EEPD1 evolved a unique DNA clamping dimer protecting reversed replication forks

Abstract

Exonuclease/endonuclease/phosphatase (EEP)-fold hydrolases are canonically monomeric phosphodiesterases exemplified by APE1, DNase I, and TDP2 nucleases. While EEP family domain containing protein 1 (EEPD1) acts in DNA stress responses, its proposed nuclease activities are enigmatic. Here, we integrate hybrid structural methods, evolution, biochemistry, cancer genomics, plus molecular and cell biology to define EEPD1 structure, assembly, and function at stalled DNA replication forks. Results imply EEPD1 surprisingly requires both unique EEP domain dimer and distinctive tandem Helix-hairpin-Helix [(HhH)₂] domains to clamp double-stranded (ds) DNA at reversed DNA replication forks for fork protection. Small-angle X-ray Scattering (SAXS), crystal, and cryo-EM structures unveil an unprecedented tryptophan handshake dimer, conserved interface di-Trp-Pro pocket, and adjustable “wrist” enabling an open-closed conformational switch. EEPD1 dimer cooperatively binds complex dsDNA replication fork intermediates but alone lacks nuclease activity due to loss of key EEP catalytic residues during Metazoan evolution and atmospheric oxygen buildup. Instead, EEPD1 prevents nucleolytic degradation of reversed replication forks by MRE11. Furthermore, cancer bioinformatics support oxidative damage-dependent EEPD1 association as a significant modulator of overall patient survival. Collective findings uncover unexpected EEP dimer and fork protection function in clamping, not cleaving, reversed replication forks for metazoan oxidative stress responses controlling genome stability and cancer outcomes.

Graphical Abstract

Graphical Abstract.

Introduction

The evolution of multicellularity in metazoans, which emerged approximately 590 million years ago, involved key innovations enabling robust responses to a constantly changing tissue microenvironment. These include novel transcription factors, signaling pathways, and components of cell adhesion and cytoskeletal regulation [1]. Notably, many of them are now implicated in cancer including those regulating transcription and DNA damage responses (DDR), which can directly link tissue-level signaling to nuclear DDR enzymes to suppress oncogenesis and maintain genome integrity. A relatively understudied but exemplary metazoan-exclusive gene identified as a DDR protein that may influence cancer outcomes is EEP family domain containing protein 1 (EEPD1) [2–6].

EEPD1 is a member of the ancient exonuclease/endonuclease/phosphatase (EEP) superfamily, whose members include nucleases like apurinic/apyrimidinic endonuclease 1 (APE1) and its bacterial homolog ExoIII, tyrosyl-DNA phosphodiesterase 2 (TDP2), and deoxyribonuclease-1 (DNase I) [7–11], as well as many pathogen nucleases [12]. Notably EEPD1 is a relatively enigmatic member of this vast EEP superfamily that encompasses thousands of distinct sequences across dozens of recognized functional families that span all three domains of life (archaea, bacteria, and eukarya). Unlike other superfamily members, EEPD1 uniquely has two sets of tandem helix-hairpin-helix [(HhH)₂] domains in its N-terminus. The even more ancient (HhH)₂ domain can bind double-stranded (ds) DNA via its two hairpins and has known roles in DNA replication and repair in all domains of life as seen in bacterial RuvA and in archaeal and eukaryotic RAD51 [13–16]. EEPD1 is critical for genome stability and replication fork restart in cells [2, 4, 17]. Consistently, EEPD1 expression is altered in several cancers, including colorectal [2, 18, 19], breast [20], brain [17, 19], and prostate [5]. EEPD1 is transcriptionally regulated by sterol-responsive liver X receptors (LXR) [21, 22] in macrophages and post-transcriptionally by microRNAs that modulate cancer [23] or cholesterol transport [24], suggesting it could have as yet unreconciled connections between metabolism and cancer. Yet, the nuclease activity inferred for EEPD1 in vitro is orders of magnitude lower than that of established EEP nucleases such as APE1 and DNase I [17, 25, 26]. This significant disparity raises questions about the role of this presumed nuclease activity and the specific structural and biochemical basis for EEPD1 function in the cellular context.

Here, we provide the first atomic structures of EEPD1’s hydrolase domain, as determined by X-ray crystallography, and reveal an unprecedented obligate dimeric configuration. Both X-ray scattering and cryogenic electron microscopy (cryo-EM) identify it as the first dimer of the EEP hydrolase superfamily. Importantly, we show that the EEPD1 dimer, mediated by a unique “tryptophan handshake” interface, is the functional unit required for DNA binding. Provocatively, we find no detectable nuclease activity: structures and sequence evolution connect inactivity to loss of a conserved catalytic residue essential for EEP superfamily enzyme hydrolysis. Notably, our evolutionary analyses show that despite EEPD1 having lost the hallmark EEP domain catalytic residue, the dimer interface and DNA-binding domains are highly conserved, underscoring their ancient and fundamental importance. Moreover, we find that in parallel to its established association with plasma membrane [21, 27], EEPD1 enters the nucleus, localizes to reversed replication forks in response to oxidative stress, and protects stalled replication forks from nucleolytic degradation by MRE11 nuclease. These data support an EEPD1 functional role in the metazoan replication stress response and cancer cell survival by reducing replication instability. Our collective findings suggest that metazoan EEPD1 has evolved to act as a molecular clamp for reversed DNA-replication fork binding and protection against nucleolytic attack rather than making cuts, as suitable to link multicellular stress sensing from the tissue microenvironment to nuclear DNA replication stress responses. The EEPD1 dimer’s role as a nonhydrolytic but potent DNA-binding clamp supports a replication fork protection molecular function in oxidative stress responses to maintain genome stability as a satisfying explanation for the implied functional impact of dimer interface mutations found in cancer patients.

