Modeling Cancer Genomic Data in Yeast Reveals Selection Against ATM Function During Tumorigenesis

The DNA damage response (DDR) comprises multiple functions that collectively preserve genomic integrity and suppress tumorigenesis. The Mre11 complex and ATM govern a major axis of the DDR and several lines of evidence implicate that axis in tumor suppression. Components of the Mre11 complex are mutated in approximately five percent of human cancers. Inherited mutations of complex members cause severe chromosome instability syndromes, such as Nijmegen Breakage Syndrome, which is associated with strong predisposition to malignancy. And in mice, Mre11 complex mutations are markedly more susceptible to oncogene-induced carcinogenesis. The complex is integral to all modes of double strand break (DSB) repair and is required for the activation of ATM to effect DNA damage signaling. To understand which functions of the Mre11 complex are important for tumor suppression, we undertook mining of cancer genomic data from the clinical sequencing program at Memorial Sloan Kettering Cancer Center, which includes the Mre11 complex among the 468 genes assessed. Twenty five mutations in MRE11 and RAD50 were modeled in S.cerevisiae and in vitro. The mutations were chosen based on recurrence and conservation between human and yeast. We found that a significant fraction of tumor-borne RAD50 and MRE11 mutations exhibited separation of function phenotypes wherein Tel1/ATM activation was defective while DNA repair functions were mildly or not affected. At the molecular level, the gene products of RAD50 mutations exhibited defects in ATP binding and hydrolysis. The data reflect the importance of Rad50 ATPase activity for Tel1/ATM activation and suggest that inactivation of ATM signaling confers an advantage to burgeoning tumor cells. Author Summary A complex network of functions is required for suppressing tumorigenesis. These include processes that regulate cell growth and differentiation, processes that repair damage to DNA and thereby prevent cancer promoting mutations and signaling pathways that lead to growth arrest and programmed cell death. The Mre11 complex influences both signaling and DNA repair. To understand its role in tumor suppression, we characterized mutations affecting members of the Mre11 complex that were uncovered through cancer genomic analyses. The data reveal that the signaling functions of the Mre11 complex are important for tumor suppression to a greater degree than its role in DNA repair.


Abstract
The DNA damage response (DDR) comprises multiple functions that collectively preserve genomic integrity and suppress tumorigenesis. The Mre11 complex and ATM govern a major axis of the DDR and several lines of evidence implicate that and RAD50 were modeled in S.cerevisiae and in vitro. The mutations were chosen based on recurrence and conservation between human and yeast. We found that a significant fraction of tumor-borne RAD50 and MRE11 mutations exhibited separation of function phenotypes wherein Tel1/ATM activation was defective while DNA repair functions were mildly or not affected. At the molecular level, the gene products of RAD50 mutations exhibited defects in ATP binding and hydrolysis. The data reflect the importance of Rad50 ATPase activity for Tel1/ATM activation and suggest that inactivation of ATM signaling confers an advantage to burgeoning tumor cells.

Author Summary
A complex network of functions is required for suppressing tumorigenesis.
These include processes that regulate cell growth and differentiation, processes that repair damage to DNA and thereby prevent cancer promoting mutations and signaling pathways that lead to growth arrest and programmed cell death. The Mre11 complex influences both signaling and DNA repair. To understand its role in tumor suppression, we characterized mutations affecting members of the Mre11 complex that were uncovered through cancer genomic analyses. The data reveal that the signaling functions of the Mre11 complex are important for tumor suppression to a greater degree than its role in DNA repair.

Introduction
The Mre11 complex, consisting of dimers of Mre11, Rad50, and Xrs2 in budding yeast (or Nbs1 in fission yeast and other eukaryotes) plays a central role in the DNA damage response (DDR). It is a primary sensor of DNA double strand breaks (DSBs) and thus situated activates the transducing kinase Tel1/ATM. In addition, the complex is required for both homology directed DNA repair (HDR) and nonhomologous end-joining (NHEJ) [1,2].
The Mre11 complex has an elongated structure characteristic of the structural maintenance of chromosomes (SMC) protein family [3,4] suggests that those conformational states mediate distinct Mre11 complex functions [2,5].
Perhaps consistent with this interpretation, the complex's role in activating Tel1/ATM can be genetically separated from its DSB repair functions [6][7][8], suggesting that their underlying mechanisms are distinct. For example, Rad50 L1237F , a mutation found in an urothelial tumor, (rad50-L1240F in yeast) selectively impairs Tel1/ATM activation [9]. The altered residue is located in the Rad50 ATP binding domain, supporting the idea that Rad50 ATPase activity and the structural transitions attendant to ATP binding and hydrolysis are integral to Tel1/ATM activation [2,5]. In addition, mutations that impair the association of the Mre11 complex with Tel1/ATM or Mre11 complex DNA binding can also selectively impair Tel1/ATM activation [10][11][12].
The identification of Mre11 complex tumor alleles and modeling of their consequences in mice and yeast have been invaluable for understanding Mre11 complex roles in the DDR [13]. For example, phenotypic analysis of cells established from affected persons revealed the role of the Mre11 complex in activating ATM mediated DNA damage signaling.
We have shown in mouse models that the Mre11 complex is critical for suppressing oncogene induced breast carcinogenesis [14]. Having identified the Rad50 L1237F mutation as the underlying cause of an extraordinary response to chemotherapy, likely due to its selective inability to activate Tel1/ATM, we reasoned that cancer genomic data could offer a rich source for understanding Mre11 complex functions. The principle being that the development of malignancy could be viewed as a genetic screen for mutations affecting the processes that suppress malignancy.
Herein, we modeled twenty-five RAD50 and MRE11 tumor alleles in yeast and in vitro in an attempt to shed light on the tumor suppressive function(s) of the complex.
We prioritized recurrent mutations that affected conserved residues in MRE11 and RAD50. We found that ten of the modeled alleles, including one that occurred in 16 distinct cancers, were defective in Tel1/ATM activation. Collectively these data suggest that selection against ATM activation occurs during the progression of malignancy.

