Molecular basis of UV lesion binding and repair inhibition by ETS-family transcription factors

Abstract

Mutation hotspots in melanoma frequently occur at DNA binding sites of E26 transformation-specific (ETS)-family transcription factors, as ETS factors stimulate the formation of UV-induced cyclobutane pyrimidine dimers (CPDs) while suppressing repair at ETS-bound DNA sites. To elucidate the molecular mechanism by which ETS factors bind to damaged DNA sites and inhibit repair, we investigated the binding of members from the three major classes of the ETS superfamily (Ets1, ELF1, and PU.1) to cognate DNA containing a cis-syn TpT CPD. These site-specific CPDs modulated ETS recognition and repair by a model repair enzyme in a position-dependent manner. Specifically, a deaminated CPD located in a damage hotspot in the ETS binding motif consistently stimulated binding and inhibited T4 PDG (a CPD repair enzyme) by all three paralogs. Co-crystal structures of PU.1 reveal that CPDs and mismatches are recognized within the framework of canonical ETS/DNA complexes. Molecular dynamics simulations in explicit solvent show that CPD introduces compensatory structural dynamics to both the free and ETS-bound states that strongly modify the underlying thermodynamics of recognition. The results offer a molecular basis for how ETS factors induce mutation hotspots in skin cancers and other UV-exposed tissues by binding to CPD-containing sites and inhibiting their repair.

Graphical Abstract

Graphical Abstract.

Introduction

Ultraviolet (UV) light is the primary etiological agent for skin cancers such as melanoma by inducing mutagenic damage to DNA [1, 2]. The most common type of UV photodamage is the cyclobutane pyrimidine dimer (CPD), which is formed by a (2 + 2) cycloaddition reaction between the C5–C6 double bonds of neighboring pyrimidine bases [3]. CPDs are repaired by the nucleotide excision repair (NER) pathway [4], which plays a critical role in preventing carcinogenic UV-induced mutations. Xeroderma pigmentosum patients with inherited defects in NER have >1000-fold increased risk of skin cancer [5]. Previous genome-wide surveys of NER activity in human cells using the excision repair-sequencing (XR-seq) method have indicated that the rate of CPD repair is remarkably variable across the genome, with slow repair associated with locations where access to DNA damage is restricted by chromatin architecture or DNA-bound proteins [6, 7]. Since the efficiency of repair significantly impacts UV-induced mutation rates, it is important to elucidate the molecular mechanisms that regulate lesion accessibility and repair activity.

Somatic mutations in melanoma are highly enriched at DNA sites bound by transcription factor proteins [8–10]. Some of these mutation hotspots have been identified as noncoding driver mutations, including, for example, recurrent mutations that create binding sites for E26 transformation-specific (ETS) family transcription factors in the promoter of the telomerase reverse transcriptase (TERT) gene [11, 12]. However, there are also recurrent somatic mutations at many other transcription factor binding sites in melanomas that are not linked to carcinogenesis [13]. Instead, analysis of genome-wide repair of CPD lesions in yeast and human cells indicates that DNA binding by certain classes of transcription factors may promote UV mutagenesis by inhibiting NER [8, 14, 15]. For example, the CCCTC-binding factor (CTCF) inhibits the repair of CPDs at CTCF binding sites across the genome, which correlate with elevated somatic mutation rates in melanoma at these binding sites [8, 16, 17]. In parallel, we have shown that purified CTCF protein binds with only slightly lower affinity to a CPD-containing DNA binding site in vitro, and inhibits CPD cleavage by T4 pyrimidine dimer glycosylase, a model CPD repair enzyme [18]. Other transcription factors vary in their ability to tolerate CPD lesions in their cognate binding sites. Early studies identified several prominent examples (p53, AP-1, NF-Y, E2F4, and NFκB) with significantly decreased binding affinities to site-specific CPDs [19]. Recent high-throughput characterization has indicated that impaired transcription factor binding to damaged binding sites is widespread [20], and that many transcription factors, including the ETS-family members ETS1 and ELK1, show altered DNA-binding specificity for UV damaged DNA. However, the molecular mechanisms that regulate transcription factor binding to damaged DNA sites remain unclear.

DNA binding sites of ETS-family members are among the most highly mutated noncoding elements in melanoma [9, 21–27]. These mutation hotspots occur at least in part due to induced CPD formation at specific locations in ETS-bound DNA sites [24–28], as ETS transcription factor binding alters the DNA structure so that C5–C6 double bonds of neighboring pyrimidine bases have distances and torsion angles favorable to CPD formation [26]. One of these damage and mutation hotspots occurs at the central TpC step in the central 5′-(A/T)TCC-3′ ETS binding consensus. Notably, a recent genome-wide analysis indicates the TpC CPDs at this location in the ETS motif rapidly deaminate to uracil (i.e. to a TpU CPD) and are repaired slowly by the NER pathway in vivo [14]. A recent study indicates that the Class I ETS transcription factor ETS1 may inhibit repair by blocking accessibility of CPD lesions to the NER damage sensor UV-DDB, at least in vitro [20]. However, the molecular mechanism by which ETS factors accommodate a CPD lesion in their DNA binding site is unclear. This is particularly the case for Class II and III ETS-family members, which comprise more than half of the 28 human ETS paralogs [29] and are also normally [30] as well as pathologically [31] expressed in epidermal and dermal tissues with exposure to UV radiation. While the structures of CPD-containing DNA alone [32] or in complex with DNA repair factors such as photolyases [33, 34], UV-DDB [35], or xeroderma pigmentosum C [36] has been previously characterized, how ETS proteins or other transcription factors accommodate a CPD lesion in their DNA binding site is not known.

To gain insight into these questions, our objectives in this work were three-fold. First, we determined the DNA-binding properties of archetypal members from three ETS-family classes (I–III), to test whether binding and repair inhibition at CPD-bearing DNA sites represents a class property of the ETS family. Second, we solved high-resolution co-crystal structures to gain insight into the impact of CPD and downstream Watson-Crick mismatches (due to CPD deamination) on ETS/DNA complexes. Third, we dissected the structural thermodynamics of ETS binding to these DNA substrates to probe the energetic and dynamic basis of CPD formation and recognition by ETS proteins. The results provide a deep characterization of how a bulky helix-distorting lesion is accommodated within ETS transcription factor binding sites and interferes with subsequent repair, thereby promoting mutation hotspots at ETS binding sites in skin cancers.

Materials and methods

Nucleic acids

Synthesis of the CPD oligos is detailed in Supplemental methods. Non-CPD oligos were purchased from Integrated DNA Technologies (Midland, IA). Dried oligos were dissolved in aqueous 1.0 M NaCl and desalted on G-25 resin (HiTrap Desalting columns; Cytiva) in water under the control of an ÄKTA start (Cytiva) system. The salt-free high-MW fraction was lyophilized and redissolved in 30 mM HEPES (pH 7.5) containing 100 mM potassium acetate. Duplexes were annealed from equimolar amounts of the desired complementary sequences (in ~200 μl volume) by heating to 90°C followed by passive cooling to room temperature. Duplex concentrations were determined by UV absorption at 260 nm using the nearest-neighbor extinction coefficients for the WC sequence [37].

