Eukaryotic cells have evolved complex mechanisms to deal with environmental stresses. Signal-transduction pathways are rapidly activated after exposure to DNA-damaging agents and other cellular stresses, and these pathways affect processes such as gene transcription and cell-cycle progression1, 2, 3. ATM, the gene that is mutated in the human disease ataxia–telangiectasia (AT), is crucial for the initiation of signalling pathways in mammalian cells following exposure to ionizing radiation (IR) and to other agents that introduce double-strand breaks into cellular DNA4, 5. Cells from AT patients typically lack detectable ATM protein, contain abnormalities in telomere morphology and have abnormal responses to IR, including increased cell death, increased chromosomal breakage and cell-cycle checkpoint defects6. In addition, AT patients exhibit progressive cerebellar ataxia, immune deficiencies, gonadal atrophy, oculocutaneous telangiectasias, radiation sensitivity, premature ageing and increased risk of cancers, particularly lymphomas.
The ATM gene encodes a 370-kDa protein that belongs to the phosphoinositide 3-kinase (PI(3)K) superfamily7, but which phosphorylates proteins rather than lipids8, 9. The 350-amino-acid kinase domain at the carboxy terminus of this large protein is the only segment of ATM with an assigned function. Exposure of cells to IR triggers ATM kinase activity, and this function is required for arrests in G1, S and G2 phases of the cell cycle5. Several substrates of the ATM kinase participate in these IR-induced cell-cycle arrests. These include p53, Mdm2 and Chk2 in the G1 checkpoint8, 9, 10, 11, 12; Nbs1, Brca1, FancD2 and SMC1 in the transient IR-induced S-phase arrest13, 14, 15, 16, 17, 18, 19; and Brca1 and hRad17 in the G2/M checkpoint20, 21.
The mechanisms by which eukaryotic cells sense DNA strand breaks remain to be elucidated, but the rapid induction of ATM kinase activity following IR suggests that it acts at an early stage of signal transduction in mammalian cells8, 9. Transfected ATM is a phosphoprotein, raising the possibility that ATM kinase activity is modulated by post-translational modification. We have now identified an important site of ATM phosphorylation induced by IR; we have shown that it results from intermolecular autophosphorylation, and have proceeded to explore the functional significance of the phosphorylation event. This phosphorylation event does not directly regulate the activity of the kinase, but instead disrupts ATM oligomers, which in turn allows accessibility of substrates to the ATM kinase domain. The rapidity and stoichiometry of the reaction indicate that ATM is not activated by binding directly to DNA strand breaks, and we present data that support a model in which DNA damage rapidly causes changes in higher-order chromatin structures that initiate ATM activation.
Ionizing radiation induces autophosphorylation of ATM
Exposure of cells to IR significantly increased the incorporation of 32P-orthophosphate into both transfected ATM (Fig. 1a) and endogenous ATM (Fig. 1b). No such increase was observed after the irradiation of cells that had been transfected with kinase-inactive ATM. Incorporation of radioactive phosphate into ATM also occurs in in vitro assays of ATM kinase activity9, 22. This in vitro phosphorylation of ATM is not seen with the kinase-inactive ATM protein, is inhibited by exposure to 30 nM wortmannin, is dependent on the addition of manganese, and is not dependent on the addition of exogenous DNA (Fig. 1c). Identical properties are characteristic of phosphorylation of target substrates by ATM9, 22, 23. These observations are all consistent with a model in which ATM phosphorylates itself. To identify the site(s) of phosphorylation in ATM, radioactively labelled ATM protein was digested with trypsin, and tryptic phosphopeptides were evaluated by sequential two-dimensional electrophoresis and chromatography. A single major de novo phosphorylated peptide was identified in both transfected (Fig. 1d) and endogenous (Fig. 1e) ATM isolated from irradiated cells. Phospho-amino-acid analysis showed the target amino acid to be a serine (data not shown).
Figure 1: Metabolic labelling of ATM in response to ionizing radiation.
a, ATM phosphorylation is increased in response to ionizing radiation (IR) and is dependent on ATM kinase activity. After transfection of Flag-tagged wild-type (WT) or mutant (KD) ATM, 293T cells were irradiated (10 Gy) and then incubated with 32P-orthophosphate for 30 min. Radiolabelled ATM was immunoprecipitated with anti-Flag, and evaluated by autoradiography (upper panel) and by immunoblot with anti-Flag antibody (lower panel). b, Phosphorylation of endogenous ATM is increased in response to IR. GM00536 cells were irradiated (10 Gy) and then metabolically labelled with 32P-orthophosphate for 30 min. ATM was immunoprecipitated (IP) and evaluated by autoradiography (upper panel) and by immunoblot with anti-ATM antibody (lower panel). c, WT-ATM autophosphorylates in vitro. After transfection into 293T cells, Flag-tagged WT-ATM or KD-ATM were immunoprecipitated with anti-Flag and used in in vitro kinase assays under different conditions. Wortmannin was added to the samples indicated, on ice, 15 min before the kinase reaction, and 5 
M manganese and 1 g sonicated calf thymus DNA were present or absent in the kinase reaction, as indicated. d, e, A single principal tryptic phosphopeptide in ATM is detected after IR. Flag-tagged WT-ATM or endogenous ATM labelled following IR was digested on the nitrocellulose membrane with trypsin. Peptides were purified and spotted at the origin, and subjected to electrophoresis (x axis) and thin-layer chromatography (TLC; y axis). f, The principal tryptic phosphopeptide contains a glutamic acid residue. Peptides generated by V8 protease digestion of the principal tryptic ATM phosphopeptide were spotted (Ori) on the plate, left to right, as follows: purified tryptic peptide, V8 digested tryptic peptide and a mixture of the two samples. g, h, Edman degradation. The principal tryptic peptide and the V8/tryptic peptide were purified and subjected to the noted number of cycles of manual Edman degradation. Phenyl isothiocyanate (PITC) derivatization of the lysine 
-amino group results in changed mobility of the peptides after one cycle. Free phosphate is released from the tryptic or V8/tryptic phosphopeptides after eight to nine cycles or after two cycles, respectively. i, The predicted IR-induced site of phosphorylation, serine 1981, is conserved in mouse and Xenopus. -ve, negative; +ve, positive.
