CSB promoter downregulation via histone H3 hypoacetylation is an early determinant of replicative senescence

Cellular senescence has causative links with ageing and age-related diseases, however, it remains unclear if progeroid factors cause senescence in normal cells. Here, we show that depletion of CSB, a protein mutated in progeroid Cockayne syndrome (CS), is the earliest known trigger of p21-dependent replicative senescence. CSB depletion promotes overexpression of the HTRA3 protease resulting in mitochondrial impairments, which are causally linked to CS pathological phenotypes. The CSB promoter is downregulated by histone H3 hypoacetylation during DNA damage-response. Mechanistically, CSB binds to the p21 promoter thereby downregulating its transcription and blocking replicative senescence in a p53-independent manner. This activity of CSB is independent of its role in the repair of UV-induced DNA damage. HTRA3 accumulation and senescence are partially rescued upon reduction of oxidative/nitrosative stress. These findings establish a CSB/p21 axis that acts as a barrier to replicative senescence, and link a progeroid factor with the process of regular ageing in human.

S enescence, which is a process that limits proliferation of damaged cells in response to various types of stress, has been associated with organismal ageing 1 . Accumulation of senescent cells in tissues, as well as withdrawal of senescent stem cells from the regeneration process, are marks of tissue deterioration linked to ageing 2 and may contribute to organ degeneration and aged-related diseases. Senescence also acts as a barrier to cancer 3 and plays a physiological role in normal development and tissue homeostasis 4-6 . Senescence is characterized by a proliferative arrest of metabolically active cells 7 , which is distinct from other forms of cell cycle arrest (e.g., quiescence) by its essentially irreversible nature, and by acquisition of specific features, including increased cell size, β-galactosidase activity, persistent DNA damage-response (DDR), presence of heterochromatin foci, and highly active secretory activity (senescence-associated secretory phenotype, or SASP) 8 . Senescence triggered by replication-induced DNA damage, or telomere shortening that determine a persistent DDR, and resulting in the stabilization of the transcription factor p53 and expression of the cyclin-dependent kinase inhibitor p21 (coded by p21 Waf1 ) 9,10 , is called replicative senescence. Stimuli such as genomic and epigenomic damage, oxidative and proteotoxic stress, and mitogenic signals that engage either the DDRdependent (p53/p21 Waf1 ) pathway or activation of the CDKN2 locus through expression of the tumor suppressor p16 (encoded by p16 Ink4 ) also induce senescence 11 . High levels of p16 are observed late during replicative senescence following the p21 trigger, probably to maintain the senescent state 12 .
Senescence allows tissue repair and remodeling through the SASP 11 . Moreover, induced senescence promotes cell reprogramming in various tissues through paracrine factors 13,14 . Conversely, exhaustion of the stem cell pool results from paracrine induction of senescence in non-damaged cells 15 . Furthermore, defective clearance of senescent cells, which is particularly relevant during ageing, induces a persistent and deleterious SASP 16 . In this context, specific elimination of p16 Ink4 -positive cells in a progeroid mouse model delays the onset, or limits the progression, of aged-related symptoms 17 .
Premature cellular senescence has been described in mesenchymal stromal cells differentiated from Werner syndrome-induced pluripotent stem cells (iPSCs) 18 , and in other progeroid syndromes, such as the Hutchinson-Gilford progeria syndrome 19 and Ataxiatelangiectasia 20 . Similarly, human primary fibroblasts depleted of BLM, WRN, or RECQ4 helicases, which are responsible for precocious ageing in the Bloom syndrome, Werner syndrome, and Rothmund-Thomson syndrome, respectively, display multiple markers of senescence compared to normal fibroblasts. Moreover, keratinocytes from progeroid Cockayne syndrome, mutated in CSA and with impaired DNA repair capacity, display early senescence 21 . Although suspected of being implicated in normal ageing, defects observed in progeroid syndromes have not been formally demonstrated to drive physiological ageing.
These findings raise the possibility that factors responsible for Cockayne syndrome (CS), namely CSA and CSB, are effectors of cellular senescence, and also question which activity of these multifunctional proteins is determinant. In this context, we identified a novel stress-induced protease activity that improperly degrades functional mitochondrial proteins in cells from CS patients 22 . Notably, in CS cells combined oxidative and nitrosative stress (ROS and RNS) promote overexpression of the HTRA3 protease, which results in degradation of the mitochondrial DNA polymerase POLG1 responsible for replication of the organelle genome, thereby triggering mitochondrial dysfunction. The cognate HTRA2 protease is also overexpressed in CS cells, but it is unclear whether POLG1 degradation is promoted by HTRA3 alone or in combination with HTRA2. Although the mechanism leading to HTRA3 (and HTRA2) overexpression in CS cells remains undefined, it is likely uncoupled from the well documented CSA/CSB involvement in the repair of ultraviolet (UV)-damaged DNA through transcription-coupled nucleotide excision repair (TC-NER) 23 . Indeed, these proteins are also transcription factors and have been implicated in chromatin remodeling in vitro (in particular CSB) 24 . CSA/CSB have also been detected in mitochondria where they repair oxidative damage of mitochondrial DNA 25 . Critically, scavenging of oxidative and nitrosative species with a porphyrine derivative, MnTBAP, led to rescue of the cellular and molecular defects in progeroid CS patient cells 22 .
Here, we report that the CS alterations of HTRA3, HTRA2, POLG1 and mitochondria are recapitulated during replicative senescence of human primary fibroblasts. We also show that these alterations and replicative senescence are triggered by downregulation of CSB, through reduced histone acetylation of its promoter. Consequent CSB depletion results in the exposure of the p21 Waf1 promoter to activation, which leads to senescence, and this activity of CSB is independent of its function in UV-induced DNA repair.

