Nuclear NFκB Activity Balances Purine Metabolism in Cellular Senescence

Upregulation of nuclear factor κB (NFκB) signaling is a hallmark of aging and major cause of age-related chronic inflammation; however, its physiological functions and mechanisms remain unclear. By combining mathematical modeling and experiments, we show that dysfunction of negative feedback regulators of NFκB, IκBα and A20, alters the NFκB nuclear dynamics from oscillatory to sustained, promoting cellular senescence by remodeling epigenetic regulation and metabolic landscape. Sustained NFκB activity by IκBα downregulation enhanced inflammation- and senescence-associated gene expression through increased NFκB-DNA binding and slowed the cell cycle by upregulating purine catabolism via mTORC2/AKT pathways. Notably, IκBα knockdown combined with A20 overexpression resulted in lower NFκB amplitude, cytokine expression, and SA-β-gal activity than IκBα knockdown alone. IκBα downregulation is correlated with hypoxanthine phosphoribosyltransferase 1 (HPRT1) expression in the purine salvage pathway in aged mouse hearts. Our study suggests that nuclear NFκB homeostasis is critical for balancing purine metabolism associated with chronic inflammation and tissue aging.


INTRODUCTION
Aging is a complex and irreversible phenomenon that reflects lifelong biochemical responses to the environment. Recent research has shown that age-related dysfunction occurs from the genome to organ levels 1,2 . Inflammatory aging, which refers to aging-associated chronic inflammation, critically impacts the induction of a wide range of age-related human diseases 3 , and its regulation is becoming increasingly important in aging societies. Elderly individuals exhibit low-grade and persistent chronic inflammation in their tissues along with elevated levels of circulating inflammatory factors that activate nuclear factor κ B (NFκB) signaling, such as interleukin-6 and tumor necrosis factor a (TNFα) 4,5 . Meanwhile, the accumulation of senescent cells during aging produces various pro-inflammatory cytokines that represent a senescenceassociated secretory phenotype (SASP), which increases the risk of developing age-related diseases 6 . As NFκB is the master transcription factor for SASP genes 7,8 , NFκB is considered a positive feedback regulator in inflammatory aging and cellular senescence. However, several studies have also shown that NFκB prevents cellular senescence, therefore it remains unclear which properties of NFκB activity is related to cellular senescence 9,10 . Herein, we speculate that a quantitative rather than qualitative understanding of NFκB activity is necessary to unravel the role of NFκB in cellular senescence and inflammatory aging.
NFκB regulates the expression of genes involved in immune responses and cell proliferation, apoptosis, and senescence [11][12][13] . Excessive NFκB activity induces pathological inflammation and contributes to the progression of various human diseases 14 . Notably, NFκB also induces transient, sustained, or oscillatory nuclear dynamics associated with different biological functions and cell fates 15 . The quantitative features of NFκB activity, including the peak amplitude, duration, speed, and oscillation, are associated with the discrimination of immune 3 threats by macrophages 16 , and it encodes a short-term history derived from prior pathogenic and cytokine signals, affecting later cell functions 17 . These studies demonstrate that the homeostasis of NFκB nuclear dynamics is important for the maintenance of cellular functions.
The canonical NFκB pathway undergoes negative feedback regulation in response to environmental factors and exhibits stimulus-specific oscillatory dynamics. In particular, NFκB inhibitor alpha (IκBα), which is rapidly expressed upon NFκB activation, acts as a strong negative feedback regulator by binding to NFκB and inhibiting its nuclear translocation, thereby inducing characteristic NFκB oscillatory patterns 18 . Notably, knockout (KO) or knockdown (KD) of the IκBα gene in TNFα-stimulated cells results in sustained NFκB nuclear activation and upregulation of immediately expressed genes that show distinct patterns generated from oscillatory NFκB [19][20][21] . However, the physiological significance of NFκB dynamics and their long-term implications on signaling crosstalk, epigenetics, and metabolic regulation have not been fully addressed. In this study, we investigated the impact of changes in TNFα-induced NFκB dynamics on cell functions and found that sustained NFκB induces activation of purine catabolic pathways via the mTORC2/AKT pathway and prolongs the cell cycle period, rewires epigenetic states by enhanced NFκB-DNA binding to promote inflammatory gene expression. These results indicate that sustained nuclear NFκB activity significantly alters the landscape of the cellular network to induce cellular senescence and inflammatory aging.

