ATM Regulates Differentiation of Myofibroblastic Cancer-Associated Fibroblasts and Can Be Targeted to Overcome Immunotherapy Resistance

ATM signaling supports the differentiation of myoCAFs to suppress T-cell infiltration and antitumor immunity, supporting the potential clinical use of ATM inhibitors in combination with checkpoint inhibition in myoCAF-rich, immune-cold tumors.


Introduction
Immune-checkpoint blockade elicits durable antitumor clinical responses in an increasing number of malignancies. However, its success is limited to a fraction of cancer patients, highlighting the need to identify targetable resistance mechanisms to widen its clinical effectiveness (1). Analysis of human cancers and mouse models has shown that nonresponsiveness to immune-checkpoint blockade can result from limited T-cell infiltration, mediated by the cells of the tumor microenvironment (1).
Among the heterogeneous CAF population, a distinct myofibroblastic phenotype has been found in most types of solid tumors (2). Myofibroblastic CAF (myoCAF) are analogous to myofibroblasts found in wound-healing and tissue fibrosis. These contractile, secretory cells are regulated through TGFb1 signaling, express a-smooth muscle actin (SMA), and secrete collagen-rich extracellular matrix (ECM; refs. 2, 3). Studies in multiple cancer types have shown that tumors with an SMA-positive, myoCAF-rich stroma are associated with poor prognosis (4)(5)(6) and that myoCAF contribute to many hallmarks of malignancy (7). Transcriptomic analyses of tumors that fail to respond to anti-PD-1/PD-L1 have identified upregulation of myoCAF genes (8)(9)(10)(11)(12)(13), suggesting that myoCAF mediate resistance to checkpoint blockade. Consistent with this, myoCAF-rich cancers contain low levels of infiltrating T cells (8,10,14) and functional studies in murine models have confirmed that myoCAF promote resistance to different types of immunotherapy (8,13,14), in part by promoting T-cell exclusion from tumors. This has led to the emergence of myoCAF as potential therapeutic targets (15). However, a limited understanding of CAF heterogeneity and of the mechanisms regulating the accumulation of different CAF phenotypes has resulted in unsuccessful clinical trials, illustrating challenges to effectively target these cells (16)(17)(18)(19).
TGFb/SMAD signaling is recognized as the central pathway regulating myofibroblast/myoCAF differentiation, although other pathways also influence the myoCAF phenotype (3). We found previously that during TGFb1-induced differentiation, myofibroblasts upregulate a number of genes associated with DNA repair, suggesting that the DNA damage response (DDR) pathway may be triggered during the process (20). The DDR is a complex cellular network of coordinated pathways that maintain genome integrity by detecting and repairing DNA lesions (21), orchestrated by DNA damage-sensing kinases ataxia-telangiectasia mutated (ATM), ataxia-telangiectasia mutated and Rad3-related (ATR), or DNA-dependent protein kinase (DNA-PKc; ref. 22). TGFb/SMAD signaling has been reported to contribute to the cellular DDR induced by genotoxic insults (23,24) or by radiation/stress-driven bystander effects (25)(26)(27). However, it remains to be determined if and how the TGFb/SMAD pathway affects DNA repair during the acquisition of the myofibroblast/ myoCAF phenotype.
The myofibroblast differentiation process is associated with the generation of intracellular reactive oxygen species (ROS), and we have previously described a role for the ROS-generating enzyme NADPH oxidase 4 (NOX4) in promoting and maintaining the myoCAF phenotype (28). Notably, the inactive noncovalent ATM homodimer is typically activated by the Mre11-Rad50-Nbs1 (MRN) complex in response to DNA double-strand breaks (DSB; refs. 29, 30) becoming a monomer through autophosphorylation. However, ATM can also be activated as a covalent dimer through oxidation and independent of DNA damage (31), suggesting a possible link between NOX4-generated oxidative stress, ATM activation, and myofibroblast differentiation.
Here we sought to investigate the role of the DDR signaling in regulating myoCAF phenotype and function using in vitro/ex vivo culture of myofibroblasts/myoCAF, analysis of human tumors, and CAF-rich murine tumor models that recapitulate the stromal morphology, immune microenvironment, and immunotherapy resistance found in CAF-rich human tumors (14).
Multiplex images were generated using a stain clearing-based method: nuclear counterstain (hematoxylin) and registration markers (CD31 using DAB) remain permanent, followed by sequential staining and clearing of a transient stain (SMA, pATM, and CK using AEC), as previously described (14).
SMA IHC (Sigma-Aldrich; A2547) on FFPE tissues was performed as previously described (14). Quantification of SMA staining in mouse tumors was carried out using color thresholding and deconvolution in Fiji (Supplementary File S1 for the macro in Fiji). At least three independent fields of view (FoV) of at least three tumors were randomly selected by a consultant pathologist (GJT) to generate a mean value per tumor representing the SMA-positive area percentage, which was compared between treatment groups.