Materials and methods

Materials and reagents

MGC human EEPD1 complementary DNA (cDNA) was purchased from Dharmacon (GenBank: BC065518). pET StrepII TEV LIC cloning vector (1R) was a gift from Scott Gradia (Addgene #29 664). Enzymes for subcloning and Q5 site-directed mutagenesis kit were purchased from New England Biolabs. Primers used in subcloning were purchased from Integrated DNA Technologies (See Supplementary Table S1). Uracil–DNA glycosylase (UDG) was purchased from New England Biolabs. APE1 is a gift from Dr Sankar Mitra.

Oligonucleotides for incision assay were purchased from Midland Scientific or MWG.

Cyanine-5-tagged oligonucleotides and their complementary strands for binding assay were purchased from Integrated DNA Technologies (see Supplementary Table S1). ATTO488-NHS ester for labeling EEPD1 was purchased from Sigma–Aldrich (41698).

U2OS and 3KO (SMARCAL1, ZRANB3, and HLTF triple knockout in U2OS) cells are gifts from Dr David Cortez. ON-TARGETplus human EEPD1 siRNA SMARTPool was purchased from Horizon.

Cancer analysis

For the analyses of differential gene expression data in TCGA patients, we used an in-house pipeline, as reported [28]. Scripts are available at https://zenodo.org/records/17550102. For TCGA tumor types that lack “tumor normal,” we compared expression in tumor from TCGA with normal tissue expression from GTEx, which are hosted and reprocessed by UCSC Xena [29]. Subgroups for lower-grade glioma (LGG) patients by genotypes were sorted in the Xena browser. Kaplan–Meier survival curves are generated in Xena. Raw data were downloaded and reproduced in GraphPad Prism (version 10.6).

Hierarchical clustering of linear R values for the correlation between steady-state RNA levels of EEPD1 and 10 reactive oxygen species (ROS)-related genes in 33 TCGA tumors used the Euclidean distance method and average linkage clustering approach.

Protein expression and purification

Coding sequences for crystallization constructs were engineered using restriction enzyme NotI/NdeI into pET28b-pps expression vector so the recombinant protein was expressed with N-terminal hexahistidine (His₆) tag and PreScission protease cleavage site (see Supplementary Table S1). Transformed Rosetta2^™ strain was grown in auto-induction media at 37°C to OD₆₀₀ ∼0.6–0.8 and then expressed protein at 18°C overnight. Liters of cultures were harvested by centrifugation. Cell pellet was resuspended in NiNTA buffer A (25 mM sodium phosphate pH 6.5, 0.5 M NaCl, 20 mM imidazole, 5% glycerol, and 5 mM βME) supplemented with 0.1% TritonX-100, 1 mg/ml lysozyme, 1 mM phenylmethylsulfonyl fluoride (PMSF), and Pierce^™ protease inhibitor tablet (Thermo Scientific). Homogenized lysates were sonicated (Qsonica) and clarified by centrifugation in a Beckman-Coulter JA25.50 rotor at 18 000 rpm (26 581 × g) for 30 min. Filtered lysate was loaded onto HisTrap HP 5 ml column (Cytiva) on an Akta Pure FPLC system. The column was washed with 5 column-volume (CV) of NiNTA buffer A, and tagged protein was eluted isocratically by NiNTA buffer B (NiNTA buffer A with 0.4 M imidazole). The eluate was pooled and exchanged in dialysis buffer (25 mM sodium phosphate pH 6.5, 0.5 M NaCl, 2.5% glycerol, and 5 mM βME) and treated with PreScission protease to remove the fusion tag at 4°C overnight. Tag-cleaved protein was repassed through HisTrap column to remove free tag, dialyzed in Heparin buffer A (25 mM sodium phosphate pH 6.5, 60 mM NaCl, 5 mM βME, and 5% glycerol). The solution was passed through Heparin HT 5 ml (Cytiva), and bound protein was eluted by gradient elution of 0%–100% (in 10 CV) Heparin buffer B (25 mM sodium phosphate pH 6.5, 1 M NaCl, 5 mM βME, and 5% glycerol). The eluate containing target protein was pooled, concentrated and exchanged into SEC buffer (10 mM MES pH 6.5, 100 mM KCl, and 1 mM TCEP) by passing through equilibrated HiLoad Superdex 75 16/600 (Cytiva). Target fractions were pooled, concentrated, scanned by UV–Vis spectrophotometry to measure concentration. The stock was either used directly for setting up crystallization or added 5% glycerol, aliquoted, flash frozen in liquid nitrogen, and stored at -80°C.