Sustained NFκB activity induces growth arrest and cellular senescence.
First, we confirmed that IκBα KD alters oscillatory NFκB nuclear activity to sustain activity in MCF-7 cells stimulated with TNFα ( Figures 1A, S1A, and S1B). The change in NFκB dynamics suppressed cell proliferation after 24 hr of TNFα stimulation ( Figure 1B) and led to a flattened cell shape and significantly increased cell size (Figures 1C, S1C, and S1D). These features are similar to the senescence phenotype 22 . Therefore, we examined SA-β-gal activity, one of the indicators of cellular senescence 23 . Although TNFα alone did not promote SA-β-gal activity, under IκBα KD/KO conditions, TNFα increased the number of SA-β-gal positive cells after 24 hr (Figures 1D and S1E-S1G). The number of SA-β-gal positive cells was suppressed by KD of RELA, a subunit of NFκB, in IκBα KD+TNFα conditions ( Figures 1E and S1H), thus confirming that the phenomenon is NFκB-dependent. As senescent cells typically exhibit a SASP 6,23 , we next examined the effects under IκBα KD+TNFα conditions. The expression of various SASP genes, including IL8 and CCL2, was drastically elevated (Figures 1F and 1G). The amount of Cyclin D1, which is essential for the G1/S transition, was decreased compared to that with TNFα alone ( Figure 1H). Furthermore, a time-course cell cycle analysis using FUCCI live cell imaging revealed that IκBα KD+TNFα increases the number of the cells at the G2/M phase after 24 hr and prolongs the duration of the G2/M phase (median values: control, 3.8; IκBα KD, 8.3; P < 0.05) (Figures 1I, S1I, and S1J). These results indicate that the sustained NFκB activity mimicked by IκBα KD+TNFα modulates cell cycle checkpoints and inhibits cell proliferation.
Of note, a previous study showed that IκBα not only plays a negative role in NFκB activity but also positively affects the induction of several NFκB target genes 19 . Therefore, we sought to 4 determine whether the sustained NFκB dynamics were responsible for the induction of cellular senescence rather than IκBα itself. In addition to IκBα, A20 has been shown to act as a negative feedback regulator of the NFκB pathway by inhibiting upstream IKK activity and TTR complex (TRAF2, TRADD and RIP) binding to the TNFα receptor and modulating the oscillatory dynamics of NFκB 24 . Although the regulatory mechanisms of IκBα and A20 on the NFκB pathway are different, downregulation of each molecule leads to sustained NFκB dynamics 24,25 . Therefore, we investigated whether A20 KD promotes cellular senescence. A20 KD+TNFα indeed exhibited sustained NFκB activity (Figures 2A and S2A), increased SA-β-gal staining and SASP gene expression (IL8 and CCL2) ( Figures 2B and 2C), and slightly reduced Cyclin D1 expression ( Figure 2D) in MCF-7 cells; however, these effects were weaker than those of IκBα KD ( Figures 1A and 1D-1H).
To evaluate the quantitative difference in the effects of IκBα and A20 on NFκB nuclear dynamics, we constructed a mathematical model of the NFκB pathway ( Figures 2E, S2B and Tables S1-3) modified from earlier studies 26,27 and performed simulation analyses. We trained the model with time-course western blot data obtained from TNFα-stimulated MCF-7 cells ( Figures S2C-S2F) and analyzed the NFκB dynamics altered by IκBα and A20 KD. The simulated downregulation or overexpression (OE) of IκBα or A20 showed that IκBα KD has a greater influence on changing nuclear NFκB activity from oscillatory to sustained, thereby inducing a higher total amount of nuclear NFκB activity ( Figures 2F and S2G). Sensitivity analysis (see the Methods) also confirmed that IκBα-related reactions presented slightly higher sensitivity to the sum of nuclear NFκB activity than the A20 reactions ( Figure S2H), suggesting that IκBα has a more critical effect on nuclear localization of NFκB. The result was consistent with the experimental results of NFκB nuclear dynamics ( Figures 1A and 2A) and the extent of cellular senescent phenotypes ( Figures 1D-1H and 2B-D) between IκBα and A20 KD. These results indicate that the total nuclear abundance of NFκB associated with prolonged NFκB activity may be critical in the induction of cellular senescence. To further evaluate the importance of these quantitative features of NFκB (e.g., oscillation or abundance) and determine whether A20 OE could compensate for the effect of IκBα downregulation to recover oscillation dynamics to inhibit cell senescence, we first performed simulation analyses of nuclear NFκB dynamics by combining IκBα KD and A20 OE ( Figure S2I). Under A20 OE, IκBα KD still exhibited sustained NFκB activity to some extent; however, its overall amplitude was reduced. To confirm this result experimentally, we established A20 OE MCF-7 cells ( Figure S2J) and examined the effect of IκBα KD on NFκB dynamics. As computationally predicted, A20 OE reduced the amplitude of NFκB activity in IκBα KD+TNFα conditions but failed to fully alter the NFκB dynamics from sustained to oscillatory ( Figure S2K). However, under the same conditions, the increase in the number of SA-β-gal positive cells associated with IκBα KD ( Figure 1D) was mitigated by A20 OE (Figures 2G), which also reduced IL8 and CCL2 expression ( Figure 2H). These results suggest that the amount of nuclear NFκB level is more critical for the induction of cellular senescence.

