ATM suppresses c-Myc overexpression in the mammary epithelium in response to estrogen

SUMMARY ATM gene mutation carriers are predisposed to estrogen-receptor-positive breast cancer (BC). ATM prevents BC oncogenesis by activating p53 in every cell; however, much remains unknown about tissue-specific oncogenesis after ATM loss. Here, we report that ATM controls the early transcriptional response to estrogens. This response depends on topoisomerase II (TOP2), which generates TOP2-DNA double-strand break (DSB) complexes and rejoins the breaks. When TOP2-mediated ligation fails, ATM facilitates DSB repair. After estrogen exposure, TOP2-dependent DSBs arise at the c-MYC enhancer in human BC cells, and their defective repair changes the activation profile of enhancers and induces the overexpression of many genes, including the c-MYC oncogene. CRISPR/Cas9 cleavage at the enhancer also causes c-MYC overexpression, indicating that this DSB causes c-MYC overexpression. Estrogen treatment induced c-Myc protein overexpression in mammary epithelial cells of ATM-deficient mice. In conclusion, ATM suppresses the c-Myc-driven proliferative effects of estrogens, possibly explaining such tissue-specific oncogenesis.


INTRODUCTION
One in eight women has breast cancer (BC) in their lifetime, and approximately 67%-80% of BCs are estrogen receptor positive (ER + ). 1,2 Approximately 5%-10% of BCs are hereditary, with mutations in the BRCA1, BRCA2, and ATM (ataxia telangiectasia mutated) genes accounting for most cases. [3][4][5][6] Approximately 0.5%-1% of the general population carries a germline mutation in the ATM gene. 7 Women carrying these germline mutations develop BC with high penetrance after loss of their intact ATM allele. 4,5,8,9 Mechanistically, this loss causes ER + BCs, while loss of the intact BRCA1 or BRCA2 allele causes ER-negative BCs. 6,10 Considering that the loss of heterozygosity (LOH) events occurs at extremely low frequency (approximately 10 −5 per base) even in cancer cells, the loss of ATM in a small number of cells increases the risk of BC development by five times in females aged younger than 50 years. 11 The molecular mechanism underlying this dramatic increase in BC development after ATM loss remains unclear.
ATM controls the cellular response to DNA double-strand breaks (DSBs). In response to DSBs, the ATM kinase is immediately activated 12 to stabilize the p53 tumor suppressor protein via phosphorylation 13 and stimulate the DNA damage checkpoint pathway. 14,15 ATM promotes homology-directed repair (HDR), which functions as the dominant DSB repair pathway during DNA replication. [16][17][18][19] Although the damage checkpoint and HDR prevent mutagenesis in all cycling cells, the mechanism by which the phenotype resulting from defective ATM is manifested selectively in estrogen-regulated tissue remains a central unresolved question in ATM biology.
Estrogens bind to ERs α/β and strongly stimulate the proliferation of normal mammary epithelial cells and BCs by inducing c-MYC oncogene expression. [20][21][22] Because of the extremely high potential for estrogens to stimulate cellular proliferation, ER antagonists are currently the first-line therapy for patients with ER + BC. 23 Ligand-activated ERs bind to enhancers, leading to the transient transcription of multiple genes. [24][25][26][27] 17β-Estradiol (E2) exposure causes the recruitment of the FoxA1 transcription factor to genomic DNA independent of ERs. Various growth signals, including estrogens, transiently induce c-MYC oncogene expression by controlling clustered enhancer elements known as superenhancers. [28][29][30] c-Myc protein is closely associated with malignant tumor aggression. 31,32 Even a modest increase in c-Myc expression can significantly enhance oncogenesis in mice. 33,34 The early transcriptional response to various extracellular signal molecules, including estrogens, cytokines, and neurotransmitters, is achieved by the activation of topoisomerase II (TOP2) catalysis at promoters and possibly enhancers. [35][36][37][38][39][40][41][42][43][44][45] Of the two known TOP2 enzymes, TOP2A is expressed in cycling cells, while TOP2B is expressed ubiquitously and plays a role in transcription in post-mitotic cells (reviewed in ref. 35 ). TOP2 catalyzes strand passage reactions, which involve the movement of one intact double-stranded DNA duplex through a transient enzyme-bridged break in another (gated helix). 28,[45][46][47] This transient break is also known as a TOP2-DNA cleavage complex (TOP2cc), in which TOP2 is covalently bound to the 5′ end of the DNA DSB, which is repaired by the intrinsic ligation activity of TOP2. The reaction catalyzed by TOP2 occasionally becomes "abortive," leading to the spontaneous generation of stalled TOP2ccs 48,49 (reviewed in ref. 45,50,51 ). Etoposide, a widely used anti-cancer agent, inhibits this ligation step, generating "stalled" TOP2ccs. 52 Stalled TOP2ccs are repaired by non-homologous end-joining (NHEJ), the dominant DSB repair pathway in G 0 /G 1 phase cells. 45 The direct ligation of stalled TOP2ccs by NHEJ requires prior removal of 5′ TOP2 adducts 53,54 by tyrosyl DNA phosphodiesterase-2 (TDP2) 55 and the MRE11 endonuclease 49,53,56,57 (reviewed in ref. 45,58 ). Data suggest that MRE11 collaborates with BRCA1 and CtIP to remove 5′ TOP2 adducts from DSB ends to facilitate NHEJ. 53,[59][60][61][62][63][64][65][66][67] ATM and TDP2 function independently of each other in the repair of stalled TOP2ccs in primary mouse embryonic fibroblasts at G 1 phase. 68 It is unclear whether ATM is required for the repair of stalled TOP2ccs in other cells and tissues when TDP2 is present.
During the early transcriptional response, highly active TOP2 frequently generates spontaneously arising stalled TOP2ccs at promoters. [35][36][37][38][39][40][41][42][43][44][45]69 The resulting DSBs are ligated via NHEJ. 38,60,70,71 Whether a similar pattern of stalled TOP2ccs formation occurs at enhancers in this phase is unclear. The effect of these DSBs at enhancers on the transcriptional response when stalled TOP2ccs are left unrepaired is also unknown; investigations of this issue are hindered by difficulties in measuring the activation kinetics of enhancers during the response to extracellular signals.
While genome-wide analyses of epigenetic markers have un-covered many constitutively active enhancers, 72 the ability to identify enhancers that activate target genes only transiently in response to extracellular stimuli remains a challenge. Nevertheless, the activity of extracellular stimuli-dependent enhancers can be measured by identifying enhancer RNAs (eRNAs), which are bidirectionally transcribed from active enhancers, as their activity correlates with eRNA expression. [73][74][75][76][77][78] eRNAs can be detected using an approach known as cap analysis of gene expression (CAGE), [73][74][75][76][77][78] in which RNAs are sequenced from the 5′ RNA cap. 79 However, its sensitivity is limited by the short average half-life of eRNAs (only 1 min), which is approximately 50 times shorter than that of mRNAs. 80 As a new CAGE method, native elongating transcript-CAGE (NET-CAGE) overcomes this problem by selectively examining nascent RNAs complexed with RNA polymerase II (Pol2) and undergoing productive elongation in vivo. 76 To prevent the release of paused Pol2 during in vitro RNA extraction (reviewed in ref. 81 ), NET-CAGE involves the extraction of RNA with the Pol2 inhibitor, α-amanitin. These methods allow NET-CAGE to accurately measure the activation kinetics of both individual enhancers and their target promoters in response to extracellular signals. 76 In the current study, NET-CAGE revealed that the defective repair of stalled TOP2ccs in cells deficient in ATM and TDP2 significantly changed the activation kinetics of enhancers in response to E2. The data led us hypothesize that stalled TOP2ccs occur frequently at estrogen-dependent enhancers of oncogenes and defective DSB repair promotes oncogenesis by causing dysregulation of the response of oncogenes to estrogen.
Here, we aimed to better understand the mechanism by which the risk of oncogenesis is dramatically enhanced upon selective LOH of the ATM gene in mammary epithelial cells.
To this end, we explored a potential role of ATM in estrogen-mediated oncogenesis in the mammary tissue. In brief, we revealed that ATM is required for repairing stalled TOP2ccs, which occur at enhancers of the c-MYC gene in human BC cells during estrogen exposure.
Defective repair of such DSBs causes overexpression of the c-Myc protein in human BC cells following E2 exposure and in mammary epithelial cells upon an intraperitoneal (i.p.) injection of E2 into mice, with the high expression continuing for 24 h. Our study highlights the role of ATM in suppressing the oncogenic effect of estrogens and preventing ER + BC development.