After failing to identify the target peptide by automated sequencing, mass spectrometry, predicted chromatographic migration, and mutagenesis of all of the serines in the ATM kinase domain, the labelled phosphopeptide was ultimately identified by secondary proteolytic and chemical cleavage of the primary phosphopeptide. Digestion of the primary tryptic phosphopeptide with V8 protease resulted in a peptide with altered chromatographic mobility (Fig. 1f), dictating the presence of a glutamic acid in the peptide. Unchanged peptide mobility after cyanogen bromide treatment in formic acid indicated that it does not contain methionine (data not shown). Manual Edman degradation of both the principal tryptic phosphopeptide and the V8/tryptic phosphopeptide revealed a secondary spot in each cycle, consistent with a derivatized C-terminal lysine -amino group peptide24 (Fig. 1g,h). The phosphorylated serine was released after the eighth or ninth cycle of Edman degradation from the tryptic peptide and after the second cycle from the V8/tryptic peptide (Fig. 1g,h). Thus, the phosphorylated serine is eight or nine amino acids to the C-terminal side of either a lysine or arginine residue in the tryptic peptide, and two amino acids to the C-terminal side of a glutamic acid residue in the V8/tryptic peptide. The only tryptic peptide in ATM that meets these sequence requirements is the 19-residue peptide 1974-SLAFEEGSPQSTTISSLSEK-1992, with serine 1981 as the predicted phosphorylated serine (Fig. 1i). Serine 1981 is conserved in mouse and Xenopus ATM (Fig. 1i), but is not found in suspected homologues of less complex metazoans, and is located in the amino terminus of the FAT domain, a region of approximately 500 amino acids with some conservation across the PI(3)K family of kinases, including Frap, ATM and Trapp25. Because the target serine is in an 'SQ' site, either an autophosphorylation event or phosphorylation by an ATM family member is indicated22.
Rabbit polyclonal antibodies were generated that specifically recognize serine 1981 only when it is unphosphorylated (anti-1981S) or only when it is phosphorylated (anti-1981S-P), and specificity was shown on peptides and full-length protein (Fig. 2a–c). Within 30 min after the irradiation of cells at 10 Gy, the recognition of wild-type ATM by the anti-1981S antisera decreased several fold, whereas recognition of kinase-inactive ATM was unchanged (Fig. 2c). Conversely, binding of the anti-1981S-P antisera increased several fold after IR, whereas recognition of kinase-inactive ATM was unaffected. The decrease in anti-1981S binding indicates that a high fraction of total cellular ATM becomes modified on serine 1981 after IR, and that anti-1981S-P binding results mirror the metabolic labelling of transfected wild-type ATM and kinase-inactive ATM in 293T cells (Fig. 1a). Phosphorylation of serine 1981 in endogenous ATM in non-transformed human fibroblasts was not apparent in unperturbed cells, but was readily detectable 1 h after exposure to IR and 5 h after exposure to both IR and ultraviolet irradiation (Fig. 2d). Recognition of ATM by the phosphospecific antibody after IR cannot simply be due to changes in cell-cycle distribution, because cell-cycle profiles do not change significantly in the first hour after IR, and this phosphorylation event is also seen in primary fibroblasts that are arrested in G0 after either IR or ultraviolet irradiation (Fig. 2c, right column).
Figure 2: Intermolecular autophosphorylation of ATM serine 1981 occurs in vivo after irradiation.
