XLF regulates filament architecture of the XRCC4·Ligase IV complex

The XRCC4 C-terminal region supports N-terminal domain interactions forming tetramers and filament-like structures

The XRCC4 high-resolution structure, spanning residues 1-203, displays a dimeric state with a tight interface between the N-terminal globular structure and an exposed coil-coil motif (Junop et al., 2000; Sibanda et al., 2001). However, the structure and function of the XRCC4 C-terminal region is unknown. To define the dynamic structure of full-length human XRCC4 and the role of the C-terminal domain, we employed SAXS (Putnam et al., 2007) combined with sample re-purification using size exclusion column chromatography (SEC) (Rambo and Tainer, 2010b). Pair distribution functions (P(r)) calculated from SAXS profiles of XRCC4 over 1-3 mg/ml indicate that the protein exists in a protein concentration dependent equilibrium of different oligomerization states in solution (Figure S1A). To characterize different oligomerization states, samples were re-purified prior to SAXS collection by a combination of SEC and multiangle light scattering (MALS), which allowed us to determine polydispersity and Mw for each fraction (Figure 1A-C). The maximal dimension (D_max) of XRCC4 in the small molecular weight fraction (Mw MALS~90kDa, Mw SAXS~75kDa) was ~170Å (Figure 1C). Calculated envelopes show the position of the bulky head domains, while the wide protrusion most probably belongs to the XRCC4 C-terminal region (Figure 1D).

To further examine the arrangement of the XRCC4 dimer and possible conformations of the C-terminal region, we employed rigid-body modeling (Pelikan et al., 2009) using rigid helices identified by secondary structure prediction (Figure S1B). Comparison of models with experimental data shows that the C-terminal region is in close contact with the coil-coil motif, with possible interactions with the N-terminal head domain (Figure 1D). Although the secondary structure prediction suggests that the C-terminal region of XRCC4 is disordered, a recent study using a de novo algorithm predicted that this region of mouse-XRCC4 exhibited globular characteristics (Recuero-Checa et al., 2009). However, an XRCC4 model with a globular C-terminus does not match the SAXS data (χ²~8), indicating that the globular conformation of the XRCC4 C-terminus likely does not exist in solution (Figure S1C). Although our SAXS data leads to a different model from that previously proposed (Recuero-Checa et al., 2009), the EM 2D projection of the mouse XRCC4 showed similar features to our SAXS reconstructions (Figure 1D). Notably, the folded back helical stalk indicated from our SAXS data is reminiscent of the structure of XLF (Andres et al., 2007; Li et al., 2008).

Given the predicted conformational disorder of the C-terminal region (Figure S1B), we reasoned that the coexistence of different conformations could contribute to the experimental SAXS profile. We therefore applied a Minimal Ensemble Search (MES) in which a genetic algorithm is used to identify the minimal ensemble of conformers required to best fit the experimental data (Bernado et al., 2007; Pelikan et al., 2009). Assembly of two MES-selected conformers slightly improved the fit to the experimental data (χ²=1.4 vs. χ²=1.8). Consistent with the earlier analysis, both conformers showed the C-terminal region folded back onto and twisted around the coil-coil motif (Figure S1CD).

We next examined the structural arrangements of XRCC4 tetramers in the collected higher molecular weight fraction (Figure 1A - magenta) and a filtrated sample, which had been passed through a filter without SEC (termed simple filtration, Figure 1BCD - green). Interestingly, SEC-MALS analysis of the higher molecular weight fraction showed a broad signal between 100 and >200 kDa suggesting polydispersity and indicating that the XRCC4 tetramer is in equilibrium with dimers and higher order oligomers. This is consistent with the previously determined high dissociation constant for XRCC4 tetramers (K_D=22 μM) (Junop et al., 2000). The P(r) for the higher molecular weight fraction shows broadening with a distinct shoulder, indicating the presence of tetramers. The P(r) shoulder transforms into a maxima at r=75-100Å for the simply filtrated sample and indicates a larger fraction of tetramers and possible larger oligomers (Figure 1C). Calculated envelopes show the location of the XRCC4-dimer with the head domains in the central part of the envelope and the coil-coil motifs pointing outwards at the opposite extremity (Figure 1D). The regions corresponding to the coil-coil motif are more distinct for the simply filtrated samples (Figure 1BD), consistent with these samples containing a larger fraction of tetramers. These results demonstrate that the interaction of XRCC4-dimers within the tetramer is mediated through the head domains. The formation of larger oligomers observed at higher protein concentrations (>3.0mg/ml, Figure 1A and S1A) is consistent with repeated head-head complexation in the XRCC4 filaments (Figure 1E). Moreover, manual superposition of the XRCC4 tetrameric arrangement found in the crystal lattice (Junop et al., 2000) closely matches the SAXS envelopes (Figure 1D).