Protein purification

pET28b-based (Novagen) constructs encompassing the C-terminal ETS domains of murine Ets-1 (residues 280–440, a kind gift from Dr Lawrence McIntosh) or human PU.1 (residues 165–270) were transformed into BL21(DE3)pLysS Escherichia coli (Invitrogen). Residues 161–340 of human ELF1 encompassing its internal ETS domain were cloned into pCDF-1b (Novagen) and similarly transformed. Cultures in LB medium were induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside at OD₆₀₀ of 0.6 for 4 h at 30°C. Harvested cells were re-suspended in Buffer H [10 mM HEPES, pH 7.4, with 0.5 M NaCl and 0.5 mM Tris(2-carboxyethyl)phosphine) containing 1 mM phenylmethylsulfonyl fluoride] and lysed by sonication. The lysate was cleared by centrifugation and loaded onto a HiTrap SP HP column (Cytiva) equilibrated with Buffer H. After washing, the protein was eluted along a linear NaCl gradient under the control of a Bio-Rad NGC instrument. Samples for co-crystallization were concentrated in Amicon Ultra-15 centrifugal filters (10k MWCO) and then polished and exchanged into Buffer H on a HiLoad 16/600 Superdex 75 column (Cytiva). Protein concentration was determined by UV absorption at 280 nm based on extinction coefficients (in M⁻¹ cm⁻¹) of 39 880 for Ets1, 28 420 for ELF1, and 22 460 for PU.1.

Electrophoretic mobility shift assays

6-FAM-labeled duplex oligonucleotides (200 nM) harboring native and variously modified ETS-binding sequences of the DPH3 promoter were titrated with ETS domain in 25 mM Tris–HCl (pH 7.9) containing 0.5 mM ethylenediaminetetraacetic acid, 6 mM MgCl₂, 60 mM KCl, 10% glycerol, 10 mM dithiothreitol, and 0.12% w/v of bovine serum albumin (BSA) in a total volume of 50 μl. The DNA–protein mixture was incubated on ice for 45 min and 5 μl of the mix was loaded onto an 8% native polyacrylamide gel. Electrophoresis was performed at 30 V/cm for 30 min. Gels were imaged using a Typhoon FLA700 scanner (Cytiva). Bands were quantified using ImageQuant software (Cytiva).

T4 pyrimidine DNA glycosylase inhibition assays

From the gel mobility shift samples, 10 μl of DNA and ETS domains were treated with 10 U of T4 Pyrimidine DNA glycosylase (PDG) (New England Biolab) at 37°C. Samples were quenched after 15 and 30 min with 2 μl of 1 M NaOH and heated at 95°C for 15 min. Formamide was added to 70% v/v and the samples were further denatured at 95°C for 15 mins. From each sample, 10 μl was loaded onto a 12% polyacrylamide gel containing 8 M urea in 0.5× TBE buffer. Electrophoresis was carried out at 30 V/cm at 60°C for 20 to 25 min. Gels were imaged using a Typhoon FLA7000 imager (Cytiva) quantified using ImageQuant software (Cytiva).

X-ray crystallography

Purified protein was mixed with target DNA at 1:1 molar ratio in Buffer H to yield a complex concentration of 200 µM. Co-crystals were grown for 5 days by vapor diffusion at 293 K in a 2 μl hanging drop comprised of a 1:1 mixture of protein:DNA complex with mother liquor containing 100 mM sodium acetate (pH 4.6) and 2% PEG 3350. Prior to freezing, 2 μl of a cryoprotectant solution containing 100 mM sodium acetate, 2% PEG 3350, and 20% glycerol was laid on top of the hanging drop and the well closed for 1 h of incubation (4 μl total volume, 10% glycerol concentration). After 1 h, crystals were transferred to the above 20% glycerol solution prior to freezing. X-ray diffraction data sets were collected at the National Synchrotron Light Source II at Brookhaven National Laboratory (Upton, NY).

The diffraction data were processed using the XDS package [38] and scaled using Aimless in the CCP4 package [39]. Molecular replacement was performed using a PU.1 co-crystal complex (PDB: 8E3K) as the search coordinates in the PHENIX suite [40] via the maximum-likelihood procedures in PHASER. Rounds of refinement were then carried out using phenix.refine [40] followed by model building in Coot [41]. Refinement statistics and information regarding specific beamlines, detectors, collection wavelengths, and oscillation angles are given in Supplementary Table S1.

Fluorescence polarization titrations

A Cy3-labeled DNA probe [42] encoding a high-affinity sequence for PU.1 (5′-AGCGGAAGTG-3′; 0.5 nM) was incubated with graded concentrations of protein and unlabeled DNA to equilibrium (overnight at 25°C) in 10 mM Tris–HCl (pH 7.5) containing 0.01% BSA and 0.15 M NaCl. Steady-state fluorescence anisotropies were measured at 595/30 nm in a Molecular Dynamics Paradigm micro-plate reader with 530/30 nm excitation. Observed anisotropies, computed from buffer-subtracted plane-polarized fluorescence, were reported as means ± S.D. of three or more experiments:

(1)

where represents the anisotropy of the free DNA probe and refers to the ith bound species. F is the fractional bound probe and was analyzed as a function of total protein concentration as follows.

Model-dependent analysis of fluorescence polarization titrations

As previously established by calorimetric [43] and diffusion measurements [44] in solution, PU.1 (P) binding to a single cognate DNA site (D of species j) sequentially with negative cooperativity:

(2)

where and are the equilibrium dissociation constants describing the 1:1 and 2:1 PU.1/DNA complexes (and ):

(3)

The concentrations of the various species were solved by numerical root finding, as detailed in [42], to compute the fractional bound probe (taken as species ), :

(4)

This approach accounts for mutual depletion among all species, a detailed exposition of which is found in [45], and yields absolute binding affinities (without reliance on IC₅₀ values). To reduce the degrees of freedom in competition (j = 2) experiments, protein was deployed at sub-saturating concentrations with respect to the probe. Nonlinear least square regression of the system of equations consisting of Eqs. (1) and (3) was performed using Origin software (Northampton, MA).

Isothermal titration calorimetry

Purified DNA and ETS domain were extensively co-dialyzed in 10 mM NaH₂PO₄/Na₂HPO₄ at pH 7.4 with 0.15 M NaCl. Dialysate was used in all dilution and rinsing procedures. DNA was loaded into the cell (V = 1.4 ml) at a nominal concentration of 10 µM and protein at 200 µM in the syringe (300 μl) of a VP-ITC instrument (Malvern). DNA (or buffer) was titrated, with constant stirring, with an initial 2.0-μl injection, followed by 30 9.0-μl injections of protein spaced at 300-s intervals at 25°C. After baseline subtraction, heat peaks were integrated and expressed as  i.e. heat per mol of injected titrant as a function of the titrant:titrate molar ratio X (with volume corrections made by the acquisition software).

Experimental per-injection heats were modeled according to an initial value problem as described previously [46]:

(5)

where the indices 1 and 2 on the enthalpy changes refer to the sequential species PD and PPD, and 0 corresponds to the baseline corresponding to the residual heat of protein dilution. The three derivatives that satisfy the sequential binding model as embodied by Eq. (2) are:

(6)

The macroscopic equilibrium constants and are as defined in Eq. (3). is transformed from experimental independent variable X as follows:

(7)

where the subscript “t” refers to the total concentrations and n is a correction factor for capturing concentration deviations from nominal values. Typically, 0.9 < n < 1.1. Numerical solution of the initial value problem by Runge–Kutta methods was implemented using the d02pvc/d02pdc functions from the NAG C (Numerical Algorithms Group, Oxford, UK) library, as previously described [46], and nonlinear regression was performed using Origin. The reported enthalpy change for the 1:1 PU.1/DNA complex (corresponding to the co-crystal structures) was taken as the fitted difference from the first transition to the final baseline at high-excess of titrant, adjusted to the experimental heats from protein-into-buffer titrations.