Comet assay
HFFF2 cells (10 Â 10 3 ) were treated with TGFb1 for 24 hours or xirradiated (2 Gy) 30 minutes before harvesting, resuspended in 0.6% low melting point agarose (UltraPure LMP Agarose from Life Technologies), placed onto precoated slides with 1% normal melting point agarose, and coverslipped. Slides were placed on ice for 30 minutes to allow gels to solidify, coverslips were removed, and slides were placed in ice-cold lysis buffer (2.5 M NaCl, 0.1 M EDTA, 10 mmol/L Tris-HCl, 1% TritonX-100 at pH 10) in the dark, overnight. Next (with light protection), slides were washed for 10 minutes with ice-cold ddH 2 O and placed in ice-cold alkaline buffer (300 mmol/L NaOH, 1 mmol/L disodium EDTA pH > 13) for 20 minutes, and electrophoresed for 20 minutes at 30 v/300 mA protected from light. Following electrophoresis, slides were incubated with neutralization buffer (0.4M Tris-HCl, pH 7.5) for 20 minutes, washed twice for 10 minutes with ddH 2 O, and placed at 37 C overnight to dry. Slides were rehydrated with ddH 2 O for 30 minutes at room temperature, stained with propidium iodide (PI, 2.5 mg/mL) for a further 20 minutes, washed with fresh ddH 2 O for 20 minutes and placed at 37 C to dry before the comet scoring using an Olympus BH-2-RFL-T2 fluorescence microscope (Komet Analysis software version 5.5; Andor Technology). The percentage of DNA in the tail of the comet resulting from total DNA breaks (single-and double-strand breaks) was calculated for each cell. Six independent experiments were carried out with 50 cells/comets analyzed per replicate.

Invasion assays
CAF, HFFF2, or shCTR/shATM HFFF2 populations were treated with 13.3 mmol/L KU55933 AE TGFb1 for 7 days (treated days 1 and 4). Cells were washed twice with PBS, and serum-free DMEM was added for 72 hours. Fibroblast-conditioned medium was collected, centrifuged, filtered (0.2 mm), normalized according to cell number, and used as a chemoattractant in the lower chamber of Transwell invasion assays as previously (4,20). Two-way ANOVA was used for statistical analysis.
Fluorescence microscopy 5 Â 10 3 HFFF2 were plated on permanox chamber slides (Thermo Scientific) and treated as described in the figure legends. Immunostaining was performed as previously (28). DAPI was used as a nuclear counterstain, and coverslips were mounted using a fluorescent mount-ing solution (Dako). Images were taken using an Olympus IX81 fluorescence microscope and the Xcellence program. DNA-damage foci were quantified from at least 8 nonoverlapping FoV with $300 cells in total per condition. Kruskal-Wallis statistical test was used to indicate the presence of a different distribution per time point in each group of foci. For collagen 1A1 deposition, 5 Â 10 3 HFFF2 or MLF were seeded into chamber slides and treated with TGFb1 AE inhibitors for a week (treated days 1 and 4). Decellularization was performed using 0.25M NH 4 OH in 50 mmol/L Tris at 37 C for 30 minutes and slides fixed with ice-cold 100% methanol at À20 C followed by a blocking step. The remainder was performed as previously described (28). Antibodies used for immunofluorescence are listed in Supplementary File 2. SMA and collagen 1A1 quantification was performed using Fiji (macro to calculate the image mean can be found in Supplementary File S1). High-resolution imaging of mouse tumors was carried out using an Axioscan.Z1 scanner (Zeiss). Immunocytochemistry for NOX4 was performed using antibodies from Novus Biologicals and Abcam with similar results (see Supplementary File S2 for antibody details).

Viability, proliferation, and toxicity assays
To assess cell viability and proliferation, HFFF2s were plated in 96well plates (5 Â 10 3 cells/well) and treated the following day with inhibitors AETGFb1. After 3 days, [H 3 ]-thymidine was added to the medium for 16 hours, and radioactivity was measured using TopCount NXT (Packard Bioscience). MTT assay was carried out using CellTiter 96 AQueous Nonradioactive Cell Proliferation Assay (Promega) according to the manufacturer's instructions. The absorbance was read at 490 nm using a Victor plate reader (Perkin Elmer). To monitor cell death, HFFF2s were plated in 6-well plates (200 Â 10 3 cells/well) and treated the following day with inhibitors AE TGFb1. After 3 days, cells were detached, resuspended in 300 mL of FACs buffer containing 1.7 ng/mL PI (Thermo Fisher) and analyzed using a Bd FACSCanto Flow Cytometer (BD Biosciences). Cells were less than 90% confluent at the end of each experiment. Cells were treated in the final 30 minutes with 10 mmol/L H 2 O 2 (for PI and thymidine assay) or 5 mg/mL cisplatin (for MTT assay) as positive controls.

Senescence and ROS assay
HFFF2s were plated in 24-well plates (1 Â 10 3 or 5 Â 10 3 cells/well) and treated with TGFb1 the following day (and then every 3 days). Senescence was quantified at days 5 and 12 from TGFb1 treatment determining the percentage of SA-b-Gal-positive cells (Sigma) of the total number of cells counterstained with Dapi (Sigma) as previously described (20). At least 200 cells/well in at least three independent experiments were counted per sample using a CKX41 Olympus fluorescent microscope. Cells were less than 50% confluent at the end of the experiment to avoid SA-b-Gal false-positive cells due to cell contact. ROS assay was performed as previously (28).