Transcriptional regulation of sustained NFκB activity in vitro and in vivo.
Subsequently, we investigated the relationship between NFκB nuclear dynamics and target gene regulation. First, we extracted differentially expressed genes (DEGs) in MCF7 cells between 5 IκBα KD+TNFα (48 hr) and TNFα conditions and identified 687 upregulated DEGs, including cellular senescence-promoting and SASP genes (e.g., CD36, BCL3, CCL2, and S100A8/A9) 1,23,28,29 (Figure 3A). These DEGs were classified into four clusters based on the time-course patterns of gene expression, which was measured every 15 min under the IκBα KD+TNFα condition 19 ( Figure 3B). To clarify whether the gene expression time-course was associated with the chromatin status, chromatin accessibility was assessed using an assay for transposaseaccessible chromatin sequencing (ATAC-seq) data. We found that the ATAC signal was elevated in the flanking regions of genes in cluster 2, which showed a monotonically increasing pattern, and their signals were stronger in the IκBα KD+TNFα condition than the TNFα or IκBα KD alone condition ( Figure 3C). Considering that the chromatin accessibility enhanced by IκBα KD+TNFα might be critical for the regulation of genes controlled by NFκB, we extracted the regions where chromatin accessibility was significantly enhanced by IκBα KD+TNFα ( Figure  S3) and assessed the NFκB binding to these regions. We analyzed NFκB (RELA)-chromatin immunoprecipitation sequencing (ChIP-seq) data and found that the RELA signal intensity in chromatin open regions proximal to cluster 2 genes was markedly increased under IκBα KD+TNFα conditions; however, it was not observed under TNFα alone ( Figure 3D). Furthermore, we investigated whether the number of NFκB bindings affects gene expression in each cluster 30 . Notably, the average number of RELA binding sites in cluster 2 genes was significantly higher than that for all NFκB target genes ( Figure 3E). Since cluster 2 genes were enriched with several cytokine signaling pathways, such as TNFα and IL17 ( Figure 3F), these results suggest that an inflammatory positive feedback loop is epigenetically enhanced by sustained nuclear NFκB localization. Senescent cells accumulate in aged tissues in a variety of mammals 23 . Given that sustained NFκB activity may be involved in promoting cellular senescence and chronic inflammation associated with aging in vivo, we examined the levels of IκBα, A20, and TNFα in young (8 weeks old) and aged (69-73 weeks old) mice ( Figures S4A-S4D). We found decreased IκBα and A20 protein expression and increased TNFα in the heart tissues of aged mice ( Figure 4A). We also found a marked increase in the NFκB-positive area in the cell nuclei of aged hearts using immunohistochemistry staining ( Figures 4B and S4E) and upregulation of the cellular senescence markers p16 and Il-6 ( Figure 4C). Since p16 is a negative regulator of CDK4/6, these data imply that cytokine signaling and cell cycle delays are increased in association with persistent NFκB activity. RNA-seq analysis of heart tissues from young and aged mice and predictions of active transcription factors using DoRothEA 31 suggested that NFκB family members NFKB1 and RELA were activated in aged tissue ( Figure 4D). The fold change relationship between genes with sustained NFκB activity in vitro (IκBα KD+TNFα vs. control MCF-7) and those showing aging in vivo (aged vs. young mouse hearts) revealed that in MCF-7 cells, the expression of cluster 2 genes, which contain cellular senescence-promoting and SASP genes, was correlated with the expression of age-related genes in the heart (R = 0.31, P < 0.0001) ( Figure 4E). In addition, we examined the relationship between the number of NFκB binding sites and the increase in gene expression with aging in the mouse heart. After predicting putative NFκB targets within ± 500 bp from the transcription start site (TSS) using the ChIP-Atlas 32 (see the Methods), we classified the genes into four groups (no, low, medium, and high upregulation) according to the degree of fold change in expression (aged vs. young mice) ( Figure 4F). Counting NFκB binding sites in the flanking regions (within ± 10,000 bp from TSS) of the genes using HOMER 33 revealed that the "high" expression group included a significantly higher 6 number of NFκB binding motifs (P < 0.0001, high vs. no upregulation) ( Figure 4G). Gene Ontology (GO) analysis of the high group genes in the aged heart ( Figure 4H) showed a high enrichment of inflammatory signals, which was similar to the GO results for the cluster 2 gene set in MCF-7 cells ( Figure 3F). These results suggested that sustained NFκB activity might contribute to inflammatory progression in the hearts of aged mice by upregulating the expression of senescence-related genes with increased DNA-NFκB binding, and this NFκB-mediated transcriptional regulation is common between aging tissue and the senescent cells mimicked by IκBα KD+TNFα.
Next, we compared the changes in the signaling and downstream cascade between TNFα and TNFα+IκBα KD and the relationship with cellular senescence. First, we focused on the decreased expression of Cyclin D1 by IκBα KD+TNFα in MCF7 cells because it is related to the suppression of cell proliferation ( Figure 1H). Cyclin D1 plays an important role in cell cycle entry at G1/S, and its expression and stability are regulated by multiple signaling pathways, including ErbB, MEK-ERK, PI3K-AKT, and NFκB 34 . Therefore, we examined the effect of those pathway inhibitors (Figures 5A and S5) and found that AKT inhibitor VIII (AKTi-VIII; AKT1/2 inhibitor), SB203580 (inhibitor for p38 and AKT), and BAY-117085 (NFκB inhibitor) abrogated the changes in Cyclin D1 expression under IκBα KD+TNFα conditions ( Figure 5A). The AKT and NFκB pathways are known to crosstalk with each other 35,36 . AKT is important for the translation and stability of the Cyclin D1 protein 37 and is also shown to be activated in G2 phase via the mTOR complex 2 (mTORC2) cascade 38 ; therefore, we analyzed the signaling pathway involved in AKT. As a result, we found that the expression of RICTOR, P-RICTOR, and P-AKT (Ser473) in the mTORC2 cascade was significantly reduced by IκBα KD+TNFα ( Figure 5B). Moreover, AKTi-VIII treatment increased SA-β-gal positive cells and abolished IκBα KD+TNFα effects on SA-β-gal staining ( Figure 5C). We also found that RICTOR KD increased SA-β-gal staining and decreased Cyclin D1 expression, as shown under the IκBα KD+TNFα condition ( Figure 5D). These results suggested that sustained NFκB activity, which is mimicked by TNFα+IκBα KD, suppresses mTORC2/AKT signaling and Cyclin D1 expression, thereby contributing to promoting cellular senescence. Moreover, the MEK inhibitor U0126 partially abolished the effect of IκBα KD+TNFα on Cyclin D1 downregulation ( Figure 5A), and IκBα KD+TNFα slightly reduced the P-ERK level ( Figure 5B). Since ERK activity is important for inducing CCND1 39 , downregulating the Cyclin D1 protein by IκBα KD+TNFα might be elicited via suppression of both the mTORC2/AKT and ERK signals and their crosstalk.

