Early growth response 2 (EGR2) is a novel regulator of the senescence program

Senescence, a state of stable growth arrest, plays an important role in ageing and age-related diseases in vivo. Although the INK4/ARF locus is known to be essential for senescence programs, the key regulators driving p16 and ARF transcription remain largely underexplored. Using siRNA screening for modulators of the p16/pRB and ARF/p53/p21 pathways in deeply senescent human mammary epithelial cells (DS HMECs) and fibroblasts (DS HMFs), we identified EGR2 as a novel regulator of senescence. EGR2 expression is up-regulated during senescence and its ablation by siRNA in DS HMECs and HMFs transiently reverses the senescent phenotype. We demonstrate that EGR2 activates the ARF and p16 promoters and directly binds to the ARF promoter. Loss of EGR2 downregulates p16 levels and increases the pool of p16- p21- ‘reversed’ cells in the population. Moreover, EGR2 overexpression is sufficient to induce senescence. Our data suggest that EGR2 is a regulator of the p16/pRB and direct transcriptional activator of the ARF/p53/p21 pathways in senescence and a novel marker of senescence.


Introduction
The limited replicative capacity of cultured human cells, resulting in senescence, was first described by Hayflick & Moorhead (1961) and has since been implicated to play an important role during in vivo ageing and age-related diseases (van Deursen, 2014). Senescence, a stable proliferative arrest, occurs in response to diverse damaging stimuli triggering up-regulation of cyclin-dependent kinase inhibitors (CDKIs), altered gene expression and subsequent nuclear and cellular morphological changes (Sharpless and Sherr, 2015). Two families of CDKIs, including p16 INK4A (p16) and p21 Cip1/Waf1 (p21), can independently initiate senescence programs by directly binding and inhibiting cyclin-CDK complex phosphorylation of retinoblastoma (RB) (Dyson, 1998).
Study of p16 regulation has revealed numerous pathways that converge to regulate p16, and by extension the INK4/ARF locus, which also encodes p15 INK4B and p14 ARF /p19 ARF (ARF) Gil and Peters, 2006;Martin, Beach and Gil, 2014). Importantly, ARF functions to inhibit MDM2 ubiquitination and degradation of p53, leading to up-regulation of p21, a transcriptional target of p53. Thus, the INK4/ARF locus forms a pivotal link between the two key senescence initiation cascades (Zhang, Xiong and Yarbrough, 1998).
Although it is well established that ETS1 mediates p16 induction in fibroblasts by the RAS/RAF/MEK cascade during oncogenic signalling, leading to oncogene-induced senescence (Serrano et al., 1997), the upstream pathways activating the INK4/ARF locus in epithelial and fibroblast senescence are not well understood. To date, overexpression of the homeobox protein, MEOX2, has been identified to induce senescence in keratinocytes and fibroblasts by directly binding to and activating the p16 promoter (Irelan et al., 2009), and overexpression Page 4 of E2F1 induces senescence in fibroblasts via increased ARF expression (Dimri et al., 2000).
Furthermore, recent evidence has suggested that senescence is a multi-step, dynamic process throughout which the senescent phenotype evolves (Kim et al., 2013). Deep senescence (DS) takes over 7-10 days to develop post-senescence induction. For example, in epithelial cells, it is defined when cultures at p16-dependent stasis undergo no further expansion upon at least two serial passages (Lowe et al., 2015, Methods). In fibroblasts, it is further characterised by additional markers of senescence, most notably the senescence-associated secretory phenotype (SASP) (Coppé et al., 2008;Rodier et al., 2009), accompanied by elevated reactive oxygen species (ROS) levels (Passos et al., 2010;Lowe et al., 2015), and a loss of lamin B1 (Freund et al., 2012). Despite our growing understanding of the elaboration of the senescent state, there is a lack of knowledge of the key regulatory pathways upstream of the p16/pRB and ARF/p53/p21 pathways in DS.
We have previously demonstrated that DS is reversible in p16-positive primary adult human mammary epithelial cells (HMECs) using p16 siRNA transfection (Lowe et al., 2015). Of relevance, p16-dependent epithelial senescence is independent of ARF/p53/p21 pathway activation (Garbe et al., 2009), whereas senescence in primary adult human fibroblasts engages both the ARF/p53/p21 and p16/pRB pathways (Alcorta et al., 1996; Figure 1A). We took note of previous work in human neonatal foreskin fibroblasts (HCA2) which demonstrated that p53 knockdown in senescence reinitiates DNA synthesis but with limited proliferation (Gire and Wynford-Thomas, 1998), and subsequent findings that p53 or pRB inactivation in neonatal foreskin fibroblasts (BJ), with low levels of p16, reversed senescence (Beauséjour et al., 2003). However, p53 inactivation or p16 shRNA knockdown followed by p53 inactivation in foetal lung WI38 fibroblasts, with higher levels of p16, did not reverse senescence, leading the authors to suggest that activation of the p16/pRB pathway may provide a dominant second barrier to senescence reversal (Beauséjour et al., 2003). Page 5 Here, we show that DS in primary adult human fibroblasts with high p16 levels can be reversed using transfection of p16 siRNA in combination with p21 siRNA. Subsequently, we perform siRNA screens in DS HMECs and human mammary fibroblasts (HMFs) in order to further understand the key regulators upstream of the p16/pRB and ARF/p53/p21 pathways which drive senescence. In this study, we present evidence that early growth response 2 (EGR2) acts as a regulator of p16 and transcriptional activator of ARF in senescence and is a novel marker of senescence.