For BS3 crosslinking and SEC–MALS–SAXS experiments on full-length EEPD1, cDNA was engineered into pESUMO expression vector so that the recombinant protein was expressed with N-terminal His₁₂ tag and a SUMO tag. Protein was produced and purified in the same workflow as described above with modifications. All purification buffers contain 25 mM HEPES pH 7.5. After clarified cell lysate with targeted protein passed through HisTrap HP 5 ml of column, the resin was washed with 10 CV NiNTA buffer A before eluted by NiNTA buffer B with 0.8 M imidazole. The His₁₂-SUMO tag was then removed by SUMO protease ULP1 at 4°C overnight. The solution was repassed through HisTrap column to remove the free tag and protease, followed by desalting dialysis and Heparin purification described above. Target fractions were further purified and buffer exchanged into SEC buffer (25 mM HEPES 7.5, 100 mM KCl, and 1 mM TCEP) by passing through HiLoad Superdex 200 16/600 (Cytiva). Target fractions were pooled, concentrated, scanned by UV–Vis spectrophotometry to measure concentration. The stock was supplemented with 5% glycerol, aliquoted, flash frozen in liquid nitrogen and stored at −80°C.

Constructs for biochemistry and biophysics assays are cloned into vector 1R using ligation independent cloning (LIC) method, so the peptides are expressed with an N-terminal StrepII tag. Mutants are designed and created by site-directed mutagenesis. Typically, each construct was expressed at 18°C overnight in 1-liter culture grown in auto-induction media. Cell pellet was collected, lysed in Streptactin buffer A (25 mM HEPES pH 7.5, 0.8 M NaCl, 5 mM βME, and 5% glycerol) with addition of 1 mg/ml lysozyme, 1 mM PMSF, Pierce^™ protease inhibitor tablet, and 0.05% TritonX-100. Clarified lysate was incubated with 1-ml Strep-Tactin XT 4Flow resin (IBA Lifesciences) with rotation. The resin was washed with 10 CV of Streptactin buffer A, and then 3 CV of low-salt buffer A (50 mM NaCl). The bound protein was batch-eluted 5–10 times, each with 1 CV of Streptactin buffer B (25 mM HEPES pH 7.5, 50 mM NaCl, and 50 mM d-biotin). The eluate was further purified by Heparin HP 5 ml and buffer-exchanged into SEC buffer (25 mM HEPES pH 7.5, 100 mM KCl, 1 mM TCEP, and 5% glycerol) as described above (Supplementary Fig. S1). Target fractions were pooled, concentrated, then mixed with equal volume of storage buffer (SEC buffer with 35% glycerol) so protein stock was in 20% glycerol. Stock was aliquoted, flash frozen, and stored at −80°C until use.

For nuclease assay, full-length EEPD1 was expressed in human Expi293F cells using a BacMam baculovirus generated from a pEZT-BM15 plasmid construct, where EEPD1 was fused with an HRV-3C/TEV protease-cleavable N-terminal TwinStrepII and a C-terminal Flag tag. For expression, cells were infected with BacMam baculovirus at multiplicity of infection (MOI) of 3 in the presence of 3 mM butyrate. Cells were then cultured at 125 rpm, 37°C, 8% CO₂, and harvested 48 h post infection. TwinStrepII-EEPD1-FLAG protein was purified by tandem affinity purification using anti-FLAG M2 beads (Sigma) followed by Strep-Tactin beads (Qiagen). The cell pellet was homogenized in a lysis buffer containing 50 mM Tris–HCl, pH 7.5, 500 mM NaCl, 0.3% NP-40, 10% glycerol, 1 mM PMSF, 1 mM βME, and EDTA-free protease inhibitor (Roche), followed by sonication. The cell extracts were clarified by centrifugation at 12 000 rpm, 4°C for 30 min. The supernatant was collected and incubated with anti-FLAG M2 beads for 1.5 h at 4°C. The beads were extensively washed with the lysis buffer. Batch elution was performed five times, each with 1 CV of 200 μg/ml 3× FLAG peptide for 15 min. The pooled eluates were then mixed with Strep-Tactin beads for 1.5 h at 4°C, and the beads were extensively washed with the same buffer. Batch elution was performed five times, each with 1 CV of 2.5 mM desthiobiotin for 5 min. Eluted fractions were analyzed by SDS–PAGE, pooled, and dialyzed in storage buffer (50 mM Tris–HCl, pH 7.5, 150 mM KCl, 0.1% NP-40, 1 mM PMSF, 1 mM βME, and 10% glycerol). The purity of the EEPD1 protein was assessed by SDS–PAGE with Coomassie staining.