Molecular dynamics simulations

Molecular dynamics simulations were performed on various PU.1/DNA complexes with and without CPD (Supplementary Table S2) as well as free DNA. DNA duplexes were constructed in B-form geometry using 3DNA [47], while co-crystal structures were used as initial coordinates for the complexes (Supplementary Table S2). Structures were prepared in CHARMM [48] and simulated with OpenMM [49] using the CHARMM36 force field [50, 51] with CPD parameters [52]. Systems were solvated in cubic boxes of 0.15 M NaCl in TIP3P water [53] with solvent margins of at least 11.5 Å. After minimization, heating, and equilibration, systems were propagated for at least 250 ns in the NPT ensemble at 1 atm and 300 K using a 2 fs time step; long-range electrostatics were treated with the particle mesh Ewald method [54] and SHAKE [55] was used for all covalent bonds involving hydrogen atoms. Heating was performed in the presence of harmonic restraints on the backbone, which were released during equilibration. To prevent fraying, the hydrogen bonds of the terminal A₁:T₃₄ and T₁₈:A₁₇ base pairs were restrained with a force constant of 1 kcal/(mol Ų) during all stages. At least three independent copies were run at least 250 ns and up to 500 ns of each system, using different random seeds for heating.

Structural analyses

H-bond and contact analyses were carried out using CHARMM [56]; visual analyses with VMD [57]. Clustering of features by single linkage was performed in the GROMACS environment [58]. DNA base, step and helical parameters were computed using 3DNA [47].

Results

Paralogs from all major ETS classes bind cognate CPD-containing motifs

To characterize how CPD lesions impact the binding of ETS-family transcription factors, we synthesized double-stranded oligonucleotides containing site-specific CPD lesions in a model ETS binding site. This site, derived from the DPH3/OXNAD1 bidirectional promoter (henceforth DPH3 promoter), is highly mutated in sequenced melanoma tumors due to ETS-induced CPD formation hotspots [9, 26–28, 59]. We synthesized oligonucleotides with a cis-syn TpT CPD starting at positions −2 (CPD₋₂) and −1 (CPD₋₁) relative to the ETS motif midpoint (Fig. 1A). Position −2 corresponds to a CPD forming between two conserved thymine residues in the ETS binding consensus, while position −1 models a deaminated CPD (i.e. TpU) initially forming at a TpC base step (Fig. 1A). Previous studies have shown that ETS transcription factor binding does not induce CPD formation at position −2 but induces a CPD hotspot at position −1 [24–26]. Importantly, TpC CPDs at position −1 rapidly deaminate to TpU in cells [14] and may play a role in mutagenesis.

Figure 1.

Complex formation by ETS transcription factors with CPD-containing and mis-matched DNA within the ETS-binding motif of the DPH3 promoter. (A,) Sequences of the DNA targets studied. For concision, only the pyrimidine-rich sequences of the target duplexes are listed explicitly. (B) Schematics of the domain structure of murine Ets-1 (Class I), human ELF1 (Class II), and PU.1 (Class III). (C) Representative gel electrophoretic mobility titrations of the DNA targets by the three ETS representatives. Colored triangles denote the mobility of the putative 1:1 ETS/DNA complex. Gel images are rendered as digitized without adjustment. DNA logos for each factor were taken from the CIS-BP database [63]. The ETS:DNA ratios (r) are as follows: Ets1:(0.5, 1, 2, 6, 11, 23, 46); ELF1:(1, 2.5, 5, 9, 17, 26, 52); PU.1:(0.5, 1.5, 2.5, 5, 13, 26, 39). Black asterisks refer to ratios (r = 2 or 2.5) used in comparing lane profiles in Panel E. Red asterisks refer to the ratios used in the T4 PDG cleavage assay in Fig. 2. Gel images for the full set of triplicate experiments are shown in Supplementary Fig. S1. (D) Apparent avidity of ETS to the DNA targets in terms of ETS:DNA ratios at 50% loss of unbound DNA. Bars represent means from triplicate experiments ± S.D. Asterisks denote significant pairwise difference from native (P < .05, ANOVA/Bonferroni). (E) Representative lane profiles at r = 2 to 2.5 for each ETS/DNA complex, aligned by the unbound(U) DNA. The 1:1 complex is taken as the discrete low-mobility species in titrations of native DNA (c.f., Supplementary Fig. S2).

To evaluate the potential for ETS paralogs from all three classes to engage these CPD-containing DNA sites and inhibit their repair, we compared the binding of the purified ETS domains of Ets1 (Class I), ELF1 (Class II), and PU.1 (Class III) (Fig. 1B) to duplex oligonucleotides encoding the native (i.e. undamaged) or CPD-modified DPH3 sequences. These ETS paralogs were selected based on their representativeness of their respective class and (to maximize the robustness of the findings) the diversity of their domain architecture [29]. Since the TpT CPD at position −1 results in a T:G mismatch (MM), we also analyzed ETS binding to an oligonucleotide with a T:G MM at the same location (Fig. 1A).

Titration of the native DPH3 promoter with the ETS domains of the three paralogs yielded a mixture of bound species by electrophoretic mobility shift in a concentration- and paralog-dependent manner (Fig. 1C and Supplementary Fig. S1). Ets1 readily formed a single discrete complex even at a sub-stoichiometric concentration of protein. This species corresponds to a 1:1 protein–DNA complex based on established characterizations [60, 61] of Ets1 binding to a single cognate site. Addition of excess Ets1 led to secondary complexes with reduced mobility. Similarly, ELF1 preferentially formed a discrete species but at higher concentrations than Ets1. Across comparable concentrations, however, PU.1 induced diffuse band shifts that migrated with higher mobility than the discrete complexes of Ets1 and ELF1. Since PU.1 prefers consensus-flanking pyrimidines, particularly the +3 positions, we tested PU.1 binding with a PU.1-specific DNA site and observed a discrete complex (Supplementary Fig. S2). The diffuse, high-mobility band shifts were thus consistent with complexes that were kinetically labile to dissociation during native electrophoresis.

On this basis, the presence of a cis-syn dithymidyl (TpT) CPD dimer perturbed ETS binding to the altered motif in a position-dependent manner. Replacement of the TT dinucleotide in the 5′-TTCC-3′ consensus by a CPD (CPD₋₂) impaired binding by all three ETS paralogs (Fig. 1C). Relative to the native sequence, the discrete Ets1-bound band was nearly abrogated, and band shifts of CPD₋₂ by ELF1 were similarly diffuse as those formed by PU.1. In sharp contrast, a centered CPD in the consensus (CPD₋₁, 5′-TTTC-3′), corresponding to the most frequently damaged position in the ETS-bound motif in UV-irradiated fibroblasts [26], produced almost identical titrations for Ets1, as did the isomeric TT MM. For ELF1, the CPD₋₁ and MM sequences not only recovered the discrete complex but did so at a lower protein concentration than with the native sequence (Fig. 1C). In the case of PU.1, CPD₋₁ and MM gave rise to discrete PU.1-bound complexes that were negligibly populated with the native sequence.

