Characterization of Reaction Intermediates of Human Excision Repair Nuclease*

Nucleotide excision repair in humans is a complex reaction involving 14 polypeptides in six repair factors for dual incisions on either sides of a DNA lesion. To identify the reaction intermediates that form by the human excision repair nuclease, we adopted three approaches: purification of functional DNA·protein complexes, permanganate footprinting, and the employment as substrate of presumptive DNA reaction intermediates containing unwound sequences 5′ to, 3′ to, or encompassing the DNA lesion. The first detectable reaction intermediate was formed by substrate binding of XPA, RPA, XPC·HHR23B plus TFIIH (preincision complex 1, PIC1). In this complex the DNA was unwound on either side of the lesion by no more than 10 bases. Independent of the XPG nuclease function, the XPG protein stabilized this complex, forming a long lived preincision complex 2 (PIC2). The XPF·ERCC1 complex bound to PIC2, forming PIC3, which led to dual incisions and the release of the excised oligomer. With partially unwound DNAs, thymine cyclobutane dimer was excised at a fast rate independent of XPC·HHR23B, indicating that a major function of this protein is to stabilize the unwound DNA or to aid lesion unwinding in preincision complexes.

In humans, nucleotide excision repair is the sole DNA repair activity for removing bulky adducts. The repair reaction involves two basic steps, damage excision by dual incisions and repair synthesis. Mutations that interfere with damage excision give rise to xeroderma pigmentosum (XP) 1 (1). XP is genetically heterogeneous; mutations in seven genes, XPA through XPG, may give rise to the disease (see Ref. 2). Important progress has been made in recent years in understanding the molecular basis of XP and the roles of XP proteins in nucleotide excision repair. These studies have culminated in reconstitution of repair activity in vitro using purified repair proteins (for review, see Refs. 3 and 4). From these biochemical experiments it was concluded that the minimal set for damage removal activity (i.e. excision nuclease) is composed of six re-pair factors. They are XPA, TFIIH (which includes XPB and XPD), XPC⅐HHR23B, XPG, XPF⅐ERCC1, and RPA (5,6). Evolutionarily and structurally related proteins perform the same function in the highly homologous Saccharomyces cerevisiae in vitro system (7,8).
In addition to identifying all of the proteins for excision nuclease activity, recent studies have also characterized the individual repair factors and their interactions with one another and with DNA in considerable detail. XPA and RPA, or a complex of the two, has been implicated in damage recognition (9 -13). TFIIH has a bidirectional DNA helicase activity conferred by its XPB and XPD subunits (14 -18). XPG has a DNA cutting activity specific for the junction of single-stranded DNA to double-stranded DNA in the 5Ј to 3Ј direction (19,20); XPF⅐ERCC1 has a DNA junction cutting activity opposite to that of XPG (20 -22). XPC⅐HHR23B binds to DNA with high affinity (23,24) and is dispensable for excising certain lesions (6) and all lesions immediately 5Ј to a run of 10 mismatched base pairs (Ref. 25 and data not shown). In addition to these functional properties of the repair factors, protein-protein interactions among the various factors have been investigated. XPA binds to XPF⅐ERCC1 (26 -28) and with lower affinity to TFIIH (29). TFIIH binds XPG with high affinity (5,30) and XPC with lower affinity (16). Finally, RPA binds to XPF⅐ERCC1 and to XPA (12). These structural and functional properties of the repair factors have led to the proposal of models on assembly and catalysis by excision nuclease. However, no specific intermediates have been detected with either gel retardation or footprinting techniques in support of the reaction mechanism models.