Reversal of deep senescence in fibroblasts
Current literature suggests that senescence is a dynamic process, and that fibroblasts in 'light' senescence (with low p16 levels) can be reversed, whereas DS fibroblasts (with high p16 levels), have entered a distinct, irreversible state (Beausejour, et al., 2003). As such, we began by asking whether fibroblast DS (with high p16 and p21 levels) is truly irreversible . Building on previous work in which we have reversed DS in p16-positive DS HMECs (Lowe et al., 2015), we hypothesised that transient knockdown using previously validated p16 (Bishop et al., 2010) together with p21 (Borgdorff et al., 2010) siRNAs in DS fibroblasts would induce a 'reversed phenotype' as characterised by a panel of senescence markers ( Figure 1A).
To investigate this hypothesis, we employed senescent HMF and human dermal fibroblasts (HDFs) that had been serially passaged to senescence and cultured for a further 21 days to ensure a deeply senescent state with high p16 and p21 levels ( Figure 1B, Figure S1, Methods), and developed an efficient protocol to introduce siRNA into these classically hard to transfect cells (Methods). Subsequently, we depleted p16 and/or p21 mRNA in DS HMFs or HDFs with potent siRNAs ( Figure S2A-B), and assessed the impact on numerous cellular and molecular markers classically associated with senescence in comparison to DS cells transfected with siGLO (a negative control targeting cyclophilin B (PPIB); 'DS+siGLO'). While depletion of p16 with siRNA in DS HMFs ('DS+p16 siRNA') did not significantly alter the arrested phenotype or cellular and molecular markers of senescence, p21 depletion ('DS+p21 siRNA') significantly increased cell number and modulated some features of senescence morphology towards an Page 6 early proliferating (EP) phenotype, namely, significantly decreased cell area, nuclear area, and nuclear elongation; and significantly increased nuclear roundness and cell elongation ( Figure   1B-D). Strikingly, depletion of both p16 and p21 in DS HMFs and HDFs ('DS+p16+p21 siRNA') stimulated a stronger reversion to an EP morphology as characterised by multiple cellular and molecular markers ( Figure 1C-E, Figure S3). Using a panel of established senescence markers, we sought to explore further the consequences of p16 and p21 knockdown. Quantification of proliferation using 5-bromo-2'-deoxyuridine (BrdU) incorporation confirmed the significantly increased cycling activity of 'DS+p16+p21 siRNA' HMFs compared to 'DS+siGLO' HMFs ( Figure   1F). Interestingly, the percentage of BrdU positive cells in 'DS+p16+p21 siRNA' HMFs was higher even than that observed in EP HMFs ( Figure S1D), indicating that a greater proportion of the 'DS+p16+p21 siRNA' HMFs progress through S phase during the 16-hour BrdU pulse than the EP HMFs. In agreement with the reversed phenotype, 'DS+p16+p21 siRNA' HMFs also displayed down-regulation of the SASP proinflammatory signature in comparison to 'DS+siGLO' HMFs, as illustrated by significantly decreased expression of the cytokine IL-6 ( Figure 1G) and decreased IL-6 secretion ( Figure 1H). In line with the literature, IL-8 expression and secretion was also investigated but found not to be a feature of the SASP in DS HMFs (data not shown; Coppé et al., 2008). We also measured levels of 8-oxoguanine, a marker of reactive oxygen species and oxidative damage, and found a significant decrease in the 'DS+p16+p21 siRNA' population compared to 'DS+siGLO' HMFs ( Figure 1I). Furthermore, investigation of senescence-associated beta-galactosidase (SA-β-Gal) activity in DS HMFs following transfection, suggested a potential decrease in SA-β-Gal activity in 'DS+p16+p21 siRNA' HMFs compared to 'DS+siGLO' HMFs ( Figure 1J). Together, our data indicate that senescence appears to be transiently reversed in the 'DS+p16+p21 siRNA' HMFs.

