NMR– and MD simulation–based structural characterization of the membrane-associating FATC domain of ataxia telangiectasia mutated

The Ser/Thr protein kinase ataxia telangiectasia mutated (ATM) plays an important role in the DNA damage response, signaling in response to redox signals, the control of metabolic processes, and mitochondrial homeostasis. ATM localizes to the nucleus and at the plasma membrane, mitochondria, peroxisomes, and other cytoplasmic vesicular structures. It has been shown that the C-terminal FATC domain of human ATM (hATMfatc) can interact with a range of membrane mimetics and may thereby act as a membrane-anchoring unit. Here, NMR structural and 15N relaxation data, NMR data using spin-labeled micelles, and MD simulations of micelle-associated hATMfatc revealed that it binds the micelle by a dynamic assembly of three helices with many residues of hATMfatc located in the headgroup region. We observed that none of the three helices penetrates the micelle deeply or makes significant tertiary contacts to the other helices. NMR-monitored interaction experiments with hATMfatc variants in which two conserved aromatic residues (Phe3049 and Trp3052) were either individually or both replaced by alanine disclosed that the double substitution does not abrogate the interaction with micelles and bicelles at the high concentrations at which these aggregates are typically used, but impairs interactions with small unilamellar vesicles, usually used at much lower lipid concentrations and considered a better mimetic for natural membranes. We conclude that the observed dynamic structure of micelle-associated hATMfatc may enable it to interact with differently composed membranes or membrane-associated interaction partners and thereby regulate ATM's kinase activity. Moreover, the FATC domain of ATM may function as a membrane-anchoring unit for other biomolecules.

tant for the DNA damage response, cell cycle checkpoints, and metabolic processes, it is considered an interesting therapeutic target with respect to cancer (18).
It has been proposed that targeted membrane localization allows spatial separation of individual signaling branches of large signaling networks, thereby potentially improving the reliability of biochemical signaling processes (19). Because PIKKs generally participate in a multitude of signaling pathways in response to ionizing radiation and other stress factors or metabolic signals (2,5,7,16,20,21), their localization may also determine the specific signaling output (22). ATM has been found to localize and function not only in the nucleus but also in the cytoplasm (12). ATM has further been localized at microsomes (23) and cytoplasmic vesicles (24), as well as at membrane-associated vesicles (25). More recently, it has been reported that PEX5 peroxisome import receptor binds ATM to localize it to peroxisomes and that ATM can localize to mitochondria (14,15).
Although the total length of PIKK amino acid sequences ranges from about 2500 to 4500 residues, all share a similar domain organization (5,22). Fig. 1A shows human ATM as a representative example. The Ser/Thr kinase domain is close to the C terminus and shows homology to lipid kinases (4). Except for TRRAP, all are catalytically active (1,4). The FRAP-ATM-TRRAP (FAT) domain resides N-terminal of the kinase (22,26). Based on sequence analysis, the FAT and the preceding N-terminal region with only low sequence similarity between different PIKKs are mostly composed of ␣-helical repeat motifs that typically form platforms for protein-protein interactions (26 -28). The linker region between the kinase and the FAT C-terminal (FATC) domain region varies significantly in length and sequence composition (4,29). It has been referred to as the PIKK regulatory domain (PRD) (1,4,29). The highly evolutionary conserved region of the FATC domain is made up by the ϳ35 C-terminal residues (PFAM domain database entry PF02660) (4, 26, 30 -32). Based on mutagenesis studies and other data, this domain generally plays an important role for the regulation of PIKK function (4,29,31,(33)(34)(35).
Obtaining high-resolution structural data for PIKKs is challenging due to their large size. For ATM, EM structural data at lower resolution provided insights into the interaction with DNA (36), the inhibitory effect of dimerization of human ATM (37), and the dimerization properties of the yeast ATM homolog Tel1 (38,39). In 2017, higher-resolution EM structures of the closed ATM dimer (PDB code 5NP0, 5.7 Å resolution) and the presumably more active open ATM dimer (PDB code 5NP1, 5.7 Å resolution for combined data) were published (40).
The FATC domains of all PIKKs are rather hydrophobic and rich in aromatic residues (ATM and TOR; see Fig. 1B). In line with this, all interact with membrane mimetics, although with somewhat different preferences for membrane features such as surface charge, curvature, or packing density, and thus may all act as membrane-anchoring units (32). Like TOR, ATM binds to all tested membrane mimetics, including dodecylphosphocholine (DPC) micelles, bicelles composed of dimyristoyl and diheptanoylphosphocholine (DMPC, DihepPC), and small unilamellar vesicles (SUVs) made of DMPC (32). An estimate of the secondary structure content based on secondary chemical shifts suggested that the micelle-immersed human ATM FATC domain contains three helical stretches (32). Here, we present the determined NMR structure and an analysis of the relative orientation on the micelle and its membrane immersion properties and dynamics based on NMR and molecular dynamics (MD) data. We further prepared mutants by replacing conserved aromatic residues that are expected to be important for the membrane affinity and tested their ability to interact with DPC and DihepPC micelles, DMPC/DihepPC bicelles, and DMPC SUVs. Overall, the data provide a better understanding of the role of the AMT FATC domain in regulating the kinase function as well as the observed localization to specific membrane compartments.