Results
We identified the Rad50 L1237F mutation in an urothelial tumor that exhibited an extraordinary response to an otherwise ineffective combination of irinotecan and Chk1 inhibition. Modeling the mutation in yeast (rad50-L1240F) and mouse embryonic fibroblasts (MEFs) revealed that rad50-L1240F is a Separation Of Function (SOF) mutation that blocks Tel1/ATM activation while leaving DSB repair largely unaffected. The therapeutic efficacy of irinotecan in this context was interpreted to reflect the coincident defects in ATM activation-through Rad50 L1237F -and Chk1 activity-via inhibition [9]. As a result of this finding, RAD50, NBS1, and MRE11 were added to the MSKCC IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets) gene list [15]. Mre11 complex genes have since been resequenced in over 40,000 tumors with mutations or copy number alterations found in roughly five percent of solid tumors [16].
We reasoned that additional functional analyses of Mre11 complex mutations arising in human cancer could provide insight regarding the mechanism(s) of Mre11 complex function including its role in tumor suppression. In total twenty-five mutations in RAD50 and MRE11 were modeled in S. cerevisiae (S1 Table). The mutations modeled were chosen on the basis of conservation (thus encompassing mostly residues within Walker A and B domain; Fig 1B), recurrence (number of tumors), and the allele frequency (percentage of sequence reads of a given tumor that contain this mutation) observed. NBS1 was not assessed in this study. The phenotypic parameters analyzed included Tel1/ATM activation, DNA repair, telomere length and the ability to produce viable spores.
As an overview, of the twenty-five mutations modeled, ten were found to be inconsequential. Of the remaining fifteen, ten exhibited a separation of function phenotype similar to that of the Rad50 L1237F mutation: Tel1/ATM activation was affected while DSB repair functions were largely intact. Three rad50 (R1217C, E1235K, and D1241N) and two mre11 alleles (A173V, D370Y), each of which appeared to be heterozygous in their respective tumor, conferred defects in DSB repair when modeled in yeast (S1 Table). These data suggest that inactivation of the Mre11 complex-ATM axis of the DDR provides a selective advantage during tumor progression, and that defects in the DSB repair functions of the complex inhibit tumorigenesis. The mutations examined and the experimental approaches that led to these conclusions are described below.

DSB Repair Functions In Tumor Modeled Mre11 Complex Mutants
Yeast RAD50 and MRE11 mutants corresponding to tumor borne alleles were integrated into a diploid yeast strain at their respective chromosomal loci, and haploid spores were derived by tetrad dissection.
Mre11 complex-mediated homologous recombination functions were inferred from cell survival in presence of methyl methanesulfonate (MMS), camptothecin (CPT) or hydroxyurea (HU) (Fig 2A). In Mec1-proficient cells, loss of Tel1 (tel1Δ) results in only mild sensitivity to the various clastogens, evident only at higher doses (Fig 2A, bottom lane). rad50-R1217C and rad50-E1235K phenocopied rad50Δ, and were inviable upon exposure to all three clastogens (see Fig 1A for numbering of human and yeast alleles), consistent with previous studies [5,17].
The Mre11 complex also promotes NHEJ. To assess that DSB repair mechanism in the modeled mutants, we created yeast strains in which a single DSB induced at the MAT locus must be repaired via NHEJ due to the lack of a homologous donor template [20]. WT cells plated on galactose-containing media to induce HO endonuclease expression exhibit 0.16% survival. The rate is at least 40fold lower in mre11Δ and 8-fold lower in rad50-K40A, the protein product of which is defective in ATP binding [21,22]. The survival of the modeled mutants was indistinguishable from WT ( Fig 2B), indicating that NHEJ is intact in those strains.
Colonies that survive chronic induction of the HO endonuclease do so because they have inactivated the HO site through Pol4-dependent addition (+CA; predominant in WT cells) or Pol2-dependent loss (ΔACA or ΔCA; predominant in rad50Δ) [20,23]. Although survival was unaffected, sequencing of the HO junctions revealed that in all the modeled mutants, survivors predominantly exhibited one to four nucleotide deletions, in contrast to the small insertions frequently seen in WT ( Fig 2C). This difference in imprecise NHEJ-junctions between WT and mutants might reflect differences in recruitment of aforementioned polymerases or of other factors that influence DSB end processing.
In meiosis, the Mre11 complex is required at two distinct steps. First, the complex must be present for Spo11 to induce the DSBs that initiate meiotic recombination, and second, the complex is required for removal of Spo11 from the DSB end [19,24,25]. Loss of either function blocks the formation of viable spores.
Diploids homozygous for the modeled mutations were sporulated and spore viability was determined by tetrad dissection. As seen in the response to MMS, rad50-R1217C phenocopied Mre11 complex deficiency, with < 0.01% tetrads formed and < 0.3% spore viability. Spore viability was similar to WT (98%) in each of the other mutants examined (Fig 2D). These data indicate that both Spo11-mediated DSB formation and the Mre11 complex-dependent cleavage activity were unaffected in these mutants. An exception was seen in rad50-L1240F, which exhibited only 63% viable spores ( Fig 2D).
Finally, Mre11 complex assembly was generally unaffected in all mutants tested, with the exception of mre11-E38K, in which Mre11-E38K protein levels were markedly reduced. The Mre11-Rad50 interaction was also intact in all mutants, based on comparable levels of Rad50 and/or Mre11 proteins recovered in Rad50 and Mre11 immunoprecipitations ( Fig 2E). That mre11-E38K was largely proficient in DSB repair indicates that Mre11 complex levels are not limiting for the DSB repair functions tested, consistent with observations from previous studies [8,26,27].