Purine catabolism is induced by sustained NFκB activity.
mTORC2 is a critical factor in metabolic regulation 40 . Since metabolic dysregulation is a hallmark of cellular senescence and aging 23 , we investigated metabolic alterations induced by sustained NFκB activation. We performed a metabolome analysis of MCF-7 cells under IκBα KD+TNFα conditions; however, principal component analysis showed similar intracellular metabolic profiles between TNFα and IκBα KD+TNFα ( Figure S6A and Supplementary Data S1). TNFα alters several metabolic pathways, including glycolysis, lipid metabolism, and amino acid metabolism 41-43 , which might mask metabolic changes associated with cellular senescence under IκBα KD+TNFα conditions. Since IκBα KD alone moderately reduced mTORC2 activity 7 ( Figure 5B), we focused on metabolic alterations induced by IκBα KD. The results showed that 62 and 35 metabolite levels were up-and downregulated, respectively ( Figure 6A). Notably, the intracellular level of hypoxanthine, a potential oxygen generator 44 , in the purine catabolic pathway was significantly elevated by IκBα KD (P < 0. 0001, FC = 3.126, vs. control), and its effects were enhanced by TNFα (P < 0. 0001, FC = 4.711, vs. control) ( Figure 6B). The extracellular hypoxanthine level in the culture medium was also increased by IκBα KD+TNFα (P < 0. 0001, FC = 17.69, vs. control) ( Figures S6B and S6C). Pathway enrichment analysis of metabolites altered by IκBα KD also confirmed the global changes in the purine metabolic pathway ( Figure 6C). Mapping of intracellular metabolite changes induced by IκBα KD showed an increase in metabolites of the catabolic pathway of nucleotides (e.g., hypoxanthine, xanthine) and a decrease in high-energy molecules (e.g., ATP and GTP) ( Figure 6D). Biosynthesis of purine nucleotides has been reported to increase the cellular growth rate and promote the G1/S transition 45 . Moreover, an imbalance of nucleotide species suppresses cell proliferation, but it still causes a transition to S phase via DNA replication stress signaling 46 . Therefore, our results suggest that IκBα KD cells inhibit normal cell cycle progression by inducing a nucleotide imbalance, including high-energy nucleotides.
More precisely, a balance between the synthesis and degradation of purine nucleotides determines the cellular level of adenylate and guanylate 45 . Two pathways have been identified for purine nucleotides synthesis: the de novo pathway and salvage pathway 47 . Detailed analysis of mRNA and protein levels in IκBα KD cells validated the promotion of purine catabolism, including the increased ecto-5'-nucleotidase (NT5E) level and decreased hypoxanthine phosphoribosyltransferase (HPRT) 1 level, which were enhanced by TNFα ( Figures 6E and  S6D). In the purine salvage pathway, NT5E catalyzes the hydrolysis of adenosine monophosphate (AMP) to adenosine 48 , and HPRT1 catalyzes the conversion of hypoxanthine and guanine to inosine monophosphate (IMP) and monophosphate (GMP), respectively 49 . Notably, several other genes and enzymes related to purine de novo synthesis were also downregulated by IκBα KD+TNFα ( Figure S6D). To address whether upregulation of purine catabolism is one of the causes of the progression of cellular senescence, we next tested the effect of NT5E and HPRT1 inhibition. As predicted, the NT5E inhibitor adenosine 5'-(α, β methylene) diphosphate (APCP) attenuated the increase in SA-β-gal positive cells by IκBα KD+TNFα while siRNA-mediated HPRT1 KD augmented SA-β-gal staining ( Figures 6F, 6G, and S6E). Furthermore, RICTOR knockdown decreased HPRT1 expression but did not affect NT5E expression ( Figure 6H), indicating that HPRT1 expression is controlled by RICTOR. MCF-7 cells under the IκBα KD+TNFα condition downregulated Cyclin D1 expression and prolonged the cell cycle ( Figures 1H and 1I). This may be due to the depletion of nucleotides and high-energy metabolites, such as ATP, GTP, dATP, and dGTP, which are necessary for protein translation and cell cycle progression 47 .
Next, we examined whether similar changes in purine catabolism mediated by NFκB pathway are also observed in the hearts of aged mice in vivo. Protein expression of RICTOR and HPRT in aged hearts was significantly lower than that in young hearts ( Figures 7A 7B, and S7A). The protein levels of RICTOR (R = 0.7676, P < 0.001) and HPRT (R = 0.7676, P < 0.001) were highly correlated with the IκBα levels in the hearts of individual young and aged mice, respectively (Figures 7A, right and 7B, right). Metabolome analysis of the hearts of aged mice 8 also showed enhanced purine catabolic metabolism (xanthosine, xanthine, and GMP) and reduced high-energy molecules (ATP and ADP) ( Figures 7C, S7B, and Supplementary Data 2). Interestingly, mass spectrometry imaging revealed that ATP levels in the hearts of aged mice were markedly reduced in the entire tissue ( Figures 7D and S7C). As the aged heart tissues exhibited a higher amount of NFκB and lower amount of IκBα ( Figures 4A and 4B), these metabolic features presumably represent a mechanism underlying the ability of highly nuclearlocalized NFκB to interfere with the synthesis of high-energy metabolites.

DISCUSSION
The duration of nuclear NFκB activity has various effects on gene expression by regulating mRNA half-life or epigenetic status 20,21 . In this study, we show that the aging-associated phenotypic changes could be explained by a core regulatory mechanism stemming from sustained nuclear NFκB activity. That is, increased nuclear NFκB enhances NFκB-DNA binding for inflammatory gene expression, alters the mTORC2/AKT pathway to promote purine catabolism, and inhibits cell cycle progression.
Our study showed that the loss of two negative feedback regulators, namely, IκBα and A20, significantly elevated the nuclear abundance of NFκB and induction of cell senescence. However, a combination of IκBα KD and A20 OE did not strongly alter the NFκB dynamics from sustained to oscillatory but rather reduced the amplitude of nuclear NFκB activity and the expression of IL8 and CCL2 genes, which were upregulated by IκBα KD alone ( Figures 2C and  2H). These results suggest that the nuclear NFκB level is critical for inflammatory aging. Indeed, the decrease in IκBα protein levels under oxidative stress in MCF-7 cells ( Figure S7D) indicates that decreases in IκBα may occur over a lifetime for the same reason. Meanwhile, previous studies have demonstrated that defects in the feedback regulation of IκBα in the NFκB pathway cause Sjögren's syndrome and shortened life span 50 and that mice expressing vascular-specific IκBα overexpression have a prolonged life span 51 . However, further studies are needed to determine the mechanism between sustained NFκB activity and its effect on life span.
Notably, inhibition of purine nucleoside phosphorylase, which is one of the rate-limiting enzymes for purine catabolism, suppresses cellular senescence in vivo 52 and promotes the synthesis of nicotinamide adenine dinucleotide, an anti-aging molecule 53 . On the other hand, hypoxanthine, a purine catabolic metabolite and a common DNA damage base that has mutagenic potential due to its function as a potential radical generator 44,54 , was significantly elevated under the IκBα KD+TNFα condition. Therefore, hypoxanthine may be a key molecule in NFκB-associated cell senescence and likely contributes to accelerating the aging process.
Our study showed that the cell cycle period is prolonged during the G2/M phase under IκBα KD+TNFα. However, issues regarding the G2/M delay mediated by metabolic imbalance remain unresolved. Although the shortage of high-energy metabolites and nucleotides severely affect the G1/S transition, the effect of downregulating purine metabolism at the G2/M phase was not clear 45 . An earlier study showed that NFκB activation in response to ionizing radiation or etoposide arrested cells at the G2/M phase for a prolonged time 55 . Since hypoxanthine potentially 9 generates free radicals, G2/M cell cycle arrest under IκBα KD+TNFα may involve hypoxanthine upregulation. In addition, mTORC2 and AKT activity are required for G2/M cell cycle progression 38,55 , and our data show that RICTOR, P-RICTOR, and P-AKT (Ser473) expression in the mTORC2 cascade was significantly reduced by IκBα KD+TNFα. An imbalance of nucleotide species has been reported to suppress cell proliferation, although the cells still pass through S phase by the activation of DNA replication stress signaling 46 . Therefore, inhibiting or downregulating the cell cycle speed under IκBα KD+TNFα may be caused by either upregulation of hypoxanthine via HPRT1 and mTORC2 or downregulation of purine metabolism for high-energy metabolite synthesis. Whether another mechanism underlies the cell cycle regulation observed in this study requires further investigation. Identifying a mechanism that can distinguish between cellular senescence and cell death induced by TNFα-dependent CDK regulation would be also important in future study 56 . Overall, in this study, we show that temporal NFκB dynamics affect an unexpectedly broad spectrum of cell physiological factors and regulate cellular senescence. These findings provide unprecedented insights into the molecular basis of cellular senescence.