Results
HTRA3 overexpression during replicative senescence. To assess whether HTRA3, which is considered a prevalently mitochondrial protease 26 , was expressed during cellular senescence, we examined population doubling of three independent IMR-90 serially passaged human embryonic fibroblasts (Fig. 1a). Cells at passage numbers (PN) indicated with an arrow were selected for in-depth investigation, and are representative of distinct phases: proliferative PN16, PN19, PN23; the end of exponential growth, PN27; pre-senescent PN31; and senescent PN35. Senescenceassociated beta-galactosidase staining (SA-β-gal, Fig. 1b and Supplementary Fig. 1a), as well as increased cell size (Supplementary Fig. 1b, c), confirmed pre-senescence at PN31 and senescence at PN35.
Interestingly, at PN31 and PN35 HTRA3 immunofluorescence (IF) was twofold higher compared to earlier passages, independent of increased cell size ( Fig. 1c and Supplementary Fig. 1b, d). Super-resolution structured illumination microscopy (SIM) of early-passage and senescent fibroblasts revealed that HTRA3 is not prevalently mitochondrial (several mitochondria do not display HTRA3 and a large part of the labeling remains extramitochondrial), and this condition does not appear to change upon senescence (Fig. 1d). Increased HTRA3 levels during replicative senescence were confirmed by Western blot (WB), in particular the short isoform at PN31, and cleaved products of the long isoform at PN35 (Fig. 1e and Supplementary Fig. 1e). During stress, long-HTRA3 undergoes autocleavage that generates shorter L-forms 27 . Real-time quantitative PCR (RT-qPCR) analysis confirmed high levels of HTRA3 transcripts, in particular the long form, in senescent cells at PN35, together with the established senescence markers p21 Waf1 , p16 Ink4 , and the SASP marker IL-6 ( Fig. 1f). The levels of HTRA3 (short) and p21 Waf1 transcripts were 1.5-and twofold higher, respectively, also in presenescent PN31 cells compared to earlier passages. Increased levels of HTRA3 were not dependent on declined cell proliferation, since slow dividing/non-dividing early-passage fibroblasts at confluence, assessed by decline of the cell cycle markers cyclin A2 and PCNA, did not display higher levels of HTRA3 (RNA and protein) compared to cells undergoing robust proliferation ( Supplementary Fig. 2a-c). Absence of senescence in the abovementioned cells was verified by unaltered levels of p21 and p21 Waf1 as well as p16 and p16 Ink4 . Thus, HTRA3 is expressed at high levels during replicative senescence, and as early as established senescence markers, in particular the principal effector p21 Waf1 .
In CS cells, also the HTRA2 protease was overexpressed 22 and this was also the case for HTRA2 transcripts ( Supplementary  Fig. 3a) and protein (Fig. 1g, i and Supplementary Fig. 3c) in senescent cells (PN35).
Mitochondrial impairment during replicative senescence. In cells from CS patients, accumulation of HTRA3 (alone or in combination with HTRA2) was associated with depletion of the mitochondrial DNA polymerase POLG1, despite high levels of POLG1 transcripts, suggesting degradation of this polymerase 22 .
Accordingly, we observed reduced levels of POLG1 by IF ( Fig. 1h and Supplementary Fig. 3d) and WB (Fig. 1i) in pre-senescent (PN31) and senescent (PN35) cells, despite unchanged or increased levels of transcripts ( Supplementary Fig. 3b). Cells kept at confluence for 1-2 days displayed slightly increased levels of HTRA2 and reduced levels of POLG1 ( Supplementary Fig 2a-c), suggesting that these proteins are to some extent dependent on factors other than replicative senescence. In CS cells, POLG1 depletion was associated with increased ROS and reduced mitochondrial ATP production 22 . Senescence ( Supplementary Fig. 4a-d) was  associated with increased levels of oxidative stress, measured by reduced glutathione (GSH), a strong scavenger of ROS, and its ratio with oxidized glutathione (GSSG) 28 ( Supplementary Fig. 4e), and to some extent mitochondrial ROS ( Supplementary Fig. 4f, g). Senescent cells displayed reduced ATP production by mitochondrial oxidative phosphorylation (OXPHOS), and decreased levels of mitochondrial complexes I, III, and IV, which were also reduced during pre-senescence ( Supplementary Fig. 4h, i). Thus, senescent cells recapitulate cellular and mitochondrial alterations observed in CS patient cells.
CSB depletion is an early event in replicative senescence. We then asked whether altered HTRA3 and POLG1 levels during replicative senescence were a consequence of CSB impairment, since CSB mutation resulted in these defects in CS cells. We observed a progressive and dramatic decrease of CSB transcripts from PN27 to PN35 (from twofold to eightfold, respectively, Fig. 2a), confirmed by WB at the end of the exponential phase (PN27) (Fig. 2b), and by IF in pre-senescent and senescent fibroblasts (Fig. 2c, d). CSB depletion was not observed in slowly dividing/non-dividing early passages fibroblasts (Supplementary Fig. 2a-c). Thus, reduced expression of CSB was detected earlier than the appearance of senescence, preceding the established senescence marker p21 Waf1 and SA-β-gal + staining, as well as accumulation of HTRA3/HTRA2, and depletion of POLG1. Interestingly, transcripts of CSA, the other factor mutated in CS, did not decline during senescence (Fig. 2a), whereas the protein was reduced by less than 50% in pre-senescent (PN31) and senescent (PN35) cells, therefore at a lower extent and later PN than CSB (Fig. 2b).