To further examine this XRCC4-tetramer arrangement and the conformation of the C-terminal regions, we used rigid body modeling, as described above. The reconstructed atomic model gave an excellent match to the experimental data (χ² =1.8), indicating a back-folded C-terminus similar to that observed in the dimeric state. A recently proposed model in which tetramers were formed via interactions between the coil-coil regions (Dahm, 2008; Junop et al., 2000; Modesti et al., 2003) does not agree with our experimental data, suggesting that the “coil-coil” tetramer does not exist in solution (Figure S1D).

To investigate the role of the C-terminal region in the formation of tetramers and filaments, we examined scattering results of XRCC4 truncations: XRCC4 1-140 and XRCC4 1-161. These constructs predominantly formed dimers in solution at protein concentrations approximately 4 times higher than the concentration at which the full length XRCC4 formed tetramers. Only a small proportion of the protein existed in the filament form at this concentration under standard buffer conditions (pH 7.0) (Figure 1E), and a higher proportion of filaments were observed at pH 8.0, as indicate by larger distances in the P(r) function (Figure 1E). A similar effect of pH was observed for full length XRCC4 (Figure S1E). These results indicate weak head-head interactions for the truncated XRCC4 and stabilization of this interaction by the C-terminal region in the full-length protein.

Together, our results are consistent with the C-terminal region of XRCC4 (residues 141-334) being important for XRCC4 tetramer formation. These analyses provide structural insights into full-length XRCC4, showing folding back of the disordered C-terminal region, similar to that observed in XLF, and a head-to-head interaction between XRCC4 dimers in tetramers and filaments that forms at high protein concentrations (Figure 1E and S1A).

LigIV tandem BRCT domains destabilize XRCC4 filaments

LigIV has the unique ability to catalyze ligation in NHEJ, and its function requires physical interaction with XRCC4 (Critchlow et al., 1997; Grawunder et al., 1997). Within the XRCC4·LigIV complex, interactions map to the central coil-coil region of XRCC4 and to the linker region between the two BRCT domains (Grawunder et al., 1998; Sibanda et al., 2001). The recent high-resolution structure of XRCC4 in complex with the LigIV BRCT domains reveals an extensive binding interface formed by a clamp-shaped helix-loop-helix motif spanning both BRCT1 and BRCT2 domains (Wu et al., 2009). To test the effects of the LigIV-BRCT domains on XRCC4 structural arrangements in solution, we combined XRCC4 with an excess of the tandem BRCT domain and examined scattering results of the re-purified XRCC4·BRCT assembly (Figure 2A).

MALS of the SEC peak fraction, in which both proteins co-eluted (Figure 2A), shows a low polydispersity peak with Mw~120 kDa, revealing the formation of an XRCC4·BRCT complex with a molar ratio of 2:1 (XRCC4:BRCT). The calculated P(r) function had a D_max~220Å and differed from XRCC4 alone in the presence of a new maxima at r~110Å, which corresponds to the average distance between the head domains of XRCC4 and the BRCT domains (Figure 2CD). SAXS reconstructions reveal a protrusion in the central coil-coil region of XRCC4 and show BRCT complex formation. The broad-tail in the P(r) function (r=125-220Å) generates a slight bend in the extension relative to the central coil-coil region and is consistent with straightening and kinking of the XRCC4 coil-coil region observed upon BRCT complex formation (Wu et al., 2009). These SAXS results suggest further extension and stabilization of the C-terminal helices upon BRCT complex formation. The reconstructed atomic model closely matches the experimental data (Figure 2B) and shows displacement of the C-terminal helical stalks from the coil-coil region (Figure 2DE).