a, Anti-1981S and anti-1981S-P specific antibodies were generated. Dot blots were prepared with 1 g spots of 367S, 1981S, 367S-P and 1981S-P peptides, as indicated in the schema, and were immunoblotted with either anti-1981S or 1981S-P specific antisera. To confirm specificity, 1981S or 1981S-P peptide (20 g ml-1) was added as a blocker to the antisera, as indicated, for 1 h before immunoblotting. b, Anti-1981S-P recognizes a single protein in crude extracts prepared from exponentially growing primary fibroblasts that is induced by IR (arrow). Whole-cell extract was prepared 30 min after 10 Gy IR, and 10 g of each sample was resolved and immunoblotted with anti-1981S-P antibody. c, 1981S is phosphorylated in vivo. Thirty minutes after irradiation (10 Gy), transfected Flag-tagged WT-ATM, KD-ATM or S1981A-ATM was immunoprecipitated from 293T cells and immunoblotted with anti-1981S, anti-1981S-P or anti-Flag antibodies. d, ATM is phosphorylated de novo in primary fibroblasts following exposure to IR and ultraviolet light. Exponentially dividing and G0 primary fibroblasts were exposed to either 10 Gy IR or 10 J m-2 ultraviolet, and whole-cell extracts were prepared after 1 h and 5 h. ATM was immunoprecipitated and the samples were blotted with anti-1981S-P and anti-ATM antibodies. Levels of p53 serine 15 phosphorylation and total p53 were determined by immunoblot. e, ATM phosphorylates a glutathione S-transferase (GST) fusion protein containing 1981S in vitro. GST fusion proteins containing 1981S or 1981A were generated, purified and used as substrates in in vitro kinase assays with either WT or KD Flag-ATM. Upper panel, 32P incorporation; middle panel, 1981S-P immunoblot; lower panel, Flag-ATM immunoblot. f, IR-induced serine 1981 phosphorylation is inhibited by 20 M wortmannin. Primary fibroblasts were incubated with wortmannin for 1 h and exposed to 10 Gy IR; serine 1981 phosphorylation was assessed in ATM immunoprecipitated from whole-cell extracts that were prepared 30 min later. g, ATM kinase activity is required for serine 1981S phosphorylation after IR. Flag-tagged WT-ATM, KD-ATM or S1981A-ATM was expressed in ataxia–telangiectasia (AT) fibroblasts. Thirty minutes after 10 Gy IR, ATM was immunoprecipitated and immunoblotted with anti-1981S-P and anti-Flag antibodies. Complementation of IR-induced p53 phosphorylation was also assessed (bottom panels).
ATM autophosphorylation had been suggested by the dependence of the enhanced incorporation of phosphate after IR on ATM kinase activity in vivo and in vitro, by its wortmannin sensitivity, and by the fact that the target serine was an 'SQ' site. A polypeptide containing the appropriate serine residue at 1981 was an excellent in vitro substrate of the ATM kinase (Fig. 2e), showing that ATM could phosphorylate this site. In addition, wortmannin concentrations of 20 M or more effectively inhibited phosphorylation of serine 1981 after the irradiation of human diploid fibroblasts (Fig. 2f). This concentration of wortmannin inhibits both ATM and DNA-dependent protein kinase (DNA-PK), but not ATR (ataxia–telangiectasia and Rad3 related), in vivo23. The kinetics and levels of IR-induced phosphorylation of serine 1981 were identical in matched cell lines with and without DNA-PK activity (data not shown), thus excluding DNA-PK as a potential responsible kinase.
Although the previous in vivo experiments here had shown that optimal phosphorylation of ATM was dependent on the presence of active ATM kinase, their interpretation was complicated by the fact that the cells that we used contained endogenous wild-type ATM. Even the kinase-inactive mutant of ATM can be phosphorylated to some extent in these cells (Figs 1a and 2c). To further demonstrate the importance of ATM kinase activity in the phosphorylation of serine 1981, constructs encoding wild-type, kinase-inactive and S1981A-ATM were transfected into AT cells so that the only potential source of ATM kinase activity was the transgene being used. All of these constructs were expressed at similar levels in AT cells, but only wild-type ATM was recognized by anti-1981S-P, and its binding was increased several fold within 30 min of exposure to IR at 10 Gy (Fig. 2g). Thus, although some phosphorylation of kinase-inactive ATM is observed after transfection into 293T cells (which contain endogenous ATM activity), transfection into a cell line that lacks ATM kinase activity shows no detectable phosphorylation on serine 1981. Therefore, phosphorylation of serine 1981 depends on the activity of the ATM kinase itself, and the phosphorylation of transfected kinase-inactive ATM must occur in trans.
Dissociation of ATM into active monomers
Because the various ATM constructs had been introduced into AT cells, the functional ramifications of serine 1981 phosphorylation could also begin to be assessed. Although the introduction of wild-type ATM restored IR-induced p53 phosphorylation to AT cells, neither kinase-inactive nor S1981A-ATM provided this functional complementation (Fig. 2g). By contrast, the in vitro kinase activity of S1981A-ATM was indistinguishable from wild-type ATM (Fig. 3a). Transfection of S1981A-ATM into HeLa cells inhibited both the IR-induced G2 checkpoint (Fig. 3b) and S-phase replication arrest (Fig. 3c) in a manner indistinguishable from the kinase-inactive ATM construct. Consistent with its abrogation of the IR-induced cell-cycle checkpoints, the transfected S1981A mutant also blocked the IR-induced phosphorylation of endogenous ATM protein on serine 1981 (Fig. 3d). Thus, although mutation of serine 1981 did not abolish ATM kinase activity in vitro, expression of S1981A-ATM fails to complement AT cells, and effectively inhibits the cellular activities of endogenous ATM in a dominant-inhibitory manner.
Figure 3: S1981A-ATM has kinase activity but is dominant-inhibitory in cells.
a, Flag-tagged S1981A-ATM has kinase activity. Flag-tagged WT-ATM, KD-ATM or S1981A-ATM was expressed in AT fibroblasts and immunoprecipitated for use in in vitro kinase assays using GST-p53 substrates. 32P incorporation into both GST-WT-p53 and GST-p53 mutated to alanine at the two 'SQ' sites, serines 15 and 37, was assessed. b, Flag-tagged S1981A-ATM is dominant-inhibitory in a G2/M checkpoint assay. The three Flag-tagged ATM constructs or vector alone were transfected into HeLa cells. The percentage of cells in mitosis with and without 2 Gy IR was assessed using flow cytometry by examining histone H3 serine 10 phosphorylation and DNA content. c, Flag-tagged S1981A-ATM is dominant-inhibitory in an S-phase checkpoint assay. HeLa cells pre-labelled with 14C-thymidine were transfected with the three Flag-tagged ATM constructs or vector alone, irradiated (5 Gy) and, 30 min later, pulsed with 3H-thymidine for 20 min. The relative amount of DNA synthesis in irradiated compared with unirradiated cells was determined. d, KD-ATM and Flag-tagged S1981A-ATM are dominant-inhibitory for IR-induced intermolecular autophosphorylation of endogenous ATM at serine 1981. The two mutant Flag-tagged ATM constructs or vector alone were transfected into 293T cells. Thirty minutes after exposure to IR (2 Gy), Flag-ATM was cleared from the lysate 2 
by immunoprecipitation with anti-Flag beads, and the remaining endogenous ATM was immunoprecipitated with anti-ATM antibody. Levels of immunoprecipitated endogenous ATM and levels of contaminating transfected Flag-ATM were determined by anti-ATM and anti-Flag immunoblotting, respectively.
To investigate the reasons for the different effects of serine 1981 mutation on in vitro versus cellular activities of ATM, biochemical studies of ATM domains and protein–protein interactions were pursued. Although a peptide containing the ATM kinase domain was insoluble in bacteria when co-transfected with glutathione S-transferase (GST) alone or with a GST fusion with the ATM target peptide p53(1–101), a significant amount of the peptide was stabilized and solubilized in the presence of a GST fusion protein containing ATM amino acids 1961–2046 (Fig. 4a, top panel). The kinase domain was also solubilized in the presence of the GST 1961–2046 fusion protein containing S1981A, but remained insoluble in the presence of phosphorylation-mimic fusion peptides S1981D and S1981E. Furthermore, the soluble kinase domain co-purified on glutathione-agarose with the GST 1961–2046 fusion proteins containing wild-type sequence or S1981A (Fig. 4a, third panel). These results show that the kinase domain and phosphorylation domain can stably bind to one another, and that sequences flanking serine 1981 are crucial for this interaction. The fact that the interaction is prevented by mutation of serine 1981 to either aspartic acid or glutamic acid, both of which have charged side chains that mimic serine phosphorylation, indicated that phosphorylation of serine 1981 would prevent the interaction of this domain with the kinase domain (see below).
Figure 4: Interaction of ATM domains with proteins.
a, A recombinant ATM kinase domain is stabilized in Escherichia coli by GST fusion proteins containing either 1981S or 1981A. Recombinant 6 His-tagged ATM kinase domain and GST alone, GST fusion proteins containing either 1981S, 1981A, 1981D or 1981E, or GST-p53 were expressed in E. coli BL21. Whole-cell extracts were prepared, and soluble and insoluble fractions separated. The amount of soluble (upper panel) and insoluble (second panel) recombinant ATM kinase was determined by anti-6 His immunoblotting. Binding of GST-1981S and GST-1981A to the kinase domain was shown by co-purification from the soluble fraction using glutathione-agarose (third panel). Expression of GST proteins was assessed by anti-GST immunoblotting (bottom panel). b, ATM can be crosslinked into a complex that migrates considerably slower than the ATM monomer before, but not after, exposure to IR. Thirty minutes after irradiation (0 or 10 Gy), exponentially growing fibroblasts were treated with formaldehyde for 10 min. Whole-cell extracts were prepared, and immunoprecipitated ATM was immunoblotted with anti-1981S-P or anti-ATM. c, ATM is a dimer before, but not after, exposure to IR. Haemagglutinin (HA)-tagged WT-ATM was co-expressed with either Flag-tagged WT-ATM, KD-ATM or S1981A-ATM. Thirty minutes after irradiation (0 or 10 Gy), ATM immunoprecipitated with anti-Flag was immunoblotted with anti-HA antibody. Total-cell extract was also immunoblotted with anti-HA (bottom panel). d, Purified, recombinant ATM fragment (arrow) is soluble in Saccharomyces cerevisiae. Coomassie-blue stain of 6 His-Flag-ATM 1909–3056 that was affinity-purified against anti-Flag M2 sepharose, eluted with Flag peptide, subjected to size-exclusion chromatography and resolved on a 4–12% denaturing gradient gel. The single additional protein band of approximately 60 kDa was identified as an ATM degradation product by immunoblotting (right panel). e, Purified recombinant ATM fragment has kinase activity. The recombinant protein shown in part d specifically phosphorylates p53 at serine 15 and itself at serine 1981. Autophosphorylation is not stimulated by the addition of exogenous DNA ends, and both phosphorylation events are inhibited by wortmannin. Anti-ATM immunoblot of the yeast ATM protein fragment reveals a band migrating at a size consistent with a homodimer (top panel, arrowhead).
If the serine 1981 domain of ATM (in the unphosphorylated state) binds to the kinase domain of ATM, this could occur either in cis or in trans. To determine whether ATM protein exists in a protein complex in cells, diploid human fibroblasts were exposed to a range of formaldehyde concentrations to attempt to covalently crosslink endogenous ATM into the most minimal complex feasible. A prominent complex containing ATM could be immunoprecipitated from the cells following their treatment with 0.5 mM or 1 mM formaldehyde for 10 min (Fig. 4b). This complex migrated electrophoretically considerably more slowly than the denatured ATM monomer, which runs at 370 kDa, and beyond the range of conventional markers of molecular mass. This ATM-containing complex was not present if the cells had been exposed to IR and was not recognized by the anti-1981S-P antibody (Fig. 4b). To establish whether at least two ATM molecules are present in a cellular complex and whether dissociation of the complex is dependent on the phosphorylation of serine 1981, haemagglutinin (HA)-tagged ATM was transfected into 293T cells along with wild-type, kinase-inactive or S1981A Flag-tagged ATM. HA-ATM was immunoprecipitated by anti-Flag sepharose in association with each of the three Flag-tagged ATM proteins (Fig. 4c). After irradiation at 2 Gy, HA-ATM was immunoprecipitated with both kinase-inactive and S1981A Flag-ATM, but was no longer bound to wild-type Flag-ATM. Consistent with the concept that an unphosphorylated serine 1981 is required to bind to the kinase domain and enable it to fold correctly, ATM constructs with serine 1981 mutated to aspartic acid or glutamic acid could not be expressed in cells (data not shown). Together, these results show that ATM exists as a dimer or a higher-order multimer in unperturbed cells, and that intermolecular ATM autophosphorylation on serine 1981 is required for the dissociation of the complex after DNA damage.
These results provide a mechanistic explanation for the dominant-negative cellular activities of both the kinase-inactive ATM and the S1981A ATM mutant. Because the ATM kinase domain interacts stably with the domain containing the autophosphorylation site, we reasoned that a truncated recombinant ATM molecule that included the phosphorylation and kinase domains might fold correctly and possibly have kinase activity. A plasmid containing 6 His-Flag-ATM 1923–3056 was expressed in Saccharomyces cerevisiae, and a soluble ATM fragment was recovered and purified by anti-Flag affinity and size-exclusion chromatography (Fig. 4d). In an in vitro kinase assay, the highly purified ATM fragment had wortmannin-sensitive activity and became autophosphorylated on serine 1981 (Fig. 4e). Further supporting the concept that the ATM protein homodimerizes in cells, some of this ATM fragment migrated at the size of a dimer even under the denaturing conditions of the SDS–PAGE (polyacrylamide gel electrophoresis) gel (Fig. 4e, top panel).
Cellular signals involved in ATM activation
The kinetics, dose responsiveness and stoichiometry of ATM autophosphorylation after IR were examined in primary human fibroblasts. Phosphorylation of serine 1981 was detectable immediately after collecting cells following the 30 s exposure required to deliver 0.5 Gy of IR, was maximal by 5 min, and remained stable and detectable for at least 24 h thereafter (Fig. 5a,b). Dose-dependence studies showed that phosphorylation of serine 1981 was detectable after 0.1 Gy IR, that it was maximal at 0.4 Gy at a 15 min time point, and that no further increase was seen between 1 and 9 Gy (Fig. 6b,d). Exposure to IR at 0.1 Gy should, theoretically, cause just four double-strand breaks in the genomic DNA of a human diploid cell26, 27. The extraordinary sensitivity of the detection of DNA strand breaks by ATM phosphorylation was supported by the observation that ATM phosphorylation could also be detected by the introduction of a small number of DNA breaks introduced by a restriction enzyme. ATM phosphorylation was detectable (Fig. 5f) after transfection of the restriction enzyme I-Sce1 into SV-40-transformed fibroblasts that contained two integrated copies of a plasmid containing the I-Sce1 target sequence, a site that has not been found in an unmanipulated mammalian genome28.
Figure 5: Kinetics, stoichiometry and detection sensitivity of ATM phosphorylation.
a, b, 1981S phosphorylation is maximal within 5 min after exposure to IR. Exponentially growing primary fibroblasts were exposed to 0.5 or 2 Gy IR. ATM immunoprecipitation was performed at the times indicated, and the levels of 1981S phosphorylation and total immunoprecipitated ATM were assessed by immunoblotting. c, d, 1981S phosphorylation is maximal at about 0.4 Gy IR. ATM was immunoprecipitated from exponentially growing primary fibroblasts 30 min after various doses of IR, and the levels of 1981S phosphorylation and total ATM were determined by immunoblotting. e, Over 50% of ATM is phosphorylated within 15 min following irradiation at 0.5 Gy. Exponentially growing primary fibroblasts were exposed to 0 or 0.5 Gy IR, and lysates were subjected to sequential immunoprecipitation with conventional anti-ATM or anti-1981S-P antibody followed by conventional anti-ATM antibody. The amount of ATM immunoprecipitated was determined by anti-ATM immunoblotting, and the specificity and completeness of the 1981S-P immunoprecipitation were confirmed by anti-1981S-P immunoblotting. f, The anti-1981S-P antibody can detect the presence of a few DNA breaks in the human genome. Constructs expressing Flag-tagged WT-ATM, KD-ATM or S1981A-ATM were co-transfected with a construct encoding the restriction enzyme I-Sce1 into GM00637 cells that had integrated two copies of a plasmid containing the sequence cut by I-Sce1. As a control, ATM was immunoprecipitated from irradiated cells (0.5 Gy for 30 min). 1981S phosphorylation and total ATM were assessed by immunoblotting with anti-1981S-P or anti-Flag.