To verify whether CSB depletion triggers cellular senescence, we knocked down CSB in early-passage IMR-90 fibroblasts with two shRNA (shCSB#1, shCSB#2, Fig. 3a-c). Cells transduced with control shRNA (shSCR) progressively decreased CSB transcripts prior to increased expression of p21 Waf1 and HTRA3 (black columns in Supplementary Fig. 5a-c), thereby confirming downregulation of CSB during senescence. The direct correlation observed between p21 Waf1 and HTRA3 (Fig. 3d), supports the notion that increased expression of both transcripts is strictly linked in senescence. CSB downregulation, which was maintained until PN20 and PN21 for shCSB#1 and shCSB#2, respectively, was nevertheless sufficient to promote cellular senescence. Figure 3e shows proliferation arrest of CSB-knocked down fibroblasts earlier (PN26, eight passages post-transduction of shCSB#1; PN24, after six passages for shCSB#2) compared to cells transduced with shSCR (PN28, ten passages post-transduction). Precocious senescence upon transient CSB knockdown was confirmed by a higher number of SA-β-gal + cells at each passage, and to a larger extent for the more efficient shCSB#2 ( Fig. 3f and Supplementary Fig. 5d), with an initial burst as early as PN19 and PN20. Further, the levels of HTRA3 and p21 Waf1 in CSB-silenced fibroblasts were higher than control beyond PN19-PN20 ( Supplementary Fig. 5b, c).
We then focused on PN19 and PN20, when CSB was efficiently knocked down, and observed a 15-fold increase of p21 Waf1 mRNA as early as PN19 with shCSB#2 compared to shSCR, and a fivefold increase at PN20 (Fig. 3g). Higher levels of p21 with both shRNAs at PN20 were confirmed by WB (Fig. 3i). Moreover, both shCSBs induced a significant increase of HTRA3 transcript and protein at PN20 (Fig. 3h, i). WB also revealed an increase of HTRA2 and a decrease of POLG1 levels with both shCSBs (Fig. 3i), confirming our results (see Fig. 1g-i), and consistent with the alterations observed in CS cells. Interestingly, cumulated data from shSCR and the two shCSBs ( Supplementary Fig. 5a, b) demonstrate a strong inverse correlation between CSB and p21 Waf1 expression (Fig. 3j). Taken together, these data indicate that low CSB levels trigger p21-dependent replicative senescence.
We wondered whether CSB affects p21 expression by direct interaction with its promoter. Chromatin immunoprecipitation (ChIP) assays showed that CSB directly interacts with two regions of the p21 Waf1 promoter that are target of p53 29 , and a more extended interaction in the distal (5') region was observed with the C-terminal anti-CSB than with N-terminal anti-CSB ( Fig. 4a-c). As a control, both antibodies failed to display a signal at the TATA box that is close to the transcription start site. Moreover, cells stably expressing the eGFP reporter under control of the CDKN1A (p21Waf1) promoter (Fig. 4d) expressed higher levels of eGFP upon CSB silencing (shCSB#2), assessed with IF ( Fig. 4e, f) and WB (Fig. 4g), further showing that CSB directly regulates p21 expression.
Finally, we observed that ectopic overexpression of CSB in a fraction of early-passage IMR-90 fibroblasts resulted in lower p21 expression ( Supplementary Fig. 6a, b) and a smaller percentage of SA-β-gal + cells ( Supplementary Fig. 6c, d) compared to cells carrying an empty vector, suggesting that senescence was delayed upon increased CSB levels.
CSB depletion by histone H3 hypoacetylation of its promoter. We then investigated the mechanism responsible for CSB downregulation during replicative senescence. UV-induced CSB repression has been recently described in age-related nuclear cataract through coordinated DNA hypermethylation and histone deacetylation at the CSB promoter 31 . Histone hypoacetylation is associated with compact chromatin and repressed transcription 32 . RT-qPCR of DNA methyltransferases DNMT3a and DNMT3b, which are responsible for de novo DNA methylation, did not reveal differences among the various cell passages (Fig. 5a). However transcripts of DNMT1, the enzyme responsible for DNA methylation maintenance during replication, progressively decreased at PN31 and PN35, probably due to reduced replication in pre-senescent and senescent cell populations, as suggested by lower levels in poorly dividing early-passage fibroblasts (Supplementary Fig. 2c). We then examined the methylation profile of the CSB promoter (positions in Supplementary Fig. 7) by pyrosequencing after bisulphite treatment. At all passages, we observed low methylation levels for CpG(#8), which was hypermethylated in age-related nuclear cataract, as well as the other seven CpG sites in the promoter region of CSB (Fig. 5b). Thus, methylation alterations do not occur in this region during replicative senescence. However, 24 h treatment with 10 µM, but not 2 µM, of the methyltransferase inhibitor 5-aza-2′-deoxycytidine (5-aza-dC) ( Supplementary Fig. 8a, b) resulted in enhanced transcription of CSB ( Supplementary Fig. 8c) and protein content ( Supplementary Fig. 8d), indicating that global DNA hypomethylation is associated with CSB overexpression.
Conversely, the levels of acetylated histone H3 decreased at the end of the exponential growth (PN27), and to a larger extent in pre-senescent (PN31) and senescent (PN35) cells (Fig. 5c). The decrease in histone H3 acetylation largely exceeded the global depletion of H3, and this was not dependent on cell cycle arrest since levels of acetylated H3 histone were not altered in slowly dividing/non-dividing IMR-90 ( Supplementary Fig. 2a). Accordingly, mRNA levels of histone deacetylase HDAC1 (and of highly homologous HDAC2) increased transiently at PN27 (Fig. 5d), and were reduced at later passages. Remarkably, the peak of HDAC1 expression and histone H3 hypoacetylation appeared at the switch point for CSB depletion, i.e., PN27 (see Fig. 2), suggesting a role of this histone modification in the regulation of CSB expression during replicative senescence. ChIP on the CSB promoter (positions in Supplementary Fig. 7) showed reduced levels of acetylated histone H3 at PN27 and PN31 compared to the previous passages (Fig. 5e), whereas the levels of global histone H3 were not altered. Thus, hypoacetylation of histone H3 at the CSB promoter directly correlates with CSB downregulation during replicative senescence.
