Cross-linking of DNA-binding proteins to DNA with psoralen and psoralen furan-side monoadducts. Comparison of action spectra with DNA-DNA cross-linking.

We have developed a novel photocross-linking technique using free 8-methoxypsoralen and DNA furan-side monoadducts plus long wave ultraviolet light (UVA). Both sequence-specific (Max) and nonspecific (RecA and T7 RNA polymerase) DNA-binding proteins were cross-linked. The macroscopic equilibrium binding constant (∼109 M−1) and DNase I footprinting indicated that binding of Max to its cognate sequence (E-box) was unimpaired by 8-methoxypsoralen and that cross-linking occurred in normal complexes. RecA protein and T7 RNA polymerase were cross-linked to a 12-mer DNA furan-side monoadduct with UVA. Cross-link yields were directly proportional to the UVA dose. Cross-links were stable to 8 M urea, 1-10% SDS, commonly used alcohols, and mild acids (5% trichloroacetic acid). The DNA in cross-links was reversed with 254 nm UV (photoreversal) or with hot base (base-catalyzed reversal), consistent with (2 + 2) cycloaddition via the 4′,5′-furan of the psoralen. Comparative action spectra for DNA-DNA cross-linking and DNA-protein cross-linking revealed that the latter occurred maximally at 300 nm, while the former occurred maximally at 320 nm. This 20-nm blue shift suggested a higher potential energy surface for an excited psoralen participating in protein-DNA cross-linking as compared with DNA-DNA cross-linking. As with DNA-DNA cross-linking, DNA-protein cross-linking is a two-photon process. Absorption of the first photon formed a 4′,5′-adduct with DNA, which then absorbed a second photon, leading to cross-linking to protein. Based on the action spectra and the known excited states of psoralen, it is suggested that the triplet n,π* transition localized in the C-2=O of psoralen may be involved in protein-psoralen photoreactions.

The control of gene expression is achieved principally through the regulatory activities of DNA-binding proteins. Modern methods of x-ray crystallography and NMR have revealed the nature of the interactions that govern specificity in DNA-protein recognition. Techniques of protein-DNA cross-linking have been used either to confirm high resolution structures in solution or to obtain structural information where no high resolution structures are available (e.g. see Refs. [1][2][3][4][5]. A number of photoreactive agents such as carbenes, nitrenes, thio derivatives, benzophenones, halogenated pyrimidines, or simply 254-nm UV, are routinely employed to achieve protein-DNA photocross-links (reviewed in Refs. 6 -10). Psoralens ( Fig. 1) possess some advantages. Psoralens and their adducts are highly photostable in ambient room light and stable for months when frozen, and as such are easy to use. Psoralens can be site-specifically attached to DNA at pyrimidines (preferentially at 5Ј-TpA sites), and their absorption bands extend well beyond (up to 410 nm) those of either protein or DNA, enabling crosslinking without damage to either protein or DNA. In an ongoing effort to develop photocross-linking methods and to understand protein-DNA recognition in various systems, we focused on psoralen as a probe for DNA-protein interactions. Psoralens are potentially useful because they can react with both DNA and proteins (11)(12)(13)(14). Psoralens (Fig. 1) are furocoumarins used in the treatment of vitiligo, psoriasis mycoses fungoides, and several other skin diseases (see Refs. [15][16][17] for reviews). Upon exposure to long wavelength UV light (UVA; 310 -400 nm), 1 psoralens react with DNA or RNA to form covalent adducts (12,17,18). In DNA, psoralens react primarily with thymidine and to a lesser extent with cytosine. Psoralens photoalkylate DNA by (2 ϩ 2) cycloaddition to the 5,6-double bond of pyrimidines. Monoaddition occurs via either the 4Ј,5Ј-double bond of the furan (Fig. 1, III) or the 3,4-double bond of pyrone of psoralen (see Fig. 1 of Ref. 19 for the molecular structures of psoralen adducts in DNA). Absorption of a second photon by furan-side monoadducts ( Fig. 1, III), after a 1-s relaxation, results in the cross-linking of the furan-side adduct to the thymine on the complementary strand (20). Recently, NMR-derived structures of the DNA furan side and the interstrand cross-link have been solved (19).