In this investigation, we set out to study the events taking place from the damage recognition step to the release of the damage in an oligonucleotide. Specifically, we wished to detect the excision reaction intermediates of repair protein-DNA complexes and the conformation of DNA within these complexes. Using a biotin-tagged substrate and purified repair factors, we show that the first protein-DNA complex detectable by the streptavidin pull-down assay requires XPA, RPA, XPC⅐HHR23B, and TFIIH proteins as well as ATP hydrolysis. This complex is termed Preincision Complex 1 (PIC1). PIC1 is relatively unstable and forms a tighter, more stable complex called Preincision Complex 2 (PIC2) in the presence of XPG, even when XPG has lost its 3Ј-junction nuclease activity because of an active-site mutation. XPF⅐ERCC1 does not bind to PIC1, but it does bind to PIC2 containing either wild-type or mutant XPG. We call the complex that forms with all six repair factors PIC3. When PIC3 contains the active-site mutant XPG, only the 5Ј-incision is made by this complex (31). However, when the complex is formed with wild-type XPG, it immediately leads to dual incision and release of the excised oligomer of 24 -32 nucleotides in length. Permanganate footprinting reveals that in PIC1, the DNA is unwound on both sides of the lesion by 5-10 nucleotides. In PIC2, the unwound region remains the same length as in PIC1 but becomes more stable. PIC3 does not promote further unwinding but leads to dual incisions that expose a 24 -32-nucleotide region in the strand complementary to the damaged strand to permanganate modification. Using model substrates containing mispaired DNA 3Ј, 5Ј or 3Ј and 5Ј to the lesion, it was confirmed that lesion unwinding is on the reaction pathway and that the main function of XPC⅐HHR23B is to aid in DNA unwinding.
Pull-down Experiments-We used a 136-base pair duplex with a T [6 -4]T photoproduct. The complete nucleotide sequence of the T [6 -4]T substrate is identical to the T [6 -4]T substrate employed previously (34) except that the 5Ј-terminal guanine of the T[6 -4]T-containing strand was replaced by a cytosine in this study. The oligonucleotide containing a site-specific photoproduct (5Ј-GTAT [6 -4]TATG-3Ј) was provided by X. Zhao and was prepared by the method of Smith and Taylor (35). This oligomer was 5Ј-phosphorylated with [␥-32 P]ATP (7,000 Ci/mmol, ICN) using T4 kinase and annealed with other five partially overlapping oligonucleotides. After ligation, the full-length 136-mer with an internal 32 P label at the fourth phosphodiester bond 5Ј to T [6 -4]T was purified through an 8% denaturing and a 5% nondenaturing polyacrylamide gel. Since the substrate has a protruding guanine and a cytosine at either 3Ј-end, dGTP and biotinylated dCTP were used to fill in the one-base protruding ends, resulting in the biotinylation of the 3Ј-terminus of the T [6 -4]T-containing strand. Subsequently, the biotinylated substrate DNA was incubated with streptavidin covalently linked to magnetic beads (Dynabead M-280, Dynal) and became linked to magnetic beads. Following pulling down using a magnet, the substrate was ready for the pull-down experiments.
As illustrated in Fig. 1, the pull-down experiments were initiated by incubating a subset of the six excision nucleases constituents and the magnetic bead-attached substrate (50 fmol) in 25 l of excision reaction buffer (30 mM Hepes-KOH, pH 7.9, 50 mM KCl, 4 mM MgCl 2 , 200 g/ml bovine serum albumin) with or without 2 mM ATP at 30°C for 60 min. The substrate DNA was then pulled down by holding a magnet beneath the reaction vial and washed twice with 100 l of excision reaction buffer containing 2 mM ATP to remove loosely bound proteins. The washed DNA pellet was then resuspended in 25 l of excision reaction buffer with 2 mM ATP containing the repair proteins that were omitted in the first pull-down incubation and incubated at 30°C for 90 min. To examine if dual incisions had occurred in this second incubation, the reaction mixture was deproteinized, ethanol-precipitated, and analyzed using an 8% denaturing polyacrylamide gel.
