Cellular Senescence in Brain Aging

Aging of the brain can manifest itself as a memory and cognitive decline, which has been shown to frequently coincide with changes in the structural plasticity of dendritic spines. Decreased number and maturity of spines in aged animals and humans, together with changes in synaptic transmission, may reflect aberrant neuronal plasticity directly associated with impaired brain functions. In extreme, a neurodegenerative disease, which completely devastates the basic functions of the brain, may develop. While cellular senescence in peripheral tissues has recently been linked to aging and a number of aging-related disorders, its involvement in brain aging is just beginning to be explored. However, accumulated evidence suggests that cell senescence may play a role in the aging of the brain, as it has been documented in other organs. Senescent cells stop dividing and shift their activity to strengthen the secretory function, which leads to the acquisition of the so called senescence-associated secretory phenotype (SASP). Senescent cells have also other characteristics, such as altered morphology and proteostasis, decreased propensity to undergo apoptosis, autophagy impairment, accumulation of lipid droplets, increased activity of senescence-associated-β-galactosidase (SA-β-gal), and epigenetic alterations, including DNA methylation, chromatin remodeling, and histone post-translational modifications that, in consequence, result in altered gene expression. Proliferation-competent glial cells can undergo senescence both in vitro and in vivo , and they likely participate in neuroinflammation, which is characteristic for the aging brain. However, apart from proliferation-competent glial cells, the brain consists of post-mitotic neurons. Interestingly, it has emerged recently, that non-proliferating neuronal cells present in the brain or cultivated in vitro can also have some hallmarks, including SASP, typical for senescent cells that ceased to divide. It has been documented that so called senolytics, which by definition, eliminate senescent cells, can improve cognitive ability in mice models. In this review, we ask questions about the role of senescent brain cells in brain plasticity and cognitive functions impairments and how senolytics can improve them. We will discuss whether neuronal plasticity, defined as morphological and functional changes at the level of neurons and dendritic spines, can be the hallmark of neuronal senescence susceptible to the effects of senolytics.

Information processing in the brain is a highly complex process; however, it relies on the activity of neurons interconnected at synapses. Strength and efficiency of those neuronal network connections change in response to environmental stimulation enabling the brain to maintain the information, process it and initiate the appropriate response. Modification of synaptic transmission is called synaptic plasticity and could manifest as morphological, electrophysiological changes as well as changes in synaptic protein content. The number of functional connections in the brain translates into the ability of the neuronal network to store and process information. Quantitative and qualitative measurement of neuronal morphology at the level of dendrite arborisation and morphology of dendritic spines that host excitatory synapses is a tool widely used to assess synaptic plasticity. Hippocampus together with Prefrontal Cortex (PFC) are brain regions critical for cognitive abilities that have been extensively studied in the context of aging-associated changes. Most of the excitatory synapses are located on spines placed all along the dendrites. Dendritic complexity and length is critically important for the regulation of neuronal function. In humans, hippocampal neurons appear to retain the size and complexity of their dendritic arborisation throughout life (Flood, 1993 ). However, there are also studies suggesting that hippocampal dendritic trees of some subregions, such as CA1, could actually extend with age (Turner and Deupree, 1991 ). In contrast, in cortical neurons of animal and humans, numerous studies showed regression of dendritic arbors in cortical neurons with age. Total dendritic length and complexity decrease with age for apical and basal dendrites (de Brabander et al., 1998 ). Animal models of accelerated aging partially replicate the changes observed in aged animals and humans. Thus, SAMP8 mice, with accelerating aging exhibit thinner and 45% shorter apical dendrites in medial PFC (Shimada et al., 2006 ). Both the density and morphology of dendritic spines may undergo