Mouse experiments
All mice were purchased from the Japan SLC and maintained under specific pathogen-free conditions at Graduate School of Frontier Biosciences Osaka University. Female 8-weeks-old (young) and 69-73-weeks-old (aged) C57BL/6 N mice were used in experiments. All animals were separately caged with a 12:12-h light-dark cycle and had free access to water and chow throughout the study. The excised tissues were immediately frozen in liquid nitrogen and stored at −80°C until use. All experiments were approved by the Animal Care and Use Committee of Osaka University and were performed following institutional guidelines.

Quantitative RT-PCR (qRT-PCR) analysis
Real-time PCR analysis was conducted as described previously 57 . Briefly, RNA was isolated from cells or tissues using NucleoSpin RNA Plus (Macherey-Nagel, Duren, Germany, Cat. No. 740984.250) and subjected to cDNA synthesis using ReverTra Ace qPCR RT Master Mix (TOYOBO, Osaka, Japan) and quantitative PCR using a KOD SYBR qPCR kit (TOYOBO) with CFX96 Real-Time PCR System (Bio-Rad Laboratories, Hercules, CA, USA) according to the manufacturer's protocol. The Δ Δ Cq method was used to quantify gene expression, using RPL27 or Rplp0 expression as an internal reference 58 . All experiments were performed in triplicate. The primers used for real-time PCR are listed in Supplementary Table S4.

Western blotting
Immunoblot analysis was performed as described previously 57  For the separation of nuclear fraction, the nuclear and cytoplasmic protein extraction kit (Biovision Inc., Milpitas Boulevard, Milpitas, CA, USA, K266) was used according to the manufacturer's procedures.

RELA immunocytochemistry
Cells were fixed with 4% paraformaldehyde (Thermo Fisher Scientific) in phosphate buffered saline (PBS) for 15 min, rinsed with PBS, and permeabilized with 0.2% Triton X-100 (Nakalai Tesque) in PBS for 10 min. Cells were blocked using 1% goat serum Thermo Fisher Scientific, #16210-064) in PBS for 15 min and incubated with RELA antibodies (1:1600; Cell Signaling Technology, #8242) diluted in 1% goat serum/PBS at 4 °C overnight. After rinsing in PBS, cells were incubated with Alexa Fluor 488-conjugated secondary antibodies (1:2000; Thermo Fisher Scientific, A32731), 0.125 ng/mL CellMask TM (Thermo Fisher Scientific) as a cell body stain, and 1 µg/mL 4',6-diamidino-2-phenylindole (DAPI; Nacalai Tesque) as a nuclear stain for 1 h at 25 °C in the dark. Fluorescent images were acquired using IN Cell Analyzer 2500HS (Cytiva, Preston, UK) at 20× magnification. Images of 30 fields were taken for each condition. CellProfiler (ver. 3. 1. 9) was used to segment cellular regions from CellMask TM images, to segment nuclear regions from DAPI images, and to quantify RELA signal intensities for each cell 59 . The nuclear-to-cytoplasm signal ratio for RELA was calculated based on the integrated signal density.

Analysis of cell morphology
Cell morphology analysis was conducted as described previously 57 . Briefly, Cells were fixed using 4 % paraformaldehyde for 15 min and permeabilized with 0.1 % Triton-X100 for 5 min. Cell membrane and nucleus were stained with 0.125 ng/mL CellMask TM (Thermo Fisher Scientific) and 1 μ g/mL DAPI (Nacalai Tesque) for 30 min at 25 °C in the dark, respectively. Fluorescent images were acquired using IN Cell Analyzer 2500HS (Cytiva, Preston, UK) at 20× magnification. CellProfiler (ver. 3. 1. 9) was used to extract geometrical features from the cell images 59 . DAPI and CellMask TM staining was used to visualize the segmentation of nucleus and cell body, respectively. The following morphological parameters of the cells were evaluated. Area, Form Factor, Solidity, Extent, Orientation, Eccentricity, Compactness, major axis length, minor axis length, maximum ferret diameter, minimum ferret diameter, Perimeter, and Mean radius (total 13 parameters). PCA based on the 13 parameters was performed by the "prcomp" function in the "stats" R package.

β -galactosidase (SA-β-gal) staining
The senescent cells were stained by a senescence β -galactosidase staining Kit (#9860, Cell Signaling Technology), according to the manufacturer's instructions. Briefly, cells in 12-well plates were washed with PBS, fixed for 10 min in a fixative solution, washed twice with PBS, and incubated in SA-β-gal staining solution overnight at 37 °C. Images were acquired with a Keyence BZ-9000 microscope on 20× magnification. Four random views per well were selected to count the number of cells stained positive for SA-β-gal. Each independent experiment was repeated three times and the percentage of SA-β-gal positive cells per field from a total of 12 fields was analyzed with Image J software. DAPI staining was used to identify nuclei and count the number of cells per field. The threshold for identifying senescent cells was set by a positive control (etoposide treatment).