Permanganate Probing for Single-stranded Thymines in the Reaction Intermediates-The protocol of permanganate footprint method was adapted from Refs. 36 and 37. The 5Ј-terminally 32 P-labeled T [6 -4]T substrate was incubated with the indicated amounts of repair proteins at 30°C for 60 min in excision reaction buffer with 2 mM ATP in a total volume of 25 l. Subsequently, KMnO 4 was added to a final concentration of 5 mM. After 1 min at room temperature, ␤-mercaptoethanol was supplemented to 1 M to stop the permanganate reaction. Following deproteinization by proteinase K and ethanol precipitation, the DNA was resuspended in 50 l of 1 M piperidine and incubated at 90°C for 10 min. After evaporation under vacuum to remove piperidine, the permanganate-modified thymines were revealed by resolving the DNA products on 8% denaturing polyacrylamide gels.
Bubble Substrates Mimicking Unwound Intermediates-Internally 32 P-labeled bubble substrates containing TϽϾT were synthesized as described by Mu and Sancar (25). The 32 P label was in the fourth phosphodiester bond 5Ј to the photodimer. The sequences of oligonucleotides 1-7 (as indicated in Fig. 9A) used to construct the substrates have been published (25); oligonucleotides 8 and 9 are listed in Table I. The excision reactions of these bubble substrates were performed in the absence of the indicated repair factor under reaction conditions reported by Mu and Sancar (25). All excision reactions were processed as described above and resolved on 8% denaturing polyacrylamide gels. The levels of damage excision were quantified by scanning the dried polyacrylamide gels using PhosphorImager (Molecular Dynamics).

RESULTS
To study the assembly of human excision nuclease we used the following general strategy (Fig. 1). Individual repair proteins or combinations of them were mixed with a magnetic bead-linked substrate, and then the DNA and DNA-bound proteins were isolated using a magnet. To the purified DNA⅐protein complexes, the remaining repair factors were added, and repair was measured by the excision assay. In preliminary experiments we failed to isolate complexes with one to three repair factors and proceeded to conduct experiments with combinations of four and five repair factors. We were able to isolate complexes with four and five repair factors which we named Preincision Complex 1 and 2, respectively. These complexes are most likely on the pathway for formation of excision nuclease complex because they can lead to dual incisions upon encountering the missing components.  GGGGCGAATTCGAGCTCGCCCGGGATCCTCTAGAGTCGACCTGCTGCAGCCCAAGCTTGCGCGAGGTATCCCTCGCAAA TGGCCAGCTGGCGCAGATCTGGCTCGAGGATATCGAATTCCGTACGTGTTCAGGTCC PIC1-Results of pull-down experiments conducted with two factor omissions are shown in Fig. 2. As apparent the only two-factor omission that produced a nucleoprotein complex purifiable by the pull-down assay is the one formed in the absence of XPF⅐ERCC1 and XPG proteins (Fig. 2, lane 2). All other pull-down experiments involving pairwise omissions of repair factors failed to produce excision products (data not shown). The formation of this complex (named PIC1) with XPA, RPA, XPC⅐HHR23B, and TFIIH is ATP-dependent (compare lanes 2 and 3). Most importantly, the addition of the XPG and XPF⅐ERCC1 nucleases to buffer-washed PIC1 resulted in the dual incision typical of human excision nuclease, indicating that the first four factors formed a structure conducive to assembly of active excision nuclease. Hence it is reasonable to conclude that the complex formed with these four repair factors is on the pathway to dual incision.