ATM prevents estrogen-induced genomic instability in mouse mammary epithelial cells
We first explored whether ATM promoted the repair of E2-induced DSBs. To this end, we immuno-stained the DSB marker 53BP1 in mouse mammary luminal epithelial cells ( Figure 1A). The i.p. injection of E2 (i.p.) into wild-type C57BL/6 mice increased the number of 53BP1 + epithelial cells by 1.6-fold, while injection of E2 into Atm-deficient C57BL/6 mice (hereafter ATM −/− B6 mice) caused a 6-fold increase (p < 0.005, unpaired two-tailed t test) in this population ( Figure 1B). Thus, ATM suppresses the genotoxic effect of E2. To exclude the effect of the genetic background (C57BL/6) on this genotoxic effect, we analyzed ATM −/− mice on a mixed C57BL/6 3 129/Sv genetic background (hereafter ATM −/− B6;129 mice). Here, we obtained consistent data despite the differential genetic background of each mouse ( Figures 1C and S1B). The mammary gland morphology was similar in wild-type and ATM −/− B6;129 mice ( Figure S1A). E2-induced DSBs were left unrepaired even at 24 h after E2 injection in approximately 35% of the ATM −/− epithelial cells (p < 0.005, unpaired two-tailed t test) (Figures 1C and S1B), with more than 85% of the cells in the G 0 /G 1 82 phases. Collectively, ATM prevents the accumulation of E2-induced DSBs in mouse mammary gland epithelial cells in the G 0 /G 1 phases.
Compared with the injection of E2 alone, co-injection of the ATM inhibitor (ATMi), KU-55933, and E2 into wild-type B6 mice also caused a 5-fold increase (36%-6%)/(10%-4%) (p < 0.005) in the number of 53BP1 + epithelial cells detected at 6 h, with more than 50% E2-induced DSBs remaining unrepaired even at 24 h ( Figures 1D and S1C). Despite this increase in the number of 53BP1 + epithelial cells, there was no increase in the number of ERα-negative non-epithelial cells 22 ( Figure S1D). These data indicate that ATM has a crucial role in repairing E2-induced DSBs in mouse mammary epithelial cells.

ATM promotes the repair of E2-induced TOP2-dependent DSBs in human BC cells
Having determined the prominent genotoxicity of E2 in Atm-deficient mice, we next examined whether this finding was relevant to human ER + BC cells and the nature of E2-induced DSBs. To do so, we analyzed E2-induced DSBs in serum-starved MCF-7 human BC cells in the G 0 /G 1 phases (Figures S2A and S2B; Key Resource Table). We pulsed cells with E2 for 2 h before incubating them in E2-free media for an additional 2 h and 22 h and then counting the 53BP1 foci ( Figure 2A). Similar to the wild-type and ATM −/− mouse mammary epithelial cells (Figure 1), only ATM −/− , but not wild-type BC cells showed  (Figures S2E and S2F). These data indicate that ATM promotes E2-induced DSB repair in human BC cells and murine mammary epithelial cells in the G 0 /G 1 phases, independent of TDP2.
We previously showed that E2 induces stalled TOP2ccs in BRCA1-deficient MCF-7 cells. 60,84 Here, the loss of TOP2B suppressed E2-mediated DSB induction in ATMitreated cells (p < 0.005) ( Figures 2C and S2C). We next investigated whether functional ERs were required for E2-induced DSB formation in ATM −/− cells. The ER inhibitor, fulvestrant, completely repressed E2-induced DSB formation (p < 0.005) ( Figures 2D and  S2G). These findings indicate that both TOP2 and functional ERs are required for the E2-dependent DNA damage in ATM −/− cells in the G 1 phase, as reported previously in BRCA1-deficient cells. 60,84 Furthermore, this suggests that stalled TOP2ccs form because of the transcriptional response to activated ERs.