Crystallization, data collection, and refinement

Crystals of the EEPD1^254-543 (EEP²⁵⁴) were grown by hanging-drop vapor-diffusion in 0.1 M Tris–HCl pH 8.5, 0.3 M MgCl₂, and 29% (w/v) PEG 3350 [modified from 96-well crystallization screen Index^™ HT (Hampton Research) hit condition H1]. Protein stock of 10–15 mg/ml was mixed with equal volume of reservoir buffer and incubated at 15°C. Crystals were harvested, briefly soaked in cryoprotective solution (crystallization condition plus up to 15% ethylene glycol), and then flash cooled in liquid nitrogen. Synchrotron X-ray diffraction data were collected at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 12–1 at SLAC National Accelerator Laboratory. The best crystal diffracted to ∼2 Å. Resolution extended to 1.6 Å after X-ray data were processed with XDS (ver. 5 February 2021) [30] at the beamline and Aimless (ver. 0.7.7) in CCP4 suite (ver. 7.1.014) [31], and structures were solved by molecular replacement with the Phaser (ver. 2.8.3) [32] module in PHENIX (ver. 1.21.2_5149) [33]. The truncated EEP domain model predicted by AlphaFold2 [34] was used as the search model. Helices and loops were first rebuilt by AutoBuild module in PHENIX. The structures were iteratively refined with PHENIX, and model building was done in Coot (ver. 0.9.8.95) [35].

For the second construct, EEPD1^245-543 (EEP²⁴⁵), 10 mg/ml protein stock was mixed with equal volume of reservoir liquor by Mosquito (SPT Labtech) to set up crystallization screenings. Crystals grew from Index^™ HT condition D6 [0.1 M BIS–TRIS pH 5.5, 25% (w/v) PEG 3350]. Crystals were cryoprotected in well liquor with 27.5% (w/v) PEG 3350 and 15% (v/v) glycerol and flash cooled in liquid nitrogen. Diffraction data were collected at the SSRL beamline 12–2. The crystal diffracted to 2.6 Å and resolution extended to 2 Å after data processing. For molecular replacement, truncated EEP²⁵⁴ structure was used as a searching model instead, and two molecules were found in the asymmetric unit. Model building and refinement were done in the same fashion as for the first crystal structure. X-ray and refinement statistics are included in the Supplementary Table S2.

Structural coordinates and crystallographic structure factors have been deposited with the Protein Data Bank as 9YSF (EEP²⁵⁴) and 9YXY (EEP²⁴⁵). Molecular visualization and analyses were done in PyMOL (Schrödinger, LLC) or UCSF ChimeraX [36]. Crystallographic and molecular visualization softwares were accessed through the SBGrid Consortium [37].

BS3 crosslinking assay

Protein stock was diluted with SEC buffer to 20 μM in 20-μl reaction and equilibrated at ambient temperature for 10 min. 1 μl of 25 mM BS3 crosslinker (Thermo Scientific) in dimethyl sulfoxide (DMSO) was added into the mix and the reaction was further incubated for 30 min before being terminated by addition of 1 μl of 1 M Tris–HCl buffer (pH 8.0), and 5× SDS–PAGE loading buffer. Samples were briefly heated at 95°C before being analyzed by SDS–PAGE.

Small-angle X-ray scattering and multi-angle light scattering data acquisition in line with SEC (SEC–MALS–SAXS)

We performed small-angle X-ray scattering (SAXS) experiments at the SIBYLS beamline 12.3.1 at the Advanced Light Source. This beamline has inline instrumentation and detectors coupled to a size exclusion column [38, 39]. 100 μl of 3–5 mg/ml sample was suspended in SEC running buffer (50 mM HEPES pH 7.0, 150 mM KCl, and 1 mM TCEP). The X-ray wavelength was 1.127 Å and the sample-to-detector distance 2100 mm. This combination gives scattering vectors (q) ranging from 0.01 to 0.47 Å⁻¹. The scattering vector q = 4πsin(θ)/λ, where 2θ is the scattering angle. The SAXS flow cell was connected to an Agilent 1260 Infinity HPLC system. EEPD1 was loaded onto a Shodex 802.5 SEC column preequilibrated with SEC running buffer at a flow rate of 0.65 ml/min. The eluent was subjected to the following: (i) ultra-violet light (UV) at 280 nm, (ii) multi-angle light scattering (MALS), (iii) quasi-elastic light scattering (QELS), (iv) SAXS, and (v) a refractometer. For SAXS measurements, 2-s X-ray exposures were collected continuously during a 25-min elution. All frames for analyses were subtracted by the buffer SAXS signal (averaged from selected regions of the same run).