PIC2-Previously, we reported that with the pull-down assay a preincision complex could be isolated from an XPF mutant cell extract, which upon incubation with purified XPF⅐ERCC1 protein led to excision of damage (6). However, pull-down experiments with a rodent XP-G mutant cell extract failed to yield a productive preincision complex, suggesting a role for XPG in the preincision events (6). This previous result with the rodent XP-G mutant cell extract seemed to be contradictory with the present data shown in Fig. 2 (lane 2), which indicates that a productive preincision complex (i.e. PIC1) can be formed in the absence of both XPG and XPF⅐ERCC1. Hence, we conducted single factor omission experiments using purified proteins. The results in Fig. 2 (lane 4) show that in the absence of XPF⅐ERCC1, a productive complex (designated PIC2) must have been pulled down from the mixture of XPA, RPA, TFIIH, XPC⅐HHR23B, and XPG so that its subsequent encounter with purified XPF⅐ERCC1 led to damage excision. The excision was not caused by nonspecific association of repair factors with substrate DNA because all other single omission experiments failed to give dual incisions under these conditions (see Fig. 3). As in the case of PIC1, the formation of this second preincision complex, PIC2, was also ATP-dependent (data not shown). In  7). Following pull-down and washing, the magnetic beadbound material was resuspended in 25 l of excision reaction buffer containing 2 mM ATP and the repair protein omitted in the pull-down incubation. The reaction products were resolved on an 8% denaturing polyacrylamide gel. The products of the dual incisions flanking the lesion as a result of the excision nuclease activity are indicated by a bracket. Lane 1 contained 50 fmol of substrate DNA that had been processed as in lanes 2-7 in the absence of any repair proteins. The percents of substrate excised were: lanes 1-5 and 7, undetectable; lane 6, 1.6.
light of the results of the pull-down experiments using purified proteins, it is very likely that PIC1 may have been pulled down from the rodent XP-G mutant cell extract. Since in that study only the purified XPG protein was added back to the bufferwashed PIC1 (6), dual incisions failed to occur because of the absence of XPF⅐ERCC1 in the second incubation.
Significantly, comparison of the excision signals in XPG/ XPF⅐ERCC1 and XPF⅐ERCC1 omission experiments shown in Fig. 2 and in repeated experiments conducted under identical conditions indicated that the excision signal obtained with PIC2 is much stronger than that obtained with PIC1. The most likely explanation is that PIC2 is more stable than PIC1 and better survives the repeated washes during pull-down so that upon addition of the omitted factor efficient excision takes place. This interpretation was supported by the experiments that probed the structure of DNA within these complexes as described below.
Conformation of DNA in PIC1 and PIC2-The dependence of formation of PIC1 and PIC2 on ATP and the fact that TFIIH has helicase activity (15)(16)(17) suggested that the DNA within these complexes might be unwound and that either the degree of unwinding or the protein composition of the unwound complex may be responsible for the differential stability of these complexes. Hence, we probed the DNA for unwinding within these complexes by KMnO 4 treatment, which oxidizes unpaired thymines and renders them cleavable by alkali treatment.
The substrate was a duplex of 136 base pairs with a centrally located T [6 -4]T and a terminal radiolabel in the complementary strand. Fig. 4 indicates the positions of Ts near the lesion in both strands, and Fig. 5A shows the results of permanganate probing of PIC1. In the absence of any repair factor both T(ϩ1) and T(Ϫ1) were slightly reactive with permanganate, consistent with previous reports that T [6 -4]T photoproducts unwind DNA in the immediate vicinity of the lesion (38,39). Upon incubation of substrate with XPA, RPA, and XPC⅐HHR23B, the three factors that have been implicated in damage recognition (3,4,24), no additional hypersensitive sites appeared. This indicates that complexes that form with these factors do not substantially disrupt the duplex. In contrast, when TFIIH was included, the DNA was unwound in both directions as evidenced by permanganate hypersensitivity extending to T(ϩ4) and T(Ϫ5) and by the enhanced sensitivity at T(ϩ1) and T(Ϫ1) (Fig. 5A, lane 5). This unwinding was not caused by the nonspecific helicase action of TFIIH binding to the partially unwound T [6 -4]T and enlarging it by the dual (3Ј to 5Ј and 5Ј to 3Ј) helicase activities because the unwinding required XPA, RPA, and XPC⅐HHR23B in addition to TFIIH (Fig. 5A, lanes  6 -8). The reaction also required ATP hydrolysis because it was inhibited by ATP␥S; hence the unwinding had the characteristics of helicase action (lane 9). As expected, the combination of the four repair factors had no effect on the undamaged DNA control (lanes 11 and 12). Together these data lead us to conclude that XPA, RPA, XPC⅐HHR23B, and TFIIH make a specific complex with damaged DNA and unwind it around the lesion in both 5Ј and 3Ј directions more than 5 base pairs but less than 10 base pairs.