Separation of Function
In addition to its roles in DSB repair, the Mre11 complex is required for the activation of the Tel1/ATM axis of DDR signaling and cell cycle checkpoints [1]. We have shown that the DNA repair and DDR signaling functions of the complex are genetically separable [7][8][9]. In the mec1∆ sae2∆ context, Mec1-deficiency is suppressed in a Tel1-and Mre11 complex-dependent manner [28]. To assess whether Tel1 activation was affected in the modeled mutants, we crossed the mutants into a mec1∆ sae2∆ strain and assessed MMS sensitivity and Rad53 activation, both of which depend on Tel1 activity in that context.
Tel1 activation can also be queried via DNA damage induced phosphorylation of Rad53. The same mec1∆ and mec1∆ sae2∆ mutant strains were treated with MMS and Rad53 phosphorylation, as inferred from the appearance of slower migrating bands upon western blotting. Whereas MMS treatment induced Rad53 phosphorylation in mec1Δ sae2Δ and to a lesser extent in mec1∆ cells, minimal to no phosphorylation was observed in rad50-D67N/Y, rad50-R1259C, mre11-E38K and rad50-L1240F cells (Fig 3D and S1 Fig). These data clearly show that Tel1 activation is impaired in rad50-D67N/Y, rad50-R1259C, mre11-E38K as well as in rad50-L1240F cells, while the DSB repair functions of those mutants were essentially unaffected. The defect in Tel1 activation was not complete, as telomere shortening, which is observed in tel1∆, tel1-kd, and rad50∆, was less severe in these mutants ( Fig 3C).

Intragenic Suppressors of rad50-L1240F
We previously showed that phenotypes of mutations affecting the Rad50 hook domain can be suppressed by intragenic mutations in the Rad50 coiled coil domain.
This outcome is likely due to the intragenic mutations mitigating alterations in the path of the coiled coils caused by the hook domain mutations [8]. To gain insight regarding the mechanism(s) by which Tel1 activation is impaired in the rad50 separation of function alleles, we carried out a screen for intragenic suppressors of rad50-L1240F. We mutagenized the rad50-L1240F ORF to establish a rad50-L1240F* plasmid library. This library was transformed into a rad50Δ mec1Δ sae2Δ strain and the ensuing transformants were selected for MMS resistance. Four intragenic suppressors of rad50-L1240F, two in the coiled coil domain and one each in Walker A and B, were identified (S343P, A1079T, and I23V, S1247N respectively; see Fig 4C).  Fig 4B). In each rad50-L1240F suppressor strain, survival on MMS and CPT were substantially enhanced ( Fig 4A and S2A Fig). This enhanced survival was associated with slightly increased Rad53 phosphorylation in mec1∆ and mec1∆ sae2∆ strains (Fig 4B), suggesting that the intragenic suppressors partially restored the Tel1 activation defect of rad50-L1240F. However, all four rad50-L1240 suppressors also rescued the modest MMS and CPT sensitivity of rad50-L1240F observed at high clastogen doses in a Mec1-proficient setting (Fig 4D and S2B Fig).
Additionally, the mild telomere shortening phenotype of rad50-L1240F was partially suppressed by rad50-S343P and rad50-A1079T, but not by rad50-I23V and rad50-S1247N (S2C Fig). rad50-I23V and rad50-S343P alone were indistinguishable from WT in all respects (S2A, S2C, S2D Fig). These data indicate that, as with rad50 hook mutations [8], a mutation in the globular domain (rad50-L1240F) can be suppressed by mutations in the coiled coil domain, as well as also by rad50-L1240F proximal and distal mutations within the globular domain.