The covalent photobinding of psoralens to proteins was documented after the discovery of their reactivity with DNA (21)(22)(23)(24)(25). Compared with their DNA photochemistry, the reactivity of psoralens with proteins is poorly understood. Psoralen derivatives induce protein-DNA cross-links in vivo (26 -28) and in vitro (13). The structural basis of the chemical adducts among psoralen, proteins, and amino acids is unknown. Psoralens were shown to photochemically inactivate a number of enzymes. Some examples are lysozyme, glutamate dehydrogenase, 6-phosphogluconate dehydrogenase, ribonuclease, nuclear histones, and DNA polymerases (see Ref. 11 for review). Many different amino acids in the proteins were shown to react with free psoralens including Trp and Tyr (11,28). Singlet oxygen ( 1 ⌬ g ; 1 O* 2 ) was implicated in the formation of the photoadducts between proteins and psoralen (29 -33). Photoreactive mechanisms not involving 1 O* 2 have been proposed (see Ref. 11 and references therein). We have recently isolated and chemically characterized a psoralen-Tyr photoadduct. 2 One of us previously reported the development of a technique for photocross-linking DNA-binding proteins to DNA via psoralen (13,14). Using this technique, it was shown that promoter DNA binds in the cleft of T7 RNAP, and a nonspecific ssDNA binding site occurred in the fingers domain. The DNA template in an elongation complex was cross-linked to T7 RNAP by placing a site-specific psoralen furan-side monoadduct in the path of an elongating T7 RNAP (13,35,36).
Here, we further demonstrate the versatility of the psoralen cross-linking method to study both sequence-specific and nonspecific DNA recognition. As a first step toward an understanding of the different photochemistries of psoralen in DNA-protein cross-linking and DNA-DNA cross-linking, we compared relative action spectra for cross-linking.

Methods
Irradiation of DNA-Protein Complexes with Continuous Wave UVA Light-The light source was a 200-W mercury/xenon arc lamp (Oriel Corp., Stratford, CT). The lamp housing was fitted with a liquid filter to eliminate IR and a dichroic mirror with Ͼ90% reflectance in the 350 -450 nm range (Oriel Corp. model 66218). For analytical cross-link preparations, the lamp housing was fitted with a mercury line band pass filter centered at 365-nm with a 10-nm bandwidth (Oriel Corp. model 56531). A 2.8-cm fused silica lens was used to focus the UV light on to the slit of a thermostated 100-l quartz cuvette (1-cm path length) that contained the reaction mixture. The cuvette was held in a customdesigned cuvette holder thermostated by a circulating Lauda bath containing a 50:50 mixture of water/ethylene glycol. The UVA light impinging and exiting the slit window of the cuvette was measured using an international light meter (model IL1700) fitted with a calibrated UVA probe (model SFD038).
Max-DNA Cross-linking-8-MOP was added to 32 P-labeled duplex DNA at a molar ratio of 1000:1 and allowed to incubate in the dark at room temperature for 30 min in K ϩ phosphate-Mg 2ϩ buffer (50 mM K ϩ phosphate (pH 6.0), 1 mM Mg(OAc) 2 ) or Tris-Mg 2ϩ buffer (10 mM Tris-HCl (pH 8.0), 100 mM KCl, 1 mM MgCl 2 , 5 mM dithiothreitol, 10% glycerol). Both buffers gave similar results. Max was then added to the "dark" reaction at a Max monomer:DNA molar ratio of 4:1 or 12:1. The mixture (25 l) was incubated for 30 min at room temperature (ϳ27°C) and then irradiated for 3 min with the mercury/xenon arc source (see above) fitted with a dichroic mirror. To 10 l of the photoreaction, an equal volume of 2% SDS sample buffer was added and heated for 5-10 min at 95°C. The denatured samples were electrophoresed on a 20% acrylamide, SDS, Tris, Tricine gel (29:1 acrylamide:bisacrylamide). The gel was run overnight at 2 mA/cm. The wet gel was exposed to an x-ray film and subsequently to a phosphor screen.
Noncovalent Binding of Max to Cognate DNA-Gel mobility shift assays were performed as follows. One-half pmol of 32 P-labeled duplex DNA was incubated in K ϩ phosphate-Mg 2ϩ buffer or Tris-Mg 2ϩ buffer with various amounts of 8-MOP for 10 min in the dark at room temperature (ϳ25°C). Various amounts of Max (see figure legends) were added, and the incubation was continued for 30 min (25-l final volume). The reaction was mixed with 5 l of 50% glycerol and loaded on an 8% acrylamide nondenaturing gel (19:1 acrylamide:bisacrylamide). Before loading the samples, the gel was prerun for 1 h, and the tank buffer was replaced with fresh buffer. The gel (12 cm length ϫ 16 cm width) was run at 10 V/cm until the bromphenol blue dye reached 6 cm from the bottom of the wells. The gels were soaked for 20 min in 5% methanol, 5% acetic acid, 3% glycerol with agitation and then dried under vacuum at 80°C. The radiolabeled bands were visualized by autoradiography with x-ray film, and subsequently with a phosphor screen.