Western blotting
Cells were lysed in ice-cold RIPA buffer [1% NP40, 150 mmol/L NaCl, 25 mmol/L Tris-HCl pH 7.5, 0.1% SDS, 1% Na-deoxycholate, 1 mmol/L EGTA, 1 mmol/L EDTA, 1 mmol/L Na-orthovanadate, 10 mmol/L Na-fluoride, 2.5 mmol/L Na-pyrophosphate, 1 mmol/L b glycerophosphate þ freshly added 1 mmol/L PMSF, 20 mmol/L N-ethylmaleimide (NEM), 10 ng/mL Microcystin-LR, 0.12 mmol/L okadaic acid and phosphatase and protease inhibitor cocktails; all from Sigma]. Protein extracts were then centrifuged, and supernatants were quantified using the Dc-Bio-Rad protein assay kit (Bio-Rad Laboratories) as previously described (28). For nuclear-cytoplasmic protein extraction, we used NE-PER Reagents according to the manufacturer's instructions (Thermo Fisher). Equal amounts of protein were electrophoresed either in reducing (þDithiothreitol, DTT) or in nonreducing conditions in 3% to 8% or 4% to 12% SDS-PAGE gels and electro blotted as previously (28). To detect the ATM dimer, cells were detached, washed once with PBS, and then kept moving at room temperature for 10 minutes with PBS þ 100 mmol/L NEM before nuclear extraction. HSC70 was used as a loading control. Bound antibodies were detected using a chemiluminescence system (Pierce), visualized at the Fluor-S Multi-imager (Bio-Rad), quantified using Fiji, and were plotted as ratios between a given protein and HSC70 unless otherwise stated. Antibodies are described in Supplementary File S2.

Quantitative real-time PCR
For quantitative real-time PCR (qRT-PCR), RNA was extracted and retrotranscribed as described previously (14), and qRT-PCR was performed as previously (28). The mRNA expression levels were analyzed using the DDC t relative quantitation method. GAPDH or HPRT were used as human or mouse housekeeping genes, respectively. Gene-specific primers were designed using http://www.ncbi.nlm.nih. gov/tools/primer-blast/. Primer sequences are given in Supplementary File S2.

Collagen gel contraction assays
Fibroblasts (5 Â 10 5 ) were added to 1 mL of ice-cold 3 mg/mL type 1 collagen (Millipore) gels containing 10% FBS DMEM as described previously (21). Gels were incubated at 37C for 1 hour to polymerize. After adding 1 mL of 10% FBS DMEM, gels were detached using a spatula and photographed after 24-hour incubation. The gel area was measured using Fiji.

In vivo mouse experiments
All mouse experiments were performed according to national guidelines and were approved by the authors' institutional review board and the UK Home Office. Xenograft model: 6.7 Â 10 5 5PT cells AE 2 Â 10 6 HFFF2 cells were resuspended in 100 mL PBS and injected s. c. in the flank of partially immunocompromised, male RAG1 À/À C57BL/6 mice (3-5 months old). Isograft models: 0.5 Â 10 5 TC-1 cells AE 3 Â 10 5 TGFb1-treated MLF (myoMLF) or 1 Â 10 5 MC38 cells AE 5 Â 10 5 TGFb1-treated murine colon fibroblasts (myoMCF) were resuspended in 100 mL PBS and injected s.c. in the flank of C57/BL6 female mice (6-8 weeks old). Five to eight animals for both models were used per group. Tumor size was measured over time using an electronic caliper and calculated using the formula 4p/3 Â r3 [radius (r) was calculated from the average diameter, measured as the tumor width and length]. AZD0156 (ATMin; provided by AstraZeneca) was tested only on TC-1 and MC38 models (5PT cells were generated using prolonged cisplatin treatment and potentially have an altered DDR pathway (32)). Vaccination with DNA vaccine encoding tetanus fragment C domain 1 (Dom) fused to the immunodominant CD8 epitope of HPV E7 RAHYNIVTF (RAH, E749-57) p.Dom-RAH (34) was administered once via intramuscular injection (i.m.) when tumors were palpable (days 8-13; 20 mg of DNA in 100 mL PBS). P.Dom without the epitope served as a control. aPD-1 antibody (300 mg; Bio X Cell; RMP1-14) or IgG2a isotype control was given via intraperitoneal (i.p.) injection when tumors were palpable every other day, totaling 3 doses starting from days 9 to 14 from the tumor challenge. When tumors were palpable (at days 8-21 for TC-1 model and at days 8-9 for the MC38 model), mice were treated with oral gavage daily with either vehicle or 20 mg/Kg AZD0156 (ATMin) until the end of the experiment (as per the manufacturer's instructions).
For TC-1 survival analysis, mice were injected with RAH/CTR vaccines on day 14; AZD0156 was administered daily from days 15 to 47 and then twice a week until the end of the experiment (day 82). For MC38 survival analysis, aPD-1 was administered on days 14, 16, and 18; AZD0156 was given daily from days 9 to 28 and then twice weekly until the end of the experiment (day 47).
For tumor growth curves, statistical testing was performed on the mean area under the curve (AUC) using a two-way ANOVA. Following euthanasia, tumors were excised and either fixed in 10% formalin and embedded in paraffin or directly embedded in OCT matrix (Thermo Scientific) for IHC processing or disaggregated for FACS staining as previously (EF506 viability dye was used for live/dead cell staining; see Supplementary File S2 for antibodies; ref. 14).

