The Cerebro-oculo-facio-skeletal Syndrome Point Mutation F231L in the ERCC1 DNA Repair Protein Causes Dissociation of the ERCC1-XPF Complex*

Background: An F231L mutation in the DNA repair complex ERCC1-XPF causes severe disorders. Results: The observed enhanced dissociation of the mutant ERCC1-XPF helix-hairpin-helix dimer relates to an altered orientation for the interaction anchor Phe894 of XPF near Phe231 of ERCC1. Conclusion: The local structure around Phe231 is essential for complex stability. Significance: Studying mutations interfering with DNA repair is crucial to understand endonuclease-related diseases. The ERCC1-XPF heterodimer, a structure-specific DNA endonuclease, is best known for its function in the nucleotide excision repair (NER) pathway. The ERCC1 point mutation F231L, located at the hydrophobic interaction interface of ERCC1 (excision repair cross-complementation group 1) and XPF (xeroderma pigmentosum complementation group F), leads to severe NER pathway deficiencies. Here, we analyze biophysical properties and report the NMR structure of the complex of the C-terminal tandem helix-hairpin-helix domains of ERCC1-XPF that contains this mutation. The structures of wild type and the F231L mutant are very similar. The F231L mutation results in only a small disturbance of the ERCC1-XPF interface, where, in contrast to Phe231, Leu231 lacks interactions stabilizing the ERCC1-XPF complex. One of the two anchor points is severely distorted, and this results in a more dynamic complex, causing reduced stability and an increased dissociation rate of the mutant complex as compared with wild type. These data provide a biophysical explanation for the severe NER deficiencies caused by this mutation.

UV light, ionizing radiation, mutagenic compounds, and natural cellular metabolism can damage DNA and lead to mutations causing cancer or cell death and senescence contributing to aging-related disorders (1). As a countermeasure, the cell has developed a set of complementary repair mechanisms, one of which is the nucleotide excision repair (NER) 3 pathway (2,3). This multistep "cut-and-patch" repair system deals with the broad class of helix-distorting lesions, including UV-induced DNA lesions and numerous bulky chemical adducts and intrastrand cross-links. Two proteins involved in this repair process, ERCC1 (excision repair cross-complementation group 1) and XPF (xeroderma pigmentosum complementation group F) form a heterodimeric complex that functions as a structurespecific DNA endonuclease responsible for the 5Ј incision when the lesion is excised from the DNA as a 22-30 oligonucleotide in a late step of the NER reaction (4,5). There is considerable evidence for additional NER-independent roles of the ERCC1-XPF heterodimer in homologous recombination (6 -9), repair of interstrand DNA cross-links (9,10), telomere length maintenance (11)(12)(13), double strand break repair (14), and gene conversion (9,15).
The proper functioning of the ERCC1-XPF complex as an endonuclease in NER critically depends on the formation of a heterodimer via the proteins' C-terminal tandem helix-hairpinhelix (HhH) 2 domains (16 -19). This contact region of the dimer is not only important for the interaction between the monomers but also plays a role in DNA binding; the C-terminal domain of ERCC1 is essential for double-stranded DNA (dsDNA) binding (18,20,21), whereas the C-terminal (HhH) 2 domain of XPF can bind single-stranded DNA (ssDNA) (22). We proposed that this would position the ERCC1-XPF complex at an ss/dsDNA junction and place the catalytic module of XPF, which is adjacent to its (HhH) 2 domain, at the 5Ј site at some distance from the DNA lesion (22).
In classic cases of XPF deficiency, the DNA repair function is still partly retained, and UV sensitivity is only moderately increased (23)(24)(25). In recent years, however, several more severe cases of XPF-related deficiencies have been described (10, 26 -29). For instance, it was shown that mutations in XPF can cause Fanconi anemia (29). Mutational studies on a mouse homolog of human ERCC1 have shown the N terminus of ERCC1, despite being dispensable for DNA repair, to be important for heterodimer expression (7,30). Inherited deficiencies in ERCC1 are rare in humans (28,31,32), probably reflecting the lethality of many genetic mutations in this protein, which became apparent also from studies in mice perinatal mortality (33). The similarity of these symptoms to those resulting from XPF defects in mice, combined with the biochemical and cellular parallels, supports the notion that ERCC1 and XPF function as an obligate complex in vivo. Only two cases involving ERCC1 defects have been reported in humans, having in common a F231L point mutation that causes one of the severest cases of NER deficiency known until now. One case (165TOR) is a biallelic heterozygous mutation, one of which is F231L, which leads to severe degeneration of the brain and spinal cord, diagnosed as cerebro-oculo-facio-skeletal syndrome, and death at early infancy (31,34). The other case of ERCC1 deficiency is a biallelic homozygous F231L mutation, resulting in an early onset of Cockayne syndrome and the patient's death at early infancy. Both cases link the F231L mutation to NER deficiency, where the ERCC1-XPF complex is still present, but in reduced quantities, resulting in slower repair rates (28,31).
Phe 231 is conserved in ERCC1 of invertebrates and mammals. The residue is located in the C-terminal (HhH) 2 domain of ERCC1, and in the wild type protein structure, it has extensive interactions with the (HhH) 2 domain of XPF (16,18,20,21). Therefore, in principle, the loss of ERCC1-XPF functionality in the F231L mutant could be due to ERCC1 instability and/or the reduced stability of the entire ERCC1-XPF complex. Furthermore, because the phenylalanine is in close proximity to the hairpin regions of the ERCC1 (HhH) 2 domain that are important for its DNA binding role, the mutation could also affect the DNA binding properties of the ERCC1-XPF complex. In order to explore these possibilities, we produced and purified the ERCC1-XPF (HhH) 2 complex containing the F231L mutation and compared several structural and functional properties with those of wild type ERCC1-XPF. We show that whereas the mutant heterodimer preserves the overall fold of the ERCC1-XPF (HhH) 2 complex as well as its DNA binding properties, the mutation causes a side chain reorientation for residue 231 that disrupts interactions with nearby amino acids of XPF. We show that this small, local disturbance in the ERCC1-XPF heterodimer leads to a lower stability of the complex due to an increased dissociation rate.