The NMR structure of micelle-immersed hATMfatc consists of three helices with a central flexible linker and a less flexible C-terminal linker
The three-dimensional structure of the human ATM FATC domain (hATMfatc) associated with DPC micelles (PDB code 6HKA) was determined by multidimensional heteronuclear NMR spectroscopy in solution. Because cleavage of the GB1 tag is not efficient and results in very low yields, all structural restraints were derived from NMR data recorded for the GB1tagged form (hATMfatc-gb1ent) (32). Previous studies have shown that the GB1 tag does not affect the association with membrane mimetics and results only in little overlap of NMR signals of the GB1 tag and the FATC domain (41). The structural statistics are given in Table 1.
The structure of micelle-associated hATMfatc, residues Thr 68 -Val 100 in hATMgatc-gb1ent corresponding to Thr 3024 -Val 3056 in full-length human ATM, consists of three helical regions (Fig. 1, B-D). The structure is overall well-defined with 85.5% of the residues in the well-structured region (Val 72 -Ala 83 , Pro 86 -Leu 92 , Pro 94 -Val 100 ) occupying the most favored region of the Ramachandran plot (Table 1). Based on the 20 lowest energy structures, the N-terminal longer ␣-helix (␣1) is on average formed by residues Gly 73 -Ala 83 of the 100-residue fusion protein, which corresponds to residues Gly 3029 -Ala 3039 of full-length human ATM. Following a three-residue linker (Ile 84/3040 -Asp 85/3041 -Pro 86/3042 ), a second ␣-helix (␣2) is on average formed by residues Lys 87/3043 -Arg 91/3047 . Following another proline-containing short linker (Leu 92/3048 -Phe 93/3041 -Pro 94/3049 ), the C-terminal helix follows. Although tertiary contacts between the helices could not be observed, their orientation with respect to each other is not completely random. This may be explained by the presence of a proline in each linker. Based on the analysis of the conformations of short peptides based on residual dipolar couplings, prolines increase local order (42 48) in the short C-terminal helix. A more detailed description of the 15 N relaxation data is provided in the supporting information (text and Figs. S1 and S2).
Consistent with the differences in the amino acid sequence to the FATC domain of yeast TOR1 (y1fatc; Fig. 1B, bottom), the overall structural appearance is quite different. Y1fatc contains two conserved cysteines that can form a disulfide bond, and oxidized y1fatc in the absence of membrane mimetics forms already an ␣-helix that is followed by a C-terminal disulfide bonded loop (43). In contrast, hATMfatc in buffer is largely unstructured, and most backbone amide signals are not visible in the 1 H-15 N HSQC spectra (32), which hampers its resonance assignment.
Compared with micelle-associated hATMfatc, micelle-associated oxidized y1fatc adopts only a single ␣-helix that is followed by a disulfide bonded loop or bulb (Fig. 1C, right) (44). Even in the reduced micelle-associated state of y1fatc, the C terminus folds back, and the whole structure is less dynamic than that of hATMfatc (44). The { 1 H}- 15 N NOE values of oxidized and reduced micelle-associated y1fatc at 298 K are more uniform in the well-structured region and do not show residues with significantly lower values in between (44), as observed for micelle associated hATMfatc (Fig. 1E).
The surface of micelle-associated hATMfatc (Fig. 2, top) is largely hydrophobic except for a positively charged surface region formed by Lys 87/3043 , Arg 91/3047 , and Lys 97/3053 that is in the neighborhood of a small negatively charged surface region formed by Asp 85/3041 . In contrast, that of micelle-immersed y1fatc (Fig. 2, bottom) shows a large acidic region toward the N terminus and only two small positively charged regions around Lys 2448 and Arg 2458 . The C-terminal loop of y1fatc is largely hydrophobic (44).

The helical regions of micelle-associated hATMfatc can move with respect to each other
Additional insights into the dynamics as well as immersion properties (see below) of micelle-associated hATMfatc were obtained from molecular dynamics simulations. In total, three runs of MD simulations were performed using as starting structure for each one of the three lowest-energy NMR structures (Fig. 3A). The simulation time for each run was 1.3 s. The first 300 ns were discarded as equilibration time. As illustrated by the pictures of the simulated micelle-associated hATMfatc of each run (Fig. 3B), the relative orientation of the three helices with respect to each other varies in each due to the dynamic linkages between the helices, especially the one between ␣1 and ␣2. Additional information about the conformational freedom is provided based on RMSD and RMSF (global and local) values in the supporting information (text and Fig. S4 (A-C)). Consistent with the fact that no interhelical NOEs could be observed, distances below 0.5 nm between residues on different helices were only rarely and for very short simulation time periods observed (Fig. S4D). Plots of the angle between ␣1 and ␣2 as well as between ␣2 and ␣3 ( Fig. 3C and Fig. S5A) as a function of time show that these angles can vary significantly as a function of the simulation time. Even with a simulation time of 1.3 ms, such simulations usually do not converge. In summary, the 15 N relaxation data and the MD simulations suggest that the reorientation of the helices with respect to each other due to the dynamic central linkage between ␣1 and ␣2 and less dynamic linkage between ␣2 and ␣3 is the major contribution to their dynamic properties, rather than more local conformational changes.