ATM promotes 5′ TOP2 adduct removal from DSB ends in the G 1 phase
We investigated whether ATM significantly contributes to the repair of stalled TOP2ccs in TK6 cells, in which TDP2 is functional. 60,84 To monitor the DSB repair kinetics in the G 1 phase, we disrupted the ATM gene in TK6 cells ( Figures S3A and S3B) and pulse-exposed To test whether ATM promotes NHEJ and/or the removal of 5′ TOP2 adducts before repair by NHEJ, we next examined the repair of restriction enzyme-induced "clean" ligatable DSBs carrying 3′-OH and 5′-phosphate DSB ends. 53, 85 We expressed a regulatable AsiSI enzyme in TK6 cells and transiently activated AsiSI to induce DSBs and monitor the repair kinetics in the G1 phase. LIG4, but not ATM, was required to repair AsiSI-induced clean DSBs (p < 0.005) ( Figure 3B). Thus, the requirement of ATM for repairing stalled TOP2ccs ( Figure 3A) suggests that ATM promotes the removal of 5′-TOP2 adducts, but not subsequent NHEJ. ATM −/− and NHEJ-deficient DNA-PKcs −/− , MCF-7 cells also exhibited a similar delay in repairing etoposide-induced DSBs in the G 1 phase (p < 0.005) ( Figure 3C). Our findings demonstrate that TK6 and MCF-7 cells require ATM for efficient repair of stalled TOP2ccs.
To confirm the ATM-dependent removal of 5′-TOP2 adducts from DSB ends, we measured the amount of stable TOP2ccs in TK6 cells and G 1 -arrested MCF-7 cells. We lysed cells and separated TOP2ccs from free TOP2 in cellular lysates by cesium chlo-ride-gradient ultracentrifugation ( Figure S3D). Etoposide-induced stalled TOP2ccs were detected as single or double dots in the third and middle fractions of the TOP2-DNA complex carrying intact TOP2 adducts and TOP2 partially degraded by the proteo-some, respectively ( Figures 3D and S3E). 60,84 LIG4 −/− cells showed no increase in the accumulation of etoposide-induced TOP2ccs ( Figures 3D and 3E). Compared with solvent-treated wild-type cells, ATM −/− cells and ATMi-treated wild-type cells exhibited 2-fold higher amounts of etoposide-induced TOP2ccs (p < 0.005) ( Figures 3F, 3G, and S3E), even when we synchronized the cells in the early G 1 phase with the CDK4/6 inhibitor, palbociclib ( Figure  S3F). 86 The result was reproduced in MCF-7 cells arrested at the G 0 /G 1 phase (Figures S3G, S3H, and S3I). While ATM −/− had DSBs carrying partially degraded TOP2, wild-type cells did not ( Figures 3G and S3I), implying the ATM-dependent removal of TOP2 adducts only after their "debulking" with the proteasome. 87 Thus, ATM promotes the removal of 5′ TOP2 adducts from DSB ends via a mechanism that is independent of its role in DSB end resection during homologous recombination.

ATM promotes the removal of 5′ TOP2 adducts by phosphorylating CtIP at T847/T859
ATM promotes DSB end resection of HDR via phosphorylation of CtIP at T847/T859 in the S/G 2 phases. [88][89][90][91][92][93][94][95][96][97] Although CtIP phosphorylation is hardly detectable in the G 1 phase, 88 undetectable or low levels of CtIP phosphorylation can still promote the removal of blocking adducts from DSB ends in a process that is independent of its role in HDR. 59,[98][99][100] To address the potential role of ATM-dependent phosphorylation of CtIP at T847/T859, we inserted the T847A/T859A point mutations into the CtIP allelic genes of TK6 cells to generate CtIP T847A/T859A cells ( Figure S3J). CtIP T847A/T859A cells were 2-fold more sensitive to etoposide (according to etoposide dose that decreases the percent survival to 10%) ( Figure S3K) and exhibited delayed repair of etoposide-induced DSBs in the G 1 phase compared with wild-type cells ( Figure S3L). The conditional depletion of CtIP and CtIP T847A/T859A mutations increased the accumulation of stalled TOP2ccs to the same extent after etoposide exposure ( Figures 3H, 3I, and S3M). CtIP T847A/T859A cells synchronized in the G 1 phase also showed increased formation of etoposide-induced TOP2ccs ( Figures 3H, 3I, and S3M). Collectively, these findings indicate that CtIP promotes the removal of 5′ TOP2 adducts in the G 1 phase via a mechanism that may require CtIP phosphorylation at T847 and T859.
We next investigated the functional interaction between ATM and the CtIP phosphorylation sites. ATM −/− and CtIP T847A/T859A cells showed a similar phenotype in colony survival assays ( Figure S3K), and ATMi treatment did not further delay etoposide-induced DSB repair kinetics ( Figure S3L) or increase etoposide-induced TOP2ccs formation (Figures 3H, 3I and S3M). These data suggest that ATM promotes etoposide-induced DSB repair by phosphorylating CtIP at T847 and T859. This conclusion was supported by data showing that ectopic expression of the phosphomimetic CtIP mutant transgene, CtIP T847E/T859E , reversed the defective DSB repair in ATM −/− and CtIP T847A/T859A cells ( Figure S3L). ATMdependent phosphorylation of CtIP thus promotes 5′ TOP2 adduct removal from stalled TOP2ccs. We propose that this phosphorylation occurs only transiently, is undetectable in the G 1 phase, 88 and inhibits excessive resection to facilitate NHEJ.