Molecular Phenotypes of rad50 SOF Mutations
Mechanisms that could account for the Tel1 activation defects observed in the rad50 SOF mutations include failure to recruit Tel1 to sites of DNA damage, failure of the mutant gene products to bind DNA, or failure to interact with Tel1. To investigate if impaired Tel1 signaling activity is due to reduced Mre11 complex or Tel1 DSB recruitment, we measured Xrs2-HA and Tel1-HA association to a HO-DSB at the MAT locus by chromatin immunoprecipitation (ChIP) and quantitative PCR (qPCR) with primers 0.6 kb and 1.6 kb from the DSB ( Fig 5A).
These data indicate that impaired Tel1 activation in the SOF mutants is not solely due to defects in Tel1-recruitment, and suggests that other molecular defects such as defective ATP binding or hydrolysis may underlie this phenotype.
Recent studies suggest that Tel1 has a structural role in stabilizing Mre11 complex DSB association that is independent of its kinase activity [10,31,32]. We found that the effects of tel1-kd versus tel1∆ on the survival of SOF mutants exposed to CPT were different, and did not strictly correlate with the ability of the mutant gene products to recruit Tel1 to DSBs. For example, whereas Tel1 ChIP signal was virtually absent in rad50-D67Y/N (Fig 5A), CPT survival of those mutants was enhanced in tel1-kd relative to tel1∆, suggesting that the rad50-D67Y/N gene products physically associate with the Tel1-KD protein ( Fig 5B). Therefore, the defect in Tel1 activation in rad50-D67Y/N is likely not attributable to a lack of Rad50-Tel1 interaction, and suggest that an activity intrinsic to Rad50 is required for Tel1/ATM activation.

Biochemical Analysis of SOF Mutant Gene Products
Structural analyses of the Mre11 complex globular domain suggest that ATP binding and hydrolysis determine whether it adopts a closed or open form. The former is proposed to mediate DSB end tethering, NHEJ, and Tel1/ATM activation [5,33]. Transition to the open state depends on ATP hydrolysis by Rad50 in which the globular domain opens to make the Mre11 nuclease active sites accessible for DNA substrates [2,34,35].
rad50-D67N/Y and rad50-L1240F mutations alter residues in the Rad50 ATPase domain. Given that ATP binding is required for DNA binding [22], we purified the WT and mutant Rad50 proteins from yeast cells ( Fig 5C) and the corresponding MRXholo complexes from insect cells (Fig 5E, left panel). We carried out assessment of DNA and ATP binding, as well as ATP hydrolysis. Rad50 ATP binding was measured using a filter binding assay. WT and mutant Rad50 proteins (2 µM) were incubated for 20 minutes at room temperature with a molar excess of ATP (0.1 µM α 32 P-ATP + 49.9 µM unlabeled ATP). Following incubation, the reactions were spotted on nitrocellulose filters, washed, and bound radiolabelled ATP was quantified by liquid scintillation counting. Under these conditions, the molar ratio of bound ATP to Rad50 was 1.1, in agreement with structural determination of ATP:Rad50 stoichiometry [36]. ATP binding was reduced by approximately two fold in the Rad50 mutants; Rad50-D67N (0.6 ATP/Rad50), Rad50-67Y (0.7 ATP/Rad50) and Rad50-L1240F (0.4 ATP/Rad50) ( Fig 5D).
To measure ATP hydrolysis, increasing concentrations (0-2 µM) of purified MRX WT and mutant complexes were incubated with γ 32 P-ATP and MgCl 2 in presence of ssDNA to stimulate Rad50 [22], and hydrolysis was assessed by thin layer chromatography ( Fig 5E and S4A Fig). We find that both MRX-D67Y and -D67N exhibited a 50-70% reduction in ATP hydrolysis relative to MRX-WT (Fig 5E), consistent with their respective decrements in ATP binding (Fig 5D). Recently we have established an assay to measure Mre11 complex-and DNAdependent Tel1 activation in vitro [30]. To assess Tel1 activation by modeled mutant Collectively these data suggest that ATM activation is selected against during the progression of malignancy. The common mechanistic underpinning of the observed separation of function phenotypes is related to ATP binding and hydrolysis.
Other defects are unique; in the case of Rad50-L1240F, defects in DNA binding also are correlated with the defect in Tel1/ATM activation whereas in in Rad50-D67Y/N, the Tel1/ATM activation defect may be partially attributable to reduced Tel1 recruitment to DSBs.

Structural insight of tumor modeled mutations derived by homology modeling and molecular dynamics simulation
Structural information regarding the eukaryotic Mre11 complex is available for the globular domain and the Rad50 hook domain [2]. We used globular domain information to carry out a combination of molecular modeling, bioinformatics and molecular dynamics simulations to gain insight regarding the molecular features of (WT), -D67N and -D67Y show that both mutants alter the interaction between the nucleotide and its binding site (S1, S2 and S3 movie). The D67 mutations resulted in increased interaction times and strength of the contacts between the adenosine base and N67/Y67, I65, and V63 (Fig 6B and 6C), while weakening contacts with K1193. The presence of single carbonyl oxygen of N67 leads to a frozen conformation of its side chain mediated by the NH proton of K69 ( Fig 6B). Therefore, the release of ADP upon hydrolysis may be impaired by the D67N and D67Y mutations (S2 and S3 movie). We speculate that impaired release of ADP in N67/