The hATM FATC domain resides largely in the headgroup region of DPC micelles
Based on the MD simulations of the three lowest-energy NMR structures of micelle-associated hATMfatc in the presence of DPC micelles (Fig. 3), the three helices immerse mostly into the headgroup region. To probe this experimentally, we recorded 1 H-15 N HSQC spectra of hATMfatc-gb1ent in the presence of DPC micelles (50 mM) and increasing amounts of paramagnetic 5-or 16-doxyl stearic acid (5-/16-SASL) ( Fig. 4A and Fig. S5B). Earlier MD simulations of DPC micellar systems in the absence and presence of protein showed that 5-/16-SASL can move in the micelle and that 16-SASL can bend such that the doxyl group is close to the headgroup region and not deep in the interior as commonly assumed (45). The MD data further confirmed that the doxyl group of 5-SASL, as expected (46), resides also near the DPC headgroups (45). Residues of hATMfatc that localize close to the headgroups should thus show the strongest spectral changes. The doxyl label of 5-or 16-SASL results in paramagnetic relaxation enhancement (PRE), which reduces the signal intensity of nearby amide groups with an r Ϫ6 distance dependence (47). To better correlate the PRE effects with the average chemical shift changes (Figs. S6 and S7) as well as with the distance from the micelle center of mass (COM) ( Because the listed residues based on the NMR data should localize near the headgroups, they should have a distance from the micelle COM that is in a similar range as the phosphorus atom of the DPC headgroup. Fig. 4C shows the average distance from the COM of each residue of hATMfatc to the micelle COM for run 1. Those for runs 2 and 3 are displayed in Fig. S8. The profiles as a function of the sequence look very similar for all runs. In line with the 5-and 16-SASL NMR data,  (26). To structurally characterize the C-terminal FATC domain (residues 3024 -3056, Uniprot accession number Q13315), it was fused to the B1 domain of streptococcal protein G (GB1, 56 residues, ϭ hATMfatc). B, superposition of the amino acid sequences of the FATC domains of hATM and yeast TOR1 and schematic representation of the secondary structure content in DPC micelles (schematic representation to the right). Aromatic residues are shown in light (Phe, His, and Tyr) and dark (Trp) green, hydrophilic ones in light blue, except Lys and Arg in blue and Asp and Glu in red, and those containing methyl groups in dark blue. Prolines are colored purple and glycines in orange. C, the NMR structure of hATMfatc in the presence of micelles (150 mM DPC) consists of three helices with a central dynamic linker and a less dynamic C-terminal one ( Table 1, PDB code 6HKA, BMRB accession number 27167 (67)). The structure is overall different from that of the micelle-immersed oxidized yeast TOR FATC domain (y1fatc, PDB code 2KIO) (44). For the ribbon representations, the 20 lowest-energy structures have been superimposed for the indicated regions. D, each helical stretch of micelle-associated hATMfatc is shown in line mode. The backbone of the helical stretches is depicted in red. The color coding of the side chains is the same as in B except that charged side chains are also colored light blue. All structure pictures were made with the program MolMol (86). E, information about the backbone dynamics from a plot of the { 1 H}- 15

Structure of membrane-associating ATM FATC
residues that show strong PRE effects (Fig. 4B) have mostly distances to the micelle COM that are similar to that of the DPC phosphorus atom (dotted line at 1.914 nm). In contrast, residues that show a small PRE effect, such as Thr 68/3024 at the N terminus of hATMfatc, show a larger distance to the micelle if they are more solvent-accessible or a closer one such as for example Leu 92/3046 if they reside deeper in the micelle. From a plot of the solvent-accessible surface area covered by DPC that was derived from the three MD runs (Fig. 4D and Fig. S9), it can further be estimated which residues immerse deeper and which should still be solvent-accessible. As expected, those residues showing a short distance to the micelle COM (Յ1.6 nM in Fig.  3C and Fig. S8) are 80% or more covered by DPC (e.g. for run 1 Val 72 , Val 76 , Leu 79 , Leu 89 , and Phe 93 ). As expected, these residues are mostly hydrophobic.
In summary, the NMR data using spin-labeled micelles and the MD simulations of micelle-associated hATMfatc show that it binds the micelle by a dynamic assembly of three helices with many residues localizing in the headgroup region. None of the three helices penetrates the micelle very deeply or makes significant tertiary contacts to the other helices.