siRNA screening reveals novel regulators of senescence
We next sought to identify novel genes that regulate the senescent phenotype. Initially, we interrogated our previously published gene expression datasets to identify genes whose expression was significantly up-regulated in HMEC DS relative to EP HMECs, and downregulated following p16 siRNA knockdown (Figure 2A; Lowe et al., 2015;GEO: GSE58035, q<0.05). In order to distinguish between the genes driving senescence and downstream 'passenger' genes, a siRNA screen of the top 190 genes was performed in DS HMECs (Supplementary Table 1). Each gene was targeted by a pool of three siRNAs (30nM Ambion).

Page 7
To determine the effect on a panel of senescence markers for each of the 190 siRNAs, the siGLO transfected control provided a baseline for Z score generation. Using high-content analysis, 28 siRNAs (14.7%) were identified to strongly induce reversal in the DS HMECs as defined by an increase in cell number and the loss of a panel of senescence markers (i.e. mimicking the HMEC phenotype generated by p16 siRNA). Accordingly, these 28 genes were classified as potential regulators of senescence ( Figure 2B).
To further investigate the relationships between these potential 28 regulators of senescence, we constructed a protein interaction map. Briefly, these 28 genes were probed for protein interactors using the BioGRID database ( Figure S4). Using Panther, KEGG pathways and Gene Ontology (GO) bioinformatics tools, 61 genes emerged (the 28 previously identified regulators, which includes p16, and 33 protein interactors) which grouped into six functional categories: immune response; cell adhesion/cytoskeleton; metabolism; transcription; growth/proliferation; and protein/vesicle trafficking ( Figure S5). We next asked whether the siRNA hits that emerged from the initial HMEC screen could also play a role in senescence in DS HMFs using this extended protein interaction network. As DS HMF reversal was found to require siRNA knockdown of both p16 and p21, we hypothesised that the regulators identified in the DS HMEC screen may additionally require knockdown of either the p16/pRB or the ARF/p53/p21 pathway to induce reversal in the DS HMFs.
Using the same approach as described for the DS HMEC siRNA screen, a hit list was generated for each of the three conditions (Groups 1, 2, and 3) ( Figure 2D-E). One siRNA transfected individually (Group 1) was defined as a hit, namely early growth response 2 (EGR2), a transcription factor involved in several cellular processes including cell cycle and proliferation (Parkinson et al., 2004;Srinivasan et al., 2012). Two siRNAs in combination with p16 siRNA (Group 2), fraser extracellular matrix complex subunit1 (FRAS1) and ring protein 20 (RNF20), an E3 ubiquitin ligase, were defined as hits ( Figure 2D-E). Finally, 45 of the 60 siRNAs in Page 8 combination with p21 siRNA (Group 3) were defined as hits. Strikingly, eight of these 45 siRNAs induced an increase in cell number similar to the 'DS+p16+p21 siRNA' DS HMF control, including EGR2 and S100A4 siRNA. As the 28 regulator siRNAs in the screen were identified as hits for senescence reversal in p16-dependent DS HMECs, it is perhaps unsurprising that 21 of these siRNAs were identified as hits requiring additional knockdown of the ARF/p53/p21 pathway to reverse senescence in DS HMFs. Furthermore, 24 of the 33 interactors investigated in this screen were also identified as Group 3 hits, highlighting the utility of the bioinformatics approach.
The top candidates from Group 1 (EGR2) and Group 2 (FRAS1), together with an additional 12 candidates from Group 3 were selected for further investigation (HIF1A, HSP90AA1, S100A4, BHLHE41, FN1, ACTG1, PPFIA1, JUP, CD9, PDCD6IP, MYL12A, and DAPL1). We performed a more detailed, independent screen with these 14 siRNAs using multi-parameter analysis of senescence-associated morphological markers with four conditions: 30nM siRNA individually (Group 1), 15nM siRNA in combination with 15nM p16 siRNA (Group 2); or 15nM siRNA in combination with 15nM p21 siRNA (Group 3) ( Figure S6). In addition, the impact of an increased individual siRNA dose (60nM, Group 1B) was performed to identify the most potent reversed phenotype ( Figure S6). Strikingly, 11 of the 14 siRNAs transfected individually significantly decreased cell area in a dose-dependent manner (Group 1, Group 1B; Figure S7). Of these, six siRNAs transfected individually also significantly decreased nuclear area in a dose-dependent manner (Group 1, Group 1B) and EGR2 was the only siRNA transfected individually (Group 1, Group 1B) to also significantly increase cell elongation in a dose-dependent manner. As such, EGR2 was the only siRNA that did not require knockdown of p16 and p21 to significantly increase cell number ( Figure 2) and significantly alter three senescence-associated morphologies towards a reversed phenotype in a dose-dependent manner ( Figure S7). Taken together, these data suggest that EGR2 may be acting upstream of p16 in epithelial DS and p16 and p21 in fibroblast DS. To our knowledge, no direct relationship between EGR2 and senescence has previously been described, and thus we sought to explore this finding in more detail.