Furthermore, treatment of early-passage fibroblasts (PN16) for 24 h with increasing concentrations of anacardic acid (AA), an histone acetyl transferases (HAT) inhibitor ( Fig. 5f), resulted in reduced CSB transcript and protein, concomitantly with a global reduction of acetylated histone H3 (Fig. 5g, h). A specific reduction of H3 acetylation of the CSB promoter was then observed by ChIP upon AA treatment (Fig. 5i). Interestingly, at high AA concentrations HTRA3 and p21 levels increased after further 24 h in culture and upon drug withdrawal (Fig. 5h), consistently with the late upregulation of these senescence markers following CSB depletion (see above, Fig. 1c, e, f). Under these conditions, H3 acetylation and CSB levels were restored.
level ( Supplementary Fig. 8f) and specifically at the CSB promoter ( Supplementary Fig. 8h), did not alter CSB levels ( Supplementary Fig. 8f), suggesting that global as well as local hyperacetylation of H3 does not increase CSB expression. Taken together, these experiments show that CSB depletion in replicative senescence is triggered by hypoacetylation of its promoter.
CSB-dependent replicative senescence is independent of p53. Increased levels of p21 during replicative senescence depend on p53 activation. p53 protein, but not transcript (except at PN35) levels increased towards senescence, in agreement with previous findings 33 , suggesting stabilization of the protein (Fig. 6a, b). P53 silencing (Fig. 6c, d) alone had little or no effect on the population doubling compared to control shSCR (Fig. 6e). Silencing of p53 followed by silencing of CSB reduced the population doubling as much as silencing of CSB alone, suggesting that CSB depletion triggers replicative senescence independently of p53. This result was confirmed by treating fibroblasts in culture with pifithrin-alpha (PTF-α), a p53 inhibitor that blocks p53-mediated p21 activation 34 , in the presence and in the absence of CSB silencing (Fig. 6f). Indeed, treatment with PTF-α reduced the population doubling slightly compared to untreated cells (Fig. 6g), but did not overtly affect the fraction SAβ-gal + cells (Fig. 6h, i) and p21 Waf1 expression (Fig. 6j, k), whereas CSB silencing strongly affected these readouts. PTF-α slightly increased the population doubling, reduced the fraction of SA-β-gal + cells and p21 Waf1 expression (thus confirming the   e Representative confocal acquisitions of cells stably expressing the CDKN1A (p21 waf1 ) promoter-eGFP reporter upon shCSB#2 silencing or control shSCR, immunostained for GFP (green) to increase the signal of endogenous GFP, and counterstained with Hoechst (blue) after maximum intensity projection with Imaris; scale bar = 50 µM, and f quantification of the GFP mFI/cell; n = 127 cells/condition from three independent experiments, mean ± SEM; unpaired t-test (two-tailed) (t = 4.555, DF = 253), p-value vs. shSCR. g WB of GFP and GAPDH (loading control) in cells not expressing (Mock GFP) or stably expressing (Mock shRNA) the CDKN1A (p21 waf1 ) promoter-eGFP reporter, not silenced (shSCR) or silenced for CSB (shCSB#2). Cells that express GFP display a band at 27 kDa, and also an upper band at around 60 kDa, in particular upon CSB silencing, corresponding to dimers observed in other conditions 70 . Source data are provided as Source Data files. CSB-dependent senescence is specific to the DDR/p21 pathway.
Replicative senescence of IMR-90 fibroblasts is correlated with activation of the p21 Waf1 pathway ( Fig. 1f and Supplementary  Fig. 5b), which is DNA damage-response (DDR)-dependent 3 . To assess the extent of endogenous DNA damage during replicative senescence we enumerated γ−H2AX and 53BP1 foci at critical PNs. Figure 7a and Supplementary Fig. 9a show a significant increase of both markers at PN27, concomitantly with CSB depletion (see above, Fig. 2), suggesting a link between these two events.
To verify that CSB downregulation and HTRA3 overexpression are specific to DDR-induced senescence, we triggered senescence by irradiating (10 Gy) early-passage IMR-90 fibroblasts (PN17) and examined cells after 10 days 35 . Notably, an increased number of γ−H2AX and 53BP1 foci in irradiated fibroblasts (Fig. 7b-d) correlated with a dramatic increase in SA-β-gal + cells (89.2 ± 2.3% compared to non-irradiated cells 2.7 ± 1.1%; Fig. 7e).   Senescence following irradiation was verified by high levels of the DDR-dependent senescence marker p21 (Fig. 7f). Conversely, no accumulation of the DDR-independent and senescence marker p16 was observed (Fig. 7f) including at earlier and later time points (7 days and 15 days post-IR, respectively, Supplementary  Fig. 9b), differently from a study performed with cells at a later PN and displaying apparently lower basal levels of p16 before irradiation than in our case 36 . Importantly, we observed decreased CSB, and increased HTRA3 and HTRA2 levels upon irradiation (Fig. 7f), as it was the case for replicative senescence (see Figs. 1 and 2). Decreased CSB levels upon irradiation resulted in the loss of CSB interaction with the p21 Waf1 promoter (ChIP assays, Supplementary Fig. 9c-f), supporting the notion that this interaction is a barrier to senescence. We note that POLG1 was not depleted after irradiation (Fig. 7f), suggesting that the levels of the mitochondrial DNA polymerase rely also on factors other than DDR-dependent replicative senescence, in agreement with our previous observations (see Supplementary Fig. 2a, b). Senescence can be induced in a DDR-independent manner through activation of the p16 Ink4 pathway 37 . Palbociclib, a compound that acts downstream of p16 Ink4 induces senescence by inhibiting cyclin-dependent kinases CDK4/6, and inducing permanent cell cycle arrest 38 , see Fig. 7g. Early-passage (PN20) IMR-90 cells were treated with increasing doses of palbociclib (0.2-5 μM) for 7 days followed by 1 day of drug withdrawal. A dose-dependent increase of the fraction of SA-β-gal + cells verified senescence (Fig. 7h, i), however, this did not result in increased γ −H2AX and 53BP1 foci number ( Supplementary Fig. 9g-i), confirming that palbociclib-induced senescence is DDRindependent 39 . We observed decreased HTRA3 and increased CSB transcripts at all concentrations of the drug compared to untreated control (Fig. 7j, k), in clear contrast with replicative senescence (see Figs. 1c and 2a). Increased levels of CSB were confirmed by WB (Fig. 7m), which also showed unaffected HTRA3 levels despite reduced transcription. Palbociclib treatment did not increase the levels of p16 Ink4 transcript and protein, as expected since the drug acts downstream of p16 (Fig. 7l, m) 38 . Upon palbociclib treatment, p21 RNA and protein were upregulated (Fig. 7l, m), in agreement with a previous study 40 . Taken together, these data indicate that p16-dependent senescence does not correlate with depletion of CSB and overexpression of HTRA3, as it was the case for DDR-dependent/p21 replicative senescence.