age-associated changes. Similarly to dendritic arborisation, the number of dendritic spines on hippocampal neurons is more stable than on cortical ones. Spine numbers in rat and human CA1 hippocampal region generally stay unchanged with age (Dickstein et al., 2013 ), whereas regional specific decrease in dendritic spine densities has been reported in CA3 (Adams et al., 2010 ) and subiculum (Uemura, 1985 ). A decrease in dendritic spine densities in cortical neurons of aged animals or humans, compared to young controls, has been reported and ranged from 23 to 55% depending on species, cortical region and age (Shimada et al., 2006 ). The picture is even more complex when the morphological variability of dendritic protrusions is taken into consideration. Though spines display a wide morphological continuum, there are four main classes of spine shapes: thin, filopodia (long), stubby (short and wide), and mushroom (large head with thin neck), which represent a different stage of maturation and stability, with thin being the least stable. Other classification is based on the structure of spine postsynaptic density (PSD) of excitatory synapses. PSD is an element of the postsynaptic membrane visible in the electron microscope as a thick plate consisting of densely located receptors for neurotransmitters. Large spines have large heads and often are described as mushroom-like. Those may have PSDs with distinct breaks that are called perforated. Thin spines with small heads typically have small, uniform, non-perforated PSDs (Petralia et al., 2014 ). In cortical regions, the least stable spines (non-perforated population) have been shown to be most vulnerable to aging-associated decrease whereas the number of stable mushroom spines remained unaltered (Bloss et al., 2011 ). Aging selectively alters the number and function of hippocampal synapses in a region specific manner. CA3 cells of aged animals and humans are characterized by a decreased density of spines. This pruning with aging is selective toward less mature axo-spinous synapses (Adams et al., 2010 ). In the CA1 region, the number spines is not altered but mature perforated ones display reduction in the PSD area in aged learning-impaired rats (Nicholson et al., 2004 ). Dendrites, and particularly dendritic spines, are very dynamic structures. Maintenance of the brain neuroarchitecture represents, in fact, a certain balance between sprouting and retraction of membrane protrusions of different brain cells. Therefore, the above-mentioned results could suggest that aged cortical and hippocampal neuronal networks at CA3 region are less plastic, with a decreased ability to create new intraneuronal connections, whereas hippocampal CA1 neurons are more prone to deficits in the most stable synapses associated with cognitive abilities. Age-associated changes in synaptic plasticity at hippocampal and corti

Glial cells constitute around 50% of the total amount of cells in the brain and play key roles in regulating brain homeostasis in health and disease. Their essential functions include providing nutritional support to neurons, activation of immune responses, and regulation of synaptic transmission and plasticity (Salas et al., 2020 ). In contrast to neural cells, glial cells are proliferation-competent, and as such, are prone to undergo canonical cell senescence. Interestingly, a transcriptional study of aging-related changes in gene expression across human brain regions from 480 individuals ranging in age from 16 to 106 years, showed that glial cells, not neurons, displayed the majority of differential gene expression with aging (Soreq et al., 2017 ). Astrocyte Senescence Astrocytes comprise about 20% of cells in the brain and play an important role in brain physiology and neuronal function. They provide support for neuronal cells, regulate the content of the synaptic cleft, play an essential role in the maintenance of ion homeostasis (e.g., by shunting potassium ions from the regions of high to the regions of low neuronal activity), and maintain blood-brain barrier (BBB) integrity. Acting as a part of the tripartite synapse, they control the concentration of glutamate (Glu), and thus glutamatergic transmission, and secrete their own glial transmitters. They also play a significant role in the recovery after brain injury by reducing wound-derived excitotoxicity, limiting damage, participating in scar formation and, at later post-injury stages, by in ensuring regeneration. Astrocytes also have several immune functions, i.e., they contain several pattern-recognition receptors, and can secrete cytokines and chemokines. In response to