Experimental Procedures
Cloning, Protein Expression, and Purification-The construct for the F231L ERCC1 (HhH) 2 mutant was prepared by using an extended mutagenic 5Ј primer during PCR amplification with the wild type construct as a template. Subsequently, F231L ERCC1 was cloned into the BamHI-XhoI site of the pET28b bicistronic expression vector (18), substituting the wild type ERCC1 coding region. The construction was verified by DNA sequencing. The final construct consists of F231L ERCC1(220 -297), followed by a C-terminal flexible linker and His 6 tag GGLEHHHHHH, and XPF(823-905). The plasmid was transformed into the Escherichia coli Rosetta (DE3) strain, overexpressed, and purified using an Ni 2ϩ -nitrilotriacetic acid (NTA)-agarose column, followed by gel filtration as described for wild type ERCC1-XPF (18). During NMR analysis, it was found that the expressed protein was a 1:1 mixture with a longer than expected protein, probably due to the presence of a rare alternate start codon (35,36) next to the constructed one, which resulted in seven additional amino acids at the N terminus of ERCC1 (RIRRRYN). This alternative initiation codon was present in both wild type and mutant and did not interfere with DNA binding or heterodimer stability as judged by NMR.
Surface Plasmon Resonance-DNA binding to the F231L ERCC1-XPF (HhH) 2 heterodimer was measured by surface plasmon resonance (SPR) as described before (37). All solutions for SPR were filtered through a 0.22-m filter and degassed. The experiments were performed at 12°C in 10 mM Hepes, pH 7.5, 50 mM NaCl, and 0.005% (w/v) Tween 20 (SPR buffer) at 10 l/min using a Biacore X system (Biacore AB, Uppsala, Sweden). The mutant and wild type ERCC1-XPF (HhH) 2 complexes were exchanged to the SPR buffer using Zeba Spin Desalting Columns (Thermo Scientific). Protein concentrations were determined by UV spectroscopy using a Nanodrop instrument (Thermo Scientific) assuming ⑀ 280 ϭ 2980 mM Ϫ1 cm Ϫ1 . At low concentrations, it was additionally estimated using SDS-PAGE. Low binding tubes and tips were used to prevent the loss of sample during handling. Mutant or wild type ERCC1-XPF (HhH) 2 complex (0.05 M), in the absence or presence of double strand "bubble10" (i.e. annealed DNA GGGCG-GCGGGT 10 GGCGGGGCGG and CCCGCCGCCCT 10 CCG-CCCCGCC) with concentrations ranging from 0.01 to 100 M using a dilution factor of 2, was incubated in the SPR buffer for 20 min on ice and injected over the NTA CM5 sensor chip. DNA association was monitored for 60 s, and the dissociation was monitored for 120 s. Before each SPR experiment, Ni 2ϩ was immobilized on the second flow cell of the sensor chip (Biacore AB) while the first flow cell was used as a reference surface. Between consecutive injections, the chip was regenerated with 10 l of 0.25 M EDTA in 3.5 M guanidine, pH 8.0. Data were processed using Biaevaluation version 3.2 software. To calculate the apparent dissociation constant (K d app ) of the DNA binding to the ERCC1-XPF (HhH) 2 heterodimer, the relative amount of DNA-free protein was divided by the total DNA concentration and fitted according to a simple 1:1 binding model using the software Prism (GraphPad Software Inc.). Because the DNA prevented His tag binding of the ERCC1-XPF (HhH) 2 domains on the Ni 2ϩ -NTA surface, the relative amount of DNA-free protein was determined as response values at the end of the loading time of 60 s (R 60 ) divided by the R 60 value of the ERCC1-XPF (HhH) 2 heterodimer in the absence of DNA (fraction ϭ R 60 [ERCC1-XPF] free /R 60 [ERCC1-XPF] total ). The real-time monitoring of the interaction of ERCC1 with XPF was performed in a flow cell maintaining the laminar flow (Biacore AB) and with a Sensor Chip Ni 2ϩ -NTA (Biacore AB) (37). All experiments were performed at a flow rate of 20 l/min in an SPR buffer containing 50 or 150 mM NaCl. A set of experiments was conducted at three concentrations of wild type and mutant ERCC1-XPF (HhH) 2 heterodimer (0.2, 0.1, and 0.05 M) and three temperatures (12, 20, and 25°C). Each experiment was executed twice. The association of the ERCC1-XPF complex on the chip and the dissociation from the chip (expressed as the reciprocal of time) was monitored for 60 and 120 s, respectively. Identical amounts of proteins were loaded on the chip without preloaded Ni 2ϩ in order to subtract the baseline from each obtained sensogram before analysis. SPR curves were fitted as a single-phase decay model over 60 -180 s using the software GraphPad Prism using the following equation.
Here, R is the response in arbitrary response units, R 0 is the maximum response after association (60 s), P is the plateau at infinite times, and k off app is the sum of the off-rate of the XPF dissociation from His-tagged ERCC1 and of the His-tagged protein from the chip. The SPR chip performance was optimized using different protein concentrations combined with different Ni 2ϩ concentrations and different flow rates, based on the findings of Nieba et al. (37). The protein complex was loaded on the chip with a 12-25°C temperature gradient, and the dissociation was followed for 1000 s. No reliable data could be obtained above 25°C due to elevated dissociation of Histagged protein from the chip. The baseline for the buffer was subtracted.
Circular Dichroism-CD spectra were measured using a JASCO J-810 spectropolarimeter as described before (38,39). The thermal unfolding of wild type and mutant ERCC1-XPF-His 6 was measured in the far-UV wavelength range (190 -250 nm) under identical conditions (CD buffer: 50 mM NaPO 4 , 100 mM NaCl, pH 7.0; protein concentrations, about 25 M; cell path length, 1 mm). Ten 10-s scans were recorded, and the temperature was increased from 15 to 85°C in increments of 1°C/min. The signal intensities at 222 nm were collected for each temperature and analyzed as described previously (38).