Defective repair of stalled TOP2ccs causes dysregulation of estrogen-dependent activation of potential enhancers
We investigated whether the defective repair of stalled TOP2ccs caused dysregulation of E2-induced transcription and eRNA expression. To do so, we leveraged the NET-CAGE method, which quantifies the number of eRNAs and nascent RNAs transcribed from the promoter of protein-coding genes ( Figures S4A and S4B). 76 This method excludes the detection of RNAs derived from the release of stalled Pol2 during RNA extraction from cells ( Figure S4A). We first examined the functioning of TDP2 rather than that of ATM, as TDP2 has a specific role in removing 5′ TOP2 adducts from DSB ends. 28,[45][46][47] Considering the genome instability of MCF-7 cells, 101,102 we prepared TDP2-positive and -negative cell populations by infecting TDP2 −/− MCF-7 cells (Key resources table) with a viral vector expressing an intact TDP2 or catalytic-dead TDP2 (dTDP2); an empty vector (Mock) was used a control ( Figures S4C and S4D). NET-CAGE of the infected cell populations revealed that dTDP2 had only a modest impact on the transcriptional response to E2 ( Figure S4E), indicating that the effect of TDP2 expression reflects its catalytic activity in the removal of 5′ TOP2 adducts.
Triplicate NET-CAGE analyses revealed transient activation of enhancers for only a few hours in response to E2 ( Figure 4A). We identified a total of 2,019 eRNAs that were significantly upregulated after E2 exposure in the TDP2 −/− /mock and TDP2 −/− /TDP2 cell populations ( Figure 4B). Most genomic regions identified by eRNAs matched the DNase hypersensitivity site (DHS), H3K27ac, and histone 3 lysine 4 mono-methylation modifications peaks ( Figures S4F-S4H). Approximately 50% of these eRNA-expressing sequences were localized within 1 kb of ERα-or FoxA1-binding sites 103 ( Figure 4C). These data support that the transient eRNA expression ( Figures 4B and S4J). Considering the catalytic role of TDP2 in repairing stalled TOP2ccs, these NET-CAGE data suggest TOP2-dependent DSB formation at potential E2responsive enhancers.
Linking each enhancer to its target gene is a major challenge. To this end, we analyzed the E2-induced activation kinetics of individual transcription start sites (TSSs) of coding genes and eRNAs in our NET-CAGE data. We examined such pairwise expression correlation when TSSs were localized within 400 kb of enhancers, as previously published 75 ( Figure S4K). To interpret the data, we noted that several TSSs are usually present in individual genes, with each TSS usage being differentially controlled by the core promoter immediately upstream of each TSS. 74,76,81,[104][105][106][107][108] The usage of TSSs in each gene is regulated by distinct sets of enhancers. Early transcriptional responses to extracellular stimuli activate enhancers and their target core promoters with similar kinetics; the response of target core promoters is delayed only by 30 min or more. We also found that the usage of 6,843 TSSs changed with kinetics such as the activation kinetics of neighboring (<400 kb) enhancers in the TDP2 −/− /mock and TDP2 −/− /TDP2 cell populations ( Figure S4K), as exemplified by three pairs of potential enhancers and the TSS of neighboring genes ( Figures S4L and S4M). The similarity of the activation kinetics between eRNA-expressing sequences and TSSs implies that the former sequences control neighboring TSSs. These data support the formation of stalled TOP2ccs at E2-responsive enhancers leading to alterations in their activation kinetics, which in turn changes the expression kinetics of target promoters.

ATM loss increases the c-MYC transcriptional response to E2 in ER + human BC cells
To explore the role of ATM in the early transcriptional response to E2, we examined the effect of ATMi on eRNA expression at 2 h after E2 exposure. The addition of ATMi changed the E2-dependent expression of eRNAs and coding genes ( Figures 4D-4F). Approximately 30% of the E2-responsive eRNAs were derived from sequences near ERα-or FoxA1binding sites ( Figure 4G). These results indicate that stalled TOP2ccs at E2-dependent enhancers arise at E2-responsive enhancers and that their repair is defective in the absence of functional ATM or TDP2, causing dysregulation of eRNA expression during the early transcriptional response to E2.
Remarkably, our NET-CAGE data indicated that the addition of ATMi enhanced the E2dependent expression of c-MYC ( Figure S4O) and the eRNA expression from the c-MYC +135 kb enhancer, which possesses an ERα (ESR1)-binding site, H3K27ac, and DHS 109 ( Figure S4N and S4P). We further examined the E2-induced transcription of c-MYC over time in MCF-7 ( Figure 5A) and T47D cells ( Figure 5B). Inhibiting ATM enhanced the c-MYC response to E2, which is consistent with the increase in the c-MYC transcriptional response to E2 in the absence of NHEJ reported previously. 84 Defective repair of stalled TOP2ccs thus causes c-MYC overexpression in response to E2.
We also investigated the potential role of TOP2 in the c-MYC response to E2. To this end, we examined the c-MYC response in TOP2B −/− cells cultured in the presence or absence of charcoal-stripped serum, the addition of which allows TOP2A expression. 60 The loss of TOP2B decreased the extent of E2-dependent c-MYC induction in ATMi-treated cells ( Figure 5A), but not in cells cultured with serum ( Figure S5A). Thus, the E2-dependent overexpression of c-MYC in the ATM-deficient cells depends on either TOP2A or TOP2B. Collectively, our findings indicate that ATM prevents c-MYC overexpression by promoting the rejoining of stalled TOP2ccs arising during the early response to E2.

Unrepaired DSBs at c-MYC enhancer elements cause c-MYC overexpression in response to E2 exposure
We hypothesized that stalled TOP2ccs occur at enhancers rather than promoters since local DSBs inhibit the activity of promoters. 110,111 To test this, we explored DSB formation at a known c-MYC enhancer 26 during E2 exposure by chromatin immunoprecipitation (ChiP). We detected γH2AX ChiP signals after exposing wild-type MCF-7 cells to E2 together with DNA-PKi ( Figure 5C). γH2AX signals were undetectable in TOP2A-depleted TOP2B −/− cells and in wild-type cells without E2 ( Figure 5C). These data support the occurrence of stalled TOP2ccs at the enhancer during the early E2 response. Thus, active eRNA transcription may associate with DSB formation, as early transcriptional response causes DSBs at promoters. [35][36][37][38][39][40][41][42][43][44][45] To explore the effect of unrepaired DSBs at the c-MYC enhancer on the early transcriptional response to E2 in ATM-deficient cells, we performed two experiments at the E2 responsive c-MYC +135 kb enhancer ( Figure S4N). First, we measured the interaction between the +135 kb enhancer and the c-MYC promoter by chromosome conformation capture (3C). 112,113 The 3C data showed that addition of ATMi increased the E2-induced interaction by approximately 100-fold (p < 0.005) ( Figure 5D). Second, we investigated the effect of DNA cleavage by CRISPR/Cas9 on c-MYC expression. We infected wild-type cells with a lentivirus carrying both CRISPR/Cas9 and guide RNA to cleave the +135 kb enhancer. This cleavage had a synergistic effect with E2 exposure on the induction of c-MYC gene expression in MCF-7 ( Figure 5E) and T47D ( Figure S5B), as previously reported. 114 Thus, stalled TOP2ccs often form at the c-MYC enhancer during the early E2 response. If the breakage is left unrepaired because of an ATM deficiency, it causes c-MYC overexpression by facilitating a physical interaction between the enhancer and the c-MYC promoter.