Since the excision assay suggests that PIC2 is more stable in PIC1, we wished to learn about the structure of the DNA in this complex. Under optimal conditions for formation of PIC1, we did not detect any effect of XPG (which defines the PIC2) on DNA sensitivity to permanganate. However, upon carrying out the KMnO 4 reaction with limiting amounts of TFIIH, a clear XPG effect was observed. As shown in Fig. 6 both T(ϩ4) and TϪ5) became more sensitive to KMnO 4 oxidation with increasing concentrations of XPG. However, even with the highest concentrations of XPG used, the extent of unwinding in PIC2 was indistinguishable from that in PIC1. It thus appears that PIC1 is less stable than PIC2 and that by using high concentration of TFIIH, the equilibrium is shifted toward fully unwound conformation. The presence of XPG in the complex stabilizes it, possibly through its interaction with TFIIH (5,30) and RPA (11). However, it is also known that the PIC2 is capable of making the 3Ј-incision (6). Thus, this apparent stabilization of the complex could be caused by KMnO 4 hypersensitivity induced by a flap structure resulting from the 3Ј-incision. Hence, we wished to determine whether or not the PIC2 complex required the 3Ј-incision activity of XPG protein. For this purpose we used the XPG(D812A) mutant. This is an active-site mutant, and the D812A substitution completely abolishes the structure-specific endonuclease activity of XPG without affecting its other biochemical properties (31). When PIC2 was reconstituted with XPG(D812A), it exhibited the same properties as the PIC2 formed with the wild-type enzyme (Fig. 6, lane 9). Thus, it is concluded that binding of the XPG polypeptide to other proteins in the preincision complex stabilizes PIC1 and that the XPG nuclease activity is dispensable for this phenomenon.
PIC3-The XPG and XPF⅐ERCC1 proteins are the 3Ј-and 5Ј-nuclease factors of human excision nuclease. In contrast to XPG, the pull-down experiments failed to reveal the presence of a preincision complex containing XPF⅐ERCC1. As the pulldown assay is a rather harsh method for isolating DNA⅐protein complexes, it was conceivable that such complexes did not survive the repeated washes. The unwinding assay provided us another opportunity to test whether such complexes form. Thus, we conducted KMnO 4 probing with all repair factors except XPG using the same T[6 -4]T substrate containing a 32 P radiolabel at the 5Ј-terminus of the damaged strand. Consistent with what was observed with experiments performed with substrate radiolabeled in the undamaged strand (Figs. 5 and 6), with this substrate as well the combination of XPA, RPA, XPC⅐HHR23B, and TFIIH unwound t(Ϫ2), t(ϩ2), and t(ϩ6) (Fig. 7, lane 2). This result enabled us to narrow down the 5Ј-boundary of the excision bubble to be between t(ϩ6) and T(ϩ10). The addition of increasing amounts of XPF⅐ERCC1 did not affect the permanganate footprint, nor did it result in 5Ј-incision (lanes 3-5 and 9 -10). These data support the conclusions from pull-down experiments that XPF⅐ERCC1 does not enter the excision nuclease complex before XPG and that the presence of XPG in the nuclease complex is required for XPF⅐ERCC1 to make the 5Ј-incision. This is consistent with the results obtained when the excision nuclease was reconstituted with the active-site mutant XPG protein (31).