Thermal Shift Assay-Thermal shift experiments were performed in 96-well thin wall PCR plates sealed with optical-quality sealing tape. The presented transition temperatures are the averages of three independent experiments performed at different protein concentrations. Each sample contained 5 g of protein in 25 l of buffer containing 2.5 l of 100ϫ diluted Sypro Orange (Molecular Probes, Life Technologies, Inc.). The fluorescence changes were followed on an iCycler iQ PCR detection system (Bio-Rad) from 15 to 95°C in increments of 0.2°C/min by monitoring Sypro Orange fluorescence and analyzed (40). The wavelengths used for excitation and emission were 490 and 575 nm, respectively.
For NMR temperature series, two-dimensional 1 H, 15 N HSQC spectra were measured at temperatures ranging from 280 to 333 K using 400 M 15 N-labeled wild type and mutant ERCC1-XPF in NMR buffer. In total, 15 two-dimensional 1 H, 15 N HSQC spectra were recorded at 750 MHz, and chemical shift changes were analyzed as described before (45,46).
Proton-deuterium exchange of 15 N-labeled mutant and wild type ERCC1-XPF (HhH) 2 heterodimer was measured at 750 MHz in 100 mM NaCl, 50 mM NaP i , pH 6.8 (meter reading) in D 2 O. Prior to monitoring H/D exchange, the acquisition parameters were adjusted with wild type ERCC1-XPF prepared in NMR buffer (47). Dissolving the lyophilized sample in 2 H 2 O, transferring to the NMR tube, transfer to the NMR instrument, and temperature equilibration resulted in a total preparation time of 8 min. Thereafter, the acquisition of series of two-dimensional 1 H, 15 N HSQC spectra for following the H/D exchange was started. For each residue, the decay of the signal intensities as a function of time was evaluated as described previously (48).
Phase-modulated clean chemical exchange (CLEANEX-PM) experiments were recorded at 600 MHz as explained by Hwang et al. (49). Signal intensities were obtained from two-dimensional (CLEANEX-PM)-fast HSQC spectra recorded with a mixing time of 100 ms and for reference compared with a control fast HSQC spectrum as described previously (41).
NMR structures were calculated using CYANA version 3.0 (50, 51). Distance restraints were derived from NOESY peak intensities. Torsion angle restraints and were derived from H N , H␣, 15 N, 13 C␣, 13 C␤, and 13 CO chemical shifts using TALOSϩ (52,53) with an extra error of Ϯ10°. In each CYANA cycle, 100 structures were generated and energy-minimized using 10,000 steps of simulated annealing. After initial CYANA runs, additional NOE assignments were obtained by manual inspection of the NOESY spectra. Before water refinement, CYANA-derived NOE and hydrogen bond restraints, in addition to the torsion angle restraints derived from TALOSϩ (52,53), were used to calculate 100 structures with CNS (54,55). Each of these structures was subsequently refined in explicit water using restrained molecular dynamics using the RECO-ORD protocol (55). The 20 lowest energy conformations were validated using WHATIF, WHATCHECK, and iCing (56 -58). Molecular images were generated using PyMOL (59,60).
For the wild type ERCC1-XPF (HhH) 2 heterodimer domain, we observed inconsistencies in the precise length of helix ␣. Whereas the helix length for the F231L ERCC1-XPF mutant was consistent with that of the crystal structure of wild type ERCC1-XPF (Protein Data Bank code 2A1J), this helix was shorter in the NMR structure of wild type ERCC1-XPF (Protein Data Bank code 1Z00). Therefore, the NMR structure of wild type ERCC1-XPF was recalculated following the same protocol as for the mutant. For this, the previously determined NOEbased restraints and restraints for hydrogen bonds (18) were combined with and torsion angle restraints derived from TALOSϩ (52,53) for the calculation of an ensemble of 100 conformers with CNS (54,55). Water refinement, validation, and generation of molecular images were as for F231L ERCC1-XPF.

Results
Similar to the heterodimeric complex of the C-terminal (HhH) 2 domains of wild type ERCC1-XPF (18), the system containing the F231L mutation could only be overexpressed and purified as an obligate complex using a bicistronic expression vector. During size exclusion chromatography, there was no indication of dissociation, arguing that F231L ERCC1-XPF can form a stable heterodimeric (HhH) 2 complex, similar to the wild type situation.
DNA Binding-In wild type ERCC1, the helices of the first HhH motif make extensive contacts with the dsDNA minor groove near the ss/dsDNA junction (22). Therefore, the F231L mutation, located in this first helix, could well interfere with DNA binding of ERCC1-XPF. It is known that bubble-shaped DNA substrates containing ss/dsDNA junctions efficiently bind wild type ERCC1-XPF heterodimer (18,21). One of the highest affinity DNA probes appears to be "bubble10," where two double-stranded GC-rich segments flank a region of 10 unpaired thymine bases. Using an electrophoretic mobility shift assay (EMSA), we found that both wild type and F231L mutant ERCC1-XPF bind to DNA with comparable affinity (data not shown). However, we found that the relatively weak DNA binding of ERCC1-XPF can be more accurately analyzed using SPR. For the immobilization of the ERCC1-XPF complex to the sensor chip, we made use of the His-tagged ERCC1, because alternative coupling techniques using present protein cysteines, lysines, or the N terminus could lead to random orientation of the complex on the chip or interference with DNA binding. We have compared the ability of F231L and wild type ERCC1-XPF to bind bubble10 DNA. For this, wild type or mutant heterodimer were immobilized on Ni 2ϩ -NTA chips via the C-terminal Histidine-tag of ERCC1 followed by loading of bubble10 DNA. The addition of bubble10 DNA caused dissociation of the heterodimer from the chip, which allowed to determine the fraction of free heterodimer as a function of the DNA concentration. The recorded SPR responses (Fig. 1, A and B) of wild type and mutant complexes upon DNA binding were subsequently used to calculate the dissociation constants (K d ), as described in detail under "Experimental Procedures." Both were determined to be about 0.2 M (i.e. within error the same) (Fig. 1C). These DNA binding studies show that impaired DNA binding cannot be the explanation for the malfunctioning of nucleotide excision repair as seen for the patients with the ERCC1-XPF F231L mutation.