ATM prevents c-MYC overexpression in response to E2 in murine mammary epithelial cells
We next investigated whether ATM prevents c-MYC overexpression in mouse mammary epithelial cells. After i.p. injection of E2 into wild-type and ATM −/− C57BL/6 mice, we analyzed c-Myc expression in the mammary gland by immunostaining ( Figures S5C and  5F). All c-Myc-positive (c-Myc + ) cells were stained with cytokeratin 8, a biomarker of ERα + luminal epithelial cells 115 ( Figure S5C). E2 injection greatly increased the proportion of c-Myc + cells in wild-type and ATM −/− littermate C57BL/6 mice (3-fold and 11-fold, respectively) (p < 0.005) ( Figure 5F). We also injected E2 into ATM −/− mice bred on a mixed B6; 129 background and confirmed a 3-fold induction of c-Myc + cells in ATM −/− mice at 6 h after injection compared with wild-type ( Figures 5G and S5C). Remarkably, ATM −/− mice showed prolonged c-Myc overexpression even at 24 h ( Figures 5G and S5D).
This effect was unexpected; the E2 serum concentration returns to background levels at 6 h after injection. 84 These data indicate that ATM inactivation causes a marked increase in the percentage of c-Myc + cells for 24 h after a single injection of E2.
Co-injection of ATMi with E2 also elevated c-Myc expression even at 24 h but not for 48 h, and the data seemed to be reproducible in the ATM −/− C57BL/6 mice ( Figures 5H, S5E, and S5F). The c-Myc expression kinetics were consistent with the prolonged DSB formation after a single injection of E2 plus ATMi ( Figure 1D). In summary, the loss of ATM not only increased the E2-mediated early induction of c-MYC gene expression, but also significantly extended its expression in mammary epithelial cells. Thus, ATM is likely required for an appropriate c-MYC transcriptional response to estrogens in the mammary gland.

ATM loss causes abnormal cellular proliferation after daily injection of E2
In our final assays, we investigated the consequences of an enhanced c-Myc response to E2 in Atm-deficient mice injected daily with E2 for 3 days. To identify the cells that proliferated during this period, we co-injected E2 with a nucleoside analog, 5-ethynyl-2′deoxyuridine (EdU) and examined EdU + mammary epithelial cells on day 4 ( Figures 6A  and 6B). E2 increased the percentage of c-Myc + cells by 3% and 18% (p < 0.005) by day 4 in ATM +/+ and ATM −/− mice, respectively ( Figures S6A and S6B). E2 also increased the percentage of EdU + epithelial cells by 8% (p < 0.05) and 28% (p < 0.005) in ATM +/+ and ATM −/− mice, respectively ( Figure 6C). Similarly, E2 induced a greater increase (4-fold) (p < 0.005) in the number of PCNA + epithelial cells in ATM −/− mice compared with ATM +/+ mice ( Figures 6D and S6E). The effect of injected ATMi was essentially the same as the loss of ATM in E2-injected mice ( Figures 6E and S6F). These data indicate that ATM inactivation causes abnormal mammary epithelial cell proliferation in response to E2.
To test the requirement of c-Myc for this abnormal proliferation ( Figures 6C-6E), we co-injected B6 mice with E2 plus ATMi and the Myc inhibitor, KJ-Pyr-9. 116 Inhibiting c-Myc decreased the number of PCNA + epithelial cells by 2.5-fold (p < 0.005) ( Figures 6F  and 6G). As expected, co-injection of fulvestrant (an ER inhibitor) completely suppressed the abnormal proliferation induced by E2 and ATMi ( Figures 6F and 6G). These findings indicate that ATM prevents the overgrowth of mammary epithelial cells by inhibiting c-MYC overexpression in response to estrogens.