Delayed senescence with anti-ROS/anti-RNS treatment. We showed previously that HTRA3 overexpression and POLG1 depletion in CS cells that lack CSB can be restored after 24 h treatment with MnTBAP, a superoxide dismutase mimetic and peroxynitrite (RNS) scavenger 22 . Incubating IMR-90 fibroblasts at various passages with MnTBAP for 24 h reduced SA-β-gal + cells by 13.9 ± 6.0% at PN35 (Supplementary Fig. 10a). This reduction took place at constant cell number ( Supplementary  Fig. 10b), suggesting that it was not due to cell depletion. In agreement with our model, 24 h treatment with MnTBAP decreased the levels of HTRA3 compared to untreated controls, de facto blocking or containing HTRA3 overexpression in presenescent cells (PN27, PN31) and senescent cells (PN35), respectively ( Supplementary Fig. 10c, d). Importantly, MnTBAP had essentially no effect on CSB levels ( Supplementary   Fig. 10e, f), in agreement with the rescue of HTRA3/POLG1 levels in CS patient cells that are impaired in CSB.
Long-term treatment with a tenfold lower dose of MnTBAP resulted in increased cumulative population doubling of IMR-90 fibroblasts at PN31 and PN32 compared to untreated cells, suggesting a later entry into senescence in the presence of the drug ( Supplementary Fig. 10g), which was confirmed by a decrease of SA-β-gal + cells at PN29 and PN31 (≈35% and ≈9%, respectively; Supplementary Fig. 10h). Both effects were less pronounced than in cells treated with rapamycin, which extends replicative lifespan in vitro 41 (positive control). Thus, MnTBAP treatment delays replicative senescence of human fibroblasts without intervening on CSB levels.

Discussion
The study of human ageing is expected to take advantage from model systems of monogenic progeroid syndromes such as Hutchinson-Gilford progeria syndrome (HGPS), Werner syndrome (WS), and Cockayne syndrome that mimic to some extent characteristics of normal ageing 42 . Indeed, in vitro and in vivo experiments performed on precocious ageing models showed that cellular and molecular hallmarks of ageing such as telomere dysfunction, epigenetic changes, metabolic defects or cellular senescence, recapitulate those of physiological ageing 43 . Moreover, despite in vitro reprogramming HGPS-or WS-derived iPSCs appeared to erase the pro-senescent phenotype, cells underwent again rapid and premature senescence after differentiation 18,44 . Based on our recent discovery that linked progeroid CS to improper degradation of functional mitochondrial proteins (POLG1 depletion via HTRA3/HTRA2 overexpression) 22 , we proposed an alternative hypothesis, that CSspecific defects are implicated in the senescence of normal cells.
Serially passaged primary human fibroblasts display upregulation of the HTRA3 and HTRA2 proteases and downregulation of mitochondrial POLG1, suggesting a recapitulation of CSspecific defects 22 during replicative senescence. These defects include reduced mitochondrial respiration, in agreement with DDR-induced senescence 45 , and depletion of mitochondrial complexes I, III, and IV. HTRA2 upregulation and POLG1 depletion were, however, not exclusive to cellular senescence but appeared to some extent also in long-term confluent cultured cells. This observation is in agreement with the finding that HTRA2 can induce cell cycle inhibition following serum withdrawal 46 . Conversely, HTRA3 upregulation appeared contemporary to the induction of the major effector of replicative senescence p21, with comparable kinetics, thereby establishing this protease as a bonafide marker of cellular senescence.
HTRA3 is defined as a prevalent mitochondrial protein that is translocated from mitochondria upon cytotoxic stress, based on subcellular fractionation and colocalizaton experiments 26 . Our analysis with super-resolution SIM microscopy rather indicates that only a fraction of HTRA3 is mitochondrial. This condition is maintained in senescent cells, where accumulation of HTRA3 takes place in a mitochondrial-impoverished cytoplasm. HTRA3 accumulation during senescence concerned mainly the cleaved isoform of this protein, which is the more active isoform 27 , suggesting an increase of its auto-proteolytic activity during senescence. Secretion of HTRA3 was reported in the context of remodeling of the extracellular matrix or acting as a proapoptotic agent 26,47 . These observations raise the possibility that this protease can function as part of the SASP, and participate in tissue remodeling 11 or paracrine induction of senescence in neighboring cells 48 .