Bioinformatic analysis
Laser capture microdissected (LCMD) stroma from Gene Expression Omnibus (GEO) data sets [GSE45001 (liver), GSE35602 (colon), GSE19632 (esophagus), and GSE40595 (ovary)] were confirmed to be   genesets was made nonredundant (Supplementary File S5) and used to analyze the DDR enrichment in the GEO data sets mentioned above.

Data availability
The data generated in this study are available within the article and its supplementary data files. Expression profile data analyzed in this study were obtained from GEO at GSE45001, GSE35602, GSE19632, and GSE40595

Statistical analysis
All experiments were performed at least twice and data are expressed as the mean AE standard error unless otherwise stated (SD ¼ standard deviation). Data are presented as a group with the number of biological replicates shown in the figure and the number of technical replicates described in the figure legends (n br or n tr , respectively) and summarized by mean AE 95% confidence intervals in all figures. The appropriate use of parametric versus nonparametric tests was determined using the D'Agostino and Pearson omnibus normality test, and where sample sizes were too small for this test, a normal distribution was assumed. The Grubbs test was used to identify outliers. Where data did not follow a normal distribution pattern, Kruskal-Wallis, Mantel-Cox (log-rank) or Spearman tests were used. For normal distributions, two-way ANOVA, or homo/paired/heteroscedastic Student t test were used as described in the figure legends. Statistical tests were carried out using GraphPad Prism v. 6-8, and were two-sided; a P ≤ 0.05 was considered significant (ns, nonsignificant; Ã , P ≤ 0.05; ÃÃ , P ≤ 0.01; ÃÃÃ , P ≤ 0.001; ÃÃÃÃ , P ≤ 0.0001).

ATM activation during myofibroblast differentiation
To investigate the DDR pathway during myofibroblast differentiation, we treated fibroblasts with TGFb1 and examined the activation of DDR kinases over time alongside myofibroblast markers [SMA, fibronectin EDA (FnEDA)]. We found activation of ATM (but not ATR or DNA-PKcs) during differentiation ( Fig. 1A and Supplementary Fig. S1A-S1E), whereas some ATM targets (H2AX and CHK2) were activated following TGFb1 treatment, others (p53 and KAP1) were not ( Fig. 1B and Supplementary  Fig. S1E), indicating that only a subset of ATM targets typically phosphorylated during genotoxic stress are activated during myofibroblast differentiation. We also found that TGFb1 treatment resulted in increased (albeit low) levels of pATM/pH2AX-positive nuclear foci and DNA breaks (Fig. 1C-E). This low level of DNA damage was associated with nonsignificant trends for reduced DNA replication and increased cell death; there were no changes in cell viability (monitored 3 days after TGFb1 treatment) or senescence (5 days; Fig. 1F and G). However, as previously shown (35), continuous TGFb1 administration resulted in increased senescence levels over time (% range SA-b-Gal-positive cells at 12 days ¼ 15.8-23.3; Fig. 1g), suggesting overall that accumulation of DNA damage induced by TGFb1 may ultimately promote cell senescence in a proportion of myofibroblasts.
We next examined CAF cultured ex vivo and human tumors in vivo for evidence of ATM activation. CAF from non-small cell lung cancer (NSCLC) and head and neck (HNSCC) showed ATM activation in vitro, which correlated with SMA expression (Spearman correlation ¼ 0.96, P ¼ 0.0028; Fig. 1H). GSEAs of microdissected tumor stroma data sets showed that myoCAF stroma of esophageal, ovarian, colorectal, and liver cancers are enriched for genes associated with the ATM/DDR pathway (Fig. 1I and Supplementary File 3). Multiplex immunochemistry (MxIHC) confirmed that the majority of SMA-positive CAF in NSCLC and HNSCC express pATM (Fig. 1J-N and Supplementary Fig. S1F-S1I), indicating, overall, that myoCAF in vitro and in vivo display activated ATM signaling.