DISCUSSION
The known role of ATM in HDR and the p53 activation at DSBs does not explain a dramatic increase in BC development upon the LOH in women bearing a germline mutation in the ATM gene. We, thus, aimed to elucidate why the loss of ATM promotes carcinogenesis specifically in mammary glands. We revealed that ATM inactivation causes c-MYC overexpression in E2-treated human BC cells ( Figures 5A and 5B). E2 exposure induced DSBs at c-MYC enhancers in cells ( Figure 5C), and CRISPR/Cas9 cleavage at the enhancers caused c-MYC overexpression in DSB-repair-proficient cells ( Figure  5E). Our data suggest the frequent occurrence of DSBs at c-MYC enhancers during the early transcriptional response to E2 and that a delay in rejoining can cause c-MYC overexpression. Like human BC cells, ATM deficiency in mice caused prolonged c-MYC overexpression even at 24 h after a single E2 injection ( Figures 5G and 5H). This overexpression led to an abnormal proliferative response to estrogens in Atm-deficient epithelial cells ( Figure 6). We propose that ATM suppresses oncogenesis selectively in mammary epithelial cells by inhibiting c-Myc overexpression in response to estrogens. This newly identified role of ATM, together with the ubiquitous function of ATM and the activation of p53 following the formation of DSBs, explains the high penetrance of ER + BC formation in women carrying germline mutations of the ATM gene. [4][5][6]8,9 Our study indicated that increased DSBs in ATM-deficient BC cells are caused by the formation of stalled TOP2ccs. We previously demonstrated the formation of stalled TOP2ccs during the early response to E2 and androgen in the mammary and prostate epithelial cells, respectively, of TDP2-deficient mice. 60,117 In this study, we revealed that the loss of ATM also increased E2-induced DSB formation in murine mammary epithelial cells ( Figure 1) and human BC cells ( Figure 2). DSB induction depends on TOP2 in ATM-deficient G 0 /G 1 cells (Figure 2), and ATM promotes the removal of 5′ TOP2 adducts from stalled TOP2ccs, independent of its role in HDR (Figure 3). E2-induced DSBs at c-MYC enhancer depended on TOP2B ( Figure 5C). Taken together, we conclude that ATM promotes the rejoining of stalled TOP2ccs arising in early E2 response.
Based on previously reported evidence that MRE11 collaborates with CtIP to remove 5′ TOP2 adducts in the G 1 phase, 59 we predicted that ATM participates in 5′ TOP2 adduct removal to facilitate NHEJ by phosphorylating CtIP. Here, we show that ATMi increases etoposide-induced TOP2ccs formation in wild-type, but not CtIP T847A/T859A , cells in the G 1 phase (Figures 3H, 3I, and S3M). Ectopic expression of the phosphomimetic CtIP mutant transgene, CtIP T847E/T859E , reversed the defective repair of ATM −/− and CtIP T847A/T859A cells ( Figure S3L). These data imply that ATM contributes to the removal of 5′ TOP2 adducts by phosphorylating CtIP at DSBs. Nonetheless, the DSB-dependent phosphorylation of CtIP is undetectable in the G 1 phase. 88 Conceivably, such limited phosphorylation of CtIP by ATM prevents excessive resection, but still promotes the MRE11-mediated removal of 5′ TOP2 adducts. It has been reported that TOP1 adducts suppress ATM activation at singleended breakage points 118 ; therefore, it can be speculated that the limited phosphorylation of CtIP might result from TOP2-blocking adducts. Collectively, we conclude that ATM promotes the removal of 5′ TOP2 adducts to facilitate NHEJ most likely by phosphorylating CtIP.
Our data suggest the frequent occurrence of stalled TOP2ccs at many E2-responsive enhancers after E2 exposure. Subsequent studies indicated the occurrence of DSBs at enhancers in the absence of TDP2 altered the activation kinetics of 273 enhancers carrying ER-binding sites ( Figure 4B). Furthermore, ATM inactivation also changed the E2-dependent induction of eRNAs from 587 potential enhancers ( Figure 4D). These data suggest the formation of stalled TOP2ccs during E2 exposure since ATM and TDP2 have a role in repairing stalled TOP2ccs. Our data support the existence of E2-induced DSBs at E2-responsive enhancers of the c-MYC gene in wild-type, but not in TOP2-deficient, cells ( Figure 5C). Considering the requirement of TOP2 in transcription, 28,[45][46][47]119 the induction of eRNA after exposure to E2 leads to the spontaneous formation of stalled TOP2ccs at enhancers and flanking sequences. Further research is now needed to show the occurrence of DSBs at E2-responsive c-MYC enhancers in normal human and murine mammary epithelial cells. A genome-wide analysis of DSB formation during exposure to E2 is also required. Nevertheless, we propose that stalled TOP2ccs form frequently at enhancers in the early transcriptional response to estrogens. The consequence of unrepaired DSBs at enhancers is c-MYC overexpression in human BC cells (Figures 5A and 5B) and mouse mammary epithelial cells (Figures 5F-5H).
The role of DSB repair in suppressing the early c-MYC response is unexpected, as it is widely believed that unrepaired DSBs suppress transcription from constitutively active promoters near the cleavage site. 110,120-123 Transcription from DSB ends and the resulting RNA-DNA hybrid formation also causes promoter repression of neighboring genes. 124 Thus, it is surprising that, although the loss of TDP2 and addition of ATMi increased the occurrence of E2-dependent breakage, this also caused increases in the expression from the promoters of many protein-coding genes ( Figures 4B and 4D). The presence of many upregulated genes during the early transcriptional response to E2 implies that most of the TOP2-dependent breakages occur at enhancers, rather than promoters, in this phase. As unrepaired DSBs resulted in both upregulation and downregulation of many genes in DSB-repair-deficient cells ( Figures 4B and 4D), we propose that the effect of unrepaired DSBs on the transcriptional response to extracellular signals is far more complex than that on constitutively active promoters near the DSBs. 110,[120][121][122][123] Conceivably, unrepaired DSBs at enhancers distantly localized from promoters differentially affect target promoters, depending on the function of enhancers, whether DSBs arise within or outside enhancers, the timing of rejoining, and the distance between broken enhancers and target promoters. While CRISPR/Cas9-mediated cleavage at the E2 responsive enhancer augmented the E2mediated induction of c-MYC expression ( Figures 5E and S5B), the effect of such cleavages on signal-induced transcription is likely to vary in individual enhancers and is highly unpredictable. Further genome-wide studies are required to identify hot spots of stalled TOP2ccs and their effects on functional interactions between enhancers and target genes, particularly the c-MYC locus.
Women carrying germline mutations of the ATM gene are predisposed to ER + BCs when LOH inactivates the ATM gene. [4][5][6][8][9][10] LOH events occur at extremely low frequency (approximately 10 −5 per base) in cells of malignant tumors. 125 However, once LOH occurs, the resulting ATM loss should dramatically drive selective oncogenesis in the estrogen-regulated tissues. The current study further elucidates the molecular mechanism underlying this selective oncogenesis by revealing the induction of c-Myc overexpression in response to E2 ( Figure 5). As a relatively small increase in c-Myc expression significantly enhances oncogenesis, 33,34 prolonged c-Myc overexpression ( Figures 5G and 5H) will effectively drive oncogenesis specifically in mammary epithelial cells. Moreover, known functions of ATM, the activation of p53 by DSBs and the promotion of HDR, explain the enhanced carcinogenesis of ATM-deficient cells. We propose that the combination of c-MYC overexpression, a compromised damage checkpoint, and defective HDR is responsible for the extremely efficient and selective oncogenesis in mammary epithelial cells after the loss of functional ATM via LOH.
We proposed that a defect in the repair of breakage at extracellular signal-dependent enhancers as a previously unappreciated mechanism for disease onset based on our demonstration that, in the absence of ATM, defective repair of TOP2-dependent breakage at estrogen-dependent enhancers caused c-MYC overexpression ( Figures 5C-5E). This mechanism not only explains the increased oncogenesis in mammary epithelial cells, but also the increased risk of metastatic prostate cancers in carriers bearing mutations in the ATM genes. 126,127 We propose that carriers have mutations in the BRCA1 gene since BRCA1 promotes the removal of 5′ TOP2 adducts before NHEJ. 53,60 Why carriers having mutations in the BRCA1 gene develop ER-negative BCs despite the development of ER + BCs upon the loss of ATM remains an important question. The cell origin is a controversial issue, as ER + epithelial cells can undergo de-differentiation to the ER-negative stem-like state during oncogenesis in mice [128][129][130] and humans 130 (reviewed in ref. 131 ). A positive feedback loop exists between ER and BRCA1 expression; however, BRCA1 is likely not essential for ER expression. 132 We propose that ER + BRCA1-deficient epithelial cells are converted to ER-negative BCs through the prolonged ER-dependent DSB formation in BRCA1-deficient ER + epithelial cells during E2 exposure, 60 which activates ATM and then stabilizes p53, which leads to cell senescence. 133 Thus, the loss of ER and p53 is likely to confer a considerable growth advantage on BRCA1-deficient malignant cells. In contrast, the loss of ATM weakens the capability of p53 to trigger senescence and renders ER + cells tolerant to estrogen-dependent DSB formation, thereby conferring a growth advantage in response to estrogens. In addition to the enhanced carcinogenesis, unrepaired breakage at extracellular signal-dependent enhancers can cause symptoms of ataxia-telangiectasia (A-T), the autosomal recessive disease caused by ATM mutations. 134 For example, the progressive cerebellar degeneration of people with A-T disorder may result from altered transcriptional responses to neurotransmitters, resulting in the gradual loss of neurons. If dysregulated gene expression causes BC development in carriers bearing ATM mutations and contributes to the A-T phenotype, it is possible that ER inhibitors that prevent BC development and gene therapy to correct dysregulated gene expression could relieve the severe symptoms of cerebellar degeneration.