To determine how the BRCT domain affects XRCC4 filament conformation, we incubated XRCC4 with different molar ratios of the BRCT domain, and SAXS data was collected without further purification. The narrowing of the P(r) function indicates accumulation of the XRCC4·BRCT complex and disruption of XRCC4 filaments (Figure S2). This straightening of the XRCC4 C-terminal region by BRCT interaction leads to displacement of C-terminal helical stalks from the coil-coil region and blocks stabilization of the head-head interface between two XRCC4-dimers (Figure 2E).

Dynamic nature of XLF·XRCC4 interactions

XLF is the most recently identified member of the NHEJ family of proteins. XLF and XRCC4 interact in solution with an affinity of 7.8μM and mutational analysis suggests that K63, K65, K99, of XRCC4 and L115 of XLF are involved in mediating a head-head interface (Andres et al., 2007); however, structural arrangements within the complex are unknown.

To examine the structure of the XRCC4·XLF complex, we first purified XLF 1-248 (referred to hereafter as XLF) in complex with the XRCC4 head domain (XRCC4 1-140). The SEC-MALS peak, where both proteins co-elute (Figure 3A), shows a broad signal between 100 and >200 kDa suggesting polydispersity and indicating that the XRCC4 1-140·XLF complex is dynamic in nature. We next examined scattering results from the fraction collected at the maxima of the SEC-MALS peak; however, repeated SEC of the collected peak fraction indicated that the complex underwent re-equilibration (Figure S3A). The SEC-MALS showed sample heterogeneity; however, distinct peaks at 80Å and 160Å in the SAXS profile, the P(r) function and the rod-like P(r) function with D_max~200Å (Figure 3BC) allowed us to generate low-resolution models. However, SAXS envelopes calculated for two different concentrations of XLF·XRCC4 were slightly different (Figure S3D), likely reflecting heterogeneity of the sample. To further define the XLF·XRCC4 arrangement we carried out rigid-body modeling in combination with MES (Pelikan et al., 2009). Based on previous mutational analysis (Andres et al., 2007), we positioned XLF L115 into the XRCC4-head groove formed by helices α2, α3 and beta-sheets β6, β7, thus localizing all interaction-required residues (K63, K65, K99) (Figure 3D). We performed an exhaustive search for the best-fit conformation by rotating and translating the XLF molecule in close proximity to the XRCC4 groove. The best-fit model contained two XLF homodimers interacting with the XRCC4 grooves (χ²=4.1). However, a mixture of tetramer, hexamer, as well as octamer assemblies improved the match further, giving an excellent fit (χ²=1.9), consistent with the dynamic nature of the XLF·XRCC4 interaction (Figure 3DE, model 1). The best-fit model showed a tilt angle of ~45° between XRCC4 and XLF coil-coil regions (Figure 3DE). We also generated an alternative model 2 with a similar XRRC4·XLF interface but with 0° tilt angle between the coil-coil regions. The selected ensemble of oligomers model showed significantly worse fit (model 2, χ²= 2.7), suggesting that this arrangement is not accurate (Figure 3DE). We also tested the previously proposed model (Andres et al., 2007) in which the XLF head interacted with the lower region of the XRCC4 head domain. The calculated ensemble of this oligomeric state also gave a very poor fit (model 3, χ²=7.3) (Figure 3E).

Consistent with previous results of mutational analysis (Wu et al., 2009), our model for the XRCC4·XLF head-head interface places XRCC4 residues K65 and K99 in direct interaction with XLF residue L115 (Figure 3D). Furthermore, no binding of the BRCT domain with XRCC4 1-140·XLF was observed (Figure S3), consistent with the BRCT domain interacting exclusively with the XRCC4 coil-coil motif which is not present in the XRCC4 1-140 construct.