In fact, the active-site XPG mutant enabled us to identify a third excision nuclease complex, which we call PIC3. PIC3 forms when all of the repair factors are present. The excision nuclease assembled in the presence of the five repair factors plus XPG(D812A) does not make the 3Ј-incision (31) and yielded permanganate hypersensitivity pattern essentially identical to that of the PIC2 (Fig. 8, lanes 5 and 6). Thus the presence of XPF⅐ERCC1 in the complex does not affect the size of the excision bubble. In contrast, if instead of XPG(D812A) wild-type XPG is included in the reaction mixture, dual incisions take place, and a region extending from T(ϩ18) to T(-5) becomes hypersensitive to permanganate. This result is in agreement with our earlier findings that upon dual incisions the human excision nuclease (in contrast with the bacterial excision nuclease) releases the excised oligomer and hence generates a single-stranded gap equal in size to the excised fragment (6). This explains the hypersensitivity of the T residues from T(ϩ18) to T(Ϫ5) in the excision gap (Fig. 8, lane 7).
Reactions with Model Substrates Mimicking Reaction Intermediates-The data presented so far show the formation of DNA intermediates with melted base pairs not exceeding 10 unpaired base pairs on either side of the lesion during the assembly of the excision repair nuclease. We wished to know whether by using premelted DNA (bubble structure) we could abrogate the requirement for some of the repair factors. We have shown previously that a substrate with a structure mimicking a transcription bubble stalled at a lesion, that is, a thymine dimer followed by a 10-nucleotide bubble, could be repaired in the absence of XPC⅐HHR23B (25). This finding provides support for the intermediacy of lesion-stalled transcription bubble in transcription-coupled repair and explains the XPC-independence of transcription-coupled repair (40,41).
To investigate the role of unwinding in excision, we prepared three types of substrates. TϽϾT(3Ј-10) and TϽϾT(5Ј-10) contained 10 mismatched pairs in a row immediately next to a TϽϾT either at the 3Ј-or the 5Ј-side of the photodimer, respectively. The third substrate contained 20 mispaired bases with a TϽϾT in the center of the mismatched run (TϽϾT(20)) (Fig.  9A). In light of the role of TFIIH in unwinding the DNA in the PIC1 and PIC2, we were especially interested in whether or not the requirement for TFIIH could be circumvented by using unwound substrates. Excision assays were conducted with these substrates in the absence of individual factors that are known, or thought, to unwind and stabilize unwound DNA. Photodimers in all three substrates are repaired more efficiently than a TϽϾT in a nominal duplex DNA [TϽϾT(0)]. We had reported previously that TϽϾT(3Ј-10) is repaired faster than TϽϾT(0) (25), consistent with the notion that unwinding 3Ј to the lesion is on the pathway for repair. Here we show that TϽϾT(5Ј-10) is also repaired at a more rapid rate than TϽϾT(0), indicating that unwinding 5Ј to the lesion is also important for repair (Fig. 9B). Not surprisingly, TϽϾT (20) was also excised efficiently, further confirming the KMnO 4 assay results that both 5Ј and 3Ј unwinding are on the pathway for repair. In addition, we observed that, as in the case of TϽϾT(3Ј-10), with TϽϾT(5Ј-10) and TϽϾT (20) substrates excision can occur without XPC⅐HHR23B. This finding raises the possibility that conditions such as replication, recombination, or other DNA dynamic events creating similar structures may enable cell to repair DNA in the absence of XPC. There is some indirect evidence that indeed such XPC-independent repair occurs in transcriptionally active chromosome domains (42). Most significantly, however, we find that TFIIH requirement can only very poorly be abrogated in substrate TϽϾT (20), which is more or less similar to the repair bubble we detected with permanganate (Fig. 9B, lane 22). Thus, it appears that TFIIH plays other roles in excision such as the interactions with XPA (29) and XPG (5,30) proteins, in addition to its helix unwinding activity.