Targeting ATM inhibits myofibroblast differentiation
To determine if ATM plays a functional role in myofibroblast differentiation, we cotreated fibroblasts with TGFb1 and an ATM inhibitor (KU55933). ATM inhibition suppressed differentiation, inhibiting expression of SMA ( Fig. 2A-C and Supplementary   Fig. S2A-S2H), contraction of collagen gels, SMA-positive stress fiber formation, and collagen1 deposition (Fig. 2D and E; Supplementary  Fig. S2I). We validated these findings using a second ATM inhibitor (KU60019) and ATM si/shRNA knockdown, which resulted in a similar reduction of SMA (ACTA2; Fig. 2B, F, and G; Supplementary  Targeting myofibroblast ATM promotes tumor CD8 T-cell infiltration and potentiates immunotherapy. A and B, Representative IHC staining (A) and relative quantification (B) of CD8 T cells in the core and periphery of TC-1 myo-rich tumors (described in Fig. 5F-J; n tr ¼ FoV ¼ 10; dotted lines, tumor margins). C and D, Representative IHC staining (C) and relative quantification (D) of CD8 T cells in the core and periphery of tumors described in Fig. 5K (n tr ¼ FoV ¼ 10). E and F, Flow cytometry analysis from TC-1 þ myo-rich tumors described in Fig. 5K (E shows two experiments; two-way ANOVA was used). (Continued on the following page.) Fig. S2J-S2L) and expression of ECM genes (COL1A1 and CTGF genes; Fig. 2G). Overexpression of a dominant-negative Mut-ATM (C2991L; refs. 31, 36) also suppressed expression of SMA and ECM genes ( Supplementary Fig. S2M-S2O). Targeting ATR or DNAPK did not reduce SMA expression (Fig. 2H-K), consistent with their lack of activation during myofibroblast differentiation ( Supplementary  Fig. S1E).
Recent single-cell transcriptomic studies have identified an inflammatory CAF (iCAF) subpopulation in several cancer types and described CAF plasticity, whereby cells in culture can be skewed between myoCAF and iCAF phenotypes by modulating TGFb1 and IL1 signaling, respectively (12,37). We, therefore, compared the effect of ATM inhibition on the expression of myoCAF and iCAF genes. KU55933 suppressed TGFb1 induction of myoCAF genes (COL11A1, ASPN, and ELN) and significantly upregulated IL1b induction of most iCAF genes (CCL2, LIF, IL6, and CXCL1; Supplementary Fig. S2P-S2S). This hints at a dual role for ATM in determining fibroblast phenotype: as promoter of myofibroblast differentiation and as inhibitor of inflammatory CAF phenotype.
We next investigated whether CHK2, a downstream target of ATM, is also involved in myofibroblast differentiation. For this, we used a CHK2-specific inhibitor CCT241533 (38) and found, similar to ATM inhibition, that it suppresses TGFb1-dependent SMA and FnEDA expression, SMA-positive stress fiber formation and collagen deposition ( Fig. 2L and M and Supplementary Fig. S2T and S2U). siRNA targeting CHK2 produced a similar effect, decreasing the expression of SMA (ACTA2), COL1A1, and CTGF ( Fig. 2N and O). Overall, these data indicate that ATM and its downstream signaling are central to the development of the myofibroblast phenotype.

ATM maintains the established myoCAF phenotype
Continued ATM activation in differentiated myoCAF in vitro and in vivo (Fig. 1H-N) suggests that it may play a role in maintaining the established myoCAF phenotype. We, therefore, examined the effect of ATM targeting on HNSCC and NSCLC CAF cultured ex vivo. We found that inhibiting ATM using KU55933 or shRNA knockdown, reverted CAF to a more quiescent phenotype, downregulating the expression of myofibroblast markers (SMA, collagen I, fibronectin; Fig. 3A-D), suppressing contractility (Fig. 3E) and inhibiting myoCAF-dependent cancer cell invasion ( Fig. 3F and G). Similarly, ATM inhibition or knockdown suppressed the invasion-promoting effects of myofibroblasts generated through TGFb1 treatment ( Supplementary Fig. S3A-S3C). Together, these data highlight that ATM can be targeted to both inhibit myoCAF differentiation and also "normalize" established myoCAF.