Knockdown experiments demonstrated that contemporary accumulation of HTRA3 and the senescence inducer p21 during replicative senescence depend on depletion of CSB. We show that HTRA3 and p21 levels are inversely correlated with CSB, therefore, CSB depletion in normal cells is the earliest trigger of  replicative senescence defined to date, acting upstream of p21 Waf1 induction. Importantly, despite loss of CSB downregulation upon passage in culture, the initial silencing was sufficient to reduce the replicative lifespan of cells by several weeks. p21 induction in senescence is normally triggered by p53, a master regulator of the response to stress and DNA damage 11 . However, the role of p53 in replicative senescence has not been fully clarified, since for instance p53 does not promote this process in normal human keratinocytes 33 . By silencing or inhibiting p53, alone or together with CSB, we showed first that CSBmediated senescence does not depend on p53. Moreover, p53 silencing alone had virtually no impact on replicative senescence, at least at early PNs. One may expect increased proliferation rate upon p53 silencing, however, it is possible that during early passages the extent of DNA damage is lower than required to trigger the p53 response and thereby cell cycle arrest.
These data indicate that CSB, and not p53, plays a master role in replicative senescence of human fibroblasts. Previous studies in cells derived from CS patients highlighted the physical interaction of CSB with p53 49 where CSB participates in p53 polyubiquitination and degradation upon UV induction 50 . Consequently, CS patient cells have higher basal and UV-induced levels of p53 compared to wild type cells. Thus, despite increased p53 levels during replicative senescence that are compatible with protein stabilization (also depends on reduced CSB), this process is largely p53-independent. The strong inverse correlation between p21 Waf1 and CSB indicates that in our model system p21-dependent replicative senescence is triggered by CSB depletion rather than p53 induction, revealing a CSB depletion/ p21 Waf1 axis in cellular senescence (Fig. 8). In agreement with this model, we showed direct interaction of CSB on the endogenous and an ectopic p21 Waf1 promoter and loss of this interaction when CSB is silenced or depleted following irradiation, suggesting that CSB protects the p21 Waf1 promoter from activation by other factors, which essentially do not include p53. Moreover, p53 is activated by DNA damage and DNA damaging signaling, which do not increase upon CSB silencing. These data point to CSB directly controlling senescence via p21 and HTRA3 upregulation rather than CSB depletion being a secondary effect of DDRinduced senescence.
CSB is a DNA repair protein but also acts as chromatin remodeler and transcription factor 24,51,52 . In addition to direct intervention at p21 Waf1 promoter, CSB may also promote p21 degradation through ubiquitylation due to its E3 ligase activity, as it has been hypothesized in CS cells 53 . In other contexts than replicative senescence, p21 can be upregulated in the presence of high levels of CSB, as it is the case upon palbociclib treatment, in agreement with a previous study where the process was reported to be p53-independent and rather relying on myc derepression 40 . However, in this case p21 upregulation did not precede senescence, and appeared in the same timeframe as SA-β-gal staining, suggesting that it rather plays a role in the maintenance of senescence.
We demonstrate that CSB downregulation depends on histone H3 hypoacethylation of the CSB promoter, perhaps performed by various HATs, rather than combined hypermethylation and histone deacetylation of the CSB promoter as in UV-induced agerelated nuclear cataract 31 . However, the DNA methylation status at other sites may affect CSB levels, since treatment with high doses of the methyltransferase inhibitor 5-aza-dC resulted in increased levels of CSB. Ectopic overexpression of CSB in a fraction of cells resulted in delayed signs of senescence, but in the long-term CSB overexpression may reduce or block p21 expression, and p21 is essential for the survival of senescent cells 54 . Compatibly, CSB overexpression was observed in cancer cells 55 that do not undergo senescence.
Replicative senescence involves permanent activation of the DDR/p53/p21 pathway 10 . We showed that CSB depletion and CSB promoter hypoacetylation coincide with increased DNA damage (DSBs) during replicative senescence. DNA damage signal though telomere attrition is unlikely to play a role, since replicative senescence of IMR-90 fibroblasts in 20% oxygen has been shown to be telomere-independent 56 . Despite CSB is also implicated in DSB repair through chromatin remodeling 57 , it is unlikely that this activity affects the early phases of replicative senescence since CSB depletion is simultaneous to increase of DNA damage. However, severe CSB depletion at later PNs may impair DSB repair, contributing to exacerbate the senescent phenotype. Our findings indicate that CSB depletion and subsequent HTRA3 overexpression in replicative senescence are specific to the DDR/p53/p21 pathway (they are also activated by radiation (DDR)-induced senescence), and not to DDR-independent/p16 senescence pathway.
Transcription of CSA, the other major protein mutated in CS, and which is not associated with the most severe form of the disease 58 , is not altered with replicative senescence, but the protein is depleted, although at a lower extent and several passages later than CSB, suggesting a different effect on senescence than CSB. Interestingly, we showed CSA depletion upon efficient silencing of CSB, suggesting that CSA levels depend, at least in part, on CSB. CSA and CSB are both involved in TC-NER and BER 59,60 . The apparently different underlying mechanisms and Fig. 7 CSB-dependent senescence is specific to the DDR/p21 pathway. a Enumeration of endogenous γ−H2AX and 53BP1 foci/cell at PNs preceding or during CSB depletion. n = 50-70 cells from three independent experiments; mean ± SEM; one-way ANOVA (γ−H2AX: F = 3.209, DFn = 3, DFd = 220, p = 0.0239; 53BP1: F = 5.603, DFn = 3, DFd = 220, p = 0.001) with post-hoc Tukey's test vs. the respective PN16. b Representative confocal acquisitions of irradiated (10 Gy) and non-irradiated (non-IR) IMR-90 immunostained for γ−H2AX (green) and 53BP1 (red), counterstained with Hoechst (blue, nuclei), after maximum intensity projection with the Imaris software; scale bar = 20 µM. Percentage of cells with 0, 1, 2, 3, or >3 (c) γ−H2AX or d 53BP1 foci per nucleus. n = 90 cells from three independent experiments. Mean ± SEM; two-way ANOVA (γ−H2AX: F = 35.77, DFn = 4, DFd = 20, p < 0.001; 53BP1: F = 57.63, DFn = 4, DFd = 20, p < 0.0001) with post-hoc Sidak's test vs. the respective non-IR. e Quantification of SA-β-gal + cells of irradiated and non-IR IMR-90 at PN17. n = 180-220 cells/condition from three independent experiments, mean ± SEM; unpaired Student's t-test (two-tailed) (t = 36.28, DF = 408), p-value vs. non-IR (f) Immunoblots of CSB, HTRA3, p21, and POLG1, HTRA2, p16 (GAPDH F-C staining, loading control, under each blot) in irradiated and non-IR fibroblasts. g Scheme indicating the palbociclib target in the p16 pathway to senescence. perhaps impact of CSB and CSA in replicative senescence support the notion that other causes than just impaired DNA damage repair are responsible for the severe defects of this disease 22,53 .