Cell-viability assay
Cell viability was measured using trypan blue exclusion assay. A 2 × 10 5 cells/well was seeded in 12-well plates and incubated at 37 °C. After treatment with reagents and/or siRNAs, cells were disaggregated in 400 μ L medium, and 10 μ L of the suspension was mixed with 10 μ L trypan blue (Thermo Fisher Scientific). Viable cells were counted using a Countess Automated Cell Counter (Thermo Fisher Scientific).
Cell cycle analysis using the FUCCI biosensor MCF-7 cells stably expressing the FUCCI biosensor were obtained by the PiggyBac transposase system 60 . MCF-7 cells were transfected with the pPBbsr2-H2B-iRFP-P2A-mScarlet-I-hGem-P2A-PIP-tag-NLS-mNeonGreen (https://benchling.com/s/seq-LPZ1tLdpgnpJIYOG2ujR) and the hyPBase (https://benchling.com/s/seq-oGkw53b41IZqvzF5yQ9K) 61 plasmids using Lipofectamine LTX following manufacturer's protocol (Thermo Fisher Scientific). The plasmids, pPBbsr2-H2B-iRFP-P2A-mScarlet-I-hGem-P2A-PIP-tag-NLS-mNeonGreen and hyPBase, were a gift from Dr. K. Aoki (National Institute for Basic Biology, National Institutes of Natural Sciences). After transfection, cells were released into a drug-free medium for 48 hr followed by blasticidin (10 µg/mL) selection until single colonies were formed. Single clones were expanded, and the FUCCI biosensor signal was confirmed by fluorescent microscopy. Time-lapse microscopy images were acquired using an IN Cell Analyzer 2500HS (GE Healthcare Life Science) with a CFI S Plan Fluor ELWD dry objective lens 20× (NA: 0.45) using an excitation wavelength of 475 nm and emission wavelength of 511 nm for PIP-tag-mNeonGreen and excitation wavelength of 542 nm and emission wavelength of 597 nm for mScarlet-hGem. Cells were maintained at 37 °C in a humidified atmosphere of 5% CO 2 and imaged at 20 min intervals for 60 hr. Image processing and quantifications for a percentage of cells in each cell cycle phase were performed using tools in ImageJ-Fiji software (version 1.53t, NIH). Masks of mScarlet-hGem and PIP-tag-mNeonGreen signals were obtained by applying a fixed threshold (160, 65535). The obtained masks were identified with 'Analyze Particles' (size = 10.00-infinity, circularity = 0.00-1.00) and automatically counted as ROIs (region of interest). Cell cycle phases were determined by G1-phase: PIP-tag positive, S-phase: hGem positive, G2/M-phase: PIP-tag and hGem positive. The average fluorescence brightness of mScarlet and mNeonGreen was obtained by cell tracking software (Olympus). The threshold parameters of mScarlet and mNeonGreen for each cell were manually set. Subsequently, an algorithm for estimating the cell cycle phase assembled with R packages [brunnermunzel (v.1.4.1), multimode (v.1.4), rlist (v.0.4.6.1), tidyverse (v.1.3.0)] was applied to determine each cell cycle phase at the single-cell level. For quantifications of the G2/M phase length from single-cell data, measurements within 24-60 hr were used when the effects of TNFα were observed.

Generation of IκBα KO MCF-7 cells
IκBα knockout (KO) MCF-7 cells were generated by CRISPR-Cas9 technology. We designed three independent guide RNAs targeting IκBα (NFKBIA) with the gRNA design tool (https://chopchop.cbu.uib.no/) 62 . Two gRNAs were designed to target exon 1 and one gRNA to target intron 1 of NFKBIA (Supplementary Table S5). Primers were then designed by adding guanosine base on the 5' and sticky ends corresponding to BbsI site. The pair of primers were then annealed and ligated into BbsI-digested eCas9-P2A-Puro plasmid 63 . MCF-7 cells were transfected with IκBα gRNA-containing eCas9-P2A-Puro plasmids using Lipofectamine LTX following the protocol of the manufacturer (Thermo Fisher Scientific). After transfection, cells were released into a drug-free medium for 48 h followed by puromycin (1 µg/mL) selection until single colonies were formed. Single clones were expanded, and gene deletion was confirmed by western blotting. eSpCas9(1.1)-T2A-Puro was a gift from Andrea Németh (Addgene, # 101039).

Generation of A20-overexpressing MCF-7 cells
To establish cell lines stably overexpressing A20, the PiggyBac transposase system was used 60 . The plasmid used for A20 overexpression, pPB[Exp]-EGFP/Puro-CMV>hTNFAIP3, was constructed with VectorBuilder (Kanagawa, Japan). The vector ID is VB220607-1391zue and more information can be obtained from vectorbuilder.com. MCF-7 cells were transfected with the pPB[Exp]-EGFP/Puro-CMV>hTNFAIP3 and hyPBase 61 plasmids using Lipofectamine LTX following the manufacturer's protocol. (Thermo Fisher Scientific). After transfection, cells were released into drug-free medium for 48 h followed by puromycin (1 µg/mL) selection until single colonies were formed. Single clones were expanded, and expression of A20 was confirmed by western blotting.

Immunohistochemistry of RELA in heart tissue from aged mice
Immunohistochemistry was performed at Genostaff (Tokyo, Japan). Tissue sections were fixed with 4% paraformaldehyde for 10 min. Endogenous peroxidase was blocked with 0.3% H 2 O 2 in methanol for 30 min, followed by incubation with G-Block (Genostaff) and avidin/biotin blocking kit (Vector Laboratories, Peterborough, UK). The sections were incubated with anti-REL A rabbit monoclonal antibody (0.004 µg/mL, Cell Signaling Technology, #8242) at 4 overnight and then incubated with biotin-conjugated anti-rabbit IgG (Vector Laboratories) for 30 min at room temperature. Subsequently, peroxidase-conjugated streptavidin (Nichirei, Tokyo, Japan) was added for 5 min, and peroxidase activity was visualized using diaminobenzidine. The sections were counterstained with Mayer's Hematoxylin (Muto Pure Chemicals Co., Ltd., Tokyo, Japan), dehydrated, and then mounted with Malinol (Muto Pure Chemicals Co., Ltd.).
Stained slides were digitally scanned using a NanozoomerS210 (Hamamatsu Photonics, Hamamatsu, Japan). For histopathological assessment, five random digital images of each mouse heart tissue were captured at 40× magnification using NDP view2 software (Hamamatsu Photonics). RELA-positive staining in nuclei was quantified by the ImmunotRatio ImageJ plugin 17 64 . The brown (RELA) and blue (hematoxylin) threshold adjustments were set at 10 and 15, respectively.