The radius of gyration (R_g) was calculated for each of the subtracted frames using the Guinier approximation: I(q) = I(0) exp(−q²R_g²/3) with the limits qR_g < 1.3. The elution peak was compared to the integral of ratios to background and R_g relative to the recorded frame using the program RAW [40]. Uniform R_g values across an elution peak represent a homogeneous sample. Final merged SAXS profiles, derived by integrating multiple frames at the elution peak, were used for further analyses. We calculated the Guinier plot to provide information on the aggregation state, the volume of correlation (V_c) to estimate the molecular weight [41] (Supplementary Fig. S2). The program GNOM was used to compute the pair distribution function P(r) [42].

MALS experiments used an 18-angle DAWN HELEOS II light scattering detector connected with an Optilab refractive index concentration detector (Wyatt). A 55-μl sample of 7 mg/ml BSA monomer in the buffer noted above, and a refractive index increment (dn/dc) value of 0.18, was used for system normalization and calibration. We used light scattering experiments to perform analytical scale chromatographic separations for MW determination of the principal peaks in the SEC analysis. UV, MALS, and the differential refractive index data were analyzed using the Wyatt Astra 8 software package to monitor the homogeneity of the sample across the elution peak, which complements the SEC–SAXS signal validation.

SEC–SAXS raw data, final merged SAXS curves, P(r) functions, together with SEC–MALS reports are deposited to the SIMPLE SCATTERING database (https://simplescattering.com/). The SAXS results and accession IDs are included in the Supplementary Table S3.

Solution state modeling

For atomic modeling, the structural model of EEPD1 generated by AlphaFold server [43] and the two crystal structures of the EEP domain were used as initial models. For the EEP domain model, the missing loops and linkers in the crystal structure were first built in with the program MODELLER [44]. The full-atomic models were used as templates for rigid-body modeling in the program BilboMD [45], which applies the molecular dynamics simulation to explore the conformational space of the flexible domains. In the BilboMD strategy, the theoretical SAXS profile for each registered conformation and the corresponding fit to the experimental data were calculated using the program FoXS [46]. MultiFoXS [46] was performed to select an appropriate multistate model of EEPD1 and EEP²⁴⁵ in solution. The best multistate model was determined by minimizing the discrepancy χ² between the experimental and predicted model scattering intensity data.

Abasic site incision assay

Oligonucleotide substrates

A tetrahydrofuran (THF)-containing 57-mer with the sequence 5′-ATT ATG CTG AGT GAT ATC CCT CTG GCC TTC GAA CCC [THF]AC CTC AAC CTC TGC CCA CCG-3′ was gel purified and subsequently labeled at the 5′-end using T4 polynucleotide kinase (Thermo Fisher Scientific #EK0031) and [γ-³²P] ATP (Revvity Health Sciences #BLU502A). The oligonucleotide was then annealed to its complement containing a guanine (G) opposite the lesion. To generate an apurinic/apyrimidinic (AP) site containing ds-oligo, a uracil-containing 70-mer with the sequence 5′-GCCAGTCTGTATCGGAGCTCTCTTCTTCTGTGUACTCTTCTTCTTCTTCTTAGCTTGCTATGCTATGCCT-3′ was labeled at the 5′-end with [γ-³²P] ATP and was annealed to its complement 5′-AGGCATAGCATAGCAAGCTAAGAAGAAGAAGAAGAGTGCACAGAAGAAGAGAGCTCCGATACAGACTGGC-3′ and subsequently treated with UDG. Annealing was performed in annealing buffer containing 10 mM Tris–HCl pH 7.5, 50 mM NaCl, and 1 mM EDTA by heating at 94°C for 3 min followed by slow cooling to ambient temperature. For all of the DNA substrates the unincorporated [γ-³²P] ATP was removed by a Sephadex G25 column (Cytiva).

Incision assay

10 nM 5′-[γ-³²P] labeled duplex DNA containing THF (57 nt) or AP-site (70 nt) were incubated with increasing concentrations of EEPD1 (105, 210, and 315 nM) or APE1 (180 nM) in a reaction buffer containing 50 mM HEPES pH 7.5, 100 mM KCl, 4 mM MgCl₂, and 1 mM dithiothreitol (DTT). Reactions were carried out at 37°C for 30 min and terminated by addition of formamide loading dye (95% deionized formamide, 20 mM EDTA, 0.02% bromophenol blue, and 0.02% xylene cyanol). Samples were run on 15% denaturing polyacrylamide gel containing 8.3 M urea in 0.5× TBE buffer. Gels were exposed on a screen and scanned with Typhoon FLA7000 (GE).

Electrophoretic mobility shift assay

Cyanine 5 (Cy5)-labeled oligo and its complementary oligos (see Supplementary Table S1) were annealed in annealing buffer (30 mM Tris–HCl pH 7.5 and 100 mM KCl) by heating to 95°C and 1°C/min gradient cooling programmed on a thermocycler. 20 nM substrates and 0–1 μM StrepII-EEPD1 (or truncated constructs, mutants) were mixed in 50 mM HEPES pH 7.5, 10 mM KCl, 100 ng/μl BSA, 0.5 mM TCEP, and 4% glycerol. Protein–DNA mixture was incubated at ambient temperature for 30 min, before being separated on nondenaturing 6% DNA retardation gel (Invitrogen), in 0.5× TBE buffer, at 4°C.