DISCUSSION
The basic reaction mechanism of human nucleotide excision repair and the proteins required for carrying out this reaction are relatively well understood (3,4). However, the structures of DNA⅐protein complexes at various stages of the reaction, damage recognition, dual incision, and release of the excised oligomer are not known. Two general models have been advanced. One model proposes that all of the excision repair factors are assembled in the form of a repairosome capable of carrying out the entire excision reaction in yeast (43) or even the entire excision repair process including excision, repair synthesis, and ligation in humans (44). However, the repairosome proposed for yeast is incapable of carrying out damage excision, perhaps because of lack of RPA in complex with the RAD proteins. Indeed, a systematic study of the question has revealed the ready separation of repair proteins in yeast and provided strong evidence for sequential assembly (7,8). Similarly, the protein preparation called human repairosome (44) contained many unrelated proteins and only a small fraction of the repair proteins (44). Again, the relative ease with which the repair proteins could be separated from human cell extracts (5,6,45) is evidence that sequential assembly is the modus operandi of excision repair in humans. Indeed, the best efficiency of excision has been accomplished by a reaction carried out with individually purified repair factors (6).
Although the preponderance of both experimental evidence and theoretical consideration are in favor of sequential assembly, the individual steps are not well understood. In particular, the damage recognition step remains rather ill defined (51). It has been reported that XPA (10), RPA (9,13), or the combination of XPA and RPA (11,12) is the damage recognition factors. Interestingly, none of these is capable of discriminating between cyclobutane thymine dimer-containing DNA and undamaged DNA. As thymine dimer is the most abundant lesion produced by UV and as it is repaired rather efficiently by human excision nuclease (6,34), clearly XPA, RPA, or the XPA plus RPA combination cannot be considered the damage recognition factor or complex.
It seems that the optimal approach to understand the damage recognition step is to identify distinct DNA⅐protein complexes that form with various combinations of repair factors. Here we have identified three repair complexes that we have named preincision complexes 1, 2, and 3. We have been unable to detect specific binding of XPA, RPA, or the combination of the two to TϽϾT-containing DNA by DNase I footprinting (data not shown) or by permanganate footprinting. Instead, the earliest detectable complex in our study is PIC1 whose formation is absolutely dependent on XPA, RPA, XPC, and TFIIH. It is quite conceivable that XPA and RPA make a rather unstable complex that is stabilized by unwinding activity of TFIIH and the single-stranded DNA binding activity of XPC.
The second DNA⅐protein complex that was detected in this study is PIC2, which includes XPG in addition to the four repair factors required for generating PIC1. The main difference between PIC1 and PIC2 appears to be the increased stability of PIC2 conferred by the XPG protein. The fact that a normal 3Ј-incision (by XPG) can occur in this complex (6,52) suggests that this is a functional intermediate on the pathway of assembly of excision nuclease.
A surprising finding in this study was the failure to detect a strong interaction between XPF⅐ERCC1 and PIC1 which forms in the presence of the XPA protein. It has been shown that XPA binds quite tightly to the XPF⅐ERCC1 complex (26,28). In fact, this interaction is so strong and specific that an XPA affinity column is the main purification step for the XPF⅐ERCC1 complex (46). Furthermore, the XPF⅐ERCC1 complex is the subunit of the excision repair nuclease that carries out the 5Ј-incision (6,20,47). It is possible that within PIC1 the XPF⅐ERCC1 binding site of XPA is no longer accessible to these proteins. Whatever the reason for lack of binding of XPF⅐ERCC1, the practical consequence is that we cannot constitute a complex capable of making the 5Ј-incision in the absence of XPG. It is perhaps of relevance to note that in yeast the Rad14⅐(Rad1⅐Rad10) complex, which is the structural homolog of XPA⅐(XPF⅐ERCC1), can be isolated as a very stable complex. It is thought that in yeast the Rad14⅐(Rad1⅐Rad10) complex is essential for any damage-dependent nicking to occur, and hence an uncoupled 3Ј-incision is not observed in the absence of Rad1⅐Rad10 in the yeast excision nuclease system (7,8). It appears that despite extensive structural and functional homologies between the human and yeast excision repair systems, some important differences in the mechanistic details do exist.