an acute injury, there is an increase in reactive astrocytes: cells undergo various alterations including swelling, hypertrophy, proliferation, and increased expression of a cytoskeletal protein, GFAP (Verkhratsky et al., 2010 ). Interestingly, the increase in the expression of GFAP, the major intermediate filament protein, which is also a specific marker of astrocytes, is the most common change observed in astrocytes with aging. Another intermediate filament, vimentin, which was for a long time considered to be present only in newly-generated and immature astrocytes, also increases with age (reviewed by Salminen et al., 2011 ). Transcriptomics analysis revealed dysregulation of genes associated with the cytoskeleton, proliferation, immune response, apoptosis, and ubiquitin-mediated proteolysis that occurs in the aging brain (Simpson et al., 2011 ). Moreover, aging astrocytes may lose their capacity to regulate glutamate level during synaptic activity due to a reduction in the level of astrocyte-specific glutamate transporters. They may also lose their neuronal support and reduce interactions with synapses. The outcome would be altered synaptic function and plasticity, increased neuronal oxidative and proteolytic stress, and mitochondrial dysfunction, which in the hippocampus would lead to dysfunction in memory retrieval and consolidation (Ojo et al., 2015 ). The very important question is whether astrocytes can undergo senescence similarly to many other proliferation-competent cells. Indeed, astrocytes have been shown to undergo cellular senescence in vitro , both replicative and induced by different external stressors (Cohen and Torres, 2019 ). Below are some examples. In response to oxidative stress or proteasome inhibitor, murine, and human astrocytes showed reduced proliferation capability and changes in several established markers of senescence, namely, SA-β-gal activity and in the level of p21 WAF1/CIP1 , p53, and p16 INK4A proteins (Bitto et al., 2010 ). Evans et al. ( 2003 ) showed that human astrocytes underwent senescence, which was telomere-erosion independent and p53-dependent. The same group documented an increased p16 INK4A level in the frontal cortex area of old and AD people in comparison with young subjects. It was also shown that the senescence process could be triggered in astrocytes by amyloid-β, which induced activation of SA-β-gal, p16 INK4A expression and secretory phenotype connected with p38 MAP kinase (Bhat et al., 2012 ). Others showed that ammonia induced senescence in cultured astrocytes in a glutamine synthesis- and oxidative stress-dependent manner through p53-dependent transcription of cell cycle regulatory genes, p21 WAF1/CIP1 and GADD45a. Increased p21 WAF1/CIP1 , p53, and GADD45a mRNA expression levels were also found in postmortem brain samples from patients with liver cirrhosis and hepatic encelophaty (Gorg et al., 2015 ). In another studies, carried out on long-term rat astrocyte cultures, the generation of reactive oxygen species was enhanced and mitochondrial activity decreased. Simultaneously, there was an increase in proteins that stained positively for nitrotyrosine. The expression of Cu/Zn-superoxide dismutase (SOD-1), heme oxygenase-1

Once mature, most of our neurons last our entire lifetime and show high potential for plasticity, which allows them to continuously modulate and adapt their complex synaptic network to changing conditions. The concept of neuronal senescence is relatively new even though it is well-established that neuronal functions decline with aging (Tan et al., 2014 ). As the hallmarks of cell senescence encompass more than just irreversible growth arrest, the studies to identify markers of neuronal senescence, that is senescence of proliferation-incompetent cells, were recently performed both in vivo and in vitro (Tan et al., 2014 ; Kritsilis et al., 2018 ; Wengerodt et al., 2019 ). However, data concerning this subject are still scarce, probably due to the existing paradigm telling that cellular senescence is an irreversible/permanent cell cycle arrest. It seems that depending on the type of cell senescence, this arrest can occur either in the G1 or G2 phase of the cell cycle (Bielak-Zmijewska et al., 2014 ). Since neurons are post-mitotic, non-cycling cells (permanently in the G0 phase of the cell cycle), neuronal senescence, like that of other post-mitotic cells, must rely on other foundations than proliferation arrest. Some markers of cellular senescence were observed in long-lasting neuronal cultures without additional inducer. Generally, it takes 30–40 DIV (days in vitro ) to develop a senescence