Mutation of Phe 93/3049 and Trp 96/3052 to alanine abrogates the association of hATMfatc with neutral SUVs but not with micelles and bicelles
Tryptophans but also other aromatic residues are known to play an important role for the affinity between membranes and proteins in general (48,49). For the FATC domain of TOR, we have shown previously that replacement of up to 6 -7 residues did not result in a significant abrogation of the association with A, ribbon representations of the three lowest-energy structures that were used as starting structures for three independent MD simulation runs. B, representative structure pictures of the MD simulations of hATMfatc in the presence of DPC micelles. The structures shown are the conformations of hATMfatc, which are the closest to the middle of the most populated cluster of structures for a given run. The DPC micelle is shown as a transparent averaged density iso-surface, and the peptide is shown in a ribbon representation. In both A and B, negatively charged side chains or residues are shown in red, positively charged ones in blue, polar ones in green, and aliphatic ones in gray. For simplicity, only the residue numbering as used in the PDB entry (PDB code 6HKA) is given. The N and C termini are indicated by the letters N and C. C, information about the movement of the three helices with respect to each other from an analysis of the variation of the angle between helices 1 and 2 as well as the one between helices 2 and 3 for run 1 (for runs 2 and 3, see

Structure of membrane-associating ATM FATC
micelles or bicelles. However, replacement of only one tyrosine (Tyr 2463 ) or one tryptophan (Trp 2466 ) resulted in an impairment of the interaction with SUVs (50). To find out whether aromatic residues of hATMfatc at equivalent positions (Phe 93/3049 and Trp 96/3052 ) have a similar effect, we mutated them individually and at the same time to alanine residues. The free states of the mutant proteins show even fewer peaks and thus appear to be more dynamic than WT hATMfatc (Fig. S10, left). This may be explained with the smaller size of alanine compared with tyrosine or tryptophan. In the presence of 50 mM DPC, they adopt as the WT a more ordered structure, which reduces the backbone dynamics such that the backbone and side chain amide signals become clearly visible (Fig. S10, right). The interaction with neutral membrane mimetic micelles composed of DPC or the short chain diacyl lipid DihepPC, neutral bicelles composed of DMPC for the planar region and DihepPC for the rim, and small unilamellar vesicles composed of DMPC was further monitored by recording 1 H-15 N HSQC spectra in the absence and presence of these membrane mimetics (Fig. 5 and Figs. S11 and S12). As indicated by the observed spectral changes, neither of these mutations abolished the interaction with micelles or bicelles ( Table 2). Whereas the single mutants F93A/3049A and W96/3052A could still interact with neutral DMPC SUVs (Figs. S11 and S12, green plus signs), a lack of spectral changes for the double mutant F93/3049A-W96/ 3052A (Fig. 5, bottom right, red minus sign) indicates that it could not ( Table 2).

Discussion
ATM is not only involved in DNA repair; it has also been suggested that ATM plays a role in the oxidative stress response and as a link between genome stability and carbon metabolism (9,10). In line with this, it was recognized that ATM localizes not only in the nucleus but also at cytoplasmic vesicles (24) and more specifically peroxisomes (14) and microsomes (23), the plasma membrane (25), and mitochondria (15). NMR-and CDmonitored interaction studies showed that the ATM FATC domain can interact with all tested membrane mimetics and thus may, besides protein-protein interactions, also mediate direct membrane interactions of ATM (32). The presented NMR structure of the micelle-associated state of the FATC domain of human ATM (Figs. 1, B-D and 2) and the characterization of its dynamic and immersion properties from MD and further NMR data (Figs. 3 and 4) show that membrane association is mediated by three rather flexibly linked helices that immerse largely in the headgroup region and slightly below and that do not significantly interact with each other. This dynamic assembly may enable specific interactions with other membrane-embedded proteins. Recruitment to the plasma membrane involves interactions of the protein CKIP-1 (casein kinase-2 interaction protein-1) that is involved in muscle differentiation and the regulation of the actin cytoskeleton with the ATM C-terminal region, including the catalytic and the FATC domain (51). In addition, interactions with the PEX5 peroxisome import receptor localize ATM to peroxisomes (14).
The presence of a C-terminal positively charged surface patch next to a smaller negatively charged one in the micelleassociated ATM FATC domain (Fig. 2) may not only mediate interactions with the positively and negatively charged headgroup components of neutral phosphatidylcholines or phosphatidylethanolamines but additionally enable interactions with negatively charged phosphatidylserine in the PM or phosphorylated phosphoinositol lipids at mitochondria (52). The folding-upon-binding to a membrane mimetic, which has, for example, also been detected for the 30-residue-long glucagonlike peptide 1 (GLP-1) (53), and the resulting dynamic helix assembly of membrane-associated hATMfatc may further be helpful to interact with specific receptors or membranes with different packing densities, which depends, for example, on the amount of cholesterol or the ratio of saturated versus unsaturated fatty acids in the lipids (52). Based on the membrane catalysis hypothesis, initial interactions of signaling peptides with the bilayer increase their local concentration and allow them to adopt a structure that is recognized by the target membraneresident receptors (53). The same may be true for specific protein domains or segments.
Based on mutagenesis studies, the FATC domains of all PIKKs play an important role for their regulation (26,30,31,(33)(34)(35)54) and share the ability to interact with different membrane mimetics, albeit with varying preferences for specific membrane properties (32,44). The observed evolutionarily conserved sequence differences (e.g. Fig. 1B, ATM and TOR) may result in different membrane-associated conformations that determine the specific membrane-targeting behavior (32). In line with this, micelle-associated hATMfatc adopts a rather dynamic assembly of three helices with a hydrophobic N-terminal region and a more polar C-terminal one, whereas the structure of the yeast TOR1 FATC domain consists of a single helix followed by a loop that in the oxidized state contains a disulfide bond and where the N-terminal region is acidic and the C-terminal one largely hydrophobic (Figs. 1C and 2). Consistent with this, the FATC domain of ATM cannot replace the one of ATR (ataxia-and Rad3-related) (29) or TOR (33).
The FATC domains of all PIKKs share conserved sequence features (e.g. Fig. 1B, ATM and TOR) that may be important for the regulatory function and especially for directly mediating membrane interactions (32). The conserved glycine near the C terminus of TOR (Gly 2465 of y1fatc (Fig. 1B), Gly 2544 in human