Metabolome analysis
Metabolome analysis was conducted as described previously 65 . Briefly, the cell samples (approximately 1 × 10 6 cells), conditioned media (50 µL), and heart tissues (approximately 30 mg) were mixed with 1 mL of cold methanol containing 10-camphorsulfonic acid (1.5 nmol) and piperazine-1,4-bis (2-ethanesulfonic acid) (1.5 nmol) as internal standards. Each sample was vigorously mixed by vortexing for 1 min, followed by 5 min of sonication. The extracts were then centrifuged at 16,000 × g for 5 min at 4 °C, and the resultant supernatant (400 μ L) was collected. Protein concentrations in the pellet were determined using a Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific, MA). After mixing 400 μ L of supernatant with 400 μ L of chloroform and 320 μ L of water, the aqueous and organic layers were separated by vortexing and subsequent centrifugation at 16,000 × g and 4 °C for 5 min. The aqueous (upper) layer (500 μ L) was transferred into a clean tube. After the aqueous layer extracts were evaporated under vacuum, the dried extracts were stored at −80 °C until the analysis of hydrophilic metabolites. Before analysis, the dried aqueous layer was reconstituted in 50 μ L of water.
Two liquid chromatography high-resolution tandem mass spectrometry (LC/MS/MS) methods for hydrophilic metabolite analysis were employed 65,66 . Anionic polar metabolites (i.e., organic acids, sugar phosphates, nucleotides, etc.) were analyzed via ion chromatography (Dionex ICS-5000+ HPIC system, Thermo Fisher Scientific) with a Dionex IonPac AS11-HC-4 μ m column (2 μ m i.d. × 250 mm, 4 μ m particle size, Thermo Fisher Scientific) coupled with a Q Exactive highperformance benchtop quadrupole Orbitrap mass spectrometer (Thermo Fisher Scientific) (IC/MS/MS). Cationic polar metabolites (i.e., amino acids, bases, nucleosides, etc.) were analyzed via liquid chromatography (Nexera X2 UHPLC system, Shimadzu) with a Discovery HS F5 column (2.1 mm i.d. × 150 mm, 3 μ m particle size, Merck) coupled with a Q Exactive instrument (PFPP-LC/MS/MS). The two analytical platforms for hydrophilic metabolite analysis were controlled using LabSolutions, version 5.80 (Shimadzu) and Xcalibur 4.2.47 (Thermo Fisher Scientific). Metabolome analyses and data processing were performed at the Division of Metabolomics of the Medical Institute of Bioregulation at Kyushu University. Quantitative levels of metabolites were calculated using peak areas relative to an internal standard (10camphorsulfonic) and corrected for the total protein amount (analysis for MCF7 cells) or weight (analysis for mouse heart tissues) of each sample.

Analysis for ATP imaging in the mouse heart using MALDI-MSI Standard sample analysis
To detect ATP, 9AA was used as a matrix. The 9AA was dissolved in distilled water and methanol (3:7, v/v) at a concentration of 5 mg/mL. Each concentration standard solution and matrix solution were equivalent volume mixed. The mixture was spotted on ITO-coated glass slides, dried, and measured.

MALDI-MSI analysis
Frozen serial 8 μ m sections of mice heart tissues were cut at −20°C with a Microtome (CM1950; Leica, Nussloch, Germany) and mounted on ITO-coated glass slides for MALDI-MSI and on coated glass slides for HE staining. The sections were dehydrated in a 50 ml conical tube containing silica gel and stored at −80°C.
The prepared matrix solution was applied over 200 µL/section (9AA) with an airbrush (GSI Creos, Tokyo, Japan). After spraying, the samples were immediately measured.
MALDI-MSI experiment was performed on a MALDI ion trap time-of-flight mass spectrometer (iMScope TRIO; Shimadzu, Kyoto, Japan) equipped with a 1-kHz Nd:YAG laser (λ = 355 nm). The laser spot size was approximately 15 µm, and each pixel was irradiated 80 times at a repetition rate of 1 kHz. For m/z calibration, Polyethylene Glycol 600 Sulfate (Tokyo Chemical Industry, Tokyo, Japan) was used. Mass spectra were acquired in the negative ion detection mode. The target m/z values were 505.99 derived from [M-H] -. In the imaging experiment, the interval of data points was 40 µm in the lateral and axial directions. After sample analysis, ion images were reconstructed based on data extracted from m/z ranges of the target m/z values ± 0.02 Da using Imaging MS solution (Shimadzu, Kyoto, Japan). Signal intensities from region-of-interest were extracted using Imaging MS solution.

Mathematical Model
We constructed the mathematical model of TNFR signaling pathway by extending the previous model 67 . A comprehensive diagram of the TNFα signaling network is shown in Figure. S3a. The mathematical model describes the biochemical reactions using ordinary differential equations, from activation of TNF receptors following transcription of negative feedback regulators (IκBα and A20) by NFκB. The model constitutes 17

The cosine similarity between
, is defined as where CS is cosine similarity, n = 3 is the number of molecules used in parameter estimation (IKK, IκBα, and NFκB), is a vector for the activity of the molecules at each time point in the simulation, is vector for activities of the molecule at each experimental time point. x is the inner product of , and | · | denotes the magnitude of a vector. A low cosine similarity means that the dynamic behaviors of 19 the molecules in simulation are similar to the ones in the experiment. We used 30 parameter sets for our model simulation and analysis.

Sensitivity analysis
The sensitivity coefficient of each reaction parameter, S i , is defined as where p i is an i-th parameter, P is parameter vector P = (p 1 , p 2 , …), and q(P) is a target function, for example, sum of signals, and oscillatory amplitude and period. The sum of amplitudes of peaks, and average of periods from the first to the second were also used as target functions for the analysis. The sensitivity was calculated by increasing the parameter values by 1%.

RNA-sequencing (RNA-seq) of MCF7 cells and mouse heart tissues
Total RNA was isolated using NucleoSpin RNA Plus (Macherey-Nagel, Duren, Germany, Cat. No. 740984.250). Library construction and RNA seq were performed using the Illumina sequencing platform (Rhelixa). Poly(A) RNA was prepared with the Poly(A) mRNA Magnetic Isolation Module (New England Biolabs, Ipswich, MA, USA), the sequencing library was generated with the NEBNext® Ultra™II Directional RNA Library Prep Kit (New England Biolabs) and paired-end sequencing was processed with the NovaSeq 6000 (Illumina, San Diego, CA).

Acquisition of time course RNA-seq data
Time course bulk RNA-seq data sets (DRA011742 and DRA011743 of DNA Data Bank of Japan) in which gene expression was measured every 15 min after TNFα stimulation in IκBα KD MCF-7 cells were obtained from our previous study 19 .