Biophysical Characterization-Next, the effect of the F231L mutation on the stability of the ERCC1-XPF (HhH) 2 complex was analyzed. We compared the temperature-dependent behavior of F231L and wild type ERCC1-XPF (HhH) 2 heterodimers using a thermal shift assay (ThermoFluor) (40), which measures the increase in Sypro Orange fluorescence upon exposure of hydrophobic patches during protein unfolding. The assay is commonly used to follow the unfolding of a protein with increasing temperatures, but it can also probe the dissociation of a protein complex. Previously, we studied the temperature stability of wild type ERCC1-XPF and homodimer (XPF) 2 using CD spectroscopy (38). In these studies, we found that ERCC1 dissociates from ERCC1-XPF and unfolds at a temperature of ϳ50°C and that at the same time, XPF converts into a homodimeric (XPF) 2 . Subsequently, by further increasing the temperature, the (XPF) 2 homodimer unfolds at ϳ75°C. In the ThermoFluor experiment of ERCC1-XPF, also two transitions upon temperature increase were found (Fig. 1D), similar to what was observed previously using CD spectroscopy (38). As before, the first transition at ϳ50°C is interpreted as the dissociation of the ERCC1-XPF heterodimer followed by ERCC1 denaturation, and the second transition at ϳ75°C is explained as the unfolding of subsequently formed XPF homodimers. However, we noted a significant difference in the first transition temperatures between the wild type and F231L complexes, with onsets of dissociation at 52.0 Ϯ 0.5 and 47.5 Ϯ 1.0°C, respectively. Although the difference in transition temperatures for the heterodimers can be interpreted as directly caused by a difference in unfolding of mutant versus wild type ERCC1-XPF, it could also be the result of structural distortions in ERCC1, in its binding partner XPF, and/or by enhanced dissociation of ERCC1-XPF, which could all lead to increased exposure of hydrophobic surfaces.
Far-UV circular dichroism (CD) directly probes changes in secondary structure formation, and ellipticity at 222 nm is indicative for the helical content of proteins (61). For both mutant and wild type ERCC1-XPF, again a biphasic behavior was observed upon increasing temperature, similarly as previously observed for the wild type heterodimer (Fig. 1E) (38). Also here, the first transition (at a T m of 51.0 Ϯ 1.0°C) reflects heterodimer dissociation followed by ERCC1 (HhH) 2 unfolding, whereas the XPF (HhH) 2 domain remains folded and forms a homodimer (38). The high temperature transition (with a T m of 76.0 Ϯ 1.0°C) reflects XPF homodimer dissociation and subsequent unfolding. Comparing the change in ellipticity as a function of temperature for the F231L and wild type ERCC1-XPF complex, we observe no evident differences in the first transition between the two proteins. The second transition is also very similar for both complexes, which can be expected because in both cases it probably involves the unfolding of the same XPF homodimer. Whereas the CD spectra show that the ERCC1 and XPF secondary structure elements of both wild type and mutant ERCC1-XPF start to unfold at similar temperatures, the ThermoFluor analysis shows that the mutant heterodimer is influenced already at lower temperature than wild type. Because CD monitors the unfolding of helices, whereas the ThermoFluor analysis can also signal dissociation of the heterodimer that precedes the unfolding of the ERCC1 monomer, the observed differences are indicative for the increased dissociation of ERCC1-XPF (HhH) 2 heterodimers because of the F231L mutation.
Association and dissociation of proteins forming complexes can be studied using SPR (62). To study the dissociation of the wild type and mutant ERCC1-XPF complexes directly, they were immobilized via the C-terminal histidine tag of ERCC1 on Ni 2ϩ -NTA chips. The increase in SPR response upon loading the ERCC1-XPF complex shows that the complex can be successfully loaded within 50 s (Fig. 1F). Upon buffer elution, a drop of the SPR response is observed until it reaches a plateau at about 50% of the initial response. In theory, the SPR decay could be due to a 50% loss of the ERCC1-XPF complex from the NTA chip. Alternatively, given the similar molecular weights of XPF and ERCC1, this reduction can also be explained by the dissociation of XPF from ERCC1, where His-tagged ERCC1 remains bound to the Ni 2ϩ -NTA chip. Because the F231L mutation will affect this SPR decay, as we will see below, dissociation is the more likely interpretation. Upon continued elution for 1000 s, we observe the return of the SPR response to its starting level due to dissociation of His-tagged ERCC1 from the Ni 2ϩ -NTA chip (not shown).
The dissociation of wild type and mutant ERCC1-XPF was followed by SPR at different heterodimer concentrations, temperatures, and salt concentrations (cf. Fig. 1F), as described in more detail under "Experimental Procedures." Negative response at temperatures above 25°C suggests that the SPR results are biased by ERCC1 precipitation. Data could be reliably fitted with single exponential decays, although it was evident that also slower processes contributed to the SPR response. In multiple SPR responses, a second decay with k off app of ϳ3 ϫ 10 Ϫ3 s Ϫ1 was identified, corresponding to the slow release of His-tagged ERCC1 monomer and ERCC1-XPF from the chip. At 25°C and lower temperatures, at lower ionic strength, the dissociation rates k off app for wild type and mutant seem only marginally different (Table 1). For example, at 12°C and low ionic strength, the k off app for the F231L mutant was 1.9 Ϯ The stability of ERCC1-XPF (HhH) 2 is measured using the Sypro Orange Thermofluor assay (40). Note that the concentration of mutant ERCC1-XPF and thus also XPF homodimer was slightly higher than that of wild type, causing an apparent higher transition temperature for XPF homodimer at ϳ75°C. E, thermal stability of wild type (Ⅺ) and mutant (OE) ERCC1-XPF measured by CD spectroscopy. F, dissociation of ERCC1-XPF measured by SPR at 150 mM NaCl and 12°C. The graphs show the dissociation of XPF from immobilized wild type ERCC1 (data in black, fit in blue) and immobilized F231L ERCC1 (data in gray, fit in red). RU, arbitrary response units.
0.3 ϫ 10 Ϫ2 s Ϫ1 , whereas for wild type ERCC1-XPF, k off app was 1.3 Ϯ 0.2 ϫ 10 Ϫ2 s Ϫ1 . At higher ionic strength, the k off app for the wild type ERCC1-XPF is similar to the k off app at 50 mM NaCl, with a value of 1.43 Ϯ 0.03 ϫ 10 Ϫ2 s Ϫ1 , whereas for the F231L mutant, the dissociation rate k off app significantly increases to 3.0 Ϯ 0.3 ϫ 10 Ϫ2 s Ϫ1 .