TGFb1 activates ATM via NOX4
We next investigated the molecular mechanisms underlying TGFb1-dependent activation of ATM. First, we inhibited TGFb-receptor I phosphorylation of SMAD2/3 and confirmed that this abolished ATM activation in TGFb1-treated fibroblasts (Fig. 4A). NOX4 is a SMAD downstream target (39), which we have shown previously regulates myoCAF differentiation (28); it has been reported that the ROS generated by this enzyme can cause DNA damage, suggesting a possible nuclear localization for NOX4 (27). We found that ATM phosphorylation and NOX4 expression increased with similar kinetics during the myofibroblast differentiation process (Fig. 1A and Supplementary Fig. S1A-S1C) and both localized to the cell nucleus ( Fig. 4B and Supplementary  Fig. S4A-C) but did not specifically colocalize (Fig. 4C). Targeting NOX4 using siRNA or a NOX4 inhibitor (GKT137831; ref. 28) suppressed ROS production ( Supplementary Fig. S4D) and inhibited ATM phosphorylation, as well as its downstream kinase activity ( Fig. 4D and E). DNA damage monitored by pH2AX protein expression was also decreased (Supplementary Fig. S4E and S4F). Because ATM is classically activated through DNA damage recognition by the MRN complex (29,30), we examined whether Mre11 was directly involved in TGFb1-dependent activation of ATM. Targeting Mre11 using siRNA or using a specific Mre11 inhibitor (Mirin) inhibited TGFb1-driven ATM activation/activity similar to NOX4 inhibition ( Fig. 4D and E). Consistent with this, Mre11 inhibition suppressed myofibroblast differentiation (Fig. 4F-J and Supplementary Fig. S4G and S4H).
Given that ATM can also be activated as a covalent dimer by oxidative stress (31), we next examined whether NOX4-driven ROS promotes ATM dimerization. First, we confirmed ATM dimer (ATM-D) formation by oxidation (H 2 O 2 ; Supplementary Fig. S4I), before examining the effect of NOX4 inhibition. We found that NOX4 targeting resulted in a marked reduction in basal and TGFb1-induced levels of the ATM dimer ( Fig. 4K and Supplementary Fig. S4L). Targeting Mre11 did not affect ATM dimerization (Supplementary Fig. S4J and S4K), consistent with the MRN complex being involved in TGFb1 activation of the monomer. Targeting NOX4 using GKT also inhibited ATM activation and dimerization in HNSCC and NSCLC CAF ( Supplementary Fig. S4M-S4Q). To confirm the role of NOX4 in ATM activation, we overexpressed NOX4 in HEK293 cells (33). This induced ATM activation, dimerization, and activity (the latter monitored by CHK2 phosphorylation) without the requirement for TGFb1 stimulation ( Fig. 4L and M), suggesting that NOX4 activation and oxidation of ATM are not cell type specific. These data, summarized in the schematic in Fig. 4N, show that activation of ATM during myofibroblast differentiation is modulated by NOX4 both through DNA damage/MRN complex and by direct oxidation of ATM.

ATM targeting in myofibroblasts reduces tumor growth
Next, we tested whether ATM targeting affects tumor growth in vivo using different murine tumor models (14,28). First, we coinjected immunocompromised mice with 5PT cancer cells, a cell line that promotes myoCAF differentiation (28), with either control HFFF2 fibroblasts or HFFF2 with ATM shRNA knockdown (Fig. 5A). Stromal ATM targeting attenuated myoCAF accumulation and suppressed tumor growth (Fig. 5B-E). To investigate the role of fibroblast ATM targeting in immunocompetent mouse models, and to overcome the issue of low CAF content in commonly used syngeneic murine tumors, we used a syngeneic isograft lung cancer model as described previously (14), coinjecting TC-1 cancer cells with TGFb1-treated MLF, which recapitulates the myoCAF-rich stroma of human tumors ( Supplementary Fig. S5A-S5C). We found that myofibroblast ATM shRNA knockdown similarly suppressed intratumoral myoCAF accumulation and reduced tumor size in this model (Fig. 5F-J).
We then evaluated the effect of AZD0156, a clinically tested ATMspecific inhibitor (NCT02588105). We confirmed that AZD0156 suppressed myofibroblast differentiation in vitro ( Supplementary  Fig. S5D-G) and also reversed myoCAF differentiation; notably, this latter effect was maintained when AZD0156 treatment was discontinued ( Supplementary Fig. S5H). A broad analysis of myoCAF, iCAF, and other immune genes showed that most myoCAF genes were significantly downregulated by ATM inhibition, whereas the effect on iCAF genes was gene specific and generally nonsignificant (Supplementary Fig. S5I and S5J). We next tested the effect of the inhibitor on mice injected with TC-1 cells AETGFb1-treated MLF (myoMLF). In mice bearing myoCAF-rich TC-1 tumors, AZD0156 suppressed myofibroblast accumulation, slowed tumor growth, and improved overall survival (Fig. 5K-O and Supplementary Fig. S5K-S5L). However, AZD0156 had minimal effect on the growth of control (myoCAF-low) tumors, showing that the effect of AZD0156 is specific to myoCAF in this model (Fig. 5K and L). Similar results were obtained using an isogenic colorectal cancer mouse model, coinjecting MC38 cancer cells with TGFb-treated mouse colorectal fibroblasts (myoMCF; Fig. 5P-S). Together, these data show that ATM can be targeted to suppress intratumoral myoCAF accumulation, resulting in reduced tumor growth.