DNase I Footprinting of Max-DNA Complexes-One-half pmol of 32 P-labeled duplex DNA was incubated in potassium phosphate buffer with various amounts of 8-MOP for 10 min in the dark at room temperature (ϳ25°C). Various amounts of Max were added, and incubation was continued for 30 min (25-l final volume). DNase I was added to a final concentration of 140 ng/ml, and incubation was continued for 5 min. The reaction was stopped by adding EDTA to 20 mM. The reaction was then phenol-extracted and ethanol-precipitated, and the DNA was 2 S. S. Sastry, submitted for publication. resuspended in TBE (180 mM Tris borate, 2 mM EDTA buffer), 8 M urea dyes and run on a 24% acrylamide denaturing gel.
Quantitation of Cross-linking-Cross-link yields are expressed as arbitrary units of relative cross-linking, defined as (XL b /(XL b ϩ XL f )) Ϫ (C/(C ϩ C f )) ϫ 1000, where XL b is the integrated band area representing the cross-linked DNA, XL f is the band area representing the free DNA in the same lane, C is the band area corresponding to the position of cross-links in the control (no UVA), and C f is the band area corresponding to the free DNA in the same lane. All band areas were obtained first by deducting nonspecific background counts in a portion of the gel where no samples were loaded. Autoradiography was performed with a phosphor screen. Quantitation of gel bands was carried out on phosphor images of gels with the aid of the ImageQuant program using a Phos-phorImager (Molecular Dynamics, Mountain View, CA).
To estimate the cross-link yield in terms of the amounts of protein, we translated the sum total of storage phosphor counts of all crosslinked bands to the amount of 32 P using a standard curve. We constructed the standard curve using graded amounts of known [␥-32 P]ATP of known specific radioactivity. Many rows of [␥-32 P]ATP were spotted on a 3MM filter paper, dried, and exposed to a phosphor screen. The same samples were also counted for 32 P in a scintillation counter. The 32 P Cerenkov counts of these samples were plotted against the corresponding storage phosphor counts. The relationship between 32 P Cerenkov counts (or the actual amount in pmol of [ 32 P]ATP) and the corresponding storage phosphor counts was perfectly linear over a large dynamic range of radioactivity. Using the standard curve and the observed storage phosphor counts, we obtained the 32 P counts for each cross-linked and free DNA band. From the specific radioactivity of the 32 P label we calculated the amount of DNA in cross-linked bands. Assuming that one protein monomer cross-linked per DNA, we calculated the amount of cross-linked protein. The percentage of protein cross-link yield was obtained by dividing the pmol of cross-linked protein by pmol of the total amount of protein in the reaction mixture and multiplying the result with 100.
UV Absorption Spectra-All spectra were obtained in 10 mM Tris-HCl (pH 7.5), 1 mM EDTA at room temperature with a Beckman DU60 scanning UV-visible spectrophotometer using a 100-l masked quartz cuvette of 1-cm path.
Action Spectra-Many stock reaction mixtures (each 200 l) were prepared at convenient intervals of time and kept on ice during the course of these experiments. For RecA-DNA cross-linking, each stock reaction contained 0.5 pmol of 32 P-12-mer DNA or 32 P-12-Maf DNA, 7 pmol of RecA, 1 mM ATP␥S in potassium glutamate buffer (see above for RecA-ssDNA cross-linking buffer). Stock solutions (200 l) for 32 P-12mer DNA/8-mer DNA cross-linking contained 0.5 pmol of 32 P-12-mer (or 32 P-12-Maf)-8-mer duplex in the same buffer as for RecA cross-linking. When needed, HMT was added to 3 g/ml. Without added HMT, the absorbance of the solutions at 300 nm was negligible. With HMT, it was below 0.2 between 280 and 300 nm. This minimized inner shielding effects and light scattering. Fifty-l aliquots from stock reactions were irradiated at 30°C for 30 min at each wavelength. To maintained a relatively uniform absorption cross-section, the solution was periodically mixed quickly during irradiation using a micropipette. A Bausch and Lomb monochromator (model 33-86-01, serial number 2324BK, with a grating of 2700 grooves/mm and 3.2-nm bandwidth at each wavelength) was attached immediately after the IR filter of the 200-W mercury/xenon arc housing (see above). No further attachments were made. Monochromator entrance and exit slits were set at fully open positions, and a light spot of ϳ2-3 mm was focused on the entrance slit of the quartz cuvette. A thermopile probe (Scientech, Boulder, CO, model 380101, serial number 4579) coupled to a microvoltmeter (Keittley Instruments) was positioned in line behind the cuvette exit slit, and the light energy was measured at 5-min intervals during sample irradiation. The sensitivity of the thermopile probe was independent of wavelength. The light energies at each wavelength were measured with and without samples in the cuvette. The following were the measured absolute energy distributions ( ϭ W/cm 2 ): 280 nm, 0.02; 290 nm, 0.04; 300 nm, 0.05; 310 nm, 0.13; 320 nm, 0.1; 330 nm, 0.02; 340 nm, 0.02; 350 nm, 0.01; 360, 0.06; 370 nm, 0.11; 380 nm, 0.04; 390 nm, 0.01. The sample geometry and the photon dose at each wavelength were the same for each sample. After the irradiation, the samples were denatured by heating at 95°C for 5 min and concentrated by heating in a speed vac (Savant) and then run on 10% acrylamide-SDS gels to visualize protein-DNA cross-links or 24% acrylamide-8 M urea gels to visualize the [ 32 P]DNA cross-links (13,37). Quantitation of the gel bands was carried out by PhosphorImager analysis. Action spectra were normalized for light intensities. The sample composition for each type of cross-linking experiment, photon density per sample and sample geometry and other conditions during photoreactions were the same. Action spectra are expressed in arbitrary units as relative cross-linking (defined above) of either DNA to DNA or DNA to protein.