Discussion
The Mre11 complex is required for all forms of DSB repair as well as for the initial detection of DSBs and the subsequent activation of the ATM axis of the DDR.
Rapidly accumulating cancer genomic data has revealed that Mre11 complex components are mutated in in approximately five percent of solid tumors [37]. This observation, along with data from humans and mouse model systems support the view that the Mre11 complex and the processes that it influences suppress tumorigenesis [1,14,38].
In this study, we used cancer genomic data to investigate the molecular mechanism(s) that underlie Mre11 complex-dependent tumor suppression using S. cerevisiae for genetic analysis as well biochemical analyses of recurrent mutant gene products. The mutations modeled were selected on the basis of conservation of the affected residues, allele frequency in the tumor, and recurrence. Two themes emerged. First, the mutations modeled predominantly exhibited separation of function phenotypes characterized by defective Tel1/ATM activation without substantial impairment of DSB repair. Second, the gene products of recurrent RAD50 alleles found in seventeen distinct tumors exhibited ATP binding and/or hydrolysis defects. Taken together, these themes support the interpretation that selection against Mre11 complex-dependent ATM signaling occurs during tumor progression, and that Rad50-dependent ATP metabolism is crucial for ATM activation by the Mre11 complex.
It is important to note that we did not randomly sample all Mre11 complex mutations. Filtering the mutations for conserved residues implicitly biases the analysis to the Rad50 globular domain given its higher degree of conservation.
Nevertheless, the recurrence of mutations that affect this domain provides a compelling argument for selective pressure.
Tel1/ATM activation requires DNA and the Mre11 complex, although the underlying molecular mechanism is unknown [39]. Each of the modeled Rad50 SOF mutations severely impaired Tel1 activation in vivo and in vitro (Fig 3B, 3C and 5F).
In principle, this phenotype could reflect loss of Mre11 complex DNA binding, loss of Mre11 complex-Tel1/ATM interaction, or impairment of an unknown mechanism.
Binding to naked DNA in vitro was not correlated with defective Tel1/ATM activation. Only Rad50-L1240F-containing complexes were impaired in this respect.
Similarly, ChIP analysis revealed no clear correlation between Mre11 complex engagement at a DSB site and Tel1 kinase activation. These observations suggest that the underlying molecular bases of Tel1/ATM activation defects Rad50-1240F and -D67Y/N are distinct.
A complex containing rad50-K81I, a rad50S protein that is hypermorphic for Tel1/ATM activation [40] exhibited increased DSB association and increased the abundance of Tel1 at the site as previously reported (Fig 5A) [29]. Conversely, Rad50-K40A-containing complexes, which do not bind ATP and phenocopy rad50∆ in most respects did not engage DSB sites or promote Tel1 association. However, Rad50-L1240F and -D67Y/N-containing complexes, which are equally defective in Tel1/ATM activation, had divergent effects on DSB engagement and Tel1 ChIP (Fig   5A and 5F). ATP binding and hydrolysis by Rad50-L1240F-containing complexes is more severely impaired than -D67Y/N, while DSB engagement is enhanced relative to wild type and Rad50-D67Y/N. Hence it appears that ATP binding and hydrolysis are required for Tel1/ATM activation, and that this function is distinct from the recruitment of Tel1/ATM to DNA damage.
This observation further indicates that ATP binding and hydrolysis is linked to an unknown mechanism of Tel1/ATM activation. Previous analysis of the human Mre11 complex suggested that the ATP bound "closed" form of the complex is responsible for ATM activation [33]. Our data are consistent with that view, although it is possible that the transition per se from the closed to the open form, which occurs upon ATP hydrolysis, underlies the mechanism of Tel1/ATM kinase activation rather than one form or the other. However, as ATM activation by the human Mre11 complex in vitro required ATP binding but not ATP hydrolysis [33], and ATP-binding induces multiple conformational switches in both Rad50 and Mre11 prior to ATPhydrolysis [36,41], it is tempting to speculate that a yet undefined conformational state between the closed and open complex mediates Tel1 activation. It is noteworthy that molecular dynamics simulation of ATP binding in the Rad50-D67Y or -D67N mutants suggests that the release of ADP following hydrolysis is impaired. It is thus conceivable that the transition between closed and open forms may be altered in that mutant, and may in turn account for the observed Tel1 activation defect.
Failure of the Mre11 complex to interact with Tel1 could also explain Tel1 activation defects. Tel1 ChIP signal was markedly reduced in Rad50-R1259C and Rad50-D67Y/N, while in the latter, ChIP signal of the mutant complexes to DSB sites was evident (Fig 5A), perhaps indicating impaired Mre11 complex-Tel1 interaction.
However, in the context of tel1-kd, the CPT sensitivity of each of the three was mitigated in comparison to the tel1∆ background. This suggests that the inactive Tel1 protein binds to, and enhances the functionality of the Mre11 complex via physical interaction as proposed previously [10,32,42]. Consistent with this idea, the effect of tel1-kd on CPT sensitivity was most pronounced in the Rad50-L1240F, mutant, which binds DNA in vitro at least 10 fold less well than Rad50-D67N or -D67Y ( Fig 5B). Thus, these data do not support the interpretation that the Mre11 complex-Tel1 interaction is impaired. The observation that CPT sensitivity in mre11-E38K strains was the same in tel1-kd and tel1∆ further argues that Tel1 protein primarily interacts with Rad50, as has been shown in vitro for ATM and human Rad50 [43].
Although the SOF mutants described were largely DSB repair proficient, a subtle effect on NHEJ was observed. NHEJ junctions in Rad50 SOF mutants predominantly exhibited deletions as opposed to the insertions typically observed in wild type cells (Fig 2C). The mechanistic basis for this difference is unclear, but we have previously noted a correlation between defects in Tel1 activation and deletional NHEJ. For example, in rad50 sc+h , rad50-48, and rad50∆CC mutants, which affect the Rad50 hook and coiled coil domains, deletions at NHEJ junctions are elevated.
Like the tumor borne mutations described in this study, those Rad50 alleles also exhibit SOF phenotypes in which Tel1 activation is impaired [6][7][8] and unpublished data). These observations resonate with analysis of NHEJ-mediated chromosomal translocations induced by HO cleavage. In Tel1 deficient cells, those junctions are characterized by the same ∆ACA deletions observed at "unrepairable" DSBs in Mre11 complex deficient strains and in the SOF mutants described here and previously [44]. Although the various genetic contexts alluded to above have distinct phenotypic features, attenuation or abolition of Tel1 activation is common to each.
These data thus support a role for Tel1 activity in influencing mechanistic features of NHEJ.
ATP binding and hydrolysis by Rad50-L1240F are severely impaired (Fig 5D   and 5E). As with other phenotypic outcomes, the ATP binding and hydrolysis defects of rad50-L1240F are mitigated by the S343P suppressor (Fig 5C and S4B   Fig). We proposed in a previous study that distal intragenic suppressors within the coiled coil domain influence the disposition of the globular domain via the changes in the path of the coiled coils [8]. This proposal is supported by a recent structural study of the E. coli ortholog of the Mre11 complex, SbcCD [45]. The observation that the ATPase function specified in the globular domain is influenced by a coiled coil mutation lends further support to this presumptive mechanism.
The role of the DDR in tumor suppression is multifaceted. Whereas the ATR-Chk1 axis of the DDR is required for viability of oncogene-expressing cells [46], inactivation of the Mre11 complex-ATM axis enables oncogene-driven carcinogenesis. In mouse models, Mre11 complex hypomorphism enhances the ability of the neuT oncogene to promote malignancy in mammary epithelium [14], and promotes Notch-driven leukemogenesis in the hematopoietic compartment [47].
Collectively, those observations and those presented here support the view that selection against Mre11 complex-ATM signaling in the DDR potentiates oncogene driven carcinogenesis. Implicitly, the data further suggest that the loss of fitness that would be associated with reduced DNA repair capacity is also selected against during tumorigenesis. Finally, this study predicts that mutations impairing ATM activation or signaling may sensitize tumors to clastogenic therapies as well as those that inhibit the ATR-Chk1 axis of the DDR.