Next we examined full-length XRCC4 in complex with XLF. Similar to the complex between truncated XRCC4 (1-140) and XLF, SEC showed a broad signal, indicating that the XRCC4·XLF complex is also dynamic in nature (Figure 4A). Distinct peaks at 80Å and 160Å in the SAXS profile and P(r) function also indicate that complexes formed with full-length and truncated XRCC4 adopt a similar arrangement. Higher peak intensities in the higher protein concentration samples indicates the formation of longer oligomers that contain increasingly frequent repeat units at 80Å and 160Å (Figure 4C). Calculated SAXS envelopes reveal that the protrusion from the rod-like shape is consistent with the presence of the XRCC4 coil-coil region in the full length construct, however heterogeneity of the samples did not allow us to discern more detailed features (Figure S4A). Similar to the truncated complex, we performed MES analysis of the atomic models. Selected ensembles using a model in which the XRCC4·XLF-arrangement was identical to its truncated complex (Figure 4D) provided an excellent match (model 1, χ²=1.8) to the SAXS profile (Figure 4DE). The alternative model with parallel coil-coil regions gave a similar fit (model 2, χ²=1.7) with different ratios of oligomeric states in the selected ensemble (Figure S4B). The similarity between the fit of the two different models can be explained by the fact that the flexible XRCC4 C-terminus and the larger sample heterogeneity delivered SAXS profiles with fewer features. These profiles could be fitted with two different ensembles equally well (Putnam et al., 2007). However, the model ensemble in which the XLF head group interacted with the lower region of the XRCC4 head domain gave a poor fit (model 3, χ²=4.7) (Figure 4E).

XLF tethers XRCC4·BRCT and forms an extended filament

To determine the structural arrangements of the XRCC4·XLF·BRCT assembly, we examined the complex of full-length XRCC4, XLF and BRCT after re-purification on SEC-MALS. The SEC-MALS peak, where all three proteins co-elute (Figure 5A), shows a broad signal between 150 and 200kDa similar to that observed for XRCC4·XLF and indicates the dynamic nature of the complex. Analysis of the SAXS profile obtained for the SEC peak fraction showed concentration dependence. The diluted fraction (1mg/ml) (Figure 5 BC - magenta) shows an anisotropic particle with a D_max~280Å (Figure 5C).

Next, we performed a validation of the XRCC4·BRCT·XLF atomic models using a similar approach to that described above. First, we built a model based on the described XLF·XRCC4 interface (Figure 3D) and the XRCC4·BRCT model (Figure 2D). The selected ensemble fit the data very well, in particular, the distinct spacing peak at q~0.075Å^-1 (χ²=2.4) (Figure 5BE). The alternative model 2 gave a slightly worse fit (χ²=2.5) and model 3 fit the data poorly (χ²=14.5). The selected ensemble for the best fit model shows relatively little heterogeneity, with one component accounting for 72% of the entire ensemble (Figure 5D, model 1).

This atomic model yields an XRCC4·BRCT·XLF arrangement that can be superimposed with the SAXS envelope. The reconstructed envelope shows an elongated region, consistent with the presence of two XRCC4·BRCT complexes separated by a protrusion attributable to XLF (Figure 5D and S5). Together with analysis of the XRCC4·XLF complexes (Figure 3), this XRCC4·BRCT·XLF arrangement indicates that one XLF dimer tethers two XRCC4·BRCT complexes. The model also shows that XLF and XRCC4 are tilted relative to one other. Furthermore, XLF interacts with XRCC4 in a head-to-head fashion without introducing steric clashes with the LigIV-BRCT domain (Figure 4DE). The tilt angle between XRCC4·XLF may lead to an even larger separation of the two LigIV-BRCT tethered by a central XLF molecule. In this experimental model, an XLF dimer is predicted to bind two XRCC4 dimers and each XRCC4 dimer potentially may bind two dimers of XLF, thus generating filament-like assemblies containing alternating XLF and XRCC4 dimers. Our data shows that these filament-like structures are present in solution at high protein concentrations, mimicking the effective concentrations expected for proteins tethered at a DNA double-strand break.