Limitations of the study
Our study has two major limitations. First, we did not assess tumor formation. Second, we did not show a direct link between the impact of ATM deficiency or resulting c-MYC overexpression on BC formation. The overexpression of c-MYC alone may not account for the dramatically enhanced BC risk upon LOH of the ATM gene in mammary epithelial cells. There may be multiple additional transcriptional changes in ATM-defective mammary epithelial cells upon E2 exposure, due to the persistence of stalled TOP2 lesions on the DNA. Any of these changes could be causal for BC; future studies will need to address all oncogenic changes as well as c-MYC overexpression.

STAR☆METHODS RESOURCE AVAILABILITY
Lead contact-Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Shintaro Yamada (yamada@rg.med.kyoto-u.ac.jp).

Materials availability-Mutant cells generated in this study are available from the lead contact with a completed Materials Transfer Agreement.
Data and code availability • Code availability: This paper does not report original code.
• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Animal experiments
Mice: All animal work was performed in compliance with relevant regulatory standards and was approved by the Animal Research Committee of Kyoto University and the Memorial Sloan Kettering Cancer Center Institutional Animal Care and Use Committee.
Experiments for Figures 1A-1C, S1B, 5F, 5G, S5C and S5D were performed in MSKCC. The ATM mutation 137  Immunostaining of mammary gland tissue-For the preparation of cryo-sections, the isolated mammary gland was fixed with 4% paraformaldehyde (PFA) in PBS (4°C, 15 min, on a rocking platform). The sample was washed (×3) briefly with PBS and incubated with 30% sucrose in PBS (room temperature, 1-2 h). The sample was embedded in OCT compound and frozen with liquid nitrogen. Cryo-sections (thickness 10 μm) were cut at −50°C and dried prior to fixation with 4% PFA (room temperature, 3 min) and rinsed with PBS.
For the preparation of paraffin sections, the mammary gland was fixed with 4% formaldehyde in PBS (room temperature, 48 h), dehydrated, and embedded in paraffin. Sections (thickness 5 μm) were cut and mounted on slides for either hematoxylin and eosin (HE) staining or immunostaining.
For the immunostaining of frozen mammary gland section, tissue was first fixed with 4% PFA for 5 min. No deparaffinization and epitope retrieval steps are required for frozen mammary section. Specimens (both paraffin and frozen) were then washed (33) Table). For transfection, MCF-7 cells seeded in a 6-cm dish and cultured to approximately 60% confluence and then transfected pX459-gRNA using Fugene HD according to the manufacturer's protocol. At 24 h post-transfection, puromycin (final concentration 2 μg/mL) was added to the medium and the MCF-7 cells were incubated for a further 48 h. After removing puromycin, the cells were cultured for approximately two weeks to isolate the clones. The gene-disruption events were confirmed by Western blot analysis.
ATM −/− generation, TK6 cell-To generate gene-targeting constructs, we generated left and right arms (approximately 1 kb each) of genomic sequences using the primers listed in Table S1. To generate the left arm, we added the upstream and downstream sequences derived from the ApaI site to the 5′ and −3′ ends, respectively, of the PCRamplified left arm. For this purpose, we added "5′-GCGAATTGGGTACCGGGCC" and "5′-CTGGGCTCGAGGGGGGGCC" to the 5′ end of the upstream and downstream primers, respectively, of the PCR-amplified left arm. To generate the right arm, we added the upstream and downstream sequences from the AflII site to the 5′ and −3′ ends, respectively, of the PCR-amplified right arm. For this purpose, we added "5′-TGGGAAGCTTGTCGACTTAA" and "5′-CACTAGTAG GCGCGCCTTAA" to the 5′ end of the upstream and downstream primers, respectively, of the PCR-amplified right arm. These were then inserted into the ApaI and the AflII sites of the DT-ApA/MARKER R vector using GeneArt Seamless Cloning Enzyme Mix according to the manufacturer's instructions.
The gRNA was inserted into the BbsI site of pX330 vector, which expresses gRNA and Cas9 from the U6 and chicken β-actin promoters, respectively. Exon 40 of the ATM gene was targeted for gRNA insertion, as this contains the catalytic site. TK6 cells (8×10 6 ) were then co-transfected with the two resulting targeting vectors containing different antibiotic markers (Neomycin #9368) and (Puromycin # 9-369) and pX330-gRNA (CRISPR # 9367) into 8 million. The transfected pX330 expressed the Cas9-gRNA complex, which induced DSBs at the specific locus of the genomic DNA and thus facilitated HR between the genomic locus and the arms of the targeting vectors. Details of the vectors and gRNAs are listed in Key Resource Table. CtIP T847/859A knock-in (KI) mutant generation-The CtIP T847/859A point mutant was obtained by generating a knock-in (KI) construct. Two gRNAs (gRNA#1 and gRNA#2) were in serted into the BbsI site of the pX330 expression vector. gRNA#1 and #2 (Table   S1 for sequence information) were designed in introns 17 and 19, respectively. The left arm (1,743 bp) started at 1,134 bp upstream of the gRNA#1 cutting site (but excluded the gRNA sequence) and included exons 18 and 19. The point mutation sites CtIP T847A and CtIP T859A were included in exon 18 to locate the mutations in the left arm. The right arm (1,158 bp) started downstream of the gRNA#2 cutting site ( Figure S3H). The mutation-containing cDNA vectors (# pTP2630) were a gift from Dr. Tanya Paull.
The left arm, marker, and right arm were then inserted into the expression plasmid to generate two targeting vectors (Neomycin and Puromycin containing).
For expression in human TK6 cells, the targeting vectors were stably inserted using the Neon (MPK5000) transfection method (pulse voltage: 1500 V; pulse width: 20 ms; pulse number: single).