EGR2 is a novel regulator of senescence
Page 9 As EGR2 was identified as the top hit for reversal in both the DS HMEC and HMF screens, we next wanted to explore the role of EGR2 in senescence. First, we validated mRNA knockdown for the EGR2 siRNA pool in DS HMFs ( Figure 3A), and subsequently deconvoluted the EGR2 siRNA pool (EGR2 1, 2 and 3) to determine the efficacy of each individual siRNA targeting EGR2. 'EGR2 1' siRNA was the least potent ( Figure S8), which was subsequently reflected in the phenotype ( Figure 3B-D). Using multi-parameter phenotypic analysis to control for offtarget effects, we identified 'EGR2 3' siRNA as the most potent siRNA transfected individually ( Figure S8) which significantly increased cell number and significantly reversed cell area, nuclear area and cell elongation ( Figure Figure 3D). Further characterisation of the changes to the senescence phenotype following ablation of the EGR2 in DS HMFs revealed a significant down-regulation of the SASP factor, IL-6, at the transcript level ( Figure 3E) and at the secreted protein level ( Figure 3F). As mentioned previously, IL-8 is known not to be a feature of the DS HMF SASP (data not shown, Coppé et al., 2008).
It is important to note that the human genome encodes four EGR transcription factors, EGR1-4, that share three highly homologous DNA binding zinc finger domains that can bind to the same GC-rich consensus DNA binding motif (Beckmann and Wilce, 1997). In addition, a role for EGR1 has previously been implicated in RAF-induced oncogene-induced senescence (OIS) of human BJ fibroblasts (Carvalho et al., 2019) and replicative senescence (RS) of murine embryonic fibroblasts (Krones-Herzig, Adamson and Mercola, 2003). As such, we wanted to investigate the expression of EGR family members in HMEC epithelial senescence and HMF senescence. EGR2 was the only member of the EGR family with significantly increased gene expression in DS compared to EP HMECs, and EGR2 was the only member of the EGR family whose gene expression significantly decreased in reversed HMECs (GEO: GSE58035).
Furthermore, investigation of EGR family member expression levels in EP and DS HMFs revealed a significant increase in EGR2, but not EGR1, EGR3 or EGR4 expression levels ( Figure   3G). Collectively, these data suggest that EGR2 might be the key EGR family member acting to regulate senescence in HMECs and HMFs.
Page 10 Subsequently, to further explore whether EGR2 activity and regulation is conserved across multiple senescence models and occurs in vivo in human tissues, we performed datamining of existing GEO datasets for HDF RS, bleomycin-induced stress-induced premature senescence (SIPS) in BJ foreskin fibroblasts, and RAS oncogene-induced senescence (OIS) in WI38 lung fibroblasts (Martínez-Zamudio et al., 2020) in vitro, as well as human skin and whole-blood with age in vivo (STAR Methods). The abundance of EGR2 increased during senescence across all three senescence models ( Figure S9A, p<0.05). Importantly, EGR2 expression increased in vivo in aged human skin. In addition, a recent whole-blood gene expression meta-analysis looking at over 7,000 human samples showed that EGR2 expression significantly increases with age ( Figure S9, p<0.01, Peters et al., 2015). Thus, increased EGR2 expression appears to be a feature of both in vitro senescence and in vivo ageing signatures.
EGR2 possesses a nuclear localisation signal and functions to regulate gene transcription within the nucleus, thus we hypothesised that functional EGR2 would be localised within the nucleus during senescence. Immunofluorescence staining in EP and DS HMECs revealed a significant increase of nuclear EGR2 foci in DS HMECs compared to the EP population, and in DS HMFs compared to EP HMFs ( Figure 3H-I). Further investigation of EGR2 levels in a third model of senescence, oncogene-induced senescence (OIS) in IMR90 lung fibroblasts ( Figure   S9B), identified a significant increase in nuclear EGR2 foci in OIS fibroblasts compared to the vector control ( Figure S9C). These findings support our previous mining of mRNA datasets and show that an increase in EGR2 is also observed at the protein level with the expected subcellular localisation ( Figure 3H-I), thus identifying EGR2 as a novel marker of senescence in both DS HMECs, HMFs, and OIS IMR90 fibroblasts.
Finally, to explore the potential mechanisms through which EGR2 may be driving senescence and identify a panel of genes that might be regulated by EGR2 during senescence, we asked if genes identified to be up-regulated in senescence in the HMEC gene expression array were enriched for the previously published EGR2 consensus binding sequences (ACGCCCACGCA; Jolma et al., 2013;Mathelier et al., 2016) compared to randomly sampled background gene sets (Figure S10A-C). Interestingly, there was a small but significant enrichment for EGR2 binding sites at the promoters of genes up-regulated in HMEC DS. Furthermore, ten of these Page 11 genes were identified as hits for senescence reversal in the DS HMEC screen, including p16, and nine of these were also identified as hits in the HMF siRNA screen, including the top hit S100A4, suggesting that EGR2 may act as a senescence regulator by activating the expression of these genes.