Finally, a 24 h treatment with MnTBAP limited HTRA3 overexpression in replicative senescence, but had no effect on the senescence trigger CSB, and in the long-term increased replicative lifespan of fibroblasts. This result indicates that MnTBAP acts downstream of, and partially compensates for, defects triggered by CSB depletion, in agreement with rescue of the HTRA3/ POLG1 defects in CS cells that are CSB impaired 22 . A previous study demonstrated that MnTBAP partially inhibits radiationinduced senescence in endothelial cells by rescuing complex II dysfunction 61 . These results suggest that MnTBAP limits the paracrine propagation of senescence by restricting the HTRA3/ POLG1/mitochondrial dysfunction, with implications for the treatment of CS, as well as ageing prevention therapies.
Our data reveal a role of CSB in protecting human primary fibroblasts from DDR-dependent replicative senescence, via downregulation of the senescence trigger p21, and link a progeroid factor to physiological ageing. This pathway points to CSB as a relevant functional anti-ageing regulator. In this context, cellular senescence could also participate in the precocious ageing phenotype observed in CS, as it is suggested by premature senescence of patient-derived CSA-mutated keratinocytes 21 . It should be investigated in the future at which extent this defect, if confirmed in all CS patients, is correlated with and participates to the severity of the disease, whether CSA is implicated in the same pathway as CSB, and also whether it is operating in UVSS patients that lack the progeroid phenotype and whose cells do not display the HTRA3/POLG1/mitochondrial defect 22 . Finally, these data provide a DNA repair-independent mechanistic explanation for the reported association between high CSB levels and tumor proliferation 55 , and CSB polymorphism association with longevity 62 , supporting a key regulatory role of CSB in the ageing/ cancer axis.
For radiation-induced senescence, IMR-90 (PN16) were grown to confluence and irradiated at 10 Gy with the Xstrahl RS320 irradiator (X-ray). After irradiation, the medium was refreshed and cells were incubated overnight before 1:3 dilution and cultured for 10 additional days before experiments, according to ref. 35 . When indicated, cells were harvested at 7 and 15 days post-irradiation.
Transfection of promoter reporter and expression vectors. The CDKN1A (encoding p21 waf1 ) promoter-eGFP reporter plasmid (pEZX-PF02) (HPRM51895-PF02, GeneCopoeia) was transfected in early-passage IMR-90 cells (PN16) using Lipofectamine 3000 (ThermoFisher) according to the manufacturer instructions. Forty-eight hours post-transfection, cells were cultured with 2 µg.ml −1 puromycin for >2 weeks to select stable integrations. The resultant stable cell population expressing GFP under the control of the CDKN1A promoter was then either not transduced or transduced with shSCR or shCSB#2, as described in the previous paragraph.
Ectopic expression of CSB was obtained upon transfection with Lipofectamine 3000 of early-passage IMR-90 (PN15) with the CSB (Myc-DDK-tagged) expression vector (pCMV6-Entry) (RC219020, OriGene), called here "pCSB". Transfection with pcDNA3.1 (ThermoFisher Scientific, V79020), which codes for neomycin resistance under control of a CMV promoter or "Empty vector" was used as control. The CSB coding sequence was verified by sequencing pCSB before transfection. pCSB was amplified in Stbl2 TM E.coli (ThermoScientific) at 30°C, since amplification of pCSB in classic DH5 α E.coli resulted in multiple mutations (as indicated by the supplier). Transfected cells were kept in culture in the absence of selection for 8 days, including a passage in culture, before analysis.
RNA isolation and real-time quantitative PCR (RT-qPCR) assay. Total RNA was isolated from cells using the RNAeasy ® Micro kit (Qiagen) and reversetranscribed in cDNA with SuperScript IV Reverse Transcriptase (Thermo-Fisher Scientific). Possible RNA residues were degraded by incubation for 20 min at 37°C with RNaseH (Takara Bio, 2150B). RT-qPCR was performed using PowerUp™ SYBR™ Green Master Mix (Thermo-Fisher Scientific) and the rate of dye incorporation was monitored using the StepOne Plus RealTime PCR system (Applied Biosystems). Three independent replicates were used for each condition. Data were analyzed by StepOne Plus RT PCR software v2.1 (Thermo Fisher Scientific). TBP (human) transcript levels were used for normalization of each target (=ΔCT). Realtime PCR C T values were analyzed using the 2 −ΔΔCt method to calculate the fold expression 63 . For experiments with palbociclib the fold expression was calculated individually for each replicate. qPCR primers used are listed in Supplementary  Table 1, which includes the corresponding references.
Millipore, 06-599) overnight at 4°C. After five washes in PBS containing 0.1% Tween20, the membrane was incubated with HRP-conjugated (1:10000, Thermo Fisher Scientific) or fluorophore-conjugated (F-C) secondary antibodies (1:5000, Bio-Rad and 1:25000, Biotium) for 1 h at RT and revealed by chemiluminescence or fluorescence respectively. When necessary, 0.1 M of glycine (pH 2) was used as stripping buffer for 20 min, followed by reblotting with HPR-conjugated secondary antibodies. Detection was performed using Chemidoc™ MP imaging system (Bio-Rad). Experiments were done in duplicate, and a representative Western blot was shown. Staining with ATX Ponceau S Red (Sigma-Aldrich, 09189) was used as a further marker of protein content. For Western blot analysis of the five OXPHOS complexes, cell lysates were not boiled and membranes were probed with the Total OXPHOS Human WB Antibody Cocktail (1:1000, Abcam, ab110411). Western blot bands quantification were done using the ImageJ software. Uncropped and unprocessed scan of all the blots are provided in the Source Data File.