NMR Spectral Differences between Wild Type and F231L ERCC1-XPF-NMR assignments for the F231L mutant ERCC1-XPF (HhH) 2 complex were obtained using standard triple resonance techniques and analysis of three-dimensional NOESY spectra, whereas those for the wild type complex were already available (18). The two-dimensional 1 H, 15 N HSQC spectra of the wild type and F231L mutant ERCC1-XPF (HhH) 2 complexes are similar but do show differences, both in chemical shift and in intensity (Figs. 2 and 3). Comparing the chemical shifts (Fig. 3, A and C) it is noted that differences are mostly close to the F231L mutation site (i.e. for ERCC1 helix ␣ Leu 227 , Gln 229 , Asp 230 , Val 232 , Ser 233 , and Val 235 ), but significant chemical shift differences are also found for residues farther away, both in the ERCC1 and XPF (HhH) 2 domains. Among these distant residues are Met 224 , Glu 237 , Thr 252 -Leu 260 , Leu 263 , Arg 268 , and Gly 278 in ERCC1 helices ␣ and ␥ and Phe 894 -Ser 899 in the XPF C-terminal helix. Whereas the local chemical shift differences could be merely the effect of the different chemical nature of Phe 231 and Leu 231 , the distant differences may reflect structural differences, which could influence the association and dissociation behavior of the ERCC1-XPF complex. Fig. 3C shows that the observed chemical shift differences are most pronounced for residues that locate in the wild type structure at the ERCC1-XPF interface, suggesting that the monomeric folds of ERCC1 and XPF are very similar  in both complexes and that the mutation mainly affects contacts between ERCC1 and XPF. Intensity differences between the two-dimensional 1 H, 15 N HSQC spectra of wild type and F231L mutant ERCC1-XPF were noted as well (Fig. 3B). The graph shows a plateau that coincides with regions that in the wild type ERCC1-XPF (HhH) 2 complex have stable secondary structure. F231L ERCC1-XPF shows reduced intensity for residues that are directly contacting the mutation site in the wild type structure (i.e. for residues in helices ␣, ␤, and ␥ of ERCC1 (Gln 229 , Asp 230 , Val 232 , Ser 233 , Thr 252 , Leu 253 , Thr 255 , Phe 257 , Gly 258 , and Ser 259 ) and for residues in the C-terminal helix of XPF (Thr 892 , Ser 893 , and Phe 894 )). In fact, lack of signal intensity might be the reason that the amide proton of Leu 231 could not be unambiguously assigned. The differences between the wild type and F231L ERCC1-XPF HSQC signal intensity probably reflect increased local dynamics in F231L ERCC1-XPF at the mutation site.
Amide Proton Exchange-The interface of the complex of the (HhH) 2 domains of wild type ERCC1-XPF is formed by hydrophobic interactions involving several aromatic phenylalanine residues (of which two important ones are the pseudosymmetric anchors Phe 293 (in ERCC1) and Phe 894 (in XPF)) and several intermolecular hydrogen bonds (Ser 259 HG⅐⅐⅐O Thr 892 , Glu 261 HN⅐⅐⅐O Ile 890 , Phe 293 HN⅐⅐⅐O Lys 860 , Asn 834 HN⅐⅐⅐O Phe 293 , Asn 834 HD21⅐⅐⅐O Pro 292 , His 891 HD1⅐⅐⅐OE2 Glu 261 , and Ala 895 HN⅐⅐⅐O Gly 258 ) (18,21). The importance of hydrogen bonding for the stability of the ERCC1-XPF complex and the effect of the F231L mutation on this stability were studied by amide proton/ deuterium (H/D) exchange, similar as before for homodimeric XPF (HhH) 2 (38). For this, we acquired a series of two-dimen-sional 1 H, 15 N HSQC spectra of wild type and mutant ERCC1-XPF (HhH) 2 heterodimer freshly dissolved in 2 H 2 O (Fig. 4) and compared the H/D exchange for all residues that were still visible after the dead time of the experiment (i.e. 8 min) and were non-overlapping in the spectra of the wild type ERCC1-XPF heterodimer as well as the mutant F231L. A comparison of the H/D exchange could be made for residues Cys 238 , Thr 240 -Val 242 , Leu 253 , Thr 255 , Ser 259 -Glu 261 , Leu 263 , Ala 265 , Ser 267 , Leu 271 , Ala 272 , Ala 282 , Leu 285 , His 290 , and Leu 294 of ERCC1 and the residues Asp 839 , Leu 842 , Lys 847 , Met 856 , Val 859 , Asn 861 -Ala 863 , Leu 873 , Thr 874 , Ile 876 , Leu 877 , Ala 883 , and Gln 885 -Thr 892 of XPF. Overall, we note similar H/D exchange rates for wild type and mutant ERCC1-XPF (HhH) 2 heterodimer, with slow exchanging amide protons in the core of the complex and quick exchange for residues that are more surface-exposed. Only the ERCC1 residue Leu 253 of the F231L mutant shows significantly faster exchange than in the wild type, which suggests an increased solvent accessibility for helix ␤ of ERCC1 (Fig. 4C).
Whereas H/D exchange measures slow (minutes to hours) amide proton exchange (63), CLEANEX-PM experiments can probe fast (subsecond) solvent exchange (49). Also with CLEANEX-PM, no significant exchange differences were observed between the wild type and mutant ERCC1-XPF (data not shown). The only exception for which solvent exchange was reduced in the mutant is the amide proton of Val 235 in helix ␣ of ERCC1, which is four residues apart from the F231L mutation site and hydrogen-bonded to residue 231 in the wild type structure. If anything, this suggests even a slightly increased stability of helix ␣ in F231L ERCC1 as compared with the wild type system. The red horizontal line is the average of the signal intensity for residues in secondary structure elements of the wild type ERCC1-XPF (HhH) 2 heterodimer. C, projection of chemical shift differences on the structure of the wild type ERCC1-XPF (HhH) 2 heterodimer, representing residues with maximum perturbation as red broad ribbons, residues with medium perturbation as yellow broad ribbons, and residues with small to no difference in green. The F231L mutation site is represented as gray spheres.