EGR2 regulates senescence via the p16/pRB and ARF/p53/p21 pathways
Although previous work has identified EGR2 binding to the p21 promoter (Srinivasan et al., 2012;Zheng et al., 2013), no investigation has yet been performed on other pathways of senescence ( Figure 4J). Further examination of the INK4/ARF locus revealed previously unreported hypothetical EGR2 binding sites (ACGCCCACGCA; Jolma et al., 2013;Mathelier et al., 2016) in the p16, p15 and ARF promoter regions, indicating a potential for EGR2 to bind to and regulate expression of p16, p15 and ARF. As p15 was found not to be expressed in DS HMFs ( Figure S10D), we explored the potential action of EGR2 on the p16 and ARF promoters.
To this end, we first investigated activation of the ARF promoter using transiently cotransfected U2OS cells with an expression vector encoding one of each of the four members of the EGR family or E2F1, a transcription factor known to directly up-regulate ARF which acts as a positive control (Dimri et al., 2000), together with pGL3 luciferase reporter constructs harbouring either the promoter sequence 800bp or 3.4kb upstream of the transcriptional start site of ARF (pGL3 ARF 800 or plGL3 ARF 3.4, respectively, Figure 4A). Cells transfected with the pGL3 ARF 800 or with the complete ARF promoter, pGL3 ARF 3.4, displayed a significant increase in luciferase activity following transfection with the EGR2 expression vector or E2F1 positive control, but not EGR1, EGR3, or EGR4 expression vectors, thus confirming EGR2 as a direct activator of the ARF promoter ( Figure 4A, Figure S11).
Validation of the interaction between EGR2 and the ARF promoter was performed using chromatin immunoprecipitation (ChIP) on cross-linked DNA from quiescent interleukin 2 (IL-2) dependent Kit225 human T-lymphocytes, with low levels of ARF expression, and Kit225 cells following IL-2 activation which results in increased ARF expression (del Arroyo et al., 2007). Subsequently, chromatin immunoprecipitation was performed with polyclonal antibodies against EGR2, or E2F1, which acted as a positive control. Addition of IL-2 to Kit225 cells resulted in increased binding of E2F1 and EGR2 to the ARF promoter, demonstrating that EGR2 can be detected at the endogenous ARF promoter ( Figure 4B). Page 12 In order to further explore the role of EGR2 in senescence, we introduced retroviral particles expressing EGR2 cDNA into normal human Hs68 diploid fibroblasts. In line with our previous observations that loss of EGR2 reverses senescence, stable overexpression of EGR2 was sufficient to induce proliferation arrest ( Figure 4C-D). Interestingly, p16-/-Leiden cells and p16+/-Q cells also underwent proliferation arrest following overexpression of EGR2, indicating EGR2-mediated up-regulation of ARF is sufficient to induce senescence in the absence of p16 ( Figure 4C-D).
We next explored activation of the p16 promoter and found that cells co-transfected with one of each of the four members of the EGR2 family, or E2F1, together with a pGL3 p16 construct displayed a significant increase in luciferase assay activation with the EGR2 or EGR4 expression vectors, or E2F1 positive control, confirming EGR2 and EGR4 as direct activators of the p16 promoter ( Figure 4E). As EGR4 expression is not increased in DS compared to EP HMECs or HMFs ((GEO: GSE58035, Figure 3G), we suggest that EGR2 may be important for activation of the p16 promoter in epithelial and fibroblast senescence.
If EGR2 functions to activate the p16 promoter and up-regulate p16 expression, we hypothesised that ablation of EGR2 in senescent cells would lead to a decrease in p16 levels. compared to the DS+p21 siRNA HMFs, an increase similar to that seen in the reversed DS+p16+p21 HMFs ( Figure 4H). Taken together, these data indicate that EGR2 functions to up-regulate p16 and ARF expression in senescence which is sufficient to induce proliferation arrest, demonstrating that EGR2 acts as a novel regulator upstream of p16/pRB and transcriptional activator of ARF/p53/p21 pathways in senescence ( Figure 4I).