Micro-scale thermophoresis 

Micro-scale thermophoresis (MST) measures change in movement of biomolecules along temperature gradients due to change of the hydration shell upon binding [47]. For EEPD1 dimerization, purified full-length StrepII-EEPD1 was labeled by ATTO488 dye. Briefly, 2 mg/ml protein was combined with a 2–3 molar excess of dye in reaction buffer (equal parts 25 mM HEPES, pH 7.5, 150 mM NaCl, and 0.2 M sodium bicarbonate, pH 9) and incubated in the dark for 2 h. Free dye was removed from labeled protein by desalting with a PD-10 gravity flow column or a 5 ml of FPLC HiTrap Desalting column (Cytiva). The labeled protein was aliquoted and stored at −80°C.

For EEPD1 dimerization, MST binding titrations were carried out in 50 mM HEPES pH 7.5, 100 mM KCl, and 0.05% Tween-20. Binding reactions were prepared by combining equal volume of 40 nM ATTO488-StrepII-EEPD1 and a dilution series (0–10 μM) of wild-type EEPD1 or W522A mutant. Reactions were incubated for 20 min at ambient temperature before loading into premium silica capillaries (NanoTemper). MST measurements were acquired on a Monolith NT.115 system (NanoTemper) at 25°C with 40% LED power and 30% infrared excitation for 20 s with 5-s equilibration and recovery periods. Time-averaged amplitudes were calculated over a 1-s window during excitation, and three consecutive scans were averaged to generate final values. For wild-type EEPD1 titration, data points beyond titer 0.5 μM were omitted as secondary binding was observed following the first plateau, which may correspond to a much weaker (and likely nonphysiological) association. MST amplitudes and fraction bound values were determined using NanoTemper PR. Stability Analysis 1.0.2 and exported to GraphPad Prism 10.6 for nonlinear fitting to a one-site binding model. Reported binding affinities are the average of three independent experiments.

For EEPD1–DNA interaction, MST was carried out in 50 mM HEPES 7.5, 50 mM KCl, 0.5 mM TCEP, 0.05% Tween-20, and 1.5% glycerol (except for ionic strength tests, where 30, 50, and 100 mM KCl were used). 5′Cy5-labeled substrates were annealed as described for electrophoretic mobility shift assay (EMSA). The final mixture contained 20 nM labeled substrates, and a 1:1 dilution series from 10 μM StrepII-EEPD1 (or its truncations, mutants). Reactions were incubated for 20 min at ambient temperature before loading into premium silica capillaries and measured at 25°C with 20% LED power and 30% excitation for 20 s with 5-s equilibration and recovery periods. Time-averaged amplitudes were calculated over a 1-s window during excitation, and three consecutive scans were averaged to generate final values. Outlier MST curves that exhibited signs of aggregation or convective accumulation were excluded from the analysis (typically from the highest titration concentrations). Data were exported to GraphPad Prism 10.6 for nonlinear fitting either to a one-site binding model or a specific binding model with Hill slope. Reported binding affinities are the average of three independent experiments.

Sequence conservation analysis

Evolutionary history of EEPD1 sequence was analyzed as previously described [48]. Orthologous EEPD1 sequences are selected mainly from OrthoDB v12 [49], with additional sequences from NCBI and ENSEMBL [50]. Eligible ortholog sequences are annotated by OrthoDB to contain both “IPR003583: Helix-hairpin-helix DNA-binding motif, class 1”/“IPR010994: RuvA domain 2-like” and “IPR005135/IPR036691: Endonuclease/exonuclease/phosphatase (family)”. Sequences were kept separated as clades. Sequences within each clade were aligned by MUSCLE [51] in MEGA 11 [52]. For visual simplicity, some clades were arbitrarily further collapsed for multi-sequence alignment (such as fishes). The alignment file then was used to generate sequence logo plots in WebLogo3 [53].

The groups are organized according to Open Tree of Life (https://tree.opentreeoflife.org/). Only tree form is given; branch length is not proportional to evolutionary distance.

Evolutionary trace and evolutionary action analyses

We performed evolutionary trace (ET) analyses [54, 55] using the human EEPD1 and APE1 sequences and curated orthologous sequences. We aligned the sequences using MUSCLE [51] and ran the ET analyses with the option “position-specific gap-reducing real-valued trace.” Residues with lowest coverage score (most important) are highlighted on the crystal structures or AlphaFold full-length model. Additionally, we performed the evolutionary action (EA) analysis for the human EEPD1 sequence, utilizing the same multiple sequence alignment. The resulting EA scores, which were used to predict the functional impact of missense mutations, range from 0 (benign) to 100 (pathogenic).