Protein purification

Proteins were cloned, expressed in bacteria and purified as described in Supplementary Material. Purified proteins used in this study were full-length XRCC4 (aa 1-344), XRCC4 1-140, XRCC4 1-161, XLF 1-248 and DNA LigIV 618-911.

Size-Exclusion Chromatography (SEC) and Multi-Angle laser Light Scattering (MALS)

MALS experiments were performed in-line after SEC Chromatographic separations as described in Supplementary Material. The light scattering experiments were used to perform analytical scale chromatographic separations for mass determination of the principle peaks in the SEC analysis.

Data collection and evaluation

SAXS data were collected at the ALS beamline 12.3.1 (SIBYLS) LBNL Berkeley, California (Hura et al., 2009). Wavelength λ=1.0 Å and the sample-to-detector distances set to 1.5 m resulting in scattering vectors, q, ranging from 0.001 Å^-1 to 0.32 Å^-1. The scattering vector is defined as q = 4π sinθ/λ, where 2θ is the scattering angle. All experiments were performed at 20°C and data was processed as described (Hura et al., 2009). The experimental SAXS data for different protein concentrations were investigated for aggregation using Guinier plots (Guinier and Fournet, 1955). The radius of gyration R_G was derived by the Guinier approximation I(q) = I(0) exp(-q²R_G²/3) with the limits qR_G<1.3. The program GNOM (Svergun, 1992) was used to compute the pair-distance distribution functions, P(r). This approach also provided the maximum dimension of the macromolecule, D_max. The overall shapes were restored from the experimental data using the program DAMMIF with P1 symmetry operator. (Franke and Svergun, 2009). Ten bead models obtained for each SAXS profile were averaged by DAMAVER (Volkov and Svergun, 2003) to construct the averaged model representing general structural features. Bead models were converted to volumetric format.

Atomic model building

In our rigid body modeling strategy BILBOMD, molecular dynamics (MD) simulations were used to explore conformational space adopted by XRCC4 C-terminus. The crystal structures of XRCC4 (PDBid: 1fu1) and the XRCC4·BRCT complex (PDBid: 3ii6) were used to construct the initial atomic models. The initial models were constructed by connecting XRCC4 or XRCC4·BRCT with predicted helical stacks identified by secondary structure prediction (Figure S1B). These atomic models were used as templates for rigid body modeling. For each registered conformation, the theoretical SAXS profile and the corresponding fit to the experimental data were calculated using the program FoXS (Schneidman-Duhovny et al., 2010). The flexible C-terminus in the XRCC4 and XRCC4·BRCT structures was represented by an ensemble containing 2 different conformations. The scattering from such an ensemble is readily computed by averaging the individual scattering patterns from the conformers. To select an appropriate minimal ensemble from pool of ~10000 MD - generated conformations, Minimal Ensemble Search (MES) was employed (Pelikan et al., 2009). The scattering curves from all the structures in the MD – pool were first pre-computed and the subsequent genetic algorithm-selection operators were performed using these patterns and not the structures (Bernado et al., 2007). The final model was calculated to best fit the experimental curve I(q)_experiment minimizing the discrepancy χ² between the experimental and calculated multiconformational profile. Due to the dynamic character of XRCC4, XRCC4·XLF and XRCC4·BRCT·XLF the samples represented an equilibrium of the different oligomerization states and MES was used to identify the fraction ratios of the components required to best fit the experimental data (Pelikan et al., 2009). The multicomponent scattering I(q) from such a minimal ensemble was computed by averaging the individual scattering profiles from conformers:

I_1,2,3…N (q) were the scattering profiles from the single oligomers and the momentum transfer. A MES based search was used to select an appropriate ensemble of oligomers from a pool of all generated oligomers (homodimer, tetramer, hexamer, octamer, decamer). The final model achieved the best fit to the experimental curve I(q)_experiment by minimizing the discrepancy χ² between the experimental and calculated multi-component curve. Comparison of the structural properties of the selected oligomers (two for XRCC4; three for XRCC4·XLF and XRCC4·XLF·BRCT) in the ensemble subset allowed us to distinguish the degree of heterogeneity of the experimental system.