Ethics Statement
The 12-245 Data & Tissue Usage Committee at MSKCC has reviewed this study and has no comments or concerns.

Yeast strains
All strains used in this study were in W303+ background and are listed in supplemental Table S3. All rad50 and mre11 alleles in this study were integrated at their native chromosomal locus and verified by PCR genotyping and sequencing.
Details of yeast strains and plasmid constructions are available upon request.

Damage sensitivity assays
Five-fold serial cell dilutions (250,000 cells per spot to 80 cells per spot) were spotted on YPD plates without or with S-phase clastogens and incubated at 30 °C for 2-3 days. Spores were grown in yeast extract peptone medium containing 2.6% (v/v) glycerol, 2.6% (v/v) ethanol, 1% (v/v) succinate and 1% sucrose to exponential phase. Cells were counted and plated in triplicate on plates of identical composition additionally containing either 2% galactose or 2% glucose. Cell survival is expressed as the percentage of cells growing on galactose versus glucose containing plates after 4 days incubation.

Rad53 phosphorylation
Rad53 phosphorylation was assessed as described previously [9]. Briefly, exponentially growing cells (2-4x10 7 cells/ml) in mec1Δ sml1Δ or mec1Δ sml1Δ sae2Δ background were cultured in presence of 0.15% MMS for 90 min. MMS was inactivated upon addition of 5% sodium thiosulfate (final concentration) to the cultures. TCA-extracts were prepared and 10-20 µg protein extracts were run on a 7.5% SDS-PAGE, transferred to nitrocellulose membrane and FLAG-Rad53 was detected by western blot with FLAG M2 mAb (Sigma).

Sporulation efficiency and spore viability
Diploids cells were grown overnight in YPD media, then diluted 20-fold in YPA media (yeast extract, 2% potassium acetate, 5 mg/ml adenine). After 12 hours incubation in YPA media, cells were gently pelleted, washed with water and incubated for 2-3 days in sporulation media (1% potassium acetate, 5 mg/ml adenine). The sporulation efficiency was calculated by the numbers of tetrads present among >400 of sporulated cells. Spore viability was determined by tetrad dissection of 20-40 tetrads. Two independent diploids for each genotype were assessed.

Telomere Southern blot
Freshly dissected spores were grown for 30 generations of growth and genomic DNA isolated by standard phenol chloroform extraction using glass beads [48].
Genomic DNA was either PstI-or XhoI-digested (as specified in Figure legends), run on 1.3% agarose gel and transferred on an Amersham Hybond-XL membrane (GE Healthcare) and detected by Southern blot with a 32 P-labeled telomere-specific probe (5′-TGTGGTGTGTGGGTGTGGTGT-3′) as described [49].