Discussion
Here, we show that DS can be transiently reversed in human fibroblasts using p16 siRNA in combination with p21 siRNA transfection, as characterised by the loss of a panel of senescence markers. It is important to note here that we have shown that siRNA mediated reversal of DS HMFs is transient, with population growth slowing and cells reverting to a senescence morphology by seven days post-transfection. Further investigation is required to assess the effect of long-term, stable knockdown on DS cells, including the impact on DNA damage and telomeres. However, as previous work in our group demonstrated that p16 siRNA knockdown can reverse DS HMECs, the discovery that p16+p21 siRNA knockdown can transiently reverse DS HMFs provided a unique opportunity for uncovering novel senescence regulators in epithelial and fibroblast DS. Using siRNA screening, we identified novel regulators of senescence in HMECs and HMFs, including the transcription factor EGR2, extracellular matrix protein FRAS1, E3 ubiquitin ligase RNF20, and calcium-binding protein S100A4. Further investigation of the top hit, EGR2, revealed that EGR2 ablation enables resumption of the cell cycle, reversed senescence-associated morphologies and decreased expression and secretion of the SASP component, IL-6. We demonstrate that EGR2 accumulates during in vitro senescence in DS HMECs,DS HMFs, and OIS IMR90 lung fibroblasts. Furthermore, we re-mined existing datasets to reveal an increase in EGR2 expression in RS HDFs, SIPS BJ fibroblasts, OIS WI38 fibroblasts, and in human tissue during in vivo ageing. As such, we have identified EGR2 as a novel marker of senescence across multiple senescence models, including p16-dependent epithelial DS, p16-and p21-dependent fibroblast DS, fibroblast RS, OIS and SIPS. Examination of genes differentially expressed in DS HMECs identified EGR2 binding sites in p16 and nine siRNAs found to reverse DS HMEC and HMFs, including one top reversal hit in the DS HMFs, S100A4. Further investigation of the INK4/ARF locus revealed previously unreported EGR2 binding sites in all the p16, p15 and ARF promoters. In support of this, we demonstrated that EGR2 activates the p16 and ARF promoters and that EGR2 directly binds to the ARF promoter. Furthermore, stable EGR2 overexpression was sufficient to induce proliferation arrest in the presence or absence of p16.
Lastly, we observed a decrease in p16 protein levels in DS HMFs following EGR2 knockdown Page 14 and an increase in the p16-p21-double negative subpopulation in DS HMFs following EGR2 and p21 knockdown. Given that EGR2 overexpression activates the p16 promoter, and that silencing EGR2 downregulates p16 protein levels in DS HMFs, it is likely that EGR2 directly binds to the p16 promoter. Unfortunately, discontinuation of ChIP-quality EGR2 antibodies precluded further experiments investigating this in DS cells. Further studies are needed to confirm the direct role of EGR2 activating p16 expression.
Mutations in EGR2 have been identified to lead to inherited peripheral neuropathies, including Charcot-Marie-Tooth Type 1 (Šafka Brožková et al., 2012), a demyelinating form associated with dysregulated Schwann cell proliferation and cell-cycle exit (Atanasoski et al., 2006). Accumulating evidence indicates that EGR2, a transcription factor, plays the role of regulator in these processes (Topilko et al., 1994;Zorick et al., 1996;Decker, 2006) and has been shown to directly bind to the p21 promoter in myelinating rat sciatic nerve (Srinivasan et al., 2012). In addition, a role for EGR2 as a tumour suppressor has been implicated in many tumour cell types (Unoki and Nakamura, 2003), and elevated expression of EGR2 is a favourable prognostic factor in breast cancer (TCGA, 5 year survival for high expressers = 84%; 5 year survival for low expressers = 73%; p=0.000073). Despite this, little attention has been paid to its role in senescence. In the present report, our findings indicate a functional role of EGR2 in regulation of p16 and transcriptional activation of ARF in senescence.
Importantly, whilst our data demonstrates a role for EGR2 in regulation of senescence, transient EGR2 reversal in DS cells does not delineate between the activity of EGR2 in senescence onset or maintenance. Future studies using stable EGR2 knockdown prior to senescence entry should be performed in order to dissect the roles of EGR2 in the onset and/or maintenance of senescence.