Structure of membrane-associating ATM FATC
TOR) appears to facilitate the back bending of the C-terminal region to the ␣-helix (Fig. 1C, right). Except for TRRAP, this glycine is also conserved in all other PIKKs (32) and, in the case of ATM, participates in the flexible linkage between ␣2 and ␣3 (Fig. 1C, center). The evolutionarily also completely conserved proline 3042 (ϭ Pro 86 ), aspartate 3041 (ϭ Asp 85 ), and alanine 3039 (ϭ Ala 83 ) of hATM form together with Ile 3040 (ϭ Ile 84 ) in humans but a methionine in other organisms, the flexible linker between ␣1 and ␣2. The proline and the aspartate are also found in the DNA-PKcs (DNA-dependent protein kinase catalytic subunit) amino acid sequences from most organisms (human Asp 4113 and Pro 4114 ) and, based on 13 C ␣ secondary shifts, also disrupt two helical stretches (32). Interestingly, alanine 3039 of hATMfatc is also in most other PIKKs conserved, . S11C and S12B, respectively. A green plus sign below a schematic representation of the used membrane mimetic indicates strong spectral changes and thus interactions, and a red plus sign indicates no significant changes and thus no significant interactions. To better identify the signals of the ATM FATC part, the spectrum of the GB1 tag followed by a thrombin and factor Xa site (ϭ GB1xa) is additionally shown in green at the top. Accordingly, peaks with a green peak at the top belong to the GB1 tag, which does not interact with membrane mimetics (41). In the spectrum in the presence of DPC, hATMfatc signals (red) are labeled by the one-letter amino acid code and the sequence position based on similarity to the respective WT spectrum. Small letters indicate the side chain signals. A question mark behind the label indicates tentative assignments. Note that some red peaks that appear at new positions could not be assigned in this way and thus have no label. Superpositions of the 1 H-15 N HSQC spectrum of the WT protein with that of each mutant are given in Fig. S10, and superpositions in the presence of stepwise increasing concentrations of DPC for F93/3049A-W96/3052A are given in Fig. S11A, for F93/3049A in Fig. S11B, and for W96/3052A in Fig. S12A.

Table 2 Summary of the NMR-monitored interaction analysis of mutant hATMfatc-gb1ent proteins with different neutral membrane mimetics
ϩ, strong spectral changes indicating an interaction with the respective membrane mimetic; Ϫ, no significant spectral changes/interaction.