rad50-L1240F suppressor screen
The Ycp50-rad50-L1240F plasmid was randomly mutagenized by propagating the plasmid in XL1 red bacterial mutator (mutS, mutD, mutT) strain (Agilent Technologies). rad50Δ mec1Δ sml1Δ sae2Δ spore (JPY1953) was transformed with the mutagenized Ycp50-L1240F* plasmid library. 1000 larger URA3 colonies were manually picked, diluted in water and assessed for growth on Do-Ura plates in presence of 2.5 µM CPT or 0.01% MMS. Eighteen colonies were able to grow in presence of either clastogen. Plasmids were recovered and retransformed in JPY1953, and the four plasmids promoting strongest suppression were sequenced.
The identified suppressor mutants were by site directed mutagenesis de novo inserted in rad50-L1240F::HYG integration construct and integrated at the RAD50 locus.
The powdered yeast cells were thawed, 20 ml lysis buffer (as above) was added and resuspended by pipetting to remove all clumps. The yeast extract was clarified by centrifugation (40 min 20,000 rpm, ss34 rotor) and incubated over night at 4°C in presence of 0.8 ml anti FLAG M2 agarose beads (Sigma). The FLAG beads were washed 10x with 10 ml lysis buffer. Rad50-FLAG proteins were eluted with 5x 0.8 ml elution buffer (lysis buffer containing 100 µg/ml 3xFLAG peptide). 4 ml FLAG eluate was concentrated to 0.8 ml using Amicon Ultra-4 Centrifual Filters Ultracel-30K.
Protein concentrations were determined by Lowry protein assay. 10 µl aliquots (≈10 µM concentration) were flash frozen in liquid nitrogen and stored at -80°C.

ATP binding assay
Rad50 ATP binding was determined by standard nitrocellulose filter binding assays. were prepared and measured to determine the amount of ATP bound by Rad50.

Chromatin Immunoprecipitation
Cells were cultured overnight in YPLG (1% yeast extract, 2% bactopeptone, 2% lactic acid, 3% glycerol and 0.05% glucose) at 25°C. Cells were then diluted to equal levels (5 x 10 6 cells/ml) and were cultured to one doubling (3-4 hours) at 30°C to 1 x 10 7 cells/ml. 2% galactose (final concentration) was added to the YPLG and incubated for 3 hours. Cells were harvested and crosslinked at various time points (t=0 hours and t=3hours after galactose treatment) using 3.7% formaldehyde solution. Following crosslinking, the cells were washed with ice cold PBS and the pellet stored at -80°C. The pellet was re-suspended in lysis buffer (50mM HEPES pH 7.5, 1 mM EDTA, 80 mM NaCl, 1% Triton, 1 mM PMSF and protease inhibitor cocktail) and cells were lysed using Zirconia beads and a bead beater. Chromatin fractionation was performed to enhance the chromatin bound nuclear fraction by spinning the cell lysate at 13,200 rpm for 15 minutes and discarding the supernatant.
The pellet was resuspended in lysis buffer and sonicated to yield DNA fragments (~500 bp in length). The sonicated lysate was then incubated with Dyna beads (Sheep anti-Mouse IgG from Invitrogen) with anti-HA Antibody (Santa Cruz) or unconjugated beads (control) for 2 hours at 4°C. The beads were washed using wash buffer (100 mM Tris-Cl pH 8, 250 mM LiCl, 0.5% NP-40, 1 mM EDTA, 1 mM PMSF and protease inhibitor cocktail) and protein-DNA complexes were eluted by reverse crosslinking using 1% SDS in TE buffer, followed by proteinase K treatment and DNA isolation via phenol-chloroform-isoamyl-alcohol extraction. Quantitative PCR was performed using the Applied Biosystem QuantStudio 6 Flex machine.
PerfeCTa qPCR SuperMix, ROX was used to visualize enrichment at HO2 (0.5 kb from DSB) and HO1 (1.6 kb from DSB) and SMC2 was used as an internal control.
The Ct values from the qPCR were used to estimate the amplification of the region at 0.6 kb (HO2) and 1.

Homology modeling and molecular dynamics simulation
S.cerevisiae MR complex has been generated by homology modeling of the constituent Mre11 (M) and Rad50 (R) molecules and their subsequent assembly in M2/R2 heterotetramer. Secondary structure predictions from amino acid sequences, access numbers CAA65494.1 and BAA02017.1 [50], were computed by PSIPRED [51,52].

S.cerevisiae Mre11 was built with Schrödinger Prime homology package [53] using
S.pombe crystal structure as a template (RCSB entry 4FBQ.pdb: resolution 2.2 Å, sequence homology and identity 68% and 51%, respectively), and guided by secondary structure predictions. The model has been edited to enhance its similarity with hsMre11 (RCSB entry 3TI1.pdb)

Generation of scMre11 dimer.
Two metal ions introduced into the homology model from hsMre11 structure were converted to dummy ligands. Similar dummy ligands were generated on each of two scMre11 monomers of the MR crystal structure from archaea M. jannaschii (RCSB entry 5DNY.pdb) [34]. These artificial ligand binding sites were used for two subsequent applications of Align Binding Sites tool [53], thus generating a dimeric assembly of scMre11. The conformations of the contacting loops at the interface of two scMre11 molecules have been edited by copying corresponding conformations from the crystal structure of C.thermophilum (RCSB entry 4YKE.pdb) [54]. The conformations of the loops protruding towards presumed scRad50 binding interface have been copied from the corresponding parts of the archean M2/R2 assembly (RCSB entry 5DNY.pdb).
The initial models for monomeric Rad50 were generated by SWISS-MODEL [55], using 5DA9.pdb structure from C.thermophilum as a template [54].