MCF-7 RNA-seq data analysis
For analysis of MCF7 RNA-seq, sequencing reads were preprocessed using nf-core/rnaseq pipeline (v.3.5) 69 (https://github.com/nf-core/rnaseq/blob/1.4.2/docs/output.md). Read quality was assessed using FastQC (v.0.11.9). Trim Galore (v.0.6.7) was used for adaptor trimming. The resulting reads were mapped onto the GRCh38 genome using STAR (v.2.7.6a) and then quantified by Salmon (v.1.5.2). Protein coding genes registered in GENCODE (https://www.gencodegenes.org/human/release_40.html) were used for downstream analysis. Differentially Expressed Genes (DEGs) were identified using the R package DESeq2 70 by performing a Wald significance test between gene expression levels in each condition 48 h after TNFα stimulation and calculating adjusted P-values and log 2 fold change (TNFα vs IκBαKD+TNFα). Genes with adjusted P-value < 0.01 and log 2 fold change > 0.4 were classified as upregulated DEGs, and genes with adjusted P-value < 0.01 and log 2 fold change < −0.4 as downregulated DEGs. The expression pattern of each upregulated DEG in Figure 3B was obtained using time course TPM data. Time course TPM data were Z-scored after the logarithm of 2 was taken before clustering. Clustering was performed using the partitioning around medoids (PAM) algorithm with the R package cluster. Heat maps were created using the R package complexHeatmap 71 . The creation of the regression line and the calculation of the 95% confidence intervals were calculated by the function "stat_smooth()" of the R package ggplot2 72 .

Mouse heart RNA-seq data analysis
Mapping of data to perform Gene/transcript quantification was performed using nf-core/rnaseq pipeline (v3.5.) and the default GRCm38 genome. For TF activity analysis, the DOROTHEA 73 Regulon database was used in combination with VIPER 74 . Only TFs with a confidence score of A and interacting with at least 30 genes were included. Genes that were significantly upregulated with aging were identified using the R package DESeq2 70 . Criterion was set as adjusted P-values < 0.01 and log 2 fold change > 0.
Assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) Library preparation, sequencing, and mapping for ATAC-seq were performed at DNAFORM (Yokohama, Japan). Fragmentation and amplification of ATAC-seq libraries were conducted as described previously 75 . Briefly, approximately 5×10 4 cells were lysed and transposed reactions were carried out using Tn5 Transposase (Illumina, #FC121-1030) at 37 °C for 30 min. The reaction solution was purified with the Qiagen MinElute PCR Purification Kit. PCR with custom Nextera PCR primers was then performed for five cycles using NEBNext Q5 Hot Start HiFi PCR Master Mix (New England Biolabs, Ipswich, MA, USA) 76 . The number of additional PCR cycles was determined by qPCR of partially amplified products 76 . The PCR products were purified using Agencourt AMPure XP beads (Beckman Coulter: A63881) by double size selection (left ratio: 1.4x, right ratio: 0.5x) following the manufacturer's protocol. Paired-end sequencing was performed on the Illumina HiSeq sequencer.

MCF7 ATAC-seq data analysis
Mapping and peak calls were conducted by ENCODE ATAC-seq pipeline (https://github.com/ENCODE-DCC/atac-seq-pipeline . Peak annotation was conducted by HOMER (v.4.9.1) with default settings. Changes in chromatin accessibility near the transcription start site (TSS ± 5000 bp) per condition in each region and the list of differential accessible peaks that were differentially accessible under IκBα KD+TNFα conditions compared to controls were obtained using DEseq2 (v.1.20.0). The criteria for determining a peak with a difference in accessibility were adjusted P-value < 0.05 and log 2 fold change > 0.

RELA chromatin immunoprecipitation-sequencing (ChIP-seq)
ChIP was carried out using the Simple ChIP Enzymatic Chromatin IP Kit (Cell Signaling Technology, #9005) following the manufacturer's protocol. Protein/DNA was crosslinked using 1% formaldehyde. Nuclei were isolated and sheared using an ultrasonicator (Epishear probe sonicator, Active Motif, Carlsbad, CA, USA). The sheared chromatin was incubated with RELA antibody (Abcam, ab7970) and Dynabeads Protein G beads (10004D, Thermo Fisher Scientific). After elution of ChIP products, DNA was purified using MinElute PCR Purification Kit (QIAGEN, 28004), the library was prepared using the TruSeq ChIP Library Preparation Kit (Illumina), and 36-bp single-end sequencing of the libraries was processed with a HiSeq3000 system (Illumina). 21 Mapping of data to produce BigWig files and call peaks was performed using the nf-core/chipseq pipeline (v.1.2.2) 69 (https://doi.org/10.5281/zenodo.3240506) and default GRCh38 genome. BigWig files were quantified by "computeMatrix" and visualized by "plotHeatmap" and "plotPlofile" in deeptools (v3.5.1) 77 . Each peak was assigned to the TSS of the closest gene by HOMER (v.4.11) 78 . The RELA target gene was defined as the gene whose peak was within ± 5,000 bp from TSS and They were extracted from each cluster. The number of RELA peaks annotated to each target gene was counted using R.

RELA binding site analysis using public databases
The ChIP Atlas peak browser 79 was used to identify RELA binding regions in the mouse genome (mm10). The ChIP-seq peak significance threshold was set to 500 (q value < 1E-50) and ChIP-seq peaks from the groups "TFs and other" and "All cell types" and "RELA" were used. Multiple RELA ChIP peaks in a single region were merged by mergeBed in deeptools (v.3.5.1) 80 . These RELA ChIP peaks were then assigned to the TSS of the closest gene by HOMER (v.4.11). The final RELA target gene was defined as the gene whose Rela peak was within 500 bp of the TSS. The number of RELA peaks within ±10,000 bp from the TSS annotated to each target gene was then counted on R.
KEGG enrichment analysis KEGG enrichment was compared by the "comparecluster()" function in clusterProfiler 81 R package with "fun = "enrichKEGG". The result was plotted by the R package "complexHeatmap". Terms with adjusted P-values less than 0.05 were displayed.
Data and materials availability: Metabolome data are included in the Supplementary Data. The sequence data for RNA-seq, ChIP-seq, and ATAC-seq reported in this paper have been deposited in the DNA Data Bank of Japan with accession number DRA015837. Code is available at https://github.com/okadalabipr. All other data will be available upon reasonable request.

QUANTIFICATION AND STATISTICAL ANALYSIS
We performed all experiments at least twice and confirmed similar results. Statistical analyses were performed using GraphPad Prism ver. 8.0 software (GraphPad Software, Inc., La Jolla, CA, USA) and R studio ver. 4.2.2 (Rstudio, Boston, MA, USA). For in vitro experiments, data from two or more groups were analyzed using Student's t-tests and one-way analysis of variance (ANOVA), respectively. For in vivo experiments, data from two or more groups were analyzed using the Mann-Whitney U and Kruskal-Wallis tests, respectively. Data represented as mean ± SEM or ± SD; P values < 0.05 were considered statistically significant. Figure S1. Effect of IκBα knockdown/knockout on cellular senescence (related to Figure 1).