Concluding remarks
Our work adds to the growing list of pathways known to directly regulate senescence. This includes p16 transcriptional repressors, such as homeobox protein HLX1 (Martin et al., 2013) and the N-terminal fragment of the GLI2 transcription factor (Bishop et al., 2010), as well as p16 transcriptional activators such as ETS1 (Ohtani et al., 2001), and homeodomain protein MEOX2 (Irelan et al., 2009). Importantly, we have demonstrated that EGR2 functions as a Page 15 regulator of p16/pRB and direct activator of the ARF/p53/p21 pathways, thus controlling both axes of the senescence program.
It is well established that expression of p16 increases with age in human tissues (Krishnamurthy et al., 2006), senescent cells accumulate in sites of age-related diseases (Naylor, Baker and van Deursen, 2012), and selective clearance of p16-positive senescent cells in mice has been shown to improve health-and lifespan (Baker et al., 2011(Baker et al., , 2016. As such, regulation of the p16/pRB and ARF/p53/p21 pathways by EGR2 in senescence may play an important role in ageing and age-related diseases. Furthermore, ten of these genes were identified as hits for senescence reversal in the DS HMEC screen, including p16, and nine of these were also identified as hits in the HMF siRNA screen, including the top hit S100A4, suggesting that EGR2 may act as a senescence regulator by activating the expression of these genes.
Interestingly, EGR2 as a transcription factor has the potential to regulate a network of genes in senescence, and nine hits which reversed DS HMECs and HMFs were identified to possess an EGR2 binding site, thus we hypothesise that EGR2 may potentially regulate the expression of these genes in senescence, although this has yet to be investigated further. Future exploration of the transcriptome regulated by EGR2 in senescence could provide new insights into regulation of the senescence program and potentially identify essential senescence mediators, which could be exploited to eliminate senescent cells. As implications for senescence have been described in vivo for organismal ageing and age-related diseases, furthering our understanding of this network in senescence could enable identification of therapeutic targets for treatment of ageing and age-related diseases.

Cells and reagents
Normal finite life-span HMECs and HMFs were obtained from reduction mammoplasty tissues of a 21-year-old individual, specimen 184, and 16-year-old individual, specimen 48, respectively, and were cultured as previously described (Garbe et al., 2009). Independent Page 16 HMEC cultures were serially passaged from passage 6 (P6; early proliferating, EP) until p16dependent, p21-independent stasis. Deeply senescent cultures underwent no further expansion upon at least two further weeks in culture (DS HMECs; Romanov et al., 2001;Garbe et al., 2009;Lowe et al., 2015), and independent HMF cultures were serially passaged from P4 until the population reached senescence at P29. DS HMFs underwent no further expansion upon at least three further weeks in culture (P29+3). Cells were cultured at 37°C in the presence of 5% CO2 and atmospheric O2. All cells were routinely tested for mycoplasma and shown to be negative.
IMR90 ER:STOP (vector) or ER:RAS (OIS) IMR90 foetal lung fibroblasts were produced as described in (Hari et al., 2019) and were a kind gift provided by Juan Carlos Acosta. These were maintained in DMEM supplemented with 10% FBS and 2mM L-glutamine.
U2OS cells, primary human fibroblast strain Hs68, and Kit225 T-lymphocyte cell line were maintained as previously described (del Arroyo et al., 2007). Leiden and Q cells were maintained as previously described (Irelan et al., 2009).
Please also see Supporting Information.