ATM myoCAF targeting increases intratumoral CD8 T-cell infiltration and potentiates immunotherapy
We have shown previously that myoCAF confer immunotherapy resistance by excluding CD8 T cells from tumors (14). Therefore, we tested whether ATM inhibition could reverse this effect. Analysis of myoCAF-rich TC-1 and MC38 tumors by IHC showed that ATM inhibition (fibroblast shRNA knockdown and AZD0156) resulted in a significant relocation of CD8 T cells from the tumor periphery to the tumor core ( Fig. 6A-D, G, and H; Supplementary Fig. S6A and S6B). CD4 T-cell infiltration was unaffected ( Supplementary  Fig. S6C-H). Flow cytometry analysis showed that ATM inhibition did not affect CD8 T-cell phenotype/function in myoCAF-rich tumors ( Fig. 6E and F).
We next investigated whether the effect of ATM inhibition on intratumoral CD8 T cells could potentiate the response to immunotherapy. First, we tested a vaccine model using HPV E6/E7-expressing TC-1 cells, combining AZD0156 with a DNA vaccine directed against E7 (RAH; ref. 34); we have shown previously that myoCAF-rich TC-1 tumors are resistant to the vaccine response (14). Mice bearing CAFrich TC-1 tumors were treated with AZD0156 or vaccine monotherapy, vaccine/drug combination, or control. While AZD0156 monotherapy was effective in reducing intratumoral myoCAF accumulation (Supplementary Fig. S6I and S6J), the drug/vaccine combination significantly increased CD8 T-cell infiltration and reduced tumor volume compared with the single treatments (Fig. 6I-L and Supplementary Fig. S7A). AZD0156 was also tested in combination with aPD-1 in mice bearing myoCAF-rich MC38 tumors, a model where, similarly, myoCAF confer resistance to anti-PD-1 therapy (14). We found again that the combination of aPD-1 with AZD0156 produced the most significant CD8 T-cell influx and reduction in tumor growth ( Fig. 6M-P and Supplementary Figs. S6K, S6L, and S7B). Survival experiments performed over a longer time period showed that combining ATM inhibition with immunotherapy increased the overall survival compared with single treatments alone in both tumor models ( Fig. 6Q and R; Supplementary Fig. S7C and D). When tumors relapsed upon aPD1 AE AZD0156, a second treatment with the combination was administered, resulting in significantly fewer myo-CAF and increased CD8 T cells ( Supplementary Fig. S7E-S7G). Notably, despite the high levels of CD8 T cells in these tumors, mice showed only a nonsignificant trend for increased survival (Supplementary Fig. S7H-S7O). Overall, these data suggest that ATM inhibition is effective at promoting the infiltration of CD8 T cells into myoCAF-rich tumors and can be used to potentiate the early response to immunotherapy.