Given the electrophoretic heterogeneity of ETS/DPH3 complexes (discrete, super-, and under-shifted bands), we quantified the band shifts using two different approaches. We first considered binding in the aggregate by tracking the fractional intensity of the unbound DNA band (Supplementary Fig. S3) and expressing the apparent avidity as the ETS:DNA ratio r at 50% loss of unbound DNA i.e. 50% of DNA shifted (Fig. 1D). Two-way ANOVA of this metric, factored by ETS and DNA identities, showed statistically significant differences (P < 10^–4) in both factors. All post-hoc pairwise comparisons among ETS and DNA pairs were also significant (P < .05, Bonferroni) except for CPD₋₁ and MM. Relative to the native, undamaged DNA sequence, CPD₋₂ significantly decreased binding by each ETS paralog, while CPD₋₁ and MM significantly increased binding affinity by each ETS paralog (Fig. 1D, P < .05). The effect of MM on ETS binding is consistent with previous observations that mismatched DNA could enhance binding of Ets1 over fully WC-paired sequences [62]. Our data indicate that a mismatch associated with a deaminated CPD (TpT) also promotes binding by all three ETS members.

To address the discrete 1:1 ETS/DNA complex specifically, we also examined the lane profiles at a comparable sub-saturating ETS:DNA ratio (r = 2 to 2.5). The profiles (Fig. 1E) were consistent with the lost-unbound analysis and emphasized the preferential formation of discrete complexes by CPD₋₁ and MM in the case of PU.1. Taken together, the presence of CPD thus significantly modified DNA binding by ETS proteins from all three classes. Importantly, the primary ETS-induced lesion, as embodied by CPD₋₁, enhanced site-specific binding by ETS factors generally, including those (such as PU.1) that bind the native sequence poorly. The similar binding profiles for CPD₋₁ and MM further showed that strong binding by ETS factors was likely maintained following rapid deamination of the TpC CPD formed in the 5′-TTCC-3′ consensus to TpU.

ETS binding inhibits cleavage of a CPD lesion by a model repair enzyme

The position-dependent effects of CPD binding by ETS-family relatives prompted us to ask whether these perturbations could impact the accessibility of the lesion to repair. To test this possibility, we used the repair enzyme T4 pyrimidine dimer glycosylase (T4 PDG, also known as T4 endonuclease V), which specifically cleaves CPDs, as a biochemical model to probe the competitive inhibition of DNA repair by ETS proteins. The CPD substrates (0.2 µM), alone or in the presence of excess ETS proteins, were treated at 37°C with 10 U of T4 PDG for up to 30 min. The timescale of the assay was >40-fold the residence time required for cleavage based on reported values (k_cat > 0.05 s⁻¹) of T4 PDG catalytic turnover [64].

T4 PDG-mediated cleavage of CPD₋₂, which bound ETS proteins more poorly than the native sequence, was essentially complete by 15 min and was quantitatively unaffected by any of the ETS proteins (Fig. 2A), which were present at >10-fold molar excess relative to DNA (lanes marked with red asterisks in Fig. 1C). At these concentrations, saturation of the CPD₋₂ duplex ranged from ~50% (PU.1) to ~90% (Ets1) based on the gel electrophoretic shift data (c.f., Supplementary Fig. S2). Further excess of PU.1 to >80% saturation had no effect on CPD₋₂ cleavage (Supplementary Fig. S4). In contrast, cleavage of CPD₋₁, which exhibited enhanced binding by ETS proteins (Fig. 1D), was also strongly inhibited by each of the ETS paralogs. Whereas cleavage of unbound CPD₋₁ had reached equilibrium by 15 min, cleavage in the presence of ETS proteins was still occurring at nearly initial velocities at 30 min. The efficiency of repair inhibition was significantly different (P < 10^–4 by ANOVA) and followed the rank order (initial velocities in pmol/min): ELF1 (0.28 ± 0.01) > Ets1 (1.1 ± 0.1) > PU.1 (2.9 ± 0.1).

Figure 2.

T4 PDG activity at CPD substrates is sensitive to DNA sequence context and ETS binding. (A) Representative denaturing polyacrylamide gels of untreated and T4 PDG-cleaved CPD-bearing substrates sampled at the indicated time points. Gel images are rendered as digitized without adjustments. Fractional cleavage of the substrates under each ETS/DNA combination and condition (filled and open symbols for with and without protein) was analyzed as means ± S.D. from triplicate experiments. Cleavage of CPD₋₂ is quantitatively similar with and without protein at corresponding time points. Dashed lines represent the initial velocity estimated with the 15 min time point. P-values from a two-tailed t-test for the difference at 30 min are given for the CPD₋₁ sequence. (B) Lane profiles of ETS/CPD_-1 complexes under the PDG assay conditions (c.f., Fig. 1C), referenced against the native DPH3 complex at the same ETS:DNA ratio. The beige regions represent the discrete (1:1) Ets1-bound native DPH3 complex whose intensities are taken as a fraction of the total in the (gray) band-shifted regions. The 1:1 band for PU.1 was referenced against its optimal complex (c.f., Supplementary Fig. S1). (C) Initial velocity of T4 PDG cleavage of CPD₋₁ plotted against the fractional intensity of the 1:1 complex in the bound ensemble. Points represent the means ± S.D. of triplicate independent experiments.

As the levels of CPD₋₁ saturation by the three ETS paralogs were similar at ~90%, this alone could not account for the differential inhibition of T4 PDG. We therefore considered the composition of the ETS-bound DNA as resolved in the electrophoretic shift data (Fig. 2B). Taking the discrete 1:1 complex of Ets1 with the native DPH3 site as reference, we observed that the 1:1 complex was most enriched within the bound ensemble in the order ELF1 > Ets1 > PU.1. While the kinetic interplay between ETS binding and inhibition of a competing enzyme could not be simply dissected with the present measurements, the strong differences (>10-fold) in T4 PDG inhibition among the ETS paralogs did correlate with a correspondingly large difference in formation of the 1:1 complex (Fig. 2C).

In summary, corresponding concentrations of ETS proteins that inhibited repair of CPD₋₁ had no effect on repair of CPD₋₂. Biophysical differences in CPD binding by ETS proteins are therefore reflected in biochemical inhibition of the model DNA-repair enzyme T4 PDG, consistent with dynamic competition between ETS and T4 PDG for CPD substrates, and indicate that binding of ETS factors to CPD-containing cognate sites can inhibit accessibility of the lesion to repair enzymes.

Structural bases of ETS interactions with damaged and mismatched DNA

The strong binding by ETS factors at the CPD_-1 and MM DNA sites, compounded by the correlation between 1:1 complex formation and CPD repair inhibition, motivated a need for detailed structural characterizations. In extant ETS/DNA co-crystal structures, with undamaged DNA, paralogs from all classes of the ETS superfamily demonstrate strong conservation in tertiary structure as well as key DNA contacts by signature residues, principally two arginine sidechains in the recognition helix (H3) that engage the 5′-GGA(A/T)-3′ consensus. Even in complex sequence contexts [65, 66] encoding tandem copies of the consensus motif, ETS domains adopt quaternary structures that conserve the architecture of each subunit by swapping corresponding secondary elements [67]. This structural fidelity is not shared by the DNA-binding domains of other transcription factor families in the helix-turn-helix superfamily, for example the homeodomains and POU family [68].