Another noteworthy finding of our studies is that the XPC⅐HHR23B complex is dispensable for excision of TϽϾT, not only from the TϽϾT(3Ј-10) and TϽϾT (20) but also from the TϽϾT(5Ј-10) substrate. This indicates that the presence of a single-stranded region of sufficient length around the dimer, regardless of its position relative to the dimer, makes XPC⅐HHR23B unnecessary for excision. This finding, combined with the high affinity of XPC to single-stranded DNA (23,24), supports earlier suggestions that the function of XPC is to assist in DNA unwinding and stabilize the unwound structure (24). Clearly, the findings in this report constitute a useful lead on the XPC function in excision repair, which deserves further investigation.
During the course of this study, Evans et al. (48) reported on the unwinding of cisplatin 1,2-d(GpTpG) diadduct containing DNA in cell extracts of normal and excision repair-defective cell lines. The basic conclusion of that study, that excision repair in humans involves the formation of an "excision bubble," is in agreement with our findings. However, the two studies differ in several aspects. First, as summarized in Fig. 10 the size of the unwound region is less than 20 nucleotides in our study, but it was reported to be ϳ25 nucleotides by Evans et al. (48). The discrepancy could be due to the extensive unwinding caused by the cisplatin 1,3-d(GpTpG) lesion alone and the fact that the sequences around the lesion in the two studies are different. Second, Evans et al. (48) found unwinding around the lesion even in XPA mutant cell extracts whereas we find that unwinding in our defined system is totally dependent on XPA. Conceivably, DNA-binding proteins or helicases in the XPA cell extract not related to excision repair caused the unwinding of the DNA already unwound due to the cisplatin 1,3-d(GpTpG) lesion. Finally, Evans et al. (48) did not detect any effect of XPG on the unwinding reaction. However, the cell extract they used most likely contained a mutant XPG protein of full length (49), which could assemble with the other repair factors but was unable to carry out the incision reaction. Our studies with wild-type and active-site mutant XPG protein demonstrate the plausibility of this explanation and point to the advantage of studying the assembly reaction with purified proteins rather than cell extracts.
Regarding the assembly of human excision nuclease our results allow us to propose the following minimal scheme (Fig.  11) involving at least three intermediates prior to the, usually concerted, dual incisions. It must be pointed out, however, that although we know the precise repair factors required for making the various preincision complexes, we do not know whether all of the proteins required for making a specific complex are present in that complex. It is conceivable that some of the proteins required for formation of PIC1 actually dissociate from the DNA⅐protein complex before PIC2 or PIC3 can form. In the Escherichia coli excision nuclease system, the (UvrA) 2 ⅐(UvrB) complex binds to DNA, forming an intermediate that is analogous to "PIC1" detected for the human counterpart in this study. Subsequently UvrA dissociates, leaving behind a stable UvrB⅐DNA complex (PIC2), which in turn binds to UvrC (PIC3) to initiate the dual incision (for review, see Ref. 3). The function of UvrA is to promote the formation of a productive UvrB⅐DNA complex. This activity has been termed "molecular matchmaker" (50). It is not unreasonable to expect that one or more of the human excision nuclease constituents may act as molecular matchmakers. Further work on identifying the protein compositions of PIC1 to PIC3 is required to address this possibility. FIG. 10. Location and extent of the repair bubble and its relation to dual incision sites. The 5Ј-boundary of the observed bubble lies between T(ϩ10) and t(ϩ6), whereas the 3Ј-border is between T(Ϫ5) and T(Ϫ11). Thus the unpaired region as drawn reflects an approximation. The dual incision sites of this substrate have been determined in a previous study (34).