Cryogenic electron microscopy 

Sample preparation

Full-length StrepII-EEPD1 stock was diluted to 0.2 mg/ml (∼ 6 μM) and mixed with equimolar 45-bp dsDNA in 1× TE buffer (10 mM Tris–HCl pH 8.0 and 0.1 mM EDTA). Quantifoil R 1.2/1.3 Au 300 mesh grids were glow discharged for 1 min at 15 mA in a PELCO easiGlow Glow Discharge Cleaning System. 4 μl of EEPD1:dsDNA sample was applied to the grid, blotted for 3 s with a blot force of −10, and immediately vitrified in liquid ethane using Vitrobot Mark IV (Thermo Fisher Scientific) operated at 4°C with 100% humidity.

Cryo-EM data acquisition

Cryo-EM data were acquired at Pacific Northwest Center for Cryo-EM (PNCC), using a Titan Krios transmission electron microscope (Thermo Fisher Scientific) equipped with a K3 direct electron detector. Automated data collection was performed using SerialEM, at a nominal magnification of 29 000×, corresponding to a calibrated pixel size of 0.788 Å/pixel at the specimen level. Images were acquired with a defocus range of −0.8 to −2.0 μm. Movies were recorded at a total electron dose of 50 e⁻/Ų fractionated into 50 frames. A total of 2 745 movies were collected.

Cryo-EM data analysis and model refinement

All image processing was carried out in cryoSPARC [56]. Movies were motion corrected with Patch Motion Correction followed by CTF estimation with patchCTF. Micrographs with estimated CTF resolutions worse than 8 Å were discarded. Blob picker was used for particle picking, and particles were extracted with 3× Fourier binning. After two rounds of 2D classification, 394 503 particles were used for ab initio reconstruction. Heterogeneous refinement identified 173 718 particles belonging to the best class, which were then used for 2D classification. Good 2D classes were used to train a model by Topaz [57] for improved particle picking. Particles picked by Topaz were extracted with 3× Fourier binning and cleaned by two rounds of 2D classification followed by heterogeneous refinement, which yielded 225 419 particles in the best class. These were extracted without binning and combined with the 173 718 blob-picked particles for heterogeneous refinement, with duplicates removed, the final dataset of 183 340 particles was subjected to ab initio reconstruction into three models using HR-HAIR method [58], then the best class was used for another round of HR-HAIR with C2 symmetry. A subset of 46 575 high-quality particles was used for non-uniform refinement with C2 symmetry, yielding a final cryo-EM density map of 3.6-Å global resolution.

Crystal structure of EEP domain dimer was used to fit into the density map within ChimeraX and then refined using Phenix.real_space_refine against the map. Q-scores of the refined structure were calculated using the MapQ plugin (https://github.com/gregdp/mapq) [59] in Chimera [60]. Density corresponding to the N-terminal (HhH)₂ domains or DNA was not observed. See Supplementary Figs S3 and S4 for data processing and model evaluation. Refinement statistics are included in the Supplementary Table S4. Structural coordinate and cryo-EM density map have been deposited with the Protein Data Bank as 9YI2 and with EMDB as EMD-72976, respectively.

DNA fiber assay

DNA fiber assays were performed as previously described [61–63]. Briefly, U2OS cells were seeded in a six-well plate at a density of 1 × 10⁵ cells per well. The next day, they were pulse-labeled with 50 μM 5-iodo-2′-deoxyuridine (IdU, Sigma #I7125) for 15 min, followed by 50 μM 5-chloro-2′-deoxyuridine (CldU, Sigma #C6891) for 15 min. After IdU/CldU treatment, the cells were washed 2–3 times with warmed PBS followed by media containing 4 mM hydroxyurea (HU) (Sigma #H8627) treatment for 4 h. When present, PFM39 (50 μM) [64] was added 30 min before and throughout the experiment. Cells were harvested by trypsinizing, resuspended in ice-cold PBS and spotted onto glass slides and lysed in 200 mM Tris–HCl, pH 7.4, 50 mM EDTA, and 1% SDS. DNA was allowed to attach for 5.5 min before spreading by gravity. Slides were fixed in methanol/acetic acid (3:1). Slides were air-dried and denatured in 2.5 M HCl for 30 min followed by neutralization with PBS (pH 8), and washed with PBS (pH 7.5). Slides were blocked with 5% BSA and 0.1% Triton X in PBS. IdU/CldU fibers were stained using standard immunostaining with antibodies against IdU (anti-BrdU, clone B44, Becton Dickinson #347 580, 1:50) and CldU (anti-BrdU, Genetex 1:200) followed by secondary antibodies (Invitrogen, anti-mouse Alexa Fluor 488 1:100 and anti-rabbit Alexa Fluor 555, 1:300) before mounting slides with Prolong Gold (Invitrogen #P36934). IdU/CldU Fibers were imaged using a Nikon Eclipse Ti-U inverted microscope and analyzed using ImageJ and GraphPad PRISM 10 software.