High resolution image and legend (79K)Figure 6: Activation of ATM by chromatin-active agents that do not cause DNA breaks.
a, Primary fibroblasts grown on glass slides were irradiated (0.2 or 1 Gy) and allowed to recover for 5 min or 1 h, incubated in hypotonic buffer containing 100 mM for 1 h, or treated with chloroquine (32 g ml-1) for 4 h. Cells were fixed, and immunofluorescence for H2AX
, 1981S-P and 1981S was assessed. b, Exponentially growing fibroblasts were incubated in hypotonic (100 mM or 50 mM NaCl) buffer for 1 h or in chloroquine for 4 h. ATM was immunoprecipitated, and the levels of phosphorylated ATM, total ATM and H2AX were determined by immunoblotting. No cell death was observed in chloroquine or in the hypotonic conditions, and all cells recovered when returned to isotonic conditions. c, Exponentially growing fibroblasts were incubated in DMEM containing 0.1% FBS for 24 h, and trichostatin was then added for 8 h. ATM was immunoprecipitated and levels of phosphorylated ATM, total ATM and H2AX were determined by immunoblotting. d, Schematic model of ATM activation after irradiation. DNA strand breaks lead to an alteration of chromatin structures that induce intermolecular autophosphorylations of an ATM dimer on serine 1981, and dissociation of the previously inert dimer. Active ATM monomers are then free to migrate to and phosphorylate substrates such as Nbs1 and p53.
To estimate the fraction of cellular ATM protein that becomes phosphorylated after DNA damage, sequential immunoprecipitations of ATM from irradiated primary fibroblasts were carried out with a conventional anti-ATM antibody and with the anti-1981S-P antisera. In the absence of insult, the conventional anti-ATM antibody was able to immunoprecipitate virtually all of the ATM in the first absorption from unirradiated cells, whereas the anti-1981S-P antisera brought down almost no ATM (Fig. 5e, top panel). After exposure to 0.5 Gy IR, the amount of ATM immunoprecipitated by the anti-1981S-P antisera in the first absorption (immunoprecipitation 1, sample 4) was similar to the amount of ATM immunoprecipitated by the conventional anti-ATM antibody (immunoprecipitation 1, sample 3), and was greater than the remaining cellular ATM that was brought down in the second absorption by the conventional anti-ATM antibody (immunoprecipitation 2, sample 4). The results show that well over 50% of the total ATM in an exponentially growing culture of primary human fibroblasts is autophosphorylated on 1981S by 15 min after exposure to 0.5 Gy IR (estimated 18 DNA breaks).
Given that such a high fraction of cellular ATM becomes phosphorylated so rapidly in the presence of so few DNA strand breaks, it seems highly unlikely that the ATM oligomers could require direct binding to DNA strand breaks for activation and autophosphorylation. These observations suggest that the introduction of DNA strand breaks must cause a change in the nucleus that can activate ATM at a distance from the break itself. As DNA strand breaks introduced by exposure to IR rapidly alter topological constraints on DNA29, 30, 31, alterations in some aspect of chromatin structure could meet the criteria of a rapid change that can occur at a distance in the nucleus. Chromatin and chromosome structures can be altered in the absence of DNA breaks by hypotonic conditions32, 33, by exposure to chloroquine34, 35, 36 or by treatment with histone deacetylase inhibitors37, 38, 39. Exposure of cells to mildly hypotonic buffers or either of these two chromatin-modifying drugs induced rapid and diffuse phosphorylation of ATM protein, as assessed by immunoblot and immunofluorescence (Fig. 6 and Supplementary Fig. 1).
Because no phosphorylation of histone H2AX or foci of either phosphorylated ATM or H2AX were observed with any of these treatments, there is no evidence that any of them caused DNA strand breakage40. Interestingly, these chromatin-modifying treatments induce phosphorylation of p53 (Supplementary Fig. 2), whose phosphorylation by ATM does not occur at the site of DNA breaks. By contrast, phosphorylation of both ATM and H2AX were apparent after irradiation, whether it was applied alone or in combination with the other treatments. Furthermore, all three of these agents were able to enhance the amount of ATM phosphorylation seen after the exposure of cells to submaximal (0.2 Gy) doses of IR (Fig. 6b,c). The patterns of immunofluorescent staining of phosphorylated ATM were also informative. At the earliest time points after IR, the staining was diffuse across the nucleus, but after several minutes, some foci were seen in addition to the diffuse nuclear staining. The diffuse staining, but not the foci, was seen after treatment with the other agents. These patterns are consistent with a diffuse activation of ATM and migration of a fraction of ATM protein to the sites of DNA strand breaks introduced by IR, presumably to phosphorylate substrates at the breaks. Diffuse immunofluorescence, but no ATM foci, was seen when staining the cells with the anti-1981S antibody (Fig. 6a, bottom panel).
Discussion
Characterization of an IR-induced phosphorylation event in ATM and generation of a phosphospecific antibody led to mechanistic insights into the means of signal transduction by the ATM protein kinase; our results point to a new mechanism for regulating the activity of a cellular kinase. We propose that ATM is sequestered in unperturbed cells as a dimer or a higher-order multimer with its kinase domain bound to an internal domain of a neighbouring ATM molecule containing serine 1981 (Fig. 6d). This interdomain interaction must occur for ATM to fold correctly and to remain stable in the cell; while contained in this complex, ATM is unable to phosphorylate other cellular substrates. After DNA damage, the kinase domain of one ATM molecule phosphorylates serine 1981 of an interacting ATM molecule, and the phosphorylated ATM is then dissociated from the complex and is freed to phosphorylate other substrates in the cell. The kinase-inactive and non-phosphorylatable mutants of ATM retain endogenous ATM in a complex because they cannot be phosphorylated and released after IR, thus inhibiting cellular ATM activity. This mechanism provides an explanation for the dominant-inhibitory property of kinase-inactive ATM, a property that has been used to advantage by many laboratories but has never been understood.