Generation of M2/R2 heterodimers.
Initial positioning of scRad50 over scMre11 was obtained by pairwise aligning lobe I c-alpha atoms [35] of scRad50 and mjRad50 by MR heterodimer structure of M.
jannaschii (RCSB entry 5DNY.pdb). After that, ATP analog ligands were introduced into scRad50 monomers by aligning its protein c-alpha atoms with these of ctRad50.
The new positions of Rad50 were subsequently improved by the application of Align Binding Sites tool, targeting one pair of ligands at a time (one from scRad50, one from ctRad50). Both molecules of scRad50 were next superimposed as a dimer onto the ctRad50 dimer by all c-alpha atoms. One more iteration of monomermonomer re-adjustment and re-introduction of ATP analog ligands by c-alpha atoms have improved the overall architecture of the complex and the positioning of ATP analogs in it. A few close contacts were removed by selecting appropriate rotamers of amino acid residues using Maestro graphical user interface of Schrodinger [53].
The resulting complex had no bad contacts at all; it was next energy-minimized and additionally cleaned up by 10ns molecular dynamics.

Symmetry Enhancements.
Two Mre11+Rad50 halves of the scMR complex cleaned up with energy minimization and molecular dynamics were merged into a single molecule each. A copy of one such heterodimeric half was superimposed onto another one by all its c-alpha atoms. Rad50 constituent molecules of these halves were next superimposed individually, without Mre11, and the new positions became parts of the updated complex. The ATP analogs were re-introduced in this updated complex after one more round of superimposing of a dimeric ctRad50 on the total scMR dimer, with the subsequent adjustments of monomeric scRad50 molecules onto each monomeric ctRad50.

Introduction of RBD helices.
RBD helices were homology modeled on ctMre11 (5DA9.pdb). Their position on coiled coils of scMR model was obtained by the superposition of coiled coils of ctMR and scMR.
The missing loop between K410 and L444 that connects RBD helices with Mre11 has been generated in two steps. Initially, a fragment from the RCSB entry 1DY0.pdb identified using NGL-SuperLooper [56] was introduced (A chain, residues 192-223). The standard alpha-helical six-residue segment predicted by PSIPRED for D430-K436 of scRBD domain. A shorter remaining gap between D430 and K410 was then filled with PrimeX module of Schrödinger suite.
The resulting protein structure was subjected to Protein Wizard of Schrödinger suite, energy minimizations, and molecular dynamics.
Accommodation of Y67 in the structure required conformation adjustments of the neighboring residues M102, L1198, and R1203. Two conformers of Y67 side chain have been reviewed: one directed outwards and one inwards, with the aromatic ring in the inward conformation being parallel to the adenine base of ATP ligand. For the inward mutation, the rotamers of T107, M1191, L1198, and R1203 were screened to select the one with the minimal clashes. All mutant complexes were minimized with conjugate gradient minimization for 2x10 4 steps.
L1240F mutants have their side chains positioned close to each other at the dimer interface and could potentially interfere with the Rad50 dimer assembly. In the presence of ATP molecules, the Prime module of Schrödinger suite could not fill in an artificially generated gap of three residues around residue L1240. The studied structure of the L1240F mutants was thus obtained by fitting the bulkier phenylalanine side chains at the interface already preformed, and with two ATP analog molecules already bound.
Protein complexes surrounded by a 10Å buffer in an orthorhombic box solvated in SPC water and with charges neutralized were prepared with System Builder of Maestro GUI. The relaxation protocol before production run included two steps of minimization followed by four MD runs. The first two runs 12 ps each were performed at 10K and with short 1 fs timesteps (as an NVT ensemble first and as an NPT ensemble second), with all solute heavy atoms fixed. Two subsequent runs as NPT ensembles were at 300 K and with normal 2 fs timesteps: first for 12 ps with solute heavy atoms fixed, followed by a 24 ps run with no restraints imposed.
A MD production run of 10 to 100 ns was carried out at a constant temperature and pressure with a 2 fs timestep throughout the simulation. The thermostat Nose-Hoover chain method was applied with relaxation time of 1 ps [58]. Barostat parameters were set according to Martyna et al. [59] with a relaxation time of 2.0 ps with isotropic coupling. A 9 Å cutoff was applied to Lennard-Jones interactions, and the nonbonded list was updated every 1.2 ps. The production snapshots of the coordinates were written out every 1.2 ps.
The data were analyzed by the Simulation Event Analysis and Simulation Interaction Diagram modules implemented in Schrödinger suite of programs.

Movies
Movies were prepared with Maestro by exporting frames from an entire 10 ns MD trajectory, 0.12 sec per frame.   Table S1. rad50-L1240F was previously modeled based on an RAD50 L1237F outlier patient with an extraordinary response to chemotherapy [9] (B) Rad50 primary protein structure with the modeled mutations and relevant Rad50 domains are denoted (abbreviated A, B, cc, hk). The mutations modeled in this study are highlighted in color and conserved residues in grey in the multiple sequence alignment of Rad50 proteins. The RAD50 D1238N and RAD50 Q1259K alleles previously modeled in yeast [9] are also indicated. Known motifs in Walker B domain are denoted. Circles denote residues involved in specific binding of phosphates (yellow) and magnesium (green), respectively [36]. Black circles indicate basic switch residues mutated in previous studies [5,17] Table. Some residues were also mutated in some tumors to other amino acid residues (see right side of Table), but were not assessed in this study.