Structure of membrane-associating ATM FATC
including TOR, in which the helix of the micelle-associated oxidized and reduced states is distorted around this residue (44). All PIKK FATC domains contain at least one or more tryptophans and/or one or more tyrosine or phenylalanine and several aliphatic hydrophobic residues (32). Tryptophans are often found at the interface between the apolar interior and the polar aqueous environment (48,49), and many hydrophobic residues show a positive free energy contribution for the transfer of a model peptide from a lipid bilayer to water (55). Replacement of either Tyr 2463 or Trp 2466 (Tyr 2542 and Trp 2545 in human TOR) is already sufficient to abrogate the interaction of y1fatc with DMPC SUVs (50). For hATMfatc, Phe 3049 (ϭ Phe 93 ) and Trp 3052 (ϭ Trp 96 ) at equivalent positions have to be replaced simultaneously to achieve the same effect ( Fig. 5 and Figs. S11 and S12). However, the respective mutations did not abrogate the interaction with micelles and bicelles ( Fig. 5 and Figs. S10 -S12). The different association behavior of FATC mutants with micelles and bicelles compared with SUVs can be explained 2-fold. First, in the case of micelles or bicelles, the total concentration of the used detergents or lipids and thus the available membrane mimetic surface area is much higher than for SUVs. The used concentration of DPC and DihepPC was 50 mM, and the total lipid concentration in the DMPC/DihepPC bicelle samples was ϳ240 mM. However, due to the preparation method, the total lipid concentration in the SUV samples is significantly below 50 mM because pure DMPC cannot easily be resuspended at high concentrations, and in addition, some of the lipid is lost when separating the small from larger uni-and multilamellar vesicles by centrifugation. Whereas one-dimensional 1 H NMR spectra can be used to compare the lipid content in samples with micelles and bicelles, this is due to the large size of SUVs and corresponding line broadening not possible. However, besides established methods such as dynamic light scattering, the CD signal can be used to compare different liposome preparations (50,56). A second reason for the different binding behavior of the FATC domain to micelles and bicelles versus SUVs may further be a higher affinity for curved membrane surfaces. DPC micelles are rather small, spherical particles that contain ϳ50 -60 DPC molecules, resulting in a molecular mass of ϳ19 kDa and a radius of gyration of about 17-18 Å (57, 58). Bicelles have a rather planar bilayer region that is formed by a long-chain phospholipid, such as DMPC. However, the rim that is formed by a short-chain lipid such as DihepPC shows as micelles also a high curvature. The molecular mass of bicelles is usually Ͼϳ250 kDa, and the diameter of the planar region ranges from 21 to 77 Å for q values between 0.2 and 1 (59,60). Even liposomes of the SUV type are much larger than bicelles. Although they are as micelles approximately spherical, they are less curved because the radius is much larger (around 20 -100 nm) (61). We already suggested earlier that for proteins with a rather high and broad affinity for different membrane mimetics, SUVs at the usually used lower concentrations are better suited to detect mutants that may also abrogate membrane association in localization studies in cells (50).
The characterization of the membrane association of ATM by NMR and MD simulations was done looking only at the FATC domain (Figs. 1-4). This raises the question of how accessible it is for membrane interactions in the context of the full-length protein. Based on recent cryo-EM data, ATM was suggested to be in an equilibrium between a closed, inactive dimer (resolution ϳ4 -6 Å) with restricted access to the substrate-binding site and an open, active dimer (resolution ϳ11.5 Å; if combined data used, 5.7 Å) with good access to the substrate-binding side (40). Whereas the FATC domain appears not well-accessible in the closed, inactive ATM dimer, it appears well-accessible for interactions with membranes and regulatory proteins in the open, active state. The conformation of the FATC domain in the provided cryo-EM structures is similar to the micelle-associated conformation presented here and also consists to two ␣-helices and a C-terminal helix or turnlike structure (40). Early studies suggested that DNA damage induces Ser 1981 autophosphorylation and that this results in active, monomeric ATM (62), in which the FATC domain is expected to be even more accessible than in the open dimer. Moreover, studies of oxidative activation of ATM indicated that oxidation of Cys 2991 , which is about 60 -80 residues N-terminal of the FATC domain, results in dimerization (63). Thus, future membrane interaction studies may analyze the effect of reactive oxygen species or oxidized lipids on Cys 2991 . Future localization studies of ATM in cells have to clarify how it varies in response to specific signals and which interactions with specific, especially membrane-localized, proteins such as the PEX5 peroxisome import receptor (14), as well as membrane regions and specific signaling lipids it involves. Moreover, the oligomerization state under different signaling conditions and the resulting cellular localizations have to be determined. Finally, acetylation of lysine 3016 (54) N-terminal of the highly conserved FATC region of ATM may affect the described localization at the plasma membrane, mitochondria, peroxisomes, microsomes, and other cytosolic vesicles (12,14,15,(23)(24)(25). As suggested for the FATC domain of TOR, the one of ATM may also be fused to other proteins or peptides or other substances to tether them to membrane mimetics (64).

Plasmid cloning, protein expression, and purification
The coding sequence for the FATC domain of human ATM (residues 3024 -3056, UniProt accession number 13315) was cloned into GEV2 (65). The resulting fusion protein consisting of GB1 (residues 1-56), a linker region composed of a thrombin (residues 57-62) and an enterokinase (residues 63-67) protease recognition site, and the C-terminal 33 residues of human ATM (residues 68 -100) (ϭ hATMfatc-gb1ent) was overexpressed in Escherichia coli BL21 at 37°C. Because proteolytic removal of the GB1 tag is inefficient and the GB1 tag does not disturb the interaction with membrane mimetics and its monitoring, the fusion protein was used for the described NMR studies (32,41). Uniformly 15 N-or 15 N-13 C-labeled proteins were prepared in M9 minimal medium containing 15 NH 4 Cl and/or [ 13 C]glucose as the sole nitrogen and carbon sources. The expression and the purification by a heating step (66) and IgG affinity chromatography were done as described previously (32,67).
Membrane mimetic DPC micelles were prepared by placing a defined amount of DPC from a concentrated stock solution in chloroform (0.5 M) in a glass vial and drying under a stream of nitrogen gas. The resulting DPC film was dissolved in NMR buffer containing the protein of interest. For NMR titration experiments, samples containing Ͼ30 mM DPC were prepared in the same way, whereas those with DPC concentrations Ͻ30 mM were prepared by diluting a 100 mM DPC stock in buffer with a protein stock and buffer. Note that a separate sample was prepared for each DPC concentration to avoid dilution effects and thus to keep the protein concentration constant. DihepPC micelles were prepared by diluting a 0.25 M stock solution in NMR buffer to 50 mM with the respective protein solution and buffer. Micelles form above the critical micelle concentration, which is 1.1 mM for DPC (68) and 1.4 -1.8 mM for DihepPC (69,70).
DMPC/DihepPC bicelles (q ϭ 0.2, [DMPC] ϭ 0.04 M, and [DihepPC] ϭ 0.20 M, cL 15%) were prepared as follows. An appropriate amount of a 0.25 M stock of the long-chain phospholipid (DMPC) in chloroform was placed in a glass vial, and the chloroform was blown away under a stream of nitrogen gas. To form bicelles, the right amount of the short-chain lipid (DihepPC) in NMR buffer was added, and the vial was vortexed until a clear solution was obtained. Last, the protein solution was added to the concentrated bicelles solution.
Liposomes were prepared by drying an appropriate amount of DMPC in chloroform (0.25 M) under a stream of nitrogen gas. The resulting pellet was dissolved in NMR buffer to obtain a 100 mM solution. To dissolve the pellet, it was exposed to seven cycles of freezing in liquid nitrogen, thawing by incubation in a water bath at 40°C, and thorough vortexing. To induce the formation of SUVs from large uni-and multilamellar vesicles, the DMPC suspension was incubated in an ultrasonication bath for 30 min. A clear supernatant containing only SUVs was obtained after centrifuging the lipid mixture at about 15,000 rpm in a table top centrifuge for 5 min. The resulting white precipitate at the bottom corresponds to large uni-and multilamellar vesicles. For the preparation of a protein sample in the presence of liposomes, only the clear supernatant containing small unilamellar vesicles was diluted 1:1 with protein solution. Thus, using the 100 mM stock solution, the final sample contained Ͻ50 mM DMPC.
The samples of hATMfatc-gb1ent mutants contained about 0.15 mM 15 N-labeled protein in NMR buffer. For the double mutant (F93A/W96A ϭ F3049A/W3052A), the pH was 6.5, as for the WT. The single mutants (F93A ϭ F3049A, W96A ϭ W3052A) were measured at pH 7.4 to improve their solubility and stability.
For NMR experiments with spin-labeled DPC micelles, samples containing about 0.1 mM protein and 50 mM undeuterated DPC were prepared. The concentration of 5-/16-SASL was increased by the stepwise addition of a small amount (1-4 mM) from a 0.25 M stock solution in chloroform.

NMR spectroscopy
NMR spectra were recorded on Bruker Avance 500-, 600-, and 900-MHz spectrometers at a temperature of 298 K. The 500-and 900-MHz spectrometer were equipped with cryogenic probes. Data were processed with NMRPipe (71) and analyzed using NMRView (72). Assignments for the 13 C, 15 N, and 1 H nuclei of micelle-immersed hATMfatc fusion protein were based on 2D 1 H-15 N and various 1 H-13 C HSQC spectra, partially without decoupling, 3D HNCA, HNCACB, CCONH-TOCSY, HCCH-TOCSY, HNHA, HNHB, and 15 N-and 13 Cedited NOESY spectra as described (67). The NOESY mixing times were 90 ms for the 15 N-edited NOESY and 100 ms for the aromatic and aliphatic 13 C-edited NOESY experiments. Information about backbone dynamics was derived from the analysis of 15 N relaxation data, including T 1 (spin-lattice longitudinal relaxation rate), T 2 (spin-spin or transverse relaxation rate), and { 1 H}-15 N NOE.
The immersion depth properties of hATMfatc in DPC micelles were derived by recording 1 H-15 N HSQC and 1D 1 H NMR spectra of samples containing increasing amounts of paramagnetic 5-or 16-SASL. The resulting PRE effects were derived based on the observed changes in signal intensity. Because the spectral changes with 5-SASL were overall smaller compared with 16-SASL, 1-PRE values for 2 mM were analyzed for 5-SASL, and values for 1 mM were analyzed for 16-SASL. In addition, the average chemical shift changes of the backbone amide nitrogen and proton (⌬␦(N,H) av ) were calculated based Structure of membrane-associating ATM FATC on the formula, ((⌬␦ HN ) 2 ϩ (⌬␦ N /5) 2 ) 1 ⁄ 2 . The interaction of mutant proteins with membrane mimetics was monitored by recording 1D 1 H and 2D 1 H-15 N HSQC spectra.

Structure calculation
All structure calculations were performed with XPLOR-NIH (73) using molecular dynamics in torsion angle and Cartesian coordinate space and the standard force field with the parameter files topallhdg_new.pro and parallhdg_new.pro. Distance restraints were generated in NMRView (72) and classified according to NOE cross-peak intensities. Upper bounds were 2.8, 3.5, 4.5, and 5.5 Å. The lower bound was always 1.8 Å. For all NOE restraints, r Ϫ6 sum averaging was used. Backbone dihedral angle restraints for and as well as hydrogen bond restraints for helical regions were derived based on the determined 13 C ␣ chemical shifts, observed helix typical NOE restraints, and initial structure calculations.

MD simulations
Molecular dynamics simulations were performed with Gromacs 2016.3 software (74 -76). The FATC domain was initially placed randomly near the preequilibrated micelle (50 DPC molecules) and solvated with ϳ54,000 water molecules (box size ϳ11 nm). Na ϩ and Cl Ϫ ions were added to the system to reach 0.1 M salt concentration. The CHARMM36m force field was used for the protein and DPC (77). A 2-fs time step was used. All bonds were constrained with the LINCS algorithm (78), and water bond lengths and angles were kept constant using the SETTLE algorithm (79). Initial velocities were taken from the Maxwell distribution for 303.15 K. A constant temperature of 303.15 K was maintained with the V-rescale thermostat with a 0.1-ps coupling constant (80). The DPC micelle with the protein and the water molecules with the ions were coupled to two separate thermostats with the same parameters. Constant pressure of 1 bar was maintained by the isotropic Parrinello-Rahman barostat with a 5.0-ps coupling constant and 4.5 ϫ 10 Ϫ5 bar Ϫ1 compressibility (81). The particle mesh Ewald algorithm was used for long-range contributions to electrostatic interactions (82,83). Lennard-Jones interactions were cut off at 1.2 nm with a force-switch modifier from 1.0 to 1.2 nm.
Three independent simulations were performed using as starting structures the three lowest-energy NMR structures of the micelle-associated FATC domain of human ATM. Each simulation was 1.3 ms long. For the analysis of the secondary structure based on DSSP calculations (84), the clustering, the immersion depth, and the fraction of solvent-accessible surface area covered by the micelle, the first 300 ns of each simulation were discarded as equilibration time.
RMSD and RMSF calculations were performed on the last 500 ns of each trajectory (800 -1300 ns). RMSF analysis with the local fit of the protein structure was performed in addition to the calculations with a whole protein as a reference. The RMSD/RMSF is dominated by the contributions from a change in relative orientation of the helices, and we wanted to highlight local fluctuations in particular regions that are not the result of the change in helix-helix orientation. RMSF graphs that are marked as "local 72-92" or "local 86 -98" (Fig. S4) were calculated with the use of a short part of the protein as a reference for the progressive trajectory fitting and subsequent RMSF calculation (residues 72-92 and 86 -98; 5-ns windows with the following averaging over all 5-ns intervals). Consequently, the RMSF values will be rather small for the region used for the fit.
The analysis of the angles between the helices in Fig. 3C (Fig.  S5) starts also from 800 ns, because the relative orientation of the helices changes slowly compared with the local structure of the protein. Thus, we let the system equilibrate for a longer period of time to have less bias from the initial structure.
The distance between the micelle COM and the whole peptide COM, as well as separate peptide residue COMs, was computed using the simulation period from 300 to 1300 ns. To analyze peptide stability, the secondary structure of each peptide was computed as a function of time with the gmx do_dssp analysis program, which is part of the Gromacs package. The micelle surface for the images of the micelle-associated FATC domain was defined as an iso-surface of averaged DPC density (contour density of 0.250 atoms/Å 3 ). The pictures of the FATC domain alone are representative conformations of the protein as defined by cluster analysis of the protein RMSD performed by the gmx cluster program with the GROMOS clustering algorithm and 0.1 nm cutoff. All three trajectories (excluding the first 300 ns of each run) were combined into a single set of structures, and cluster analysis was performed on this combined trajectory. Subsequently, 10 structures, one from each of the first 10 most populated clusters, were selected as representative structures of the micelle-associated ATM FATC domain. The central structures of each cluster (in terms of RMSD distance) were selected as representative for a given cluster. The combined size of the 10 most populated clusters was equal to 69.3% of the complete set of structures sampled. For the picture of a representative structure of micelle-associated hAT-Mfatc for each run, a clustering analysis for reach run was performed. From the most populated cluster, a structure with an RMSD corresponding to the average for the whole cluster, thus being in the middle, was selected. To estimate whether there are interhelical contacts below about 0.5 nm that could give rise to NOEs, interresidue distances were calculated between the centers of mass for pairs of residues that belong to different helices (excluding adjacent residue pairs; Fig. S4D).