantage of the high-throughput nature of the single-cell transcriptomics data currently available to go beyond using a small set of individual markers to identify senescent cells. The senescent state of individual cells is evaluated by checking the transcriptional profile of sets of genes known to be altered in senescent cells. These signatures are constructed based on either a curated selection of genes from the existing literature or are empirically derived from the analysis of reference datasets^272–275. Two of the signatures most widely used by the National Institutes of Health (NIH) Cellular Senescence Network (SenNet) are SenMayo⁸⁴ and SenSig¹¹⁴.

The SenMayo senescence signature is a curated set of 125 human genes and 119 mouse genes enriched in senescent and/or senescence-associated secretory phenotype (SASP)-secreting cells across 15 studies⁸⁴. Notably, this list does not include commonly used markers such as CDKN1A (also known as p21^CIP1) and CDKN2A (p16^INK4A), which were used to validate the SenMayo signature in bone biopsies. Most of the genes in SenMayo are SASP genes. Given the heterogeneity of the SASP in different cell types and forms of senescence, the SenMayo gene set, curated from multiple studies, is likely to capture senescent phenotypes in different contexts. However, its bias towards SASP genes could also lead to higher rates of false positives in the presence of widespread inflammation; hence, it should be used with care in these contexts.

The SenSig senescence signature was derived from single-cell differential expression analysis between Cdkn2a⁺ and Cdkn2a^– cells isolated from a foreign body response-driven fibrosis model in a p16-CreER^T2;Ai14 reporter mouse. This analysis provided an empirically based set of 1,803 genes²⁷⁶. Transfer learning and senescence scoring were used to identify senescent cells in both mouse and human single-cell transcriptomics datasets. The SenSig gene set contains genes representing multiple senescent hallmarks, but also a large number of genes that might be senescence-specific in the foreign body response-driven fibrosis model but not in other settings. Nonetheless, SenSig could be adapted and extended to other contexts by applying the same transfer learning model when labelled data are available (for example, transcriptional profiles of CDKN2A⁺ and CDKN2A^– cells). However, it is important to note that, because of the many CDKN2A gene dropouts in most single-cell RNA sequencing (scRNA-seq) datasets (Box 5), we do not recommend using the CDKN2A counts to label the data, as the CDKN2A^– category would contain a large number of CDKN2A⁺ cells and negatively affect the transfer learning.

An alternative methodology has recently been developed to identify senescent cells in human brains, especially of individuals with Alzheimer disease¹¹⁴. Three gene sets were generated reflecting canonical senescence pathways, including cell cycle arrest (22 genes), genes upregulated in early senescence (48 genes) and the SASP (44 genes). Principal component analysis was used to compute a weighted average expression over all genes in each gene set, referred to as the gene set’s eigengene. Eigengene levels were then used to distinguish between senescent and non-senescent cells in single-cell transcriptomics data. Although this gene panel performed well in the Alzheimer disease study, its performance in comparison with SenMayo and SenSig in other contexts is not yet known.

A thorough comparison of the performance of these artificial intelligence methods is still lacking. Given the complementary nature of their formulations, we recommend applying and contrasting more than one method: a large discrepancy in the cells identified as senescent would indicate that one or more method might not be reliable in a given context and would warrant further investigation. It is worth noting that, although these methodologies have been used exclusively with single-cell transcriptomics data, they will be applicable to single-cell proteomics data as they become available.

Recent work has shown that emergent tools of machine learning and artificial intelligence in general can be used to identify senescent cells based on their distinct morphological features. By analysing 62 morphological features extracted from cells stained with DAPI or CellMask (a marker of the plasma membrane), it was possible to identify substantial changes in various phenotypic characteristics when comparing different types of induced senescent cells with their proliferating counterparts²⁷⁷. Unsupervised learning algorithms such as t-distributed stochastic neighbour embedding and uniform manifold approximation and projection, which are commonly used in the analysis of scRNA-seq datasets, were able to cluster cells based on their morphology and distinguish between different forms of senescence²⁷⁷.

Artificial intelligence models have shown promise in achieving high accuracy in classifying senescent cells. Convolutional neural networks trained and tested on phase-contrast images at single-cell resolution achieved a multiclass classification accuracy of 0.93 when distinguishing between hydrogen peroxide-induced, camptothecin-induced or repetitive passage-induced senescence and their respective controls²⁷⁸. DAPI-stained nuclei were used to identify senescent cells induced by ionizing radiation and replicative stress from proliferating controls using deep neural networks with up to 0.95 accuracy²⁷⁹. These morphology-based deep learning models have shown promise as sophisticated tools for screening for potential anti-senescence drugs²⁷⁸.

The primary barrier to adopting these tools is computational in nature, as they rely on basic microscopy to extract cell morphology. Because these methods require a ground truth dataset for their training, they have been mostly developed using cells in culture. It is not yet clear how these methods will perform in detecting senescent cells in different tissues, although encouraging results indicate they are applicable in tissues²⁷⁹. Hence, we recommend their use for initial screening, but more work is needed to extend their applicability to the level of cell heterogeneity that exists in tissues. These techniques are ideal for increasing the feature space in the latest-generation spatial multi-omics datasets, which provide single-cell morphological data (Box 5).

Bone marrow

The bone marrow has a dynamic tissue microenvironment that is dependent on the proliferation and differentiation of mesenchymal stem cells, haematopoietic stem cells and other cells. The physiological balance between the bone marrow’s vascular, mesenchymal and trabecular cell populations supports a proper haematopoietic stem cell niche, which is required for the production of different blood cell lineages^90,91. Studies have identified oncogene-induced, irradiation-induced and ageing-induced accumulation of senescent cells in the bone marrow^92–94. These cells have been found in different tissue compartments and have impaired functions compared with their normal counterparts, thereby contributing to disorders such as immune system decay, osteoporosis and cancer^{92,93,95–98}.

In a mouse model of acute myeloid leukaemia, qPCR of bone marrow cells showed upregulation of the senescence markers Cdkn2a, Cdkn1a and SASP factors, and downregulation of Lmnb1 (ref. 94). In an irradiation-induced senescence model, the percentage of cells positive for SA-β-gal activity, in addition to the markers Cdkn2a, Cdkn1a and several SASP factors, was upregulated⁹². (Senescence of mesenchymal cells in the bone marrow and bone is discussed below; see ‘Skeleton’.) Similar to mice, markers of bone marrow senescence have been found in samples from aged human. Needle biopsies containing tissue of the bone and bone marrow analysed by qPCR revealed increased levels of CDKN2A and CDKN1A, and the SASP increased with age⁹³. Single-cell transcriptomics showed an increase of CDKN1A, TGFB1 and SASP factors⁸⁴.

The central nervous system

Human and mouse postmitotic cells and progenitor cell types of the central nervous system, including neurons, microglia, astrocytes, oligodendrocyte lineage cells and endothelial cells, exhibit overexpression of combinations of senescence-related cell cycle regulators (CDKN2A, TP53 and CDKN1A) and SASP factors (CCL2, IL1A, IL1B, IL6, MMP1, MMP3, SERPINE1, SPP1 and TNF) in the context of ageing and disease^99–105. Increased levels of SA-β-gal activity and expression of apoptosis (Bcl2) and DNA damage (γH2AX and phosphorylated p38 MAPK) regulators, and the downregulation of LMNB1, are additional characteristics of age-associated or neurodegeneration-associated neuronal and glial cells^{102,105–108}. Mouse models of pharmacogenetic and pharmacologic clearance of senescent cells have been used to validate the modulation of these markers and corresponding aged or diseased phenotypes^{102–104,109,110}. Importantly, the complex heterogeneity of senescence in the ageing mouse and human brain warrants cell type-specific and region-specific characterization.

Senescence profiles of brain ageing are found also in tissues affected by neurodegenerative pathologies such as Alzheimer disease and Parkinson disease^{106,107,109–113}. Single-nuclei sequencing of brains of individuals with Alzheimer disease revealed markers of senescence coinciding with Tau pathology in excitatory neurons, and CDKN2D was computationally identified as the top senescence marker in Alzheimer disease¹¹⁴. Post-mortem Alzheimer disease tissues also show elevated levels of SERPINE1, CDKN2A, CDKN2D, CDKN1A, IL-1β, IL-6 and CXCL8

in endothelial cells and microglia relative to control brain tissues^112,115. Mouse models of Alzheimer disease and tauopathies display overlapping marker profiles in both neuronal and glial cells, and the disease-related phenotypes exhibit a responsiveness to senolytics^{107,109,111,112}.

Compared with brains of healthy individuals, the substantia nigra of individuals with Parkinson disease has reduced levels of SATB1 (a CDKN1A repressor) and increased levels of CDKN2A, MMP3, IL-6, IL-1α and CXCL8, suggesting an increased burden of senescence^106,116. Finally, mouse models of oxidative stress (including radiation) and metabolic and inflammatory stress (including obesity and alcoholism) recapitulate features of age-associated and disease-associated senescence in the mouse brain, as characterized by the induction of CDKN2A, CDKN2D, CDKN1A, p53, IL-1α, IL-1β, IL-6 and SA-β-gal activity^102,117,118.

Adipose tissue

Accumulation of senescent cells in adipose tissue is linked to tissue dysfunction in ageing and age-related diseases^119–122. Senescent cells in both human and mouse adipose tissue have been characterized by expression of CDKN2A and CDKN1A, higher SA-β-gal activity, DNA damage (γH2AX) foci and arrest of cell proliferation (MKI67), and by expression of SASP factors including IL-6, CXCL8 (also known as IL-8), IL-1β, TNF, CCL2, SERPINE1, MMP3 and MMP12 (refs. 119–122).

Adipose tissue-derived mesenchymal stem cells from older donors (human and mouse) exhibit senescence properties and impaired regenerative potential^123,124. A single-cell transcriptomics study identified a Cdkn1a^high senescent-like cell population from aged mouse adipose tissue-derived mesenchymal stem cells, with additional upregulation of Il6, Il15, Vegfa, Cxcl2, Ccl2 and Cxcl1 (ref. 124). Notably, there is more accumulation of Cdkn1a^high cells than of Cdkn2a^high cells in natural ageing and obesity. Importantly, Cdkn1a^high and Cdkn2a^high cells exist as two distinct cell populations^122,125, indicating the heterogeneity and complexity of senescent cell populations in adipose tissue.

Mature human adipocytes undergo senescence in obesity and following hyperinsulinaemia challenge, displaying a premature senescence transcriptome and secretory profile including upregulated gene expression of CDKN2A, CDKN1A and CCND1 and downregulated expression of HMGB1 and HMGB2 (ref. 126). Additionally, there is increased secretion of the SASP proteins CXCL8, SERPINE1, CXCL2 and MMP14. Upon ageing and obesity, SASP factors from the senescent cells upregulate CD38 in adipose immune cells, thereby decreasing NAD⁺ levels, which results in increased metabolic dysfunction and a decline in overall fitness¹²⁷. TNSF8⁺PDCD1⁺CD44^hiCD4⁺ (CD153⁺PDCD⁺CD44^hiCD4⁺) T cells accumulate in adipose tissue in obesity and show senescence features, including increased expression of SPP1, CDKN2A and CDKN1A and decreased expression of SATB1, DUSP10 and EEF1A1¹²⁸.

Furthermore, the expression in adipose tissue of certain markers, including IFI27¹²⁹, INHBA¹²¹, BDNF¹³⁰, NRIP1¹³¹ and CCAR2¹³², is positively correlated with senescence in ageing and other conditions. By contrast, markers such as KL¹³³, SIRT1¹³⁴, SREBF1¹³⁵, GDF15¹³⁶, MDM2¹³⁷, RRAD¹³⁸ and GBP1¹³⁹ are reported as negative regulators of cellular senescence in adipose tissue.

Kidneys

Senescence is observed in several human and mouse kidney cell types during healthy ageing and nephropathy. Glomerular endothelial cells¹⁴⁰ and proximal tubular epithelial cells (TECs)^140,141 become senescent with ageing in mice, showing increased SA-β-gal staining and Cdkn2a expression. Senescent glomerular endothelial cells overexpress CDKN1A, the DNA damage markers TP53BP1 and γH2AX, and SERPINE1^140,142,143. Serpine1 overexpression causes podocyte detachment and apoptosis, resulting in glomerulosclerosis¹⁴⁰. Senescent podocytes in old mouse kidneys overexpress senescence markers and SASP factors, including Irf5, Cebpb, Jdp2, Ccl2, Ccl3, Ccl5, Ccl8, Mmp12, Mmp13 and Gsk3b. Il1b, Il6 and Mmp13 are also upregulated in old mouse kidneys¹⁴⁰. Senescent cells are also observed in kidneys of mouse models of acute kidney injury, diabetic nephropathy, glomerulonephropathy, chronic kidney disease and chronic ischaemic renal injury^{140,144–150}. Different animal models of kidney disease exhibit increased SA-β-gal activity and overexpression of Cdkn2a, Cdkn1a, Trp53 and SASP components, including SERPINE1, IL-1β, IL-6, TNF and MMP13 (refs. 140,148). SERPINE1 drives glomerulosclerosis in early kidney injury in diabetes^140,148.

Studies of senescent cells in ageing kidneys in humans have mostly used what appears to be histologically normal tissues (disease-free areas of kidney) from otherwise diseased kidneys. CDKN2A, CDKN1B, CDK4, MMP1 and HSPA5 mRNA levels increase in the cortex and nuclear staining of CDKN2A is observed predominantly in distal tubules and collecting ducts, but also in glomerular parietal epithelium and podocytes¹⁵¹. CDKN2A protein expression in cortical interstitial cells and TECs correlates with age; by contrast, expression of CDKN1B protein correlates with age only in cortical tubular cells¹⁵². CDKN2A expression increases in renal arterial vascular smooth myocytes^145,151. The expression of PTGS1 and PTGS2 increases in ageing kidneys¹⁵¹. Lipofuscin is present in adult and old tubular cells in the medulla and in some cortical tubules in old kidneys¹⁵¹. Telomeres are shorter in the cortex but not in the medulla¹⁵³. SERPINE1 and CDKN2A, produced by senescent glomerular endothelial cells, increase with age in kidneys from healthy kidney donors^140,145. Most human renal nephropathies are associated with accelerated cellular senescence^{142,145,147,149,154,155}, with CDKN2A being most overexpressed in TECs. WNT9A, which is induced in CDKN2A-expressing TECs in chronic kidney disease, co-localizes with MKI67-negative nuclei (indicative of non-proliferating cells)¹⁵⁴. Furthermore, the nuclear CDKN2A level increases in glomerular and interstitial cells in individuals with glomerular disease¹⁴⁵, and in tubules and podocytes in individuals with diabetic nephropathy¹⁴⁷. Of note, a recent study demonstrated that iron accumulation, as a result of vascular and haemolytic injuries or other causes, promotes fibrosis and senescence in the kidney and other organs¹⁵⁶. Senescent cells also persistently accumulate iron, even when the surge of extracellular iron level returned to normal after injury¹⁵⁶.

Liver

Cellular senescence has been widely implicated in age-dependent liver dysfunction and the development of various chronic liver diseases such as fibrosis, cirrhosis, non-alcoholic fatty liver disease, alcoholic liver disease and chronic hepatitis. Multiple liver cell types, including hepatocytes, hepatic stellate cells, endothelial cells, Kupffer cells and cholangiocytes, exhibit senescence-associated markers during ageing and age-related diseases.

Studies on aged mouse hepatocytes have shown the presence of senescence-associated markers such as SA-β-gal activity¹⁵⁷, TAF^54,158, karyomegaly and SADS¹⁵⁹, LINE-1 expression²³ and expression of SASP factors¹⁴¹. Although less is known about hepatocyte senescence in humans than in mice, senescence-associated markers such as SA-β-gal staining have been identified in human cirrhotic livers¹⁶⁰. Additionally, the expression of CDKN1A, CDKN2A and the presence of TAF has been found in the context of non-alcoholic fatty liver disease^159,161, and TAF have been reported in livers from individuals with alcoholic liver disease¹⁶².

Mouse hepatic stellate cells have also been shown to exhibit senescence-associated markers such as SA-β-gal activity and expression of Cdkn1a, Cdkn2a and SASP components in models of liver injury^163,164. In the context of primary biliary cirrhosis, which is an autoimmune disease characterized by chronic, progressive cholestasis and liver failure, senescence-associated markers have been found in cholangiocytes¹⁶⁵.

Lungs

CDKN2A is the predominant marker associated with senescence in the lung. Various mouse models have been developed to study CDKN2A-driven lung cellular senescence in response to external stressors^155,166,167. Additionally, CDKN1A, p53 and γH2AX are also upregulated in mouse and human lungs upon senescence^168,169. Moreover, SERPINE1 was shown to induce mouse alveolar type 2 (ATII) cellular senescence through activation of the p53–CDKN1A–RB1 pathway in fibrotic lung disease¹⁷⁰. TGFβ1 has been implicated in mediating cellular senescence in the lung, as TGFβ1–IL-11–MEK–ERK signalling mediates senescence-associated pulmonary fibrosis in mice¹⁷¹. Extracellular HSP90AA1 activates TGFβ1 signalling and promotes Trp53 and Cdkn1a expression to induce mouse and human pulmonary fibroblast senescence¹⁷². In injury-induced senescence mouse models, SASP factors including cytokines (IL-1α, IL-1β, IL-10, TNF, TNFRSF1B and TNFSF12), chemokines (CXCL2 and CXCL12) and proteases (MMP8 and MMP12) are robustly upregulated¹⁶⁸.

The accumulation of senescent cells is also linked with structural changes in the lung, broadly termed ‘senile emphysema’¹⁷³. Recent research has demonstrated increased expression of Trp53 and Cdkn1a, and the presence of TAF in the airway smooth muscle cells from older individuals with asthma¹⁷⁴.

Growing evidence shows increased cellular senescence in type II epithelial cells, endothelial cells and fibroblasts from emphysematous lungs. Studies have associated increased cell cycle arrest markers (CDK1A and p53), SASP factors (CXCL5, CXCL8 and VEGFA), SA-β-gal activity and γH2AX with increased cellular senescence in chronic obstructive pulmonary disease^175,176. CDKN2A-mediated senescence has also been shown to modulate the cigarette smoke-mediated senescence response in the mouse model for chronic obstructive pulmonary disease^177,178. Similarly, senescent lung fibroblasts have a pivotal role in the pathogenesis of fibrotic lung diseases^156,172. There is an association between the upregulation of GDF15, IL-6, CRP and TNFRSF1B in plasma with the susceptibility of the occurrence of interstitial lung abnormalities¹⁷⁹. Another reported hallmark of senescent fibroblasts, in idiopathic pulmonary fibrosis, is the enhanced expression of anti-apoptotic BCL2 family proteins such as BCL2L2, BCL2 and BCL2L1 along with reduced levels of the pro-apoptotic proteins BAK1 and BAX^180,181. By contrast, recent studies provide evidence for the accumulation of iron in the senescent fibrotic tissues from both mouse and human, thereby emphasizing the role of cellular metabolism in senescence¹⁵⁶. It remains to be seen whether iron chelation therapy can reverse senescence and fibrosis or obstructive lung disease.

Pancreas

An age-dependent increase in the proportion of SA-β-gal⁺ cells was reported in human cadaveric islets. Type 2 diabetes significantly increased this proportion in all ages, accompanied by a higher intensity of the DNA damage marker TP53BP1. Mouse models of insulin resistance induced by a high-fat diet and the insulin receptor antagonist S961 exhibit elevation of SA-β-gal activity, and expression of Cdkn1a, Cdkn2a and the SASP factors Il1a, Il6, Tnf, Ccl3, Ccl4, Ccl5, Cxcl1 and Cxcl10 (ref. 182). The secretion of SASP protein factors has been confirmed through proteomics analysis of senescent (SA-β-gal⁺) and non-senescent (SA-β-gal^–) mouse and human pancreatic islet cells¹⁸³, leading to the description of a unique SASP in β-cells. In a mouse model of type 1 diabetes (NOD), the expression of several senescence markers such as Cdkn1a, Cdkn2a, H2ax, Il6, Mmp2, Serpine1, Il6 and Igfbp3 has been observed. Several of these factors were also found in human pancreatic islets from donors with type 1 diabetes¹⁸⁴. Additionally, sex-biased islet β-cell dysfunction can be caused by a type of maturity-onset diabetes of the young secondary to the MAFA S64F variant, where a single amino acid substitution in the transcription factor MAFA leads to diabetes and accumulation of senescent cells in males¹⁸⁵. These cells are characterized by expression of Cdkn1a, Ankrd1, Bcl2a1d, Ccl11, Cd68, Cxcl12 and Icam1, SA-β-gal activity, decreased Lmnb1 expression and increased γH2AX¹⁸⁵.

Markers of senescence have also been found in α-cells, such as Cdkn1a, Cdkn2a and loss of nuclear Hmgb1 in NOD mice, albeit at a lower rate than in β-cells in the same mice¹⁸⁶. However, a different study did not identify any senescent markers in α-cells from NOD mice or in the pancreas from donors with type 1 diabetes¹⁸⁷. Of the adult cell lineages of the pancreas, acinar cells show the highest plasticity, allowing them to undergo dedifferentiation or transdifferentiation to an embryonic progenitor phenotype expressing ductal markers in a process known as acinar-to-ductal metaplasia^188,189. During acinar-to-ductal metaplasia, dedifferentiated acinar cells acquire a senescent state characterized by cell cycle arrest, induction of p53, CDKN1A, CDKN2A and SA-β-gal activity^190–193. The same profile was found in individuals with chronic pancreatitis¹⁹⁴.

In a study of the post-mortem pancreas of 140 individuals without pancreatic disease, 114 (81%) exhibited ductal lesions such as mucous cell hyperplasia, typically associated with KRAS activating mutations; the ductal lesions show an increasing trend after the age of 40 years¹⁹⁵. These KRAS mutations are also found in the early stages of pancreatic ductal adenocarcinoma, preceding the appearance of senescence markers such as p53, CDKN2A and SMAD4¹⁹⁶. However, as this is a cross-sectional study, this relationship is only implied, and longitudinal studies will need to be conducted to corroborate the temporality of these events.

In a study involving 158 cases of intraductal papillary mucinous neoplasms and 10 normal controls, neither SA-β-gal activity nor SAHF formation was observed in the normal controls. However, in neoplastic tissues, particularly in cases of low-grade dysplasia, four senescence markers (CDKN2A, CDKN2B, SA-β-gal activity and CBX5-γ) were detected¹⁹⁷. Interestingly, these markers decreased in abundance during the transition to intraductal papillary mucinous neoplasm. These findings suggest that senescence is induced in early neoplastic stages and may have a role in preventing the progression of malignancy.

Skeleton

In aged mice, markers of senescence have been found in osteoblast progenitors, osteoblasts and osteocytes using magnetic-activated cell sorting followed by qPCR. Osteoblast progenitors and osteoblasts exhibit a significant increase of Cdkn2a, in addition to an increase of a limited number of SASP factors⁹³. Osteocytes have increased SADS with age, in addition to an increase of Cdkn2a, Cdkn1a and Trp53 and numerous SASP transcripts⁹³.

A recent study analysed protein levels in the skeleton of young and aged mice using mass cytometry by time of flight (CyTOF). Senescent cells in the bone microenvironment, which consisted principally of CD24⁺ osteo-lineage cells and late osteoblasts/osteocytes, were defined as CDKN2A⁺Ki67^–BCL2⁺ and CDKN1A⁺ Ki67^–BCL2⁺, and exhibited a significant increase in other senescence markers, SASP factors and DNA damage markers⁵⁶. Whether osteoclasts also undergo senescence remains unclear, although elimination of CDKN2A-expressing osteoclast progenitors did not attenuate age-related bone loss in mice¹⁹⁸.

In human bone biopsies, older women (mean age 78 years) show upregulated mRNA expression of CDKN2A, CDKN1A and senescence-associated chemokines (for example, CCL2, CCL3, CCL8 and CSF1)⁹³ compared with younger women. The SenMayo gene panel (Box 3) was generated and validated in bone biopsies from two aged human cohorts⁸⁴. In late puberty, mesenchymal stem/progenitor cells undergo normal programmed senescence, indicated by the loss of nestin expression¹⁹⁹.

Following fracture in mice, cells from the fracture callus exhibit an initial increase in Cdkn1a mRNA expression, followed by Cdkn2a and accompanied by a SASP consisting of numerous interleukins, NF-κB-downstream factors and, specifically, TGFβ1 expression^200,201. Disease states resulting from focal radiation therapy also induce skeletal senescence, characterized — similar to fracture — by early Cdkn1a and late Cdkn2a expression²⁰². Focal radiation therapy-induced bone loss can be alleviated by targeting senescent cells expressing Cdkn1a, but not Cdkn2a (ref. 203), suggesting the existence of functional heterogeneity between senescent cell subtypes.

Type 2 diabetes mellitus is also associated with cellular senescence in the skeleton. In a mouse model of type 2 diabetes mellitus (high-fat diet with streptozotocin), osteocyte-enriched samples demonstrated increased expression of Cdkn2a and Cdkn1a, along with SASP markers predominantly related to NF-κB signalling and matrix metalloproteinases²⁰⁴. In both ageing and the disease states listed above, skeletal cellular senescence can also be observed through histological staining of SA-β-gal activity²⁰³, SADS^{93,200,204,205} and TAF^200,203,204, typically in bone-embedded osteocytes²⁰⁶.

Skin

Skin ageing is influenced by exposure to environmental factors such as UV radiation, which makes it challenging to establish reliable senescence markers. Senescence in the skin is identified by expression of SA-β-gal, α-fucosidase, CDKN2A, CDKN1A and lipofuscin in both the epidermis and the dermis^207,208, and of the histone variant H2AJ and γH2AX in human epidermis, especially in the stratum spinosum²⁰⁹. This expression pattern correlates with lower MKI67 expression and only slightly with telomere length reduction²¹⁰. Chromatin aberrations reinforcing senescence are a common feature of skin senescence and are characterized by nuclear foci positive for TP53BP1 and PML^211,212. In sun-protected skin, melanocytes showed the highest rate of senescence, as identified by the expression of CDKN2A and increased number of TAF, and loss of nuclear HMGB1 (ref. 213).

Despite high variability (for example, the SASP of intrinsically aged human fibroblasts in vivo includes 27 distinct proteins²¹²), there is an association between GDF15, TNF, SPP1, TNFRSF1A, FAS, CCL3, IL-15 and activin A expression and age-related disease outcomes²¹⁴. In fibroblasts, keratinocytes and melanocytes in vitro, the classical SASP typically includes IL-6, IL-1α, IL-1β, MMP1, MMP3, MMP9 and TNF²¹¹, whereas high levels of chemerin are found in the SASP of dermal fibroblasts in aged skin²¹⁵.

Photodamage markers include the accumulation of 53BP1, H2AJ and mtDNA4977 (a common mtDNA deletion)²¹⁶. Repeated UVB exposure in human keratinocytes and mouse epidermis results in loss of LMNB1 (ref. 217). The early development of UV radiation-caused solar lentigines (liver spots) features the accumulation of melanocytes expressing CDKN2A, CDKN1A and p53, as well as increased TNF²¹⁸. In response to UVB, the SASP from primary human melanocytes has elevated levels of secreted IL-6, CXCL8, CCL1, CCL2 and HMGB1 (ref. 219). Mouse models of UV radiation show increased expression of Mmp1, Mmp2 and Mmp3, and of other SASP factors such as TNF, IL-1β and IL-6 (refs. 220–223).

In the mouse skin chemical carcinogenesis model, tumour promotion with the phorbol ester TPA features an increase in SA-β-gal⁺ keratinocytes with high CDKN2A levels, accompanied by increased levels of Mmp3, Mmp9 and Cxcl10 and downregulation of Lmnb1 (ref. 223). Mouse models of melanoma and human nevi (both oncogene driven) show accumulation of SA-β-gal⁺–CDKN2A⁺ foci^224,225. Additionally, the BRAF oncogenic mutation found in more than half of all malignant cutaneous melanomas is commonly found also in non-malignant nevi, where it causes oncogene-induced senescence exemplified by high levels of CDKN2A and SA-β-gal expression^225–227.

Mammary gland

The expression of TNFRSF11A increases during ageing²²⁸, leading to DNA damage and senescence through Cdkn2a (ref. 229). Senescence markers have been identified both in mammary glands of aged mice and in cultured primary human mammary epithelial cells. Expression of Cdkn2a is upregulated ~20-fold in mammary stromal cells of aged (25–32 months) mice²³⁰. Immunostaining reveals an increase with age of the frequency of ducts expressing the SASP factor COX2 (ref. 230). Other features of inflammation, such as macrophage crown-like structures²³⁰ and Cxcl1, Cxcl2, Cxcl16, Csf1, Csf3, Ccl5 and Il6 (ref. 231), are also elevated in mammary glands of aged mice.

Ovary

Most studies of ovary senescence have been conducted in animal models. The ovary ages earlier than other organs in the human body, as reproductive function begins declining in women in their midthirties and ceases completely at the time of menopause. This ageing process is characterized by a reduction in both quantity and quality of gametes, leading to reproductive senescence^232,233. Additionally, the ovary exhibits age-dependent inflammation and fibrosis^234–237; however, whether these are associated with senescence has not been conclusively established.

In ovaries from mice of advanced reproductive age, cells classified as senescent are characterized by increased expression of Cdkn2a (refs. 238–240), Cdkn1a, Serpine1 and Hmgb1 (ref. 239). Additionally, these cells exhibit increased lysosomal content, as indicated by accumulation of lipofuscin aggregates (known as ‘aggresomes’)²³⁹ and by SA-β-gal staining in the stromal compartment²⁴¹. The secretion of SASP factors such as Ccl8 (ref. 239), CCL5 (ref. 241) and the pro-inflammatory related gene Crlf1 (ref. 239) also increases with age.

Apart from physiological ageing, treatments that accelerate reproductive ageing through genotoxic agents, such as the chemotherapeutics cisplatin²⁴² and doxorubicin²⁴³, result in increased SA-β-gal^242,243, CDKN2A (refs. 242,243) and CDKN1A (refs. 242,243) expression. Moreover, they elicit components of the SASP such as IL-6 (refs. 242,243), Ccl2 (ref. 243) and Tgfb1 (ref. 243). In a genetic mouse model of accelerated ovarian ageing, Cdkn2a, Cdkn1a, Il1a and Il1b levels were elevated²⁴⁴. Furthermore, in a mouse model of premature ovarian insufficiency, SAHF formation was observed in the ovary, indicated by increased H3K9me3 and CBX5 staining and higher HMGA1 and HMGA2 protein expression²⁴⁵.

Single-cell expression and chromatin accessibility assays of post-menopausal human ovaries revealed that SASP-associated genes were expressed in stromal cells (SERPINE1, TIMP1, TIMP2, IGFBP2, IGFBP3 and IGFBP4), endothelial cells (IGFBP7) and perivascular endothelial cells (CDKN1A). Moreover, several CCL and CXCL genes are expressed in immune cells of the postmenopausal human ovary, potentially reversing some of the detrimental effects of senescence²⁴⁶.

Placenta

Developmental senescence is a programmed transient cellular senescence that has a crucial role in mammalian embryonic development, occurring not only in the embryo but also in the placenta. Cytotropho-blast differentiation into large mesenchymal extravillous trophoblasts and multinucleated continuous syncytiotrophoblasts demonstrates several hallmarks of senescence, including loss of replicative potential, increased expression of cell cycle inhibitors (notably CDKN1A, CDKN1C, p53 and MAPK14)^247–252, increased SA-β-gal activity and lipofuscin pigmentation^247–249, SAHF formation²⁴⁸, TAF and SASP factors such as IL-6, IL-1β, MMP9, HMGB1, GDF15 and TNF^247,253 that mediate immune surveillance and tissue remodelling²⁴⁷. Furthermore, extravillous trophoblasts display nuclear enlargement²⁴⁷, and syncytiotrophoblasts exhibit characteristics of senescence and ageing, such as increased syncytial knots and fibrinoid deposition²⁵⁴ and reorganization of their cytoskeleton²⁵⁵. Moreover, mitochondria of syncytiotrophoblasts undergo morphological and functional changes, characterized by decreased size and diminished ATP production, leading to increased ROS production to adapt to changing oxygen tension in the placenta^248,256.

In addition to trophoblast senescence, which is a physiological phenomenon that progresses with gestation and peaks at term, senescence is also extensively described in maternal decidual cells and fetoplacental membranes, where it has a pivotal role in the promotion of parturition and detachment of the placenta from the uterus at term^257–262. Recently, highly specific syncytiotrophoblast genes were identified to be induced in physiologically aged fibroblasts, as well as during replicative senescence or stress-induced cellular senescence in non-trophoblast cell lineages²⁶³. This genetic programme suggests that components of placental developmental senescence are recapitulated in other cellular senescence contexts.

The placenta can also undergo accelerated or premature senescence in adverse pregnancy outcomes such as pre-eclampsia and fetal growth restriction²⁶⁴. Telomeres from fetal growth restriction placentas are shorter and express higher levels of CDKN2A and CDKN1A protein and EEF1A1 RNA compared with placentas from normal growth fetuses²⁶⁵.

Colon

Information about senescence in the colon is limited and we did not include this tissue in our database. Several senescence markers have been reported to increase in the ageing mouse colon in homogenized tissue¹⁴¹. The senescence‐associated tissue microenvironment promotes colon cancer formation in human colon epithelial cells and in human-derived colon organoids²⁶⁶.

Conclusions, gaps in knowledge and future directions

Based on information from published human and mouse studies, we have outlined the current knowledge about senescence markers that are specific to various tissues. Although our compilation serves as a comprehensive resource, we have also pinpointed gaps in our understanding of these markers. To bridge these gaps, we suggest adopting practices that utilize multiple markers that encompass several senescence hallmarks (Box 2), in conjunction with newly developed senescence signatures (Box 3). Use of these practices should enhance the accurate identification of senescent cells within tissues.

The development of validated senescence biomarkers represents a crucial advancement in comprehending cellular senescence across developmental stages, ageing and various diseases. These biomarkers will not only contribute to identifying potential therapeutic targets within senescent cells but also help monitoring the senolytic clearance of senescent cells (Box 4). Although certain markers may share similarities with well-established senescence-associated features such as the SASP, DNA damage and inflammation, their comprehensive cataloguing is still ongoing.

Box 4. Recommendations for using circulating markers of cellular senescence.

Circulating biomarkers provide a range of powerful tools for the clinical toolbox ranging from disease discovery to informing strategies for the treatment of diseases. Blood-based biomarkers, in particular, provide the major clinical benefit of ease of repeated collections of relatively large volumes with minimal invasiveness. To translate senotherapeutics into the clinic, reliable circulating biomarkers of cellular senescence burden will be crucial in order to identify individuals who will most benefit from removal of senescent cells and to track the efficacy of the intervention in the context of a clinical trial. Several types of biomarker exist in the blood, including soluble factors in plasma and serum and the circulating cells of the immune system. Below, we discuss the state of senescence marker discovery in each of these circulating components and the work that remains to be done for the development of truly specific and sensitive biomarkers of tissue senescence burdens.

Many soluble circulating factors found in ageing and in age-related diseases share similarities with well-known senescence-associated secretory phenotype (SASP) factors. Whether they are driven by senescent cells remains uncertain. The majority of proposed circulating biomarkers were initially SASP factors reported in cell culture^{211,280–283} and subsequently introduced into in vivo senescence signatures in conditions associated with increased burden of senescent cells, such as ageing, age-related diseases and associated clinical outcomes: multimorbidity, mortality, mobility and menopause^{214,280,284–287}. Notably, the proteins GDF15, CST3, IGFBP2, TIMP1, TIMP2, STC1, CCL1, INHBA and others robustly associate with ageing-related problems in multiple studies^{214,284–287}. Proposed lipid biomarkers include bioactive lipids such as leukotrienes and 15-deoxy-Δ-12,14-prostaglandin J2 (refs. 288,289). Other potential sources of circulating biomarkers remain to be explored in detail in vivo, including senescence-associated extracellular vesicles (another bioactive component of the SASP²⁹⁰) and senescent circulating lymphocytes^86,290–294.

Owing to the heterogeneity of the SASP and the challenge of discerning the origin of molecules secreted into the blood, much work remains to validate circulating biomarkers of senescence burden. Validated senescence biomarkers ideally meet several criteria, most importantly correlation with tissue senescence burdens and a decrease following senescent cell removal, which has been explored in few studies to date^295–297.

Blood-based circulating biomarkers have the potential to provide a powerful resource for the clinical toolbox for monitoring health status, promoting disease discovery and guiding treatment strategies. However, levels of these biomarkers can be influenced by recent food or pharmaceutical intake, making animal experiments a more controlled model for research than human studies.

A substantial research challenge arises from the overlap of senescence markers with conditions or experimental models associated with inflammation or fibrosis. This overlap complicates the establishment of cause and consequence, as the presence of these markers may signify various biological processes occurring simultaneously (Box 1). The multimodal efforts of SenNet are designed to pinpoint novel and specific markers of senescence, enabling us to differentiate them from other types of cellular change that may occur during ageing. To achieve this goal, high-content and high-throughput single-cell and spatial omics technologies are deployed, which accelerate the discovery of crucial senescence biomarkers required for progress in the study of ageing and disease (Box 5). Furthermore, innovative machine learning and artificial intelligence techniques are under development to distinguish senescent from non-senescent cells based on their morphological differences (Box 3). These unbiased methodologies, coupled with the use of transgenic model organisms and advanced imaging platforms, hold great promise for improving our understanding of cellular senescence in the near future.

Box 5. Recommendations for techniques to identify and characterize senescent cells.

A comprehensive description of the techniques available to study senescent cells in tissues is beyond the scope of this Expert Recommendation and we refer the readers to a recent Review authored by the National Institutes of Health (NIH) Cellular Senescence Network (SenNet) Imaging Working Group²⁹⁸. Here, we limit our discussion to the established and novel broad categories of experimental techniques and to how they can be applied to identify senescent cells.

Three factors should guide the choice of which technique(s) is best suited to study senescent cells in tissues: senescent cells are rare in tissues (5–10% of all cells), so their identification and characterization require single-cell approaches; due to their heterogeneity, probing markers of multiple senescence hallmarks, which can vary between different cell types and contexts, requires the use of multimodal and higher-plex techniques; and to properly capture the effect of senescent cells on their microenvironment, use of spatial techniques is required. The majority of existing single-cell spatial methods rely on low-plex imaging techniques such as conventional immunofluorescence and immunohistochemistry (IHC), which can image a very limited number of markers simultaneously in the same cell (typically three or four). We recommend using higher-plex, latest-generation immunostaining techniques such as iterative indirect immunofluorescence imaging (4i) (~80-plex)²⁹⁹ and co-detection by indexing (CODEX) (>100-plex)³⁰⁰ to capture the heterogeneity of senescent cells in tissue based on their proteomics profiles.

Emerging techniques and commercial platforms for spatial transcriptomics can now probe hundreds to thousands of genes such as multiplexed error-robust fluorescence in situ hybridization (MERFISH)³⁰¹ (~500-plex), the 10× Xenium (~400-plex) and the Nanostring CosMx (~1,000-plex), and can combine transcriptomics and proteomics, although the number of proteins that can be detected is currently very limited.

Because these techniques use a preselected set of genes and/or proteins, an appropriate markers panel for a given tissue should be carefully designed. The panel should include both markers of cell identity based on the known cell types in the tissue and senescent markers that capture most of the senescence hallmarks. Prior knowledge of senescence markers (for example, senescence-associated secretory phenotype (SASP) factors) that are specific to the tissue under investigation can be used to design a more specific panel.

The main technique currently used to study senescence through single-cell transcriptomics is the 10× Genomics Chromium and its latest-generation controllers. Although this technique relies on tissue dissociation and cannot capture the microenvironment of the senescent cell, measuring the transcriptome provides a complete view of the transcriptional changes and heterogeneity of senescent cells. One of the limitations of its application to the study of senescent cells is the difficulty of probing CDKN2A (which encodes p16^INK4A). The expression level of CDKN2A is low, which typically results in a high gene dropout rate of up to 50%²⁷⁶. Therefore, it is not advisable to solely rely on CDKN2A for quantifying the abundance of senescent cells in tissues using these single-cell transcriptomics assays, as doing so would result in a considerable number of false negatives. This problem can be overcome by using the senescence signature-based methodologies described in Box 3.

Although we performed a comprehensive search of the available literature, our Expert Recommendation has limitations. Specifically, we have focused our discussion on senescence markers applied to the identification of senescent cells in individual tissues. We refer the readers to the many Review articles that cover topics such as the links between cellular senescence, metastatic cancer²⁶⁷ and endocrine diseases²⁶⁸, and the current status of the search for senolytics²⁶⁹.

We anticipate that the results generated by SenNet, as well as by the broader cellular senescence research community using spatial mapping techniques, will considerably advance our comprehension of the diversity of senescence in various cell types and its functional implications in ageing and disease. Our primary objective here was to establish a foundation for the informed development of spatial mapping panels based on the existing evidence of senescence markers in tissues. Our tissue-centric approach reflects the practical necessity of designing spatial mapping panels with a limited number of markers when studying individual tissues. This goal is crucial given the continued widespread use of low-plex experimental techniques. The database associated with this Expert Recommendation will serve as a dynamic document that SenNet will regularly update, incorporating novel findings from within the consortium and beyond. This document, available and regularly updated at https://docs.sennetconsortium.org/biomarkers/, also features an interactive graph visualization of the senescence markers database. We anticipate that, as the information in the database evolves, guidelines for using specific markers in different tissues will be updated to align with new findings.

Glossary

False-positive senescence

A cellular phenotype characterized by the presence of senescence markers without the cell being senescent.

Gene dropouts

In single-cell transcriptomics, refers to the failure to detect or quantify certain genes, typically owing low mRNA abundance.

Immunosenescence

age-dependent functional decline of immune cells, including reduced frequency of naïve cells, increased exhaustion, and dysregulation of innate immunity and signalling pathways.

Karyomegaly

Refers to enlargement of the hepatocyte nucleus, which is associated with liver-weight increase or cellular senescence.

Long interspersed element-1

(LINE-1). A transposition-competent retrotransposable element that is still active in most mammalian genomes.

Macrophage crown-like structures

Clusters of macrophages that surround dead or dying adipocytes.

Marker prioritization

The process of selecting and ranking potential markers based on certain criteria such as specificity, sensitivity, ease of measurement and clinical utility.

Oncogene-induced senescence

A state of cellular senescence triggered by the activation of certain oncogenes (for example, RAS).

Replicative senescence

A type of cellular senescence induced by the progressive shortening of telomeres during cell divisions.

Senescence-associated β-galactosidase

(SA-β-gal). A marker of increased lysosomal activity.

Senescence-associated distension of satellites

(SADS). Observed in senescent cells as a consequence of chromatin rearrangement leading to the loss of integrity of centromeric regions.

Senescence-associated heterochromatic foci

(SAHF). Observed as DAPI-dense foci in senescent cells in culture, particularly in oncogene-induced senescence; represents large-scale heterochromatin formation.

Senescence-associated secretory phenotype

(SASP). Refers to the secretion of pro-inflammatory cytokines, growth factors, chemokines, proteases and other bioactive molecules.

Senescent cells

Cells that have permanently withdrawn from the cell cycle, usually owing to the build-up of DNA damage.

Syncytial knots

Placental syncytiotrophoblast specializations characterized by remarkable accumulation of nuclei with heavily condensed chromatin; a sign of trophoblast maturation and ageing commonly seen post 32 weeks of human gestation.