In summary, between the wild type and mutant ERCC1-XPF (HhH) 2 heterodimers, differences in amide proton chemical shifts and in NMR intensities are seen for regions in direct contact with the mutation site and for residues localized near the Phe 894 anchor of the interface of ERCC1-XPF. The solvent exchange experiments indicate that the integrity of the ERCC1 and XPF monomers is hardly affected by the F231L mutation. In line with the thermal stability analysis, this also indicates that the reduced stability of ERCC1-XPF that we observed is not caused by differences in secondary structure stability but is rather the result of different association or dissociation properties of the (HhH) 2 complex.
Temperature-dependent Chemical Shift Changes-Temperature-dependent chemical shift changes for protein amide protons in hydrogen bonds are usually linear as a function of temperature, and the ⌬␦/⌬T gradients can be related to hydrogen bond strengths (45,46). Slow amide proton exchange combined with a temperature coefficient more positive than Ϫ4.5 ppb/K indicates the presence of a hydrogen bond, whereas fast exchange combined with a temperature coefficient more negative than Ϫ4.5 ppb/K indicates the absence of a hydrogen bond. The residues with significant differences between wild type and mutant ERCC1-XPF (cf. Table 2) are mostly located in the direct vicinity of the mutation site or around the nearby anchor Phe 894 , indicating that the mutation causes mainly a local structural effect. However, overall, with only a few exceptions, the temperature coefficients are more negative for the F231L mutant than wild type ERCC1-XPF, suggesting a reduced overall hydrogen bond stability of F231L ERCC1-XPF.
NMR Structure-To see whether the increased dissociation of the F231L mutant ERCC1-XPF complex is caused by struc-  tural differences between the wild type and mutant, we set out to determine the NMR structure of the F231L ERCC1-XPF (HhH) 2 heterodimer and to compare it with existing structures of wild type ERCC1-XPF. To allow direct comparison, we recalculated the earlier reported wild type ERCC1-XPF (HhH) 2 heterodimer NMR structure with exactly the same parameters as the F231L mutant. The dispersion and cross-peak positions of the two-dimensional 1 H, 15 N HSQC spectra of the (HhH) 2 complexes of wild type and F231L ERCC1-XPF are rather similar, demonstrating that the wild type and F231L mutant complexes are both well folded and structurally not very different (Fig. 2). Of 180 residues of F231L ERCC1-XPF, the amide protons of 161 residues could be assigned. No assignments could be made for five residues of the ERCC1 C-terminal His 6 tag and a few N-terminal residues of both ERCC1 and XPF (HhH) 2 domains. This leaves only three other residues for which no amide resonances could be assigned, namely Leu 231 , Gln 280 , and Ala 880 . For the point mutation Leu 231 , it should be noted that although its amide nitrogen resonance was not found, most other atoms of Leu 231 could be assigned.
As pointed out above, the differences in chemical shifts between wild type and mutant (HhH) 2 complexes are relatively small and mainly localized around the mutation site. Based on the chemical shift values for both proteins, the overall arrangement of the ␣-helices is not altered by the mutation, and in the wild type as well as the mutant complex, residue 231 is present in an ␣-helix (Fig. 5). The N-terminal helix (␣) in wild type ERCC1, appears shortened, but this may be due to lack of data because the crystal structure shows that the segment Ala 220 -Thr 240 is also ␣-helical in wild type ERCC1-XPF (21). For the F231L mutant ERCC1-XPF studied here, helix ␣ spans from residue Asp 221 to Thr 240 .
The NMR structures of the wild type and F231L ERCC1-XPF (HhH) 2 heterodimeric complexes were determined using NOE-based distance restraints and chemical shift-based torsion angle restraints (Fig. 6A and Table 3). The total number of NOEs used to calculate the structure of the F231L mutant ERCC1-XPF is lower than that for wild type. Still, the overall quality of the wild type and F231L mutant ERCC1-XPF structures is similar. The secondary structure elements in the wild type and mutant structure ensembles are very similar. The helices ␣-⑀ of wild type ERCC1 (i.e. Met 224 -Thr 240 , Lys 247 -Phe 257 , Leu 260 -Ala 265 , Arg 268 -Ala 272 , and Pro 279 -His 290 , respectively) and the helices ␣Ј-⑀Ј of XPF (i.e. Glu 829 -Glu 831 , Pro 837 -Lys 843 , Ala 849 -His 858 , Ile 862 -Ala 866 , Gln 870 -Leu 877 , Ala 880 -His 891 , and Phe 894 -Lys 900 , respectively) are almost invariably present in all conformers of the ensemble of mutant ERCC1-XPF. However, some differences are evident from the NMR structures. For the wild type heterodimer, assignments are lacking for the N terminus of ERCC1, which obviously leads to a lower number of NOEs for this part of the protein. Therefore, as pointed out above, helix ␣ of wild type ERCC1 starts at Met 224 , whereas that of F231L ERCC1 starts at Asp 221 (Fig. 6A).
Although the two NMR ensembles of wild type and mutant ERCC1-XPF are well defined and the overall fold is the same, the structures show small local differences. The closest-to-mean root mean square deviation values for all heavy atoms are 1.35 Å for ERCC1 residues 227-294, 1.15 Å for the XPF core residues 831-896, and 1.29 Å for both regions together. This indicates small but genuine differences between the two structures and that the XPF part is more similar than the ERCC1 monomer.