Tyler_Supporting Information
Supplementary Page 1

Experimental Procedures -Supporting Information siRNA screening and Z score generation
In the DS HMEC screen, p16 siRNA was amongst the 28 driver siRNAs identified to reverse HMEC p16-dependent DS (Lowe et al., 2015). However, in DS HMFs, p16 siRNA alone was found to not be sufficient to reverse senescence. With this in mind, p16 siRNA was not included as a target siRNA in the DS HMF screen, bringing the total number to 60 target siRNAs (27 drivers and 33 interactors).
Z score = (mean value of two independent experiments for experimental siRNA -mean value of two independent experiments for siGLO)/SD for siGLO of two independent experiments.
For each of the parameters analysed, significance was defined as more than one Z score away from the siGLO mean in order to increase the window for hit detection and allow as many hits to be identified as possible. Z scores are presented as a heatmap.

Tyler_Supporting Information
Supplementary Page 2 was conducted using ImageJ software. Density levels were corrected for protein loading and were expressed relative to the negative siRNA control.

Quantitative RT-PCR (RTqPCR)
Total RNA was isolated using Qiazol (Qiagen) according to the manufacturer's protocol. One microgram of total RNA was reverse transcribed by the Superscript III Reverse Transcriptase (Thermo Fisher Scientific) following manufacturer's protocol. RTqPCR reactions were performed with SYBR Green Master Mix (ABI) using a 7500 Fast Real-Time PCR System (Applied Biosystems). For siRNA knockdown experiments, RNA was extracted from DS HMFs five days post-transfection. GAPDH levels were quantified for each cDNA sample in separate RTqPCR reactions and were used as an endogenous control. Target gene-expression levels were quantified using target specific probes. Values were normalised to the internal GAPDH control and expressed relative to siGLO transfected control levels (100%). All RTqPCR reactions were run in duplicate for two independent samples.

Enzyme-linked immunosorbent assay (ELISA)
For IL-6 analysis by ELISA, equal volumes of conditioned medium were used and assay performed as per the manufacturer's instructions (R&D Systems, Human Il-6 DuoSet ELISA DY206). Each sample was represented twice on the plate. The absorbance readings were taken at 450nm and 570nm using a CLARIOstar Plus multi-mode plate reader (BMG Labtech).
Protein concentration was then estimated according to a calibration curve obtained from the absorbance values of a dilution series of the supplied standard protein control.

Database searches
Protein interaction datasets were generated using the BioGRID bioinformatics database (http://www.thebiogrid.org). These interactions were then overlaid to generate a network requiring that each interactor generated a chain with at least two other drivers, revealing a total of 33 protein interactions. Using this method, only one protein interaction network was generated. We searched KEGG pathway (http://www.genome.jp/kegg/pathway.html) and PANTHER (http://www.pantherdb.org) databases to assign functional annotations to the 28 hits that strongly induced the reversal phenotype ('drivers') in the DS HMEC siRNA screen as well as the 14 hits in the DS HMF siRNA screen.

Quantification and Statistical Analysis
An un-paired, two-tailed t-test was performed to compare the means of two groups using the Microsoft Office Excel Analysis ToolPak (Microsoft, USA). A one-way analysis of variance (ANOVA) was used to analyse the differences between the means of three or more independent groups using Prism 7 (GraphPad Software Inc., USA). A two-way ANOVA was used to analyse the differences between multiple subgroups within multiple independent groups using Prism 7. Post-hoc statistical analysis was performed using either a Dunnett's or Tukey's multiple comparison test. The Dunnett's test was used to compare every mean to a control mean, whereas the Tukey test was used to compare every mean with every other mean. Nuclear intensity thresholds were established for p16 and p21 to define positive or negative nuclei. Nuclei were classified into four subgroups: p16 and p21 negative (p16-p21-); p16 negative and p21 positive (p16-p21+); p16 positive and p21 negative (p16+ p21-); and p16 and p21 positive (p16+ p21+  Western blot depicting p16, p21, and GAPDH levels in DS HMFs following transfection with siGLO, p16 siRNA, p21 siRNA, and p16 in combination with p21 siRNA (p16+p21 siRNA).
DAPI (blue), p16 (green), p21 (red). Size bar, 100μm. Nuclear and cellular morphologies, BrdU nuclear intensity, p16 and p21 nuclear intensities, and 8-oxoguanine cellular density was quantitated. Using the secondary only control, a nuclear intensity threshold was established to define BrdU positive or negative nuclei. Nuclear intensity thresholds were established for p16 and p21 to define positive or negative nuclei.