RESULTS
Cross-linking of DNA-binding Proteins to DNA with Psoralen: A Case Study-Max belongs to a class of sequence-specific DNA-binding eucaryotic transcription factors that contain a basic/helix-loop-helix/zipper motif (38,40,41). This class of proteins participates in tissue differentiation and cell proliferation. Max binds to its cognate DNA with high affinity (ϳ10 9 M Ϫ1 ) as a homodimer in a scissors grip fashion. The co-crystal structure of Max with its cognate sequence has been solved by x-ray crystallography (38). To test our cross-linking procedure, Max is a very good paradigm for specific DNA recognition. RecA protein (molecular mass ϭ 37.8 kDa) of Escherichia coli participates in homologous recombination and DNA repair. In the presence of ATP␥S, RecA binds strongly to ssDNA (42). T7 RNAP (ϳ99 kDa), which is normally a DNA-dependent RNA polymerase, also binds ssDNAs (13,43). Both RecA and T7 RNAP bind ssDNA without sequence specificity and as such are very good candidates for testing whether our cross-linking technique worked with nonspecific DNA-binding proteins.
Case I, Sequence-specific DNA Recognition: Transcription factor/oncoprotein Max Is Cross-linked to Its DNA Recognition Sequence with 8-MOP- Fig. 2 shows cross-linking of Max to a 16-mer radiolabeled duplex DNA containing the cognate DNAbinding sequence (CACGTG).

5Ј-TAGGCCACGTGACCGG-3Ј
3Ј-ATCCGGTGCACTGGCC-5Ј SEQUENCE 1 A single band migrating at ϳ15.5 kDa, representing the cross-linked protein was observed. Cross-linking was not observed without 8-MOP. A non-DNA-binding protein, bovine serum albumin, did not cross-link to DNA in the presence of The mass of the cross-linked Max was approximately equal to the sum of the masses of one monomer of Max (10.82 kDa) and one 16-mer DNA strand (4.8 kDa), suggesting that one monomer was probably cross-linked to one strand of the duplex DNA. This result is in agreement with the co-crystal structure of Max-DNA complex, which showed that each monomer of Max makes symmetric contacts with each strand of the DNA duplex. (38). Increasing the amount of Max monomer (from 2 to 6 pmol) without changing the amount of DNA or 8-MOP or the dose of UVA, increased the yield of cross-links by about 2.5-fold (Fig. 2). Fig. 3A shows the cross-linking efficiency as a function of photon flux at a constant concentration of 8-MOP (10 3 molar excess over DNA). Cross-link yield increases with increasing amounts of light, indicating that the reaction is completely dependent on light, as expected of a photochemical process. We estimated that ϳ10% of Max is cross-linked to [ 32 P]DNA, assuming that there is one DNA per Max monomer in the photoconjugate.
In the above experiment, 8-MOP (Fig. 1, I) was added to DNA before the addition of Max. It was important to know whether psoralen perturbed the binding of Max to DNA. We conducted a series of experiments to show that Max binding to cognate DNA was unaffected by the presence of 8-MOP. Fig. 4 (Fig. 3C). The macroscopic binding constant (ϳ10 9 M Ϫ1 ) of Max to its cognate sequence was the same with or without 8-MOP. This result showed that the presence of 8-MOP did not inhibit binding of Max. To verify that Max contacted the E-box (CACGTG) in the presence of 8-MOP, we carried out DNase I footprinting. Fig. 5, A and B, shows a footprint of Max on 32 P-labeled DNA. Within the resolution of DNase I footprinting technique, the footprints of Max on the E-box CACGTG are identical, with or without 8-MOP (Fig. 5B). These results demonstrate that the presence of 8-MOP neither quantitatively (as measured by the binding constant) nor qualitatively (as indicated by DNase I footprints) perturbed Max recognition of its cognate DNA, indicating that cross-linking probably occurred in a native Max-DNA complex.
Next, we examined if Max binding to cognate DNA was disrupted by the addition of increasing amounts of 8-MOP. This information is essential for maximizing cross-link yields in scale-up cross-link preparations with the aim of determining the cross-linking sites on Max. One way to maximize cross-link yields is to photochemically saturate the cross-linkable sites in protein-DNA complexes. This can be achieved by increasing the concentration of 8-MOP in the reaction. Clearly, this cannot be achieved at the cost of binding specificity. Fig. 6 shows that Max binding to DNA was not inhibited even at a 500-or 1000-fold excess of 8-MOP over DNA. In the lanes where 8-MOP was present (Fig. 6, lanes 1-6), Max-DNA complexes migrated as one band at the same position as when 8-MOP was absent (Fig. 6, lane 11), implying that identical complexes are formed in the presence or absence of 8-MOP. At very high concentrations approaching saturating amounts of 8-MOP (36 g/ml), Max binding to DNA was inhibited by about 50 -70% (Fig. 6, lanes 7 and 8).
Case 2, Nonspecific DNA Recognition: Cross-linking of Single-stranded DNA-binding Proteins to DNA-A 12-mer DNA (5Ј-GAAGCTACGAGC-3Ј) was furan-side monoadducted at the unique dT. Fig. 7 shows a time course of cross-linking of E. coli RecA protein (Fig. 7A) and T7 RNAP (Fig. 7B). The RecA cross-links migrated at ϳ42 kDa, while the T7 RNAP crosslinks migrated at ϳ100 kDa, consistent with the molecular masses of the proteins plus a 12-mer DNA. The yield of crosslinks increases with increasing UVA dose, indicating that cross-linking is totally dependent on UV light (Fig. 8), in a manner similar to Max (Fig. 3A). Under these irradiation conditions (Fig. 7) we have estimated that 20 -60% of RecA or T7 RNAP was conjugated to DNA.
We have roughly estimated the quantum yield. represents mols of cross-linked protein/the absorption cross-section ( a ); a ϭ 2.303 (A/bn), where A is the absorbance of the sample given by A ϭ 1/2.303 ϫ ⑀ ϫ (lnL 0 Ϫ lnL t ); ⑀ for 8-MOP is 11.8 ϫ 10 3 M Ϫ1 cm Ϫ1 at 300 nm; n is Avogadro's number; b is the path length in cm. These experiments were conducted using a monochromator attached to the mercury/xenon arc source (see "Methods"). We measured the average amount of light impinging the sample (L 0 ) and the average amount of light transmitted through the sample (L t ) during irradiation. We assumed that only one psoralen monoadduct reacted per RecA. The estimated quantum yield was 0.002-0.015 between 290 and 320 nm. The highest yield (0.015) was obtained at 300 nm, consistent with our action spectra results (see below). Quantum yields were not corrected for the binding constant for the RecA/ 12-Maf interaction. This efficiency is somewhat lower than those reported for photobinding of psoralens to DNA (44,45).
Chemical Stability of Cross-links-For the isolation and purification of cross-linked peptides it is important to know the stability of the photoadducts (see Refs. 13 and 14 for method of adduct purification). We have tested the chemical stability of DNA-psoralen-protein conjugates against certain common lab-oratory reagents used during the course of conjugate purification. The adducts are stable to heat (ϳ95°C), 1-10% SDS, 8 M urea, 0.1% trifluoroacetic acid, 80% CH 3 CN, ethyl and methyl alcohols; phenol; 5% trichloroacetic acid; and modification by iodoacetamide. Some loss of DNA from the photoconjugate was observed after treatment with HCOOH (pH 1-2). Reversal of the DNA from protein was achieved by treatment with hot base (0.1 M KOH) at 95°C or with 254-nm UV (photoreversal). This procedure is essential for the determination of the masses of psoralen-conjugated peptides (13,14).
Wavelength Dependence of Photochemical Cross-linking-Action spectra provide two essential pieces of information regarding photochemical cross-linking: 1) action spectra can identify the most effective wavelength of UV at which optimal crosslink yield can be achieved; and 2) action spectra can reveal the differences in the mechanisms between protein-DNA crosslinking and DNA-DNA cross-linking. Fig. 9 shows a spectral overlap of psoralen, DNA, DNA Maf, and protein. DNA and protein effectively stop absorbing UV at wavelengths above 310 nm. Psoralen furan-side monoadducts absorb UV effectively at least up to 370 nm, and free psoralen absorption extends to at least 410 nm. The absorption spec- trum of psoralen shows two maxima: a larger maximum at ϳ250 nm (⑀ ϭ ϳ2.2 ϫ 10 4 M Ϫ1 cm Ϫ1 ) and a smaller maximum at 300 nm (⑀ ϭ ϳ1 ϫ 10 4 M Ϫ1 cm Ϫ1 ). We examined the action spectra for protein-DNA and DNA-DNA cross-linking via HMT from 220 through 380 nm. The measured absolute energy distributions of the isolated UV bands are given under "Methods." The data represent the average energies of UV passing through the empty cuvette slit recorded during the course of the experiments. Here, it should be emphasized that the monochromator, sample volumes, sample irradiation geometry, band energy distributions, and total photon fluxes per experimental sample at a given wavelength are the same. In addition, the same protein amounts, DNAs, and psoralen were used. To minimize sample to sample variation, we prepared stock irradiation mixtures from which the same amount of sample was taken for irradiation at each wavelength, and we kept the absorbance of the sample to a minimum. Each datum point in Fig. 10 is an average of three separate sample irradiations, and the error bars indicate minimal deviation. These experimental procedures allow for reliable comparisons and a straightforward interpretation of the action spectra. Fig. 10, A-D, shows the relative action spectra for crosslinking DNA to DNA and DNA to protein (RecA) via HMT. The action spectra for cross-linking protein to DNA (Fig. 10, A and  B) are significantly different from those for cross-linking DNA to DNA (Fig. 10, C and D). Cross-linking a 12-Maf to RecA protein has an action maximum at 300 nm (Fig. 10A, 1) coincident with one of the absorption maxima (ϳ300 nm) for HMT (Fig. 9). An action maximum at 300 nm is also observed (Fig.  10B, 1) when free HMT is used to cross-link RecA protein to an unmodified 12-mer DNA. There is a smaller action peak at 365 nm (Fig. 10B, 2). For cross-linking a furan-side monoadduct to RecA, the 300-nm action peak 1 is about 3 times greater than the 365-nm action peak 2 (Fig. 10A), whereas when crosslinking is done with free HMT, the 300-nm action peak 1 is only ϳ1.7 times greater than peak 2 (365 nm). These differences are small but have been consistently reproduced. This result showed that protein-DNA cross-linking by HMT is most optimal at 300 nm compared with at 365 nm. When the 12-mer DNA is cross-linked to a complementary 8-mer DNA with free HMT or with a HMT furan-side monoadduct (DNA-DNA cross-linking), a different action spectrum is seen. Two features are apparently different: 1) the action maximum (Fig. 10, C and D, 1) is now red-shifted to 320 nm, compared with the action maximum 1 at 300 nm for DNA-protein cross-linking; and 2) the ratio of action peaks 1 at 320 nm to 2 at 365 nm (Fig. 10, C and D) is ϳ1.4.
At 365 nm, psoralen-DNA photocross-linking is favored over photoreversal. At 320 nm, both DNA monoadduction and DNA cross-linking occur (Fig. 10D). The action spectra show smaller peaks at 365 than at 320 nm, probably because of the lower psoralen extinction coefficient at 365 than at 300 or 320 nm (Fig. 9). The broad HMT absorption band (with an ϳ300-nm maximum; Fig. 9) is responsible for cross-linking. From a photochemical mechanistic view point, the significant feature is the apparent 20-nm blue shift for protein-DNA cross-linking compared with DNA-DNA cross-linking (see "Discussion").

DISCUSSION
Our present work was carried out with two specific aims: 1) to demonstrate the application of psoralen as a photocrosslinking agent for studying specific and nonspecific DNA-protein recognition, and 2) to find the optimum UV wavelength for cross-linking and to move toward an understanding of the cross-linking mechanism. We have devised two complementary cross-linking approaches to the study of DNA-protein interactions: 1) for double-stranded DNA-binding sequence-specific proteins, using Max as a paradigm, and 2) for ssDNA-binding proteins, using RecA and T7 RNAP as examples.
Specific cross-linking of Max to its cognate sequence was achieved via the addition of free 8-MOP. It may also be possible to cross-link Max via furan-side monoadducts, just as with RecA and T7 RNAP. However, since there are no favored 5Ј-TpA sites within the E-box (CACGTG), cross-linking via the use of monoadducts may yield less information. Cross-linking via intercalated/stacked psoralen is advantageous because there is no preselection of cross-linking sites. The whole gamut of cross-linkable sites can be potentially examined by scale-up procedures using free psoralen. In previously published works, we mapped promoter-binding cleft and ssDNA binding finger in the three-dimensional structure of T7 RNAP using the psoralen cross-linking approach (13,14). We hope to apply our mapping procedures to Max, sterol receptor-binding protein, and other proteins. Cross-linking may occur via the favored pyrimidines in and around the E-box. Direct cross-linking with added free psoralen is a broad spectrum approach that may yield structural information about global DNA-binding regions in proteins. Irradiation of an intercalated psoralen furan-side monoadduct will cross-link the dT in the complementary strand in double-stranded DNA (13) more avidly than a bound protein, perhaps because of geometric and quantum yield factors. Therefore, the cross-linking with psoralen furan-side DNA adducts may be less useful with sequence-specific double-stranded DNA binding proteins, mainly because of competing DNA-DNA crosslinking and preselectivity of site. However, this approach may be applicable with proteins that locally unwind DNA at their binding sites, such as helicases, DNA-repair proteins, and open complexes of DNA and RNA polymerases. Indeed, we have earlier demonstrated that a T7 RNAP complex whose elongation was blocked by a site-specific psoralen monoadduct can be crosslinked to the DNA template via the psoralen (13).
When psoralen is added to DNA followed by protein, there are at least five competing intermolecular photoreactions: psoralen-psoralen cycloadditions, DNA monoaddition, DNA furan-side monoadduct-DNA cross-linking, psoralen-protein reactions, and DNA-protein cross-linking. In addition, photodestruction and quenching by solvent molecules such as O 2 also occur. The relative yields of these photoreactions are governed by a large number of factors that may include steric, kinetic, and thermodynamic factors; quantum yields of singlet and triplet excited states; and intersystem crossing. To compensate for the very large competition from the other reactions and to enhance our chances of trapping protein-DNA cross-links, we are using a large excess of psoralen. For Max cross-linking, a 1000-fold excess of 8-MOP over DNA was added. No detectable adverse effects on binding were seen by footprinting and gel shift assays. Cross-links are formed by the photoreaction of DNA-intercalated/stacked psoralen and bound protein. Photocoupling occurs through the absorption of two photons, most probably through two separate excitation events. The first absorbed photon probably leads to monoadducts with DNA. Absorption of a second photon only by furan-side DNA monoadducts leads to cross-linking with protein. Our estimated quantum yields are in the range of ϳ0.002-0.01, depending on the wavelength of UV. These quantum yields are smaller than those reported for the conversion of furan-side monoadducts to interstrand DNA cross-links (0.02-0.04) (46,47). In our photoreactions, any Max that first reacted with free psoralen may not have cross-linked with DNA for the following reasons. 1) A specific stereochemical intercalation geometry is required for a (2 ϩ 2) photocycloaddition to DNA, which is less likely to occur with a psoralen that was first photochemically attached to Max. However, it is possible that following "photochemical fixation" of Max to DNA by the first absorbed photon, other neighboring psoralens may photoreact with Max. 2) Photomodified psoralen that is first attached to protein via the pyrone side (3,4-double bond or C-2ϭO) will not absorb a long wave UVA photon to undergo a (2 ϩ 2) reaction with DNA. The suggested overall pathway for cross-linking is as follows. Here we propose a common pathway for photocross-linking with free psoralen and furan-side monoadducts, consistent with our previous results (13,14). Other mechanisms involving the C-5 or C-4 positions of the benzene nucleus or pyrone in psoralen are possible.
Intercalation of psoralen changes the helical parameters of DNA (19,48). 1) In both photochemical adducts, the psoralen is intercalated between the base pair steps. 2) The repeat length is altered (overall 11-base pair repeat at the psoralen site as opposed to 10.5 base pairs in standard B-DNA) consistent with the intercalation of psoralen. The DNA returns to normal B form outside 3 base pairs of the intercalation site of psoralen, and the helix parameters for both adducts conform to normal B-DNA outside 3 base pairs of the intercalated psoralen.
3) The normal Watson-Crick base pairing is preserved outside the intercalated site of the psoralen. 4) Hydrogen bonding of the base pairs above and below the plane of the psoralen are not completely disrupted, although in both adducts bases appear somewhat buckled. In the case of the cross-link there may be increased solvent accessibility at the site of the cross-link. 5) Overall, the DNA-helix has a very minor bend of ϳ8°. In furanside monoadduct and cross-link, the helix is unwound by ϳ27°-28°. 6) Due to the change in the hybridization of the C-5-C-6 orbitals from sp 2 to sp 3 following cycloaddition to dT, the C-5-CH 3 and C-6 -H point away from the psoralen but still point into the major groove. The 5Ј bases above and below the psoralen stack onto the CH 3 of the photoalkylated dT, coplanar with the pyrimidine ring. 7) We estimated that psoralen monoadducts cross-link to a protein that is within ϳ8 Å of the furan-side adduct. One should note that the above description is for covalent adducts of psoralen with DNA. Noncovalently bound psoralen probably has a different configuration. It may be less rigidly bound within the DNA and exchanges freely in solution. Because intercalators are generally thought to obey the "nearest neighbor exclusion principle" (49), noncovalently bound psoralens might be intercalated between every second DNA base pair. Assuming a random mode of intercalation, at equilibrium, there should be at least two sets of DNA molecules in the population that have psoralen intercalated between alternating base pairs (other combinations may also occur).
Cross-linking via furan-side monoadducts is more applicable to ssDNA-binding proteins because they are in general nonspecific DNA-binding proteins, and therefore, the psoralen monoadduct can be located at any convenient 5Ј-TpA position within the ssDNA sequence without altering binding. Moreover, in ssDNAs without self-complementarity, intercalation of psoralen (in the formal sense) is unlikely to occur. Cross-linking is more likely to occur between DNA and protein in the absence of the competing reaction with complementary DNA strand.
What excited states might have participated in protein-DNA cross-linking? Between 200 and 400 nm, psoralen shows two transitions (Fig. 11): an n 3 * transition localized in the C-2 carbonyl and the 3 * localized in the ring system (Refs. 15, 17, and 50 and references therein). The charge transfer character of the 4Ј,5Ј-furyl double bond is lower than that of the 3,4-pyronic double bond, and direct excitation of the furyl double bond has not been observed (51). The lowest excited singlet and triplet states are the 3 * and are implicated in the (2 ϩ 2) cycloaddition with pyrimidines of nucleic acids (Fig. 11) (50, 52). Based on molecular orbital calculations and luminescence spectra, the 3,4-double bond in the pyrone ring is the site for triplet (,*) reactivity (51, 53). Electronic, steric, and kinetic factors will determine whether the 1 ,* or 3 ,* participate in photoconjugation to DNA. It may be inferred that in furan-side monoadducts (Fig. 1, III) only, the pyronic 3,4-double bond and/or the C-2ϭO are the photoreactive centers for protein-DNA cross-linking. Because the furyl 4Ј,5Ј-double bond is saturated by attachment to DNA (Fig. 1, III) it is probably taken out of action for protein cross-linking. Because the 290 -320-nm absorption band (Fig. 9) is due to the coumarin in psoralen, the action peak 1 (Fig. 10, A and B) perhaps represents cross-linking via 3 n,* localized in the carbonyl C-2ϭO (Fig. 11). Carbonyl n,* states have some ,* character in them, especially in polyaromatic molecules such as psoralens because of mixing of states (54). The smaller peak 2 in Fig. 10B may represent participation of ,* transition (pyrone 3,4double bond) in protein-DNA cross-linking at higher wavelengths (365 nm). The almost exclusive action peak (Fig. 10A,  1) for protein cross-linking with furan-side monoadduct (suppression of peak 2; Fig. 10A) may be attributed to the coumarinic pyrone C-2ϭO ( 3 n,*). Thus, while 3 ,*, localized in the olefinic 3,4-pyrone or 4Ј,5Ј-double bond of furan participates in (2 ϩ 2) cycloaddition to DNA and in DNA-DNA cross-linking (in agreement with previous reports), it is proposed that the C-2ϭO ( 3 n,*) is predominantly involved in the reaction with protein. In support for this suggestion, we observed a 20-nm blue shift in the action maximum (Fig. 10, A and B, 300-nm peak 1) for DNA-protein cross-linking compared with DNA-DNA cross-linking (Fig. 10, C and D, 320-nm peak 1). This blue shift translated into a higher potential energy of 8 kcal/mol or 0.212 eV for protein cross-linking compared with DNA-DNA cross-linking. (In this context, it is worth noting that the experimentally determined ⌬E for singlet-triplet splitting of n,* states in carbonyl compounds is generally 7-10 kcal/mol (54)). This implies that protein cross-linking occurs at a higher potential energy surface ( 3 n,*) as compared with DNA crosslinking ( 3 ,*). This suggestion is in agreement with the energy diagram (Fig. 11), which shows that 3 n,* is higher than 3 ,*. (This does not mean that 3 n,* has 8 kcal/mol higher energy than 3 ,*). The quantum yield for singlet to triplet intersystem crossing is high, as evidenced by high phosphorescence:fluorescence ratios (50 -52). With most psoralens, the triplet states are long lasting (up to 1 s) (34).