After superimposing the backbones, interesting NOE-based differences are seen for the side chain orientations of the two complexes. For example, Fig. 2B shows selected NOE patterns for residues Gly 258 , Glu 261 , and Pro 828 comparing wild type and F231L ERCC1-XPF. The NOE intensities for residues Gly 258 and Glu 261 are clearly different for both complexes, consistent with local structural differences near the mutation point, whereas farther away from the mutation site (e.g. residue Pro 828 ), no structural effects occur. In wild type ERCC1-XPF, Phe 231 is embedded deeply in the core of the (HhH) 2 heterodimer structure (Fig. 6A). Its aromatic side chain and the backbone show short distances (Ͻ4 Å) to ERCC1 residues Leu 227 -Val 235 of the same helix, but the residue is also close to Ser 259 and Leu 260 of ERCC1 and to residues Phe 889 , Ile 890 , Thr 892 , Ser 893 , and Phe 894 of XPF. The aromatic side chain of Phe 231 in wild type ERCC1-XPF is packed between the aromatic rings of Phe 894 and Phe 889 , making a T-shaped arrangement with respect to them. The F231L mutation does not influence the integrity of helix ␣, despite a very different side chain orientation of residue Leu 231 as compared with Phe 231 in wild type (Figs. 6 (A-C) and 7). In the mutant, Leu 231 maintains similar distances and angles toward residues Leu 227 -Val 235 of ERCC1 and residues Phe 889 and Ser 893 of XPF. However, the mutation allows the side chain of Phe 894 to reorient. Consequently, at the ERCC1 region flanking the XPF C terminus, contacts are lost between residues Phe 894 , Gly 258 , Ser 259 , and Leu 260 , and the interactions between the segments Ile 890 -His 891 and Leu 260 -Glu 261 are weakened. The reduction in aromatic ring interactions probably explains part of the differences in chemical shifts for segments Leu 227 -Glu 237 and Asp 252 -Leu 263 of ERCC1 and the segment Thr 892 -Ser 899 of XPF (Fig. 3).
Due to the F231L mutation, the aromatic side chains of Phe 889 and Phe 894 cannot maintain their T-shaped arrangement with respect to residue 231 (Fig. 6, B and C). In the wild type complex, Phe 889 contacts residues Asp 230 , Phe 231 , Arg 234 , Val 235 , Cys 238 , and Leu 260 of ERCC1 and Met 844 , Pro 845 , the N-terminally located helix ⑀Ј (Gln 885 -Asp 888 ), and segment Ile 890 -Thr 892 of XPF. In the mutant, Phe 889 modestly changes its contacts with Leu 260 of ERCC1, whereas most interactions FIGURE 6. Solution structures of wild type and F231L ERCC1-XPF (HhH) 2 heterodimers. A, superimposed NMR ensembles of F231L and recalculated wild type ERCC1-XPF (HhH) 2 . The backbone of ERCC1 is colored blue (cyan, wild type; blue, mutant); the backbone of XPF is colored green (lemon yellow, wild type; dark green, mutant). The mutated residue Leu 231 is shown in red. Shown are a surface depiction and side chain organization for phenylalanines Phe 231 , Phe 889 , and Phe 894 of wild type ERCC1-XPF (HhH) 2 (B) and for residues Leu 231 , Phe 889 , and Phe 894 of F231L ERCC1-XPF (HhH) 2 (C). Backbones, Phe 889 , and Phe 894 are colored as in A; the side chain of Phe 231 /Leu 231 is colored in red, and the labels in gray depict residues surrounding Phe 231 /Leu 231 .
within XPF remain. The situation for the Phe 894 anchor, however, is very different. Whereas most contacts with the neighboring residues in the XPF Phe 894 interaction pocket (Thr 892 -Ser 899 ) are still present in the mutant, the intermolecular contacts with ERCC1 are strongly altered due to side chain reorientations for residues Gln 229 , Leu 231 , and Val 235 of helix ␣, residues Leu 254 and Phe 257 of helix ␤, and residues Glu 261 and Leu 263 of helix ␥. It was found that in addition to these differences in phenylalanine contacts, several other residues of the complex show side chain reorientations. Most of these residues (Glu 269 , Leu 273 , Asn 280 , Arg 284 , Tyr 833 , and Cys 852 ) are at the surface of the complex (Fig. 7), and the observed differences can be due to lack of NOEs. However, for the interface residues Lys 843 , Ile 862 , Leu 865 , and Tyr 887 , the story is different; whereas the backbones of the twenty structures are very well superimposed, the differences between the wild type and mutant consistently show considerable changes in side chain orientation (Fig. 7). Finally, the F231L mutation leads to not only a significant reduction in the intermolecular hydrophobic packing involving the Phe 894 anchor but also a slight increase (0.3 Ϯ 0.01 Å) of the distance between the backbone amide of XPF Phe 894 and the backbone carboxyl of ERCC1 Gly 258 , causing the hydrogen bond between the backbone amide of Ala 895 and the backbone carboxyl of Gly 258 to vanish completely. In addition, as a consequence of the reorientation of Glu 261 , the imidazolium-carboxyl hydrogen bond between His 891 and Glu 261 disappears. Also, the interactions between Ile 264 and XPF residue Ala 863 , between Leu 289 and Phe 840 , and between the side chain of Lys 843 and ERCC1 residues Glu 237 , Cys 238 , and Thr 241 of ERCC1 appear diminished. Overall, the small structural changes initiated by the F231L mutation lead to local, modest conformational effects in the direct vicinity of the mutation, which accumulate in a number of differences affecting the interface between ERCC1 and XPF.