Discussion
MyoCAF are associated with poor prognosis in many cancers and have been shown to promote tumor immune evasion. However, effective targeting of this cell population has not yet been accomplished (16,19), in part confounded by the fact that CAF remain a poorly defined, heterogeneous cell population (7,17). Recent singlecell transcriptomic analyses have identified novel CAF phenotypes, including iCAF and antigen-presenting CAF (apCAF), although it is not yet clear how widely these CAF subgroups are found in different cancers or how they function (40). Most research has focused on myoCAF. These are present to a greater or lesser extent in most types of solid tumors, usually abutting tumor cells, and depositing a desmoplastic stroma rich in collagens, fibronectin, and proteoglycans that has been shown to "trap" T cells and limit T-cell access to the tumor core (14,41,42). Notably, expression of TGFb-associated myoCAF ECM genes is one of the strongest predictors of immunotherapy failure, highlighting the ability of myoCAF to create an immunosuppressive tumor microenvironment (8)(9)(10)(11)(12). Here, we show that the activation of ATM plays a central role in promoting and maintaining the myoCAF phenotype and that ATM can be effectively targeted to "normalize" myoCAF, overcome myoCAF-mediated immune evasion, and potentiate response to immunotherapy.
ATM, along with ATR and DNA-PKcs, is principally known for its role as an apical kinase in the DDR pathway. However, ATM signaling has been shown to also regulate other cell functions, including glucose metabolism (43), cell homeostasis (44), and multiple differentiation processes (45). We found that TGFb1induced myofibroblast differentiation results in the activation of ATM and CHK2 in the presence of significantly increased (albeit low) DNA breaks. The localized pattern of pH2AX and pATM staining in the nuclei of cells treated with TGFb1 confirmed the presence of DNA damage in the form of DSBs (30), consistent with TGFb/SMAD signaling, promoting ATM activation and downstream signaling as a result of direct DNA damage (25)(26)(27)46). In keeping with this, we found that inhibiting MRN complex/DNA damage activation of ATM by targeting Mre11 inhibits TGFb1 induction of the kinase.
In breast CAF, ATM has also been shown to be activated by oxidative stress in the absence of DNA damage, suggesting potential activation of the ATM dimer (47). Here, we found that ATM activation results from both oxidation and DNA damage and that both are dependent on ROS generated by NOX4. We established that NOX4 (but not Mre11) activates ATM as a dimer, indicating a ROS-driven oxidation of the kinase independent of monomer activation by the MRN complex. Consistent with NOX4 being reported to promote DDR activation in several cell types (25,27), here we demonstrate that NOX4 is the source of ROS promoting ATM dimer formation during the myofibroblast/myoCAF differentiation process. This likely explains the lack of KAP1 activation typically observed during oxidative stress activation of ATM (31) and proposes a model whereby there exists a parallel activation of ATM by DNA damage and oxidative stress, both induced by NOX4.
We found that ATM was activated in myoCAF isolated ex vivo from NSCLC and HNSCC, also detected by multiplexed immunochemistry in the SMA-positive stroma of the same tumors. GSEA also showed enrichment of ATM/DDR-related genes in the myofibroblastic stroma of ovarian, esophageal, liver, and colorectal tumors, consistent with proteomic analysis of esophageal and breast CAF performed elsewhere (48,49). In support of its role in myoCAF differentiation, ATM activation has additionally been reported to contribute to various fibrotic conditions, including systemic sclerosis (50), renal (51), and hepatocellular fibrosis (52), all myofibroblast-dependent processes, suggesting, overall, that the association of ATM signaling and myofibroblast/myoCAF phenotypes holds true across different tissues and disease pathogenesis. Continued ATM activation in myoCAF isolated ex vivo suggested that it also plays a role in maintaining the myoCAF phenotype. We found that ATM inhibition normalized myoCAF, downregulating SMA and other myoCAF markers while promoting an iCAF phenotype, consistent with recent studies that have highlighted plasticity in CAF populations, showing that CAF subgroups are not fixed in the state of terminal differentiation (37,53).
We also investigated whether DDR upstream or downstream components of the ATM pathway can regulate myofibroblastic CAF phenotype and found that Mre11 and CHK2 are also involved in myofibroblast differentiation. This contrasts with previous work that identified CHK2 as a repressor of breast myofibroblastic CAF phenotype (54), although it is consistent with the increased DDR gene expression and ATM activation observed in breast CAF by several other groups (47,49) ATM signaling involvement in myoCAF differentiation is in keeping with our previous data showing that DNA damaging agents, including irradiation, can induce a contractile, SMA-positive myofibroblastic-like phenotype through an ATM-dependent mechanism (20). Additionally, we have shown that continuous TGFb1 treatment of fibroblasts over time promotes senescence (35), which is commonly triggered by DNA damage. It appears therefore that myofibroblast activation and senescence may be two linked stress responses, perhaps with TGFb1 induction of ATM signaling and the myofibroblast phenotype, followed by later senescence forming part of the same differentiation program.
The recognition that myoCAF provide a major resistance mechanism to immunotherapy has renewed interest in CAF targeting as an immunotherapy adjunct, and different therapeutic approaches have been suggested. TGFb signaling is the major pathway regulating myoCAF differentiation, and different groups have shown that coadministration of anti-TGFb with anti-PD-1/PD-L1 antibodies significantly improves response in preclinical tumor models (8,13). TGFb blockage has been shown to reduce the formation of myoCAF and promote the formation of an "interferon-licensed" CAF subpopulation with increased immunomodulatory properties, resulting in greater anti-PD-1 efficacy (53). Inhibiting TGFb, however, can be problematic; this pleiotropic cytokine has multiple roles in normal physiology as well as tumor-suppressive effects, and its targeting has led to on-target cardiac toxicities in preclinical studies, as well as the development of cutaneous tumors in human trials (18,55). Inhibiting the mechanisms by which CAF exclude T cells from tumors is an alternative approach to overcome immunotherapy resistance and perhaps the most attractive strategy to do that is to 'normalize' myoCAF, particularly since studies have suggested that certain CAF/fibroblast phenotypes may be tumor-suppressive (56). We previously identified the ROS-producing enzyme NOX4 as a key regulator of the myoCAF phenotype (14,28). Inhibition of NOX4 using the small-molecule inhibitor GKT137831/Setanaxib suppresses myoCAF differentiation and overcomes myoCAF-mediated immunotherapy resistance (14). Here, we found that ATM is activated by NOX4 and plays a similar role in regulating and maintaining myoCAF differentiation.
Drugs inhibiting ATM and other components of the DDR are generally effective in tumors with specific preexisting DNA-repair defects or are used in combination with platinum compounds or ionizing radiation where they are associated with significant toxicity (57). As monotherapy, ATM inhibitors are well tolerated, with dose escalation phase I clinical studies with the ATM inhibitors AZD0156 or AZD1390 not yet reporting to be associated with adverse effects. However, ATM inhibitor monotherapy is generally therapeutically ineffective in conventional preclinical tumor models (58,59), where the stromal component is typically absent (14). Nonetheless, recent studies have shown that ATM inhibitors can promote response to checkpoint immunotherapy through cGAS/STING signaling in tumor cells, resulting in an enhanced interferon I response and increased immunogenicity (58). Notably, we found that ATM inhibition did not affect control (CAF-low) tumors comprising TC-1 and MC38 cells alone, but its effect was limited to myoCAF-rich tumors, where it reversed myoCAF differentiation and potentiated immunotherapy response. Whether this latter effect is modulated through the downregulation of ECM proteins or altered expression of inflammatory cytokines remains to be determined (41,58), but our data suggest that the use of ATM inhibitors can be expanded for stromal targeting.
In summary, we identify ATM activation as a novel, targetable pathway regulating myoCAF differentiation. Most types of solid tumors have a myoCAF-rich subgroup; given the major role of myoCAF in suppressing response to anti-PD-1/PD-L1 checkpoint inhibitors (8,13), targeting this pathway as part of combination immunotherapy could have significant therapeutic benefit.