For expression in KI cells, the cells were first selected using both of the selection markers (neomycin and puromycin) and then checked for the correct insertion of the left and right arms by PCR amplification from the genomic DNA of resistant colonies. Insertion of both the mutations in the cDNA of the transfected cells was then confirmed by sequencing. Details of all the primers are shown in Table S1. A total of 1 mL was collected for analysis from the top to bottom of the cesium chloride gradient. For the slot blot analysis, 100 μL of each of the collected fractions were spotted onto methanol-pretreated PVDF membrane through the slot of Bio-Dot apparatus. After brief washing with 0.2 M phosphate buffer (pH 6.8), the membrane was incubated overnight at 4°C with anti-TOP2β (1:2,000) diluted in 5% skimmed milk in TBST (0.01 M Tris-HCl, 150 mM NaCl, 0.005% Tween 20, pH 8) followed by incubation with horseradish peroxidase (HRP)-conjugated anti-mouse secondary antibody (1/5000) for 1 h at room temperature. Immunoreactive spots were developed by chemiluminescence using the ECL reagent (ECL ™ Prime western blotting detection reagent). The signal was detected by exposure to X-ray film (Amersham Hyperfilm ™ MP) and scanned (EPSON) for quantification by ImageJ software. The protocol is adapted from ref. 49 Western blot analysis-MCF-7 and TK6 cells (5×10 5 ) in 50 μL of PBS were lysed by the addition of 50 μL 2× lysis buffer (120 mM Tris-HCl, 4% SDS, 0.04% bromophenol blue, 10% glycerol, and 10% 2-mercaptoethanol). After boiling the samples for 10 min, lysates were briefly centrifuged and a sample (10 μL) from 100 μL of the supernatant was separated by polyacrylamide gel electrophoresis (5%-10% gel for ATM or 5%-20% for TDP2) for several hours at 200 mV. These separated proteins were then transferred into a nitrocellulose membrane (for ATM) and PVDF membrane (for TDP2) using the semidry method. Membranes were blocked with 5% BSA (for ATM) or 5% skimmed milk (for TDP2) in TBST at room temperature for >1 h prior to incubation with primary antibody overnight at 4°C. The membrane was washed several times with TBST solution and then incubated with an appropriate HRP-linked secondary antibody (1:5,000, in blocking buffer) for 1 h at room temperature. After washing (×3), immunoreactive bands were developed by chemiluminescence using ECL reagent (ECL ™ Prime western blotting detection reagent).
The signal was detected by exposure to X-ray film (Amersham Hyperfilm ™ MP) and scanned by a scanner (EPSON). Details of the antibodies used are listed in Key resources table.
Cell survival assay-For TK6 cells, the cell survival assay was performed according to the method described by ref. 49 In brief, various amounts of etoposide were mixed with 1.5% (w/v) methylcellulose in medium (RPMI) containing 10% horse serum, by slowly rotating tubes overnight at 4°C. Fixed numbers of cells were seeded into 6-well plates containing 5 mL methylcellulose medium per well, and incubated for 2 weeks at 37°C before counting visible colonies.
For MCF-7 cell, fixed numbers of cells (500-2,000) were cultured in 10% FBS containing DMEM and then incubated at 37°C with various amount of etoposide for 2-3 weeks. After washing with dH 2 O, plates were stained with 5% Giemsa solution to visualize and count the colonies.
Transcriptome analysis by NET-CAGE-MCF-7 cells were cultured in a serum-free medium for 24-26 h before treatment with 10 nM E2. To assess the effect of ATM loss on the E2 response, 10 μM of ATMi was added to the medium together with E2 or ethanol; DMSO was added 30 min prior to the E2 treatment as a control. Three replicates of cells cultured in three 15-cm dishes (approximately 10 6 cells per dish) were treated simultaneously and pooled for analysis. Replicates were processed simultaneously throughout to avoid potential batch effects. RNA was extracted as described previously. 76 Briefly, fresh cells were lysed in the presence of α-amanitin, and the nuclear insoluble fraction was isolated. Following DNase treatment, RNA was purified using the miRNeasy Mini kit (Qiagen) according to the manufacturer's instructions. NET-CAGE library preparation, sequencing, mapping, and gene expression analysis were performed by DNAFORM (Kanagawa, Japan). In brief, the cDNAs were synthesized from total RNA using random primers.   Figures S1B and S1C, respectively. Data (B-D) represent mean ± standard deviation from triplicates. ***p < 0.005, unpaired two-tailed t test. Analysis was performed before (left) and after 2 h of E2 exposure (middle) and after 2 h of additional incubation with E2-free media (right). Scale bar, 25 μm.
(C) Percentage of G -phase 53BP1 + 1 MCF-7 cells (≥10 foci/cell) carrying the indicated genotypes. Data replotted in boxplots are shown in Figure S2C. (A and C) The DSB repair kinetics of G 1 -phase TK6 (A) and MCF-7 (C) cells after a 0.5-h pulse exposure to etoposide. Cells were cultured in etoposide-free medium after the pulse exposure. Percentage of 53BP1 + cells was measured (≥5 foci/cell and ≥10 foci/cell for TK6 and MCF-7 cells, respectively). Representative images of TK6 cells are shown in Figure  S3C. (D) Representative dot blot analysis of TOP2ccs in TK6 cells of the indicated genotypes treated with etoposide for 2 h. Genomic DNA (50 μg) was separated by cesium chloride gradient ultracentrifugation, and individual fractions were blotted onto polyvinylidene fluoride filters followed by dot blotting using an α-TOP2β antibody (see Figure S3D). The first and the second fractions represent free TOP2. The third fraction contains stalled TOP2ccs having intact TOP2.
(E-I) Quantification of TOP2ccs. The whole stalled TOP2ccs (E, F, and H) were quantified for the indicated genotypes. Representative blots are shown in (D), Figures S3E and S3M, respectively. The x axis shows the number of etoposide-induced stalled TOP2ccs relative to that in wild-type. Stalled TOP2 having degraded TOP2 was quantified in (G) and (I).
Representative western blots were shown in Figures S3E and S3M, respectively. The x axis shows the percentage of degraded TOP2ccs relative to the whole stalled TOP2ccs in (F) and (H), respectively. Data (A-C and E-I) represent mean ± standard deviation from triplicates. **p < 0.05 and ***p < 0.005. NS, not significant. Student's t test.   Figures S6E, S6F, and (F), respectively. Data (C-E and G) represent mean ± standard deviation from triplicates. **p < 0.05 and ***p < 0.005, unpaired two-tailed Student t test. Scale bar, 50 μm.