Discussion
At the moment, the most severe cases of ERCC1-XPF deficiency are described for two patients with a F231L mutation in ERCC1 (28,31). Genetic and functional studies have revealed that one patient carries a null allele and a F231L allele, and the other patient carries a biallelic F231L mutation. This F231L ERCC1 mutant in complex with XPF retains enzymatic activity. Analogous to patients with XPF mutations, this phenotype is a consequence of low levels of ERCC1-XPF (28,31). Earlier structural studies of the wild type ERCC1-XPF complex have shown that the interface of ERCC1-XPF is composed of two pseudosymmetric hydrophobic cavities tightly embracing anchor residues and a third, large hydrophobic contact region (Fig. 8) (18,21). The first anchor is formed by the aromatic ring of ERCC1 Phe 293 , which fills the cavity composed of residues Leu 294 of ERCC1 and Tyr 833 , Asn 834 , Pro 837 , Gln 838 , Leu 841 , Met 856 , Asn 861 , and Ile 862 (helices ␣Ј and ␤Ј) of XPF. The second anchor is formed by the XPF residue Phe 894 aromatic ring, which fits into the cavity comprising ERCC1 residues Phe 231 , Val 232 , Val 235 , Leu 254 , Gly 258 , and Ser 259 (helices ␣ and ␤) and XPF residues Phe 889 , Thr 892 , Ser 893 , and Ala 895 . The interactions in both anchor regions are stabilized by multiple intermolecular and intramolecular hydrogen bonds. The third contact region lies in between both anchor regions and involves the pseudosymmetrically positioned ERCC1 helices ␣ and ␥ and XPF helices ␣Ј, ␥Ј, and ⑀Ј. In this region, the aromatic side chain of Phe 840 plays a major role for intermonomer contacts. ERCC1 residues Cys 238 , Leu 239 , Thr 241 , Leu 260 , Leu 263 , Ile 264 , Leu 285 , Val 288 , and Leu 289 surround Phe 840 , and their contacts with residues Gly 836 , Pro 837 , Gln 838 , Asp 839 , Leu 841 , Leu 842 , Lys 843 , Met 844 , and Ile 862 in the pseudosymmetrically positioned region of XPF support this interaction. Furthermore, the short ␥, ⑀, ␥Ј, and ⑀Ј helices show contacts between the side chains of Ile 264 and Leu 289 in ERCC1 and Ala 863 and Ala 866 in XPF and between Leu 260 in ERCC1 and Phe 889 in XPF. In addition, the backbone carbonyl of Lys 843 contacts the side chain guanidinium group of Arg 234 , whereas the NH 3 ϩ group of the Lys 843 side chain contacts the side chain carboxyl of Glu 237 (Fig. 7).
In wild type ERCC1, Phe 231 is located in helix ␣. It is part of the cavity containing anchor Phe 894 and completely embedded in the heterodimer core. The side chain of Phe 231 is located between the aromatic rings of Phe 894 and Phe 889 with stabilizing T-shaped arrangements (64). In the F231L mutant, the aromatic side chain of Phe 889 maintains its orientation, whereas residues Leu 231 and Phe 894 reorient their side chains toward the surface of the protein (Fig. 8B). This means not only that the interactions of the benzyl group of Phe 231 are missing in the Phe 894 hydrophobic cavity but also that the benzyl group of Phe 894 can no longer function as an anchor stabilizing the complex. In addition, the side chain of Phe 894 becomes exposed, and a smaller hydrophobic interaction surface is present in the mutant (especially concerning residues in helix ␤). This and the lack of aromatic ring interactions destabilize ERCC1-XPF by increasing the dissociation rate of the complex. The reduced hydrophobic character for the interactions within the F231L ERCC1-XPF complex interface implies that the electrostatic contributions to the interactions should relatively increase. Indeed, when increasing the ionic strength, the dissociation rate for wild type ERCC1-XPF did not change much (cf. Table 1), whereas it increased for mutant ERCC1-XPF. The higher dissociation rate for the mutant indicates that the interactions at the interface of mutant ERCC1-XPF are more electrostatic in nature than that of wild type ERCC1-XPF (65,66).
A comparison of wild type and mutant ERCC1-XPF (HhH) 2 complexes shows that the majority of structural differences is limited to regions in direct proximity of Leu 231 . These include the ␣ and ␥ helices of ERCC1 and the C-terminal helix of XPF. The amides surrounding Leu 231 also show a significant decrease in 1 H, 15 N HSQC signal intensity as compared with those surrounding Phe 231 in the wild type, and the signal of the amide of Leu 231 was even completely lacking. This loss of signal intensity can be explained by the occurrence of slow internal dynamics caused by millisecond motional effects of the Phe 894 aromatic ring. The observed structural differences, mainly at the interface, could lead to instability of the ERCC1-XPF complex. Our stability analysis of the ERCC1-XPF (HhH) 2 domains using far-UV CD spectroscopy and ThermoFluor analysis shows that the lower temperature stability of the F231L ERCC1-XPF complex is not due to a reduced intrinsic ERCC1 and XPF stability but rather to a faster dissociation of the heterodimer. Indeed, using SPR, we could directly demonstrate that this dissociation is significantly faster for the mutant than for the wild type complex.
In conclusion, the replacement of Phe 231 by leucine appears to have only a small effect on the ERCC1 and XPF monomer structures and in vitro stabilities and to have no direct functional effect on the DNA binding affinity of ERCC1-XPF at an ss/dsDNA junction. The (HhH) 2 heterodimer is still formed, which agrees with the fact that ERCC1-XPF-based DNA repair is still partially functional in patients (28). The mutation influences the hydrophobic interactions between ERCC1 and XPF in the heterodimer, causes faster dissociation of the ERCC1-XPF heterodimer, and in this way reduces the ERCC1-XPF complex stability. With one of the two anchor points disrupted, this increased dissociation of the mutant F231L heterodimer can directly result in a diminished amount of ERCC1-XPF complex. In the cellular context, the separate subunits are unstable because of increased vulnerability to proteolysis. One reason for the intrinsic instability of the individual ERCC1 and XPF molecules might be that uncontrolled DNA binding and endonuclease activities of the separate subunits are a threat to regular DNA metabolism. The lower cellular concentration of the ERCC1-XPF repair complex will reduce the DNA repair capacity and may also lead to the build-up of repair intermediates that wait for the rate-limiting ERCC1-XPF step. This may also disrupt DNA metabolism, leading to enhanced cell death and FIGURE 7. Side chain orientations in wild type and F231L ERCC1-XPF (HhH) 2 heterodimer. Side chains of residues showing the largest differences between the NMR ensembles are depicted for the recalculated complex of wild type ERCC1 (HhH) 2 (top) and XPF (bottom) (A) and the calculated complex of F231L mutant ERCC1 (HhH) 2 (top) and XPF (bottom) (B). Backbone coloring is as in Fig. 6A. Amino acid side chains are shown in yellow, with polar hydrogen in gray, oxygen in red, and nitrogen in blue. The monomers of ERCC1 and XPF are separated for a better view. cellular dysfunction, contributing to the premature aging and cancer predisposition of mutants in the ERCC1-XPF complex both in humans and mice. This is most dramatically illustrated by the very severe Cockayne syndrome-like features that are exhibited by the ERCC1-deficient patients.