Several different molecular mechanisms have been identified that regulate the activity of protein kinases. Protein kinases are generally restrained in an inactive state with the acquisition of catalytic activity controlled at multiple levels, ranging from the binding of allosteric factors to changes in the subcellular localization of the enzyme41. Because all protein kinases catalyse the same reaction, their active conformations tend to be structurally similar. However, different classes of kinase have evolved distinct inactive states, and adoption of the catalytic conformation of the enzyme can be impeded in different ways. These include steric hindrance of substrate access to the catalytic domain by an activation loop that is often controlled by phosphorylation42; allosteric regulation of the activation loop via the 
C helix, as exemplified by the PSTAIRE helix in the cyclin-dependent kinase family43; pseudosubstrate inhibition of both substrate and nucleotide binding, as seen in twitchin44; and intramolecular autoinhibition by N-terminal segments that inhibit catalytic activity, as in the case of the EphB2 receptor kinase45. Our observations suggest a new mechanism of kinase activation in which the cellular activity of one ATM kinase molecule is impeded by intermolecular association with an internal domain of a partner ATM molecule. In this model, access of substrates to the catalytic domain is impeded by this association. In some ways, this regulatory model is similar to pseudosubstrate inhibition, with the main variations being that the pseudosubstrate is a domain of itself (albeit in trans) and that this partner is not a mimic, but actually becomes a substrate to release the inhibition.
The rapidity, sensitivity and stoichiometry of serine 1981 phosphorylation have ramifications for the mechanisms that must be involved in the initiation of ATM activation. The activation of ATM, as assessed by phosphorylation of this site, is detectable as quickly as cells can be collected following an insult, and is already maximal within 5 min after exposure to an IR dose as low as 0.5 Gy. In addition, phosphorylation of this site is detectable after the introduction of only a few strand breaks in the genome of a human cell, whether the DNA breaks are introduced by low-dose IR or by a restriction enzyme (Fig. 5). Furthermore, activation by the restriction enzyme shows that breaks in the DNA phosphodiester backbone, and not some other cellular effects of IR, are capable of activating ATM. Perhaps the most impressive quantification is that well over 50% of the ATM molecules in the cell become phosphorylated within about 5 min after exposure to 0.5 Gy IR, a dose that would be expected to induce only about 18 double-strand breaks in the genome of a mammalian cell. Although the addition of DNA with free double-stranded ends has been reported to augment the activity of biochemically purified ATM kinase46, this enhancement has not been seen when immunoprecipitated endogenous ATM is used in in vitro kinase assays22. The relatively high number of ATM molecules that become rapidly phosphorylated and activated, compared with the small number of DNA strand breaks that is required to induce the activation, does not seem consistent with the possibility that ATM has to bind directly to DNA double-strand breaks to become activated; rather, it suggests that the introduction of DNA breaks must somehow signal to ATM molecules at a distance in the cell.
It is conceivable that a protein complex that binds directly to DNA breaks in mammalian cells initiates a signalling pathway that causes ATM activation. However, such a model does not easily explain the impressive rapidity and the magnitude of the ATM phosphorylation reaction that is seen in mammalian cells. An attractive alternative that is consistent with the data presented here is that the introduction of DNA double-strand breaks causes a rapid change in some aspect of higher-order chromatin structure, and that this chromatin alteration initiates ATM activation. DNA in the eukaryotic nucleus is tightly packaged with several levels of organization. DNA is wrapped around nucleosomes that are arranged into ordered arrays of solenoid-like structures. In turn, these are packaged into supercoiled loops, each containing approximately 104–105 base pairs of DNA47, 48. IR-induced DNA strand breakage removes topological constraints on DNA loops29, 30, 31. The rapidity of ATM activation and the data showing that it can occur at a distance from the DNA strand breaks are consistent with a model in which IR-induced changes in chromatin structure serve as an initiating signal.
The activation of ATM by hypotonic swelling or by treatment with chloroquine or trichostatin, in the absence of detectable DNA strand breaks (Fig. 6), is consistent with this model. Furthermore, it seems that when changes in chromatin structure activate ATM in the absence of DNA strand breaks, ATM substrates that would be phosphorylated at the site of breaks (for example, H2AX) fail to become phosphorylated, whereas substrates present elsewhere in the nucleus (for example, p53) can still be phosphorylated (Fig. 6 and Supplementary Fig. 2). It is tempting to speculate that the amount of ATM phosphorylation becomes maximal at the seemingly low dose of 0.5 Gy (about 18 strand breaks on average per cell), because the number of breaks at this dose can relax the number of higher-order loops that can signal to all of the ATM contained in a cell. Because the activation of ATM is required for most, if not all, appropriate cellular responses to irradiation, these concepts suggest that future investigations of responses to DNA damage should not focus simply on events occurring at the DNA strand break, but rather should include a consideration of more complex nuclear events.
Methods
Cell culture and immunofluorescence
293T cells, HeLa cells and 1070SK primary human foreskin fibroblasts (HFFs) (less than passage 20, American Type Culture Collection) were cultured in DMEM supplemented with 10% FBS. GM00637 and GM09607 fibroblasts were grown in DMEM containing 15% FBS. GM00536 lymphoblast cells were grown in RPMI supplemented with 10% fetal calf serum. GM00637 stably transfected with the I-Sce1 restriction enzyme site were obtained from S. Powell (Massachusetts General Hospital). Southern blotting analysis demonstrated two integration sites for the I-Sce1 plasmid (data not shown). 293T cells were transfected using Fugene (Roche), and HeLa cells, GM00637 and GM09607 with Lipofectamine (Invitrogen). Metabolic labelling was carried out by pre-equilibrating cells in phosphate-free medium for 3 h before the addition of 0.5 mCi ml-1 32P-orthophosphate (NEN) for 30 min. Inhibition of DNA synthesis and analysis of the G2/M checkpoint after irradiation were assessed as described previously20. Proteins were crosslinked by incubating cells in formaldehyde/PBS for 10 min at room temperature. Formaldehyde was washed out using PBS containing 100 mM glycine before immunoprecipitation. Hypotonic swelling was carried out for 1 h in PBS containing 0.45% glucose (w/v) and 1% FBS, with the NaCl concentration reduced to either 50 mM or 100 mM. HFFs were incubated with chloroquine in DMEM containing 10% FBS for 4 h. Before an 8 h incubation with trichostatin, HFFs were grown for 24 h in DMEM containing 0.1% FBS. For immunofluorescence experiments, HFFs grown on glass slides were fixed in 50% methanol/50% acetone for 2 h at -20 °C. Cells were incubated with primary antibodies at 1:1,000 ( H2AX from Upstate Biotechnology) and secondary antibodies at 1:500 ( Cy3 anti-mouse and fluorescein isothiocyanate anti-rabbit from Jackson Immunochemicals) in PBS, 10% FBS for 1 h.
Plasmids and recombinant protein purification
Wild-type or mutant Flag-ATM have been described previously9. Wild-type Flag-ATM was mutated using the QuikChange site-directed mutagenesis kit (Stratagene). The I-Sce1 expression plasmid has been described previously49. GST fusion proteins were expressed in BL21 from pGEX 4T1 (Pharmacia). Fusion proteins were purified and GST pull-down experiments carried out by binding to glutathione-sepharose beads (Sigma) in PBS, 0.5% NP-40, and 1 mM 4-(2-aminoethyl)-benzenesulphonyl fluoride hydrochloride (AEBSF) and 1 mM dithiothreitol. Bound proteins were washed five times in the same buffer and eluted with 20 mM glutathione in 50 mM Tris-HCl (pH 8.0). The ATM kinase domain was expressed from pET28 (Novagen). Recombinant ATM with N-terminal 6 His and Flag tags was expressed from pYES2 (Invitrogen) in JEL1, a protease-deficient S. cerevisiae strain that overexpresses the transcription factor GAL4 driven by the GAL1 promoter50. After induction in 2% galactose for 16 h, yeast cells were lysed in 50 mM sodium phosphate (pH 8.0), 300 mM NaCl, 0.4 M aprotinin, 1 mM AEBSF, 1 soy trypsin inhibitor (Roche), 1.5 mM pepstatin and 42 M leupeptin by three passages through a French press. Lysates were cleared at 35K in a 45Ti rotor (Beckman). For anti-Flag M2 sepharose (Sigma) affinity purification, Tween 20 and NP-40 were added to 1% and 0.5%, respectively. Flag-ATM was eluted in 100 g ml-1 Flag peptide (Sigma) in 'buffer A': 50 mM Tris (pH 7.5), 150 mM NaCl, 1% Tween 20, 0.5% NP-40, 50 mM NaF, 1 mM AEBSF and 1 protease inhibitor mixture from Roche. Size-exclusion chromatography was carried out using a Superdex 200 HR 10/30 column (Pharmacia) in 50 mM sodium phosphate (pH 8.0) containing 150 mM NaCl.
Antibodies and ATM assays
Mammalian cell extracts were prepared in buffer A. Cleared supernatants were immunoprecipitated with anti-Flag M2 sepharose or anti-ATM D1611 (ref. 51), and protein A/G-agarose. Beads were washed twice with buffer A and twice with RIPA buffer. Co-immunoprecipitation was performed in 'buffer B': 50 mM Tris (pH 7.5), 100 M NaCl, 0.5% Tween 20, 0.2% NP-40, 1 mM AEBSF and 1 protease inhibitor mixture (Roche). For in vitro kinase assays, beads were washed twice with buffer A, twice with buffer A containing 0.5 M LiCl, and twice with kinase buffer: 20 mM HEPES (pH 7.5), 50 mM NaCl, 10 mM MgCl2 and 10 mM MnCl2. ATM kinase reactions were carried out at 30 °C for 5 min in 50 l of kinase buffer containing 10 Ci of [-32P]ATP and 1 g of GST fusion substrate. Tryptic digestion of Flag ATM immobilized on nitrocellulose, two-dimensional resolution (electrophoresis and chromatography), manual Edman degradation and V8 digestion of peptides isolated from thin-layer chromatography plates were performed as described previously24. Western blotting was performed with anti-Flag M5 (Sigma), anti-ATM MAT3 (a gift of Y. Shiloh), anti-GST (Pharmacia) or anti-6 His (Sigma). Anti-1981S and anti-1981S-P specific antibodies were generated by immunizing rabbits with the KLH-conjugated synthetic peptides SLAFEEGSQSTTISS (three animals) and SLAFEEGSpQSTTISS (six animals) (Rockland Immunochemicals), respectively.