To gain structural and biophysical insight into the recognition of CPD and mismatched consensus DNA through co-crystal structures of DNA bound by ETS factors, we have selected PU.1 for three reasons. The first reason is its strong binding to CPD₋₁ relative to Ets1 and ELF1. The second reason relates to the sensitivity of DNA structure in crystallo to packing effects, which imposes a need for ETS/DNA complexes with equivalent crystal packing. PU.1 co-crystallizes with a broad range of DNA sequences in an identical space group with the additional benefits of a single complex per asymmetric unit and favorable diffraction characteristics [69]. Third, the PU.1/DNA complex with its optimal sequence (c.f., Supplementary Fig. S2) has among the highest resolutions among ETS co-crystal structures. We confirmed that this DNA sequence was also recognized by Ets1 and ELF1 (Supplementary Fig. S5). This complex (PDB: 8E3K, herein referred to as WC) is therefore an ideal benchmark for comparative analysis of altered interactions and dynamics induced by structural perturbations such as CPD and mismatches.

Co-crystallization of the ETS domain of PU.1 with variants harboring a (TpT) CPD at the center of the 5′-TTCC-3′ (CPD; PDB 9YNZ) or the mismatch (MM; PDB 9OA4) 5′-TTTC-3′ yielded complexes (Fig. 3A and B) packing in the identical space group (P 1 2₁ 1) as the undamaged WC complex. In all the complexes, contacts from the ETS domain were made predominantly in the 5′-GGAA-3′ strand (Fig. 3C). The purine strand in CPD showed similar contacts with the ETS domain, adjusted for resolution, as WC. In the MM structure, the wobble-paired guanine (with thymine) is shifted away from the protein, and contact was maintained through an extended sidechain conformation by R230.

Figure 3.

Intermediates of dipyrimidine modifications in complex with the ETS-family transcription factor PU.1. (A) Schematic of the ETS domain of human PU.1 (residues 165–272; denoted ΔN165) and key DNA oligonucleotides used in the co-crystallization experiments. (B) Refined co-crystal structures. (C) Contacts made by the key interfacial residues Q226, R230, and R233. Dashes denote H-bonds within a distance cutoff of 3.5 Å between heteroatoms. The absence of ordered hydration in the CPD-containing structure may be related in part to resolution (1.9 Å), so judgement on water-mediated contacts are reserved (square brackets). (D) Comparison of 2mFo-DFc maps at a common relative cutoff of 2.0 σ. The variable dinucleotide steps are rendered in color. (E) B′-factors of each bound DNA mapped to the modeled structures. Two standard deviations in B′-factor (calibrated internally to each structure) are rendered to resolve the implied dynamic distribution in the DNA. (F) Per-residue ΔB′-factors for heavy atoms between the ETS-bound mismatched and CPD-bearing DNA highlight major differences in implied DNA dynamics between the CPD and MM complexes. (G) DNA structures of the CPD (green) and MM (gold) complexes, aligned to the bound protein. (H) Major and minor groove widths, taken as the distance between P atoms of the ith and i + 3th base.

Over the whole DNA, protein-bound CPD showed more distinct differences from the parent WC and MM sequences (Fig. 3D). CPD exhibited lower electron density (2mFo-DFc) over the pyrimidine dimer while coverage elsewhere was comparable with the WC sequence. We considered the possibility of photoreversion by the TpT CPD into TT, effectively converting CPD to MM, during handling. This possibility was discounted by the qualitatively distinct protein/DNA interfaces of CPD and MM (Fig. 3C), as well as the 2mFo-DFc coverage of MM showing reduced density at positions flanking the TT base step, in contrast with CPD (Fig. 3D). To test the robustness of the distinct 2mFo-DFc map in MM, we solved a second structure in which inosine replaced guanine in the mismatched base pair and made the same wobble contacts but with a weaker stack. The resultant complex (I-MM; PDB 9OB0), which crystallized to an equivalent asymmetric unit (Supplementary Fig. S6A), showed a similar 2mFo-DFc map as MM (Supplementary Fig. S6B).

To directly establish that the CPD-containing DNA (Supplementary Fig. S7A) was intact, we probed its structure by solution ¹H NMR, focusing on the imino-¹H fingerprint region. When annealed with the same complementary strand as the native sequence, the CPD duplex showed the full complement of imino chemical shifts for G and T bases. Assigned peaks at and adjacent to the CPD were distinct in comparison with MM and unmodified DNA (Supplementary Fig. S7B). Independently, we probed the CPD enzymatically with T4 PDG and observed quantitative cleavage of a ligated duplex product (Supplementary Fig. S7C–E). The CPD duplex was therefore intact at least up to crystallization.

Based on the evidence, two scenarios can account for the low bridging density of the bound CPD. The CPD might have been split by synchrotron radiation, in which case the diffraction captured a conformational state prior to relaxation to the MM conformation. The second possibility, not exclusive of the first, is that the 2mFo-DFc maps reflected in part molecular dynamics in the bound DNA. With these considerations, we compared the (z-normalized) B′-factors, which quantify isotropic disorder in the crystal, for the ETS-bound WC, CPD, and MM sequences. Except for a localized hotspot adjacent to the TpT dimer, the DNA dynamics in CPD are distributed relatively evenly around their average along the DNA-contacting segment of the DNA duplex (Fig. 3E). This contrasts with the biased distribution in MM (Fig. 3F), which is similar to the native WC structure and other non-CPD PU.1/DNA co-crystals of the same space group. Structurally, the reduced base-base distance imposed by the TpT dimer disturbs the local helical conformation (Fig. 3G), such that the major groove is locally narrowed and the minor groove enlarged just upstream of the CPD (Fig. 3H). The ETS-bound CPD therefore exhibits distinct structural and dynamic properties as manifest in crystallo by way of the distribution of electron density, implied dynamics, and helical conformation.

Probing the structural dynamics of CPD-bound ETS complexes

To gain more insight into the impact of CPD in the PU.1-bound complex, we simulated the dynamics of the co-crystallographic models as well as other sequence contexts that were not examined. We used the TpT CPD co-crystal structure as template to generate a set of reference dynamics (1TT) over 500 ns for comparison with other CPD-containing complexes. Excluding transient interactions with mobile DNA termini absent in crystallo, 1TT reproduced the core H-bonds (Fig. 4A), based on a 3.5 Å distance between heteroatoms and 30° angular cutoffs, and inter-residue contacts (Supplementary Fig. S8A) present in the co-crystal structure. The 1TT ensemble exhibited dynamic H-bonds between R230 and a proximal imino O from the CPD, which is beyond the cutoff in the CPD co-crystal (3.8 Å), in addition to several transient H-bonds with the DNA backbone (Supplementary Fig. S8B).

Figure 4.

CPD-dependent structures and dynamics of ETS complexes. (A) H-bonded contacts between PU.1 and DNA in the (TpT) CPD-bound co-crystal (PDB: 9YNZ) and their frame-averaged equivalent in simulated complexes templated from 9YNZ. Horizontal dashed lines connect the dynamic H-bond between the R230 sidechain and CPD, which is absent for 1TC. (B) Structures of the R230-CPD interaction. The 1TT and 1TC models are the central structure from a clustering procedure of the MD ensembles aligned on the two residues and their complements. (C) Comparison of major and minor groove widths of 1TT and 1TC with their non-CPD counterparts. Points represent means ± SD of one trajectory. Results from three independent trajectories are shown in shades. (D) Stagger, buckle, and tilt of ETS-bound DNA. Points and shades are as designated in panel (C). Results for additional helical parameters are given in Supplementary Fig. S9. (E) H-bonded contacts between PU.1 and DNA in the WC (8E3K) co-crystal and average values in simulated complexes. The arrow in 2TT marks the absence of R200-mediated contacts present in all other simulated complexes. (F) Models of WC, cluster-centered structures of the 1TT and 2TT ensembles. The models are aligned on the recognition helix H3. (G) Helical curvature in PU.1-bound DNA as indicated shortened end-to-end helical distance relative to ideal B-DNA (0.34 nm rise per base step, dashed). Results from triplicate simulations were analyzed by ANOVA (*P < .05 by Bonferroni pairwise comparison).

On this basis, we simulated complexes containing a TpC (1TC) or TpU (1TU) CPD that was not available experimentally. Neither construct presented significant differences from 1TT in terms of core H-bonding (Fig. 4A) or inter-residue contacts more generally (Supplementary Fig. S8A). The absence of direct H-bonding in 1TC (and WC) between R230 and the CPD was only due to the absence of a suitably spaced H-bond acceptor in the TpC CPD (Fig. 4B). The exocyclic NH₂ of the cytosine in TpC (and the non-CPD 5′-TC-3′ step in WC) remained within 4.5 Å of a charge-dipole contact with the R230 sidechain (Supplementary Fig. S8A). This analysis indicated that ETS/DNA contacts were conserved in dimerized TpY base steps at the −1 position, regardless of the base identity of Y in the CPD, suggesting that physiologically relevant TpC (and TpU) CPDs at this location have a similar mechanism of binding by ETS proteins.

To evaluate the effect of CPD identity on helical structure in the complex, we examined groove widths in 1TT and 1TC (Fig. 4C). For each CPD, the modulation of both major and minor grooves tracked more closely (albeit incompletely) with their non-CPD counterparts i.e. MM for 1TT and WC for 1TC respectively, than between the two CPD-containing homologs. CPD identity thus perturbed helical structure to conserve contacts with protein. Probing these helical perturbations more locally, we examined base and base-step parameters among which stagger, buckle, and tilt are particularly relevant given the geometry of CPDs (Fig. 4D). While these parameters showed highly localized perturbations at and adjacent to the CPDs, the perturbations in 1TT were not only different from 1TC (as expected based on groove widths) but also from MM. Other helical parameters showed similar trends (Supplementary Fig. S9). Through local base and base-step perturbations, ETS binding thus partially “corrects” CPD effects on groove widths to retain conservation of DNA contacts.

A second point of distinction between 1TT and 1TC was seen in the stronger dynamics in 1TT. Among the replicate trajectories, which were seeded from the same structure but with different velocities, 1TT but not 1TC showed more variable groove widths (Fig. 4C) as well as local helical parameters than 1TC. In 1TT, stochastic helical fluctuations at DNA positions ± 1 from its CPD did not average out over up to 500 ns of each simulation (Fig. 4D), revealing energetically isolated geometries that were not sampled by 1TC (nor MM). These base- and CPD-specific dynamics echoed the local crystallographic disorder observed in the CPD structure c.f., Fig. 3D and F.

In addition to TpY analogues of CPD at the −1 position, we also simulated a TpT analogue at the −2 position (2TT), which was isomeric to WC. We reasoned that 2TT was a meaningful construct given that PU.1 bound WC to form the 1:1 complex specifically, as qualified by the gel-shift experiments (Supplementary Fig. S2), but negligibly in the homologous sequence context in the DPH3 promoter (c.f., CPD₋₂ in Fig. 1C). Triplicate simulations showed that 2TT was not conservative relative to WC in terms of H-bonds with PU.1 (Supplementary Fig. S10). In WC, the sidechain of R220 (located in the loop connecting helix H2 and H3) dynamically H-bonded with residues along the 3′ flanks of consensus in the GGAA strand. These contacts were entirely absent in 2TT (Fig. 4E). Indeed, these contacts were present in the simulated 1TT/1TU/1TC and (CPD-free) MM complex, suggesting a conformational feature specific to a CPD specifically in the −2 position.

Structural alignment of the simulated models to WC revealed position-dependent distortions in the PU.1-bound CPD-containing DNA (Fig. 4F). In addition to a tilting of the overall helix in 2TT that shifted R220 out of register with flanking DNA contacts, the other side of the flanking DNA (contacted by the “wing”) was locally bent to yield a highly narrowed minor groove. Overall, the bound 2TT DNA was significantly bent as manifest by a shortened end-to-end distance of the helix (Fig. 4G).

In summary, the presence of a CPD resulted in structural and dynamic perturbations that were strongly position-dependent. At the consensus-centered −1 position, a CPD induced highly local perturbations that partially compensate to conserve groove widths and core contacts with protein. In contrast, a TpT CPD at the −2 position (2TT) induced nonlocal effects that propagated across the full helical turn of DNA to which the ETS domain was bound. These latter distortions were altogether CPD- and position-specific, as the DNA in 2TT was not conserved in contacts or structure with either WC (the non-CPD analogue of 2TT) or 1TC (an isomeric CPD of 2TT).

ETS binding to CPD is thermodynamically linked to CPD-induced dynamics in the free DNA

To experimentally relate the co-crystal PU.1/DNA structures to complex stability, we measured the binding affinity of PU.1 for these targets in solution using a fluorescence polarization assay (Fig. 5A). Relative to the undamaged WC, the affinity for CPD was equivalent within experimental uncertainty, while MM was ~10-fold weaker. These results revealed a role for sequence context in modulating mismatch recognition by PU.1. The weaker binding of MM in a canonical (Class III) motif by PU.1, where it bound more tightly an analogous mismatch in the context of the (Class I/II) DPH promoter biased towards Ets1 and ELF1, suggested divergent conformational penalties imposed by Class III vs. Class I/II ETS motifs. We further probed the role of the wobble G:T base pair in the MM using the I-MM sequence. In a Watson–Crick G⋮C context, G > I substitution reduces PU.1-binding affinity due to an uncompensated increase in the local conformational entropy of the wobble-paired I:C base pair in the unbound state [70]. Contrary to this, I-MM bound PU.1 with indistinguishable affinity as MM, suggesting that negligible additional conformational entropy is lost in the DNA upon PU.1 binding.

Figure 5.