ChIPs were carried out by incubating 10 to 20 μg of chromatin with 1 μg of either antibody: (α-acetlyl-Histone H3 (Millipore, 06-599), α-Histone H3 (Abcam, ab1791), (N-and C-terminal) α-CSB (Bethyl, A301-345A and Abcam, ab96089, respectively), or α-rabbit IgG (Santa Cruz Biotechnology, sc-2027) as a control. After overnight incubation at 4°C, 40 μl of protein G-Dynabeads™ (Thermo-Fisher Scientific, 10004D) were added on lysate for 2 h at 4°C. After extensive washing, DNA was isolated from the beads by successive boiling for 10 min in the presence of 10% (w/v) Chelex 100 Resin (Bio-Rad, 1421253), incubating at 55°C for 30 min in the presence of 100 µg.ml −1 of Proteinase K (Eurobio Ingen, GEXPRK01-E), and boiling again for 10 min. After centrifugation, the resulting supernatant was used as a direct template for qPCR detection of regions of CSB ( Supplementary Fig. 9c) or p21 Waf1 promoters (Fig. 4b) (primers listed in Supplementary Table 1). Immunoprecipitated chromatin with the indicated antibodies was calculated as a percentage of the input DNA after normalization with control ChIP performed with rabbit IgGs (Mock). Immunoprecipitated chromatin with acetyl-Histone H3 was finally normalized to the H3 histone. Chromatin templates were also used for PCR detection of regions of the promoter of p21 Waf1 (Fig. 4a) (primers listed in Supplementary Table 1), using the following PCR profile: preincubation for 5 min at 95°C, 40 cycles of 30 s at 95°C, 30 s at a primer-specific annealing temperature (60°C) and 30 s at 72°C, and one final 10 min incubation at 72°C. The PCR products were separated by electrophoresis through 2.0% agarose gel and detected with the Gel Doc 2000 imaging system (Bio-Rad).
Genomic DNA isolation and methylation assays. Genomic DNA (gDNA) from IMR-90 was isolated using the QIamp ® DNA mini kit (Qiagen) and 800 ng of DNA was treated with sodium bisulfite using the EZ DNA Methylation™ Kit (Zymo Research), according to supplier instructions, to convert non-methylated cytosine residues to uracyle. Universal methylated human DNA standard (Enzo Life Sciences, ENZ-45005-0001) was use as a positive control.
The region of interest corresponding to a CpG island located in the promoter of the CSB gene ( Supplementary Fig. 7) (from −603 to −396, transcription start site (TSS) as +1: chr10:50747072 (GRCh37 coordinates)) was PCR amplified using 20 ng of bisulfite-treated genomic DNA. The primers for CSB (listed in Supplementary Table 1) were designed by web-based MethPrimer 2.0 software 66 . The PCR products (208 bp) were gel extracted using the Wizard ® SV Gel and PCR Clean-Up system (Promega) following the supplier instructions and sent for direct sequencing (Cochin Sequencing Platform, Paris, France). The level of methylation, expressed as a percentage of methyl-cytosine (%Methyl-C = (Methyl-C/(Methyl-C + C)) × 100), was analyzed at the single nucleotide resolution using the 4peaks software (Nucleobytes), in a total of eight CpG sites located at −564 (#1), ROS detection. The levels of oxidative stress were assessed measuring reduced glutathione (GSH), a strong scavenger of ROS, and its ratio with oxidized glutathione (GSSG), using the GSH/GSSG-Glo TM Luminescent assay (Promega) following the manufacturer's instruction. The GSH/GSSG ratio was assessed in 20,000 cells plated on 96-well white-wall plate using the Tecan plate reader. For detection of mitochondrial superoxide ions, cells were incubated with 5 μM MitoSOX Red Mitochondrial Superoxide indicator (Thermo-Fisher Scientific, M36008) for 10 min at 37°C.
Total ATP steady-state assays. Ten thousand cells plated on 96-well white-wall plate were treated with 10 µM oligomycin for 1 h (for glycolytic ATP) or untreated (for total ATP levels); cells were then tested with the CellTiter-Glo Luminescent assay (Promega) according to supplier instructions using a Tecan plate reader.
Statistical analysis. To determine statistical significance between groups, Student's t-test (unpaired, two-tailed; for two groups) and one-or two-way analysis of variance (ANOVA; for three or more groups) with post-hoc Tukey's or Sidak's test, were used assuming normal distribution of the data. For experiments with palbociclib the statistical significance of RT-qPCR was determined using RM (repeated measures) one-way ANOVA with post-hoc Tukey's test. Linear regression was modeled for selected pairs of transcripts using the R software. All statistical details of experiments, including the statistical test used, the number of independent experimental replications, the definition of center and dispersion, the relevant values of the test summary: for ANOVA the F ratio, the corresponding degrees of freedom (DF) for the numerator (DFn) and the denominator (DFd), and the ANOVA p-value (interaction p-value for two-way ANOVA), and for the Student's t-test the t ratio and DF, are reported in each corresponding figure legend. The post-hoc p-values are indicated in the figures panels: the exact p-value is indicated with the exception of p < 0.0001 as these values are no further specified by GraphPad Prism 6.0; post-hoc p-values > 0.05 are prevalently not indicated. All statistical analyses were performed using the GraphPad Prism 6.0 Software (GraphPad software).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data generated or analyzed during this study are included in this published article (and its supplementary information files). All data is available from the corresponding author upon reasonable request.