phenotype. Although not fully specific, the most useful marker of cell senescence is the increased activity of SA-β-gal. Indeed, increased SA-β-gal was observed in the hippocampus of aging rats (Geng et al., 2010 ) and long-term in vitro cultures of hippocampal (Dong et al., 2011 ) and cerebellar granule neurons (Bhanu et al., 2010 ). We also observed stronger staining for hippocampal SA-β-gal in old than young mice. Moreover, in long-term neuroglial co-cultures (neural plus glial cells) of rat cortical cells, we observed a quite rapid increase in SA-β-gal-positive neurons and increased IL-6 production but, interestingly, no hallmarks of DNA damage (Piechota et al., 2016 ). Moreover, induced DNA damage did not increase the number of SA-β-gal-positive neurons. Similarly, in rat hippocampal neurons cultured in vitro , no increase of γH2AX foci was observed (Ishikawa and Ishikawa, 2020 ). However, Jurk et al. ( 2012 ) documented that Purkinje neurons of the cerebellum, and also cortical, hippocampal, and peripheral neurons in old mice, had severe DNA damage, activated p38 MAP kinase, high ROS production and oxidative damage, high interleukin IL-6 production, heterochromatinization and SA-β-gal activity. Moreover, they showed that these senescence symptoms were dependent on DNA damage-induced expression of p21 WAF1/CIP1 . Others showed some features of cellular senescence, such as increased SA-β-gal activity, oxidative stress, γH2AX expression, IL-6 production and astrogliosis, in mixed rat neuroglial co-cultures (Bigagli et al., 2016 ). Moreno-Blas et al. showed that in rat mixed cultures, both neuronal and glial cells became SA-β-gal-positive, but only neurons expressed p21 WAF/CIP1 and only astrocytes accumulated γH2AX foci (Moreno-Blas et al., 2019 ). Moreover, rat senescent neuronal cells in vitro were characterized by elevated SASP. In long-term cultures enriched in hippocampal neurons several cell senescence markers were found: upregulation of Cdk inhibitor p16 INK4A (Cdkn2a), but not p21 WAF1/CIP1 (Cdkn1a), increased activation of p38 MAP kinase and loss of lamin B1, which leads to chromatin changes. Neurons at 28 DIV showed upregulation of SASP gene expression including those encoding Cxcl1, plasminogen activator inhibitor-1 (Pai-1), and insulin-like growth factor-binding proteins (Ishikawa and Ishikawa, 2020 ). The authors documented that not only hippocampal but also cortical cell culture led to expression of senescence markers. Recently, we have focused our interest on senescence of neural progenitor cells (NPCs) derived from A-T (Ataxia telangiectasia) reprogrammed fibroblasts. We showed that A-T NPCs, obtained through neural differentiation of iPSCs in 5% oxygen, possessed some features of senescence, including increased activity of SA-β-gal, increased secretion of IL-6 and IL-8 and reduced sirtuin 1 level in comparison to control NPCs. This phenotype of A-T NPC was accompanied by elevated oxidative stress. Additional sources of oxidative stress, such as increased oxygen concentration (20%) or H 2 O 2 treatment, aggravated the phenotype of senescence (Sunderland et al., 2020 ). Since the ATM kinase is a key protein involved in DNA damage response, our data indicate that senescence of neural cells is not DNA damage-dependent and that rely rather on oxidative stress, similarly to rat cortical cell senescence (Piechota et al., 2016 ). In A-T, but not in control human neural cells, we observed a REST increase. REST is a transcriptional repressor that is downregulated during neuronal development in humans to de-repre

Autophagy is an evolutionally conserved intracellular process that enables cells to dispose of damaged or unnecessary molecules and organelles. Thus, autophagy is responsible for cellular homeostasis. Recent genetic evidence indicates that autophagy has a crucial role in the regulation of animal lifespan and that basal level of autophagic activity is essential for longevity (Nakamura and Yoshimori, 2018 ). Several studies demonstrate that hundreds of proteins become highly insoluble with age, in the absence of evident disease processes (Groh et al., 2017 ) and that autophagy activity in the brain decreases with age in many experimental models (Loeffler, 2019 ). One of the most striking morphological changes in neurons during normal aging is the accumulation of lipofuscin aggregates. Lipofuscin is an autofluorescent pigment formed by lipids, metals and misfolded proteins, that is especially abundant in nerve cells, cardiac muscle cells and the skin (Moreno-Garcia et al., 2018 ). Disturbances in proteostasis, along with accumulation of damaged DNA, dysfunctional mitochondria, and lysosomes, point to the failure of the protein degrading and organelle turnover system in aging. Apart from the proteasome, exclusively destined for protein degradation, autophagy is the biological system for degradation of different kinds of molecules and of whole organelles. Indeed, attenuation of both the proteasomal (Low, 2011 ) and autophagy-dependent degradation with age has been described (Loeffler, 2019 ). Down-regulation of expression of the key autophagy genes, such as ATG5, ATG7, and BECN1, was demonstrated during aging of the human brain (Lipinski et al., 2010 ). According to Yim and Mizushima, there are four kinds of autophagy: macroautophagy, chaperon-mediated autophagy (CMA), RN/DNatophagy, and microautophagy (Yim and Mizushima, 2020 ). However, only during macroautophagy the material destined for degradation is closed in newly formed vesicles called autophagosomes. All autophagy types rely on lysosomes. Thus, their dysfunction due to, for example, lysosomal membrane permeabilization (LMP), leads not only to autophagy impairment but also to leakage of the lysosomal content into cytosol ( Figure 3 ). Hence, damaged lysosomal clearance is important for cell function and survive (Gomez-Sintes et al., 2016 ). Interestingly, in the aged rat brain, increased cathepsin D activity and microtubule-associated proteins (MAPs), MAP1, and MAP2, degraded by cathepsin D-like enzyme, were found (Matus and Green, 1987 ; Nakamura et al., 1989 ). This finding suggested leakage the enzyme from lysosomes. Further investigation demonstrated an increase in other cathepsin in rat brain, as has been summarized by Stoka et al. ( 2016 ). Thus, increase in lysosomal enzymes' activity and accumulation of lipofuscin in rat brain points to perturbation of lysosomal function with age. Figure 3 Age-dependent lysosome impairment affects function of the brain cells. With age, lysosomes become large with altered membrane protein content, namely, increased LAMP1 level, at least in quiescent neural stem cells, and decreased level of LAMP2A. Reduced LAMP2A level in lysosomal membrane is believed to be the reason of declined CMA observed in old animals. Additionally, with age, lysosomal membrane permabilization (LMP) occurs resulting in leakage of protons and lysosomal hydrolases into cytosol. Uncontrolled protein hydrolysis, e.g., of microtubule-associated proteins (MAPs), affect their function and may lead to cell death. proton leakage impedes maintenance of low lysosomal pH, which is necessary for proper functioning of hydrolases. Hence, decline in lysosomal function results in autophagy impairment and, in consequence, accumulation of various aggregated proteins, lipofuscin, and damaged organelles. Indeed, changes in lysosomal activity with age were described by Gomez-Sintes et al. ( 2016 ). Aged lysosomes become large and more vulnerable to LMP ( Figure 3 ). It was suggested that changes in lysosomal membrane components resulted in lysosome fragility. Decreased LAMP2A content in lysosomal membrane with age was found in rodents and was postulated to be the reason of age-related decline of CMA (Kiffin et al., 2007 ). Aged lysosomes were also characterized by increased pH, which affects lysosomal enzyme activity. In consequence, lysosomes accumulate undigested material, including lipofuscin, which in turn, may affect lysosomal function and integrity causing LMP. Lipidation is the very initial step in autophagy regulated by many autophagic proteins. It means conjugation of the cytosolic LC3-I protein to phosphatidylethanolamine in the phagophore membrane to form LC3-II, which is one of the markers of autophagic vesicles (Yoshii and Mizushima, 2017 ). Recently, it has been demonstrated that autophagy lipidation machinery has a non-canonical, autophagy-independent function in regulating axonal microtubule dynamics (Negrete-Hurtado et al., 2020 ). ATG proteins

As we get old, neuronal complexity declines. Dendritic arborization, length, synapse number, and spine density decrease to variable degrees in cortical areas. This is accompanied by reductions in most aspects of cognitive performance including memory, awareness, and intellectual abilities (Zhu et al., 2020 ). It can be speculated that brain aging, similarly to that of other organs, is caused by senescent cells. Senescent cell burden is low in young individuals, but increases with age in all tissues, especially in the brain, adipose tissue, skeletal muscle, kidney, skin, and ovaries (Sikora et al., 2010 ; Sikora, 2013 ). Recently, it has been shown that targeting senescent cells can alleviate the effect of brain aging, similarly to other organs, which gives the hope for increasing the cognitive performance and healthspan (Baker and Petersen, 2018 ; Sikora et al., 2019 ). The just emerging new therapeutic approach, senotherapy, that makes employs senolytics (the name originates from the words “senescence” and “lytic”), which are able to directly eliminate senescent cells, showed that, indeed, health improvement is within a therapeutic range (Baker and Petersen, 2018 ; Sikora et al., 2019 ; Kirkland and Tchkonia, 2020 ). Senolytics turned out to be promising in preclinical studies of multiple conditions in mice, including metabolic disorders, cardiovascular disorders, cognitive impairment, pulmonary dysfunction, frailty, kidney and liver dysfunction, and osteoarthritis, but they also delayed cancer onset and extended the healthspan and lifespan (Tchkonia et al., 2020 ). Several transgenic mouse models have been developed enabling the visualization, assessment and eradication of senescent cells in vivo ( Figure 5A ). In these models the promoter of the p16 INK4A gene was used to drive expression of genes encoding proteins, which induced death of senescent cells upon transgene activation. The same gene promoter was used to drive expression of fluorescent proteins in some of those mice. Using these animal models it was possible to confirm that senescent cells accumulate during aging and in response to environmental stresses, tissue damage or injury as well as at the sites of age-related pathologies. In INK-ATTAC animals, the p16 INK4A promoter drives the expression of an inducible caspase-like transgene, which encodes an apoptotic protein that is activated by a small ligand, AP20187 (Baker et al., 2011 ). Another model carrying suicide transgenes for a truncated herpes simplex virus thymidine kinase (p16-3MR mouse) under control of the p16INK4A promoter, have been developed. In this cases, selective killing of senescent cells was accomplished through treatment with ganciclovir (GCV) (Demaria et al., 2014 ). There is also a study, in which a p19ARF-directed CDKN2A gene promoter sequence that regulates expression of the diphtheria toxin receptor (ARF-DTR) was used (Hashimoto et al., 2016 ). Importantly, thanks to the transgenic mouse models that facilitate selective elimination of senescent cells, it has been shown that age-related pathological symptoms could be alleviated and, consequently, the animal healthspan and even the lifespan could be improved (reviewed in Sikora et al., 2019 ). On the other hand, transplantation of even a small number of senescent cells (preadipocytes) into young mice led to spreading of cellular senescence to neighboring tissues and caused persistent physical dysfunctions. This effect, along with lifespan reduction, could be counteracted by oral administration of senolytics (Xu et al., 2018 ). Figure 5 Mouse models applied in experimental senotherpy. Two experimental approaches for the eradication of senescent cells are used: (A) transgenic mouse in which transgene expression is under control of the promoter of a senescence-specific gene coding for p16 INK4A or p19 ARF protein. Activation of the transgene via an adequate compound leads to induction of cell death in p16 INK4A or p19 ARF -expressing cells. Additionally, transgene might carry GFP protein sequence enabling identification of GFP-positive senescent cells before and after treatment; (B) senolytic treatment of wild type mice leads to death of senescent cells; death-inducing activity of senolytics stems from the ability to inhibit prosurvival or antiapoptotic pathways that are upregulated in senescent cells. The senolytic treatment, without genetic manipulations, has already been applied in many mouse models of age-related disease ( Figure 5B ). In general, senolytics can be classified into BCL family inhibitors, PI3K/AKT inhibitors, and FOXO regulators (Zhu et al., 2020 ). Briefly, the BCL family is composed of anti-apoptotic and pro-survival proteins, including BCL-2 and BCL-xL, the selective inhibition of which has been previously shown to cause cell death in some cancers (e.g., Souers et al., 2013 ). Presently, a similar strategy has been adopted to effectively remove senescent cells (Short et al., 2019 ). At present, BCL inh

ES made substantial contributions to the conception and design of the review and gave final approval of the version to be published, ES, AB-Z, MD, GM, AK, MW, and JW participated in writing the particular chapters of the manuscript and approved the final version. GM, AK, and MD prepared figures. All authors contributed to the article and approved the submitted version.