Comparative thermodynamics of ETS interactions with CPD and mismatched DNA. (A) Equilibrium binding of the PU.1 ETS domain (ΔN165) to the four DNA sequences from the co-crystal structures (WC, CPD, MM, I-MM) as determined by fluorescence polarization. Curves represent fits of the data of a mechanistc model in which PU.1 binds sequentially up to a 2:1 complex. Intrinsic affinities to the four sequences were determined by displacement of a PU.1-probe complex at sub-saturating protein concentrations. Points represent mean anisotropies ± S.D. of triplicate experiments. (B) Isothermal titration calorimetry at 25°C. Curves represent fits of the same sequential binding model to the thermal data. The change in enthalpy (ΔH; long dashes) is computed relative to the apparent heat of diluting ΔN165 into buffer. Bounds for ΔH are given in the case of I-MM because the fitted value is smaller than the error in the estimate. (C, D) Dependence of PU.1 binding to CPD and WC on mean ionic activity (NaCl) or osmolality (nonionic co-solutes). Points represent means ± S.D. of triplicate determinations of the equilibrium dissociation constant K_D. The data points at 0.5 M Na⁺ are offset horizontally to clear their overlap. Osmotic pressure data are labeled by the osmolyte used at 0.15 M Na⁺. AcNH, acetamide; GB, glycine betaine; MG, α-d-methyl glucoside; Aq, aqueous (no osmolyte added). The spline fit of the CPD data is intended only to guide the eye. (E) Conformational heterogeneity in MD-simulated free 1TT. Shown are the central structures from the two major structural clusters (aligned by the CPD and its complement, which are colored in teal). Each histogram is sourced from an independent 250-ns run from a common starting structure. For brevity, the top histogram is a representative of three (out of five total) simulations performed.

To directly evaluate the thermodynamic basis of the strong CPD binding by PU.1, we measured the enthalpy change (ΔH) of binding by isothermal titration calorimetry (Fig. 5B). As detailed in previous analyses [43], the molar heat detected in the initial phase of a protein-into-DNA titration represents the ΔH of forming the 1:1 PU.1/DNA complex as captured in the co-crystal structures. In sharp contrast to the WC complex, which was markedly exothermic (ΔH ≪ 0) at 25°C, formation of the CPD complex was endothermic (ΔH > 0), indicating that binding of the CPD target was entropy-driven. Enthalpy contributions to the MM and I-MM complexes were quantitatively minor. As published ΔH values of native PU.1/DNA complexes track with binding affinity [43], the CPD complex represents the first example of high-affinity PU.1 interactions without favorable enthalpic contributions from the noncovalent contacts.

The lack of favorable enthalpic contributions without a corresponding penalty in affinity (free energy change) for CPD binding led us to hypothesize that CPD formation involved significant disposition of solution components for the unbound states. The thermodynamic uptake or release of water, ions, and other low-molecular weight species upon protein–DNA binding is associated with significant change in translational entropy [71]. To target these dispositions, we probed the linkage of the equilibrium dissociation constant to the thermodynamic activity of salt and nonionic osmolytes (). Increased ionic activity (), as exerted by NaCl, identically decreased binding affinity by ~10-fold from 0.15 to 0.5 M NaCl for CPD and WC, respectively, indicating no net difference in the disposition of ions from electrostatic contacts between the two complexes (Fig. 5C).

To target the disposition of nonionic co-solutes, we measured the impact of glycine betaine, acetamide, and α-methyl glucoside on PU.1/DNA binding. Osmolyte activity was measured by vapor pressure osmometry and expressed as solution osmolality. In stark contrast with NaCl, nonelectrolytes strongly differentiated binding to CPD and WC (Fig. 5D). Whereas the binding free energy (which is proportional to ) to the high-affinity WC sequence decreased linearly up to 2.5 osmolal, binding to CPD exhibited a markedly convex dependence. The effects of these nonelectrolytes differed from the ionic perturbations at equivalent osmolality (0.97 osmolal for 0.5 M NaCl). Together, these features strongly suggest that the osmolytes are probing general differences in solvent exposure upon formation of the two complexes.

The linear and colligative dependence of PU.1 binding to WC sequences on osmolality has been modeled in terms of the change in molecular hydration on binding [44, 60, 72, 73]:

(8)

where the coefficient measures the hydration change (denoting net water uptake) and 55.5 represents the molal volume of pure water. Against this backdrop, the sharp reduction in CPD binding affinity at low osmolality, taken together with the measured positive ΔH, is most consistently interpreted as the net release of osmolyte upon complex formation. The convex dependence may be accommodated by a more general formulation of Eq. (8) as the coupled disposition of hydration water and osmolyte [74]:

(9)

where is osmolyte molality. Since solute activity coefficient tracks the osmotic coefficient monotonically for the three osmolytes within the molality range of the experiment, the observed convexity in by Eq. (9) indicated a coupled release of osmolyte () and water ().

For a structural interpretation, the global conservation among the complexes bound to WC and CPD-bearing sequences at the −1 position (Fig. 4A and B) suggested that the differentiated thermodynamics of CPD binding arose from differences in the unbound states. We therefore examined the dynamics of the free DNA. As reference, WC gave a single conformation ensemble with intact H-bonded base step over 250 ns of triplicate simulations, as expected for a fully paired DNA. In contrast, independent simulations of 1TT based on the same initial coordinates but different initial velocities, yielded very dissimilar trajectories (Supplementary Fig. S11A). Structural clustering of the 1TT ensembles by the CPD base pair showed revealed two quantitatively dominant conformations (Fig. 5E). Conformation I consisted of an intact duplex in which the TpT CPD was H-bonded to its complement. Conformation II was a more heterogeneous population in which the duplex unwound locally at the CPD and adjacent nucleotides unpaired and inter-digitated locally, resulting in an elongated structure. This unusual arrangement allowed the neighboring bases to the CPD (which is not aromatic) to gain stacking interactions at a cost of two net direct H-bonds. Over five replicate simulations of 1TT (which was seeded as an intact duplex), three runs were exclusively in conformation I, while conformation II (once formed) did not revert over the duration of the simulation (250 ns), suggesting substantial energy barriers between the conformations. Conformation II was uniquely due to the CPD, not a mismatched G:T base pair, as clustering of the MM (the non-CPD counterpart of 1TT) ensembles showed a single conformation resembling the full duplex in conformation A (Supplementary Fig. S11A). The 1TT dynamics are compatible with the NMR spectra of CPD and MM (Supplementary Fig. S7B) with conformation (I) as the major population in exchange with other conformations.

The conformational heterogeneity of free CPD-bearing DNA could account for the peculiar thermodynamics of CPD binding by PU.1 if the conformations were differentially hydrated and selected by co-solutes. This interpretation was supported by an altered electrostatic profile (Supplementary Fig. S11B) and enlarged solvent-exposed surface areas (Supplementary Fig. S11C) for conformation II. Since the CPD target bound PU.1 with equal affinity as WC at normo-osmotic conditions (Fig. 5A), the thermodynamics suggested that restoring conformation II to the intact duplex in the PU.1-bound complex was not associated with excess energetic cost. The favorable translational entropies from a coupled release of co-solute and hydration water, as observed by the effects of osmotic pressure, would furnish the required free energy contributions.

Discussion

DNA binding sites of ETS-family transcription factors are among the most highly mutated genomic regions in skin cancers, including melanoma [10, 13]. Previous studies have suggested that recurrent mutations at these sites are the result not only of ETS binding stimulating UV-induced CPD formation, but also because ETS binding inhibits subsequent repair of these CPD hotspots [14, 24–26]. However, the molecular mechanism by which ETS-family members persistently bind damaged sites containing CPDs and inhibit their repair has remained unclear. Here, we show that paralogs from each of the three different ETS classes (i.e. Ets1, ELF1, and PU.1) bind cognate DNA sites containing a central TpT CPD. Furthermore, formation of the 1:1 ETS/DNA complex by each ETS paralog can restrict its access to a CPD lesion by the model repair enzyme, T4 PDG. These findings provide a molecular basis for how ETS-family members accommodate a CPD lesion in their binding site to prevent access by repair enzymes and promote mutation hotspots in melanoma and other skin cancers.

NER, the main pathway for removing bulky lesions such as CPDs from DNA, is a multistep mechanism that senses, verifies, and excises lesion-induced alterations of DNA. XR-seq [75], which captures human NER excision products genome-wide and quantifies repair activity at nucleotide resolution, supports the principle that DNA-bound proteins, including transcription factors [8], interfere with NER at their binding sites, producing locally reduced repair and elevated mutation rates [7]. In this work, we use the CPD-cleaving activity of T4 PDG as a biochemical probe of lesion sensing. Topically administered T4 PDG has also been studied as a photoprotective therapeutic in patients, particularly xeroderma pigmentosum patients deficient in NER activity [76]. Although the T4 PDG-based readout is not a direct measurement of NER kinetics, our results indicate that physical denial to CPD is one negative consequence of cognate transcription factor binding on NER. Our observations with T4 PDG are consistent with reported data on binding to UV-irradiated DNA by the UV-DDB NER complex [20], which is also inhibited by Ets1, highlighting the physiological relevance of these findings.

Environmental UV radiation penetrates through the epidermis to the dermal layer of the skin to an extent dependent on wavelength and skin-related factors [77]. These layers contain cells, such as keratinocytes and fibroblasts that are responsible for the structural components of skin, as well as melanocytes and immune cells (e.g. Langerhans cells) with specialized functions in the viable epidermis. Members from Class I to III of the ETS superfamily are broadly expressed in these tissues [30], either as part of normal skin function and turnover, or in specific pathophysiology [31]. As the tested ETS relatives are highly representative of their respective classes in DNA selectivity [78], and also derive from highly diverse domain context in the full-length protein [29], the present data support high-affinity CPD interactions as a novel class property of DNA-binding ETS domain.

Previous high-throughput binding analysis has shown that UV damage alters the sequence selectivity of ETS1 and ELK1 [20]. In UV-irradiated DNA, ETS1 and ELK1 favor binding to 5′-ATCC-3′ sequences instead of sites that contain a 5′-TTCC-3′ consensus. Our binding data provide a rationale for this change in sequence specificity, as a TpT CPD at the −2 position (i.e. 5′-TTCC-3′) of the ETS binding consensus reduces binding by each of the ETS paralogs. In contrast, 5′-ATCC-3′ sites cannot form this low-affinity lesion but can form the high-affinity CPD at the −1 position (i.e. 5′-ATCC-3′). ETS1 and ELK1 are both Class I factors. Our MD results show that a TpT CPD at the −2 position is more structurally perturbing than the same lesion at the −1 position. The stimulated binding of PU.1 and ELF1 by CPD₋₁, but inhibition by CPD₋₂, suggests that similar alteration of specificity may occur with other classes of ETS factors. A key difference between CPD₋₁ and CPD₋₂ is that CPD₋₁ models a deaminated TpC CPD, resulting in a T/U:G mismatch that promotes ETS binding, consistent with a previous report showing the potential for mismatched sequences to binding Ets1 with enhanced affinity [62]. Previous analysis of CPD deamination in UV-irradiated human cells [14] indicated that ETS binding suppresses CPD deamination at a distal location (underlined) in the ETS-binding motif (i.e. 5′-YYTTCCGG-3′) but deaminated CPDs accumulate at positions corresponding to CPD₋₁ (5′-TTCCGG-3′). Our data here provide a molecular basis for demonstrating enhanced binding of deaminated TpC CPDs by ETS transcription factors.

Although the Class III member PU.1 binds the native DPH3 promoter poorly relative to Ets1 and ELF1, as expected from motif preference, the presence of a centered CPD nevertheless stimulated binding and denied access of CPD to T4 PDG in accordance with 1:1 complex formation. While the co-crystallographic evidence showed that the gross architecture and base-specific DNA contacts within the consensus remain similar to the WC complex, binding experiments (Fig. 5) unveiled a wholly different thermodynamic profile for CPD. Previously, entropy-only driven binding of ETS factors was only observed for low-affinity to suboptimal WC-matched binding sites [43, 60]. Here, CPD binding by PU.1 in an optimal sequence context is also entropy-driven only but proceeds with no penalty in free energy (affinity). Linkage thermodynamics with osmolytes revealed, in combination with all-atom MD characterization, a basis in the CPD-specific heterogeneity in the conformational dynamics of free DNA. To our knowledge, this is the first comprehensive characterization of CPD recognition by a transcription factor. As ETS factors form structurally conserved complexes with cognate DNA with distinct thermodynamics, particularly osmotic properties [60], other ETS paralogs may exhibit other thermodynamic profiles in their position-dependent CPD binding. Nevertheless, our results cast light on the importance of CPD-induced effects on the structural dynamics of the free DNA, in addition to those on the complex.

Conclusion

In this work, we provide a structural and thermodynamic explanation for how DNA complexes of ETS-family transcription factors can accommodate a helix-distorting DNA lesion in a position-dependent manner. Binding of ETS factors to CPD in the center of the ETS binding consensus rationalize previous work [14] showing that deaminated CPDs accumulate at this relative location in ETS binding sites and why repair of these CPDs is inhibited in cells. Importantly, this position precisely coincides with the location of mutation hotspots in melanoma [9, 21–27], many of which may play an important role in melanomagenesis, and can explain a prior study indicating that UV damage alters ETS sequence specificity [20]. The stimulated recruitment of PU.1 to suitably positioned CPDs at ETS motifs that are not preferred targets in their undamaged state further suggests that the pool of ETS-based inhibitors of DNA repair may exceed the complement of ETS factors that bind the undamaged site. As many other transcription factors show altered sequence-specificity in UV-damaged DNA [20], a mechanism similar to that described here may regulate the engagement of other factors to damaged DNA sites, including many (e.g. c-MYC, EGR1, and p53) that play critical roles in carcinogenesis.

Notes

Present address: Department of Molecular Biosciences, University of South Florida, Tampa, FL 33620, USA

Present address: U.S. Army Medical Research Institute of Chemical Defense, Aberdeen Proving Ground, MD 21010, USA

Present address: Department of Pharmacology, University of Michigan Medical School, Ann Arbor, MI 48109, USA

Present address: Versiti Blood Research Institute, Milwaukee, WI 53226, USA

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This investigation was supported by National Institutes of Health grants HL155178 and TR004842 (to G.M.K.P.) and ES028698, ES032814, ES035139, and ES035888 (to J.J.W.) and National Science Foundation grant MCB 2028902 (to G.M.K.P and M.W.G). Funding to pay the Open Access publication charges for this article was provided by grant MCB 2028902.

Data availability

Coordinates and validation reports for the CPD, MM, and I-MM structures have been deposited with the RCSB PDB under DOIs https://doi.org/10.2210/pdb9ynz/pdb, https://doi.org/10.2210/pdb9oa4/pdb, and https://doi.org/10.2210/pdb9ob0/pdb. Gel images, MD trajectories, and other relevant data are available from the corresponding authors upon reasonable request.