In situ protein interactions with nascent DNA replication forks (SIRF)

SIRF assays were conducted as previously described [65–68]. Briefly, cells were treated with 125 µM EdU for 8 min, washed with PBS, and followed by media containing 50 nM camptothecin (CPT; Selleck Chemicals #S1288) or 20 µM H₂O₂ (Sigma–Aldrich #H1009) for 15 min. Where applicable, siEEPD1 SMARTPool was transfected 48 h prior to experiment start using RNAiMAX (Invitrogen) according to manufacturer’s instructions. Cells were fixed, permeabilized, and click-iT reaction was performed using biotin-azide and Alexa Fluor 488 azide according to manufacturers’ instructions. Subsequently, proximity ligation assay (PLA, Sigma–Aldrich) was performed using mouse antibody against EEPD1 (1:50; Santa Cruz Biotechnology #SC-398028) and rabbit antibody against biotinylated EdU (1:200; Cell Signaling Technology #D5A7). EEPD1 SIRF values were calculated as the ratio of PLA foci number divided by EdU-Alexa488 intensity per cell. For each independent biological experiment, fluorescent intensity values were min–max normalized using the formula:

X_normalized = (X − X_minimum)/(X_maximum − X_minimum) [68]. The minimum and maximum values were defined from the full range of measurements from each biological replica. Data graphs and statistics were derived using GraphPad PRISM 10 software.

Results

Full-length EEPD1 forms an obligate dimer with extended flexible regions

Given the link between EEPD1 and cancer [2–6, 19], we sought to understand the structure and activity of EEPD1 in relation to DNA. Full-length human EEPD1 comprises two (HhH)₂ domains and one EEP domain connected by unstructured regions (Fig. 1A). We first expressed the recombinant full-length, wild-type EEPD1 with N-terminal His-tag in bacteria and isolated it to high purity by affinity chromatography and size exclusion chromatography (Supplementary Fig. S1A), with the tag removed after the initial affinity purification step. In MALS coupled to SEC, EEPD1 eluted as a single peak with slight tailing, suggesting a dominant high-order species with minimal dissociation with a mass at 124.6 kDa (Fig. 1B). Given that the theoretical molecular weight for the EEPD1 monomer is 62.4 kDa, the MALS data indicates EEPD1 forms a dimer. We assessed dimer stability by microscale thermophoresis (MST), titrating unlabeled EEPD1 to ATTO488-labeled EEPD1. Dimer formation was measured by thermophoresis that reports a change in the protein hydration shell. The observed dimer dissociation constant (K_d) was 19 nM, consistent with stable dimer being observed in our solution studies (Fig. 1C).

Figure 1.

Full-length human EEPD1 is a flexibly tethered, multi-domain protein that uniquely dimerizes via the EEP domain. (A) Domain schematic of full-length EEPD1 (569 amino acids). Two (HhH)₂ domains and one EEP domain are flanked by intrinsically disordered regions, as predicted by PONDR. (B) On SEC–MALS chromatogram, EEPD1 (dark blue) eluted as a single tailing peak, with particle mass (aqua line) measured by MALS. Annotated mass is the mean of three individual SEC–MALS experiments with standard deviation. BSA dimer (gray, theoretical mass 132.9 kDa) is a reference. (C) MST measurement of EEPD1 dimerization. Unlabeled full-length wild-type EEPD1 was titrated into ATTO488-labeled wild-type EEPD1. K_d of wild-type EEPD1 is the mean of three technical repeats with standard deviation in brackets. (D) Full-length EEPD1 SAXS data plotted as dimensionless Kratky plot. Crosshairs show where a globular, well-folded system would peak. (E) The shape of the real-space pair distance distribution P(r) of full-length EEPD1 with multiple peaks and a long tail is consistent with an elongated flexible arrangement of the protein domains. (F) Reaction of BS3 crosslinking was separated on denaturing SDS–PAGE. Successfully crosslinked higher-order assembly indicated by arrow. Split constructs used are indicated in (A).

We further collected SAXS data of EEPD1 in solution [39]. The estimated molecular weight from SAXS [41] was 134.2 kDa, validating EEPD1 as a dimer (Supplementary Table S3). The EEPD1 peak in the dimensionless Kratky plot diverges from the crosshairs, indicating that the full-length protein is not globular (Fig. 1D and Supplementary Fig. S5). The Porod exponent of 2.4 (Supplementary Table S3) indicates a high degree of flexibility; well-folded, flexible multidomain, and unfolded proteins will have Porod exponents of 4, 3, and 2, respectively [69, 70]. The shape of the pair-distance function P(r) with multiple overlapping peaks, an elongated tail and maximal dimension (D_max) ∼190 Å indicates a multidomain protein with flexible and extended arrangement of the domains (Fig. 1E and Supplementary Table S3).

The EEP domain mediates EEPD1 dimerization

Dimerization is observed in (HhH)₂-containing proteins [71, 72] but is unprecedented for the EEP domains. We mapped which protein regions are required for dimer conformation. Surprisingly, when purifying split constructs with either (HhH)₂ or EEP domain, only ones containing the EEP domain eluted in higher-order dimeric state on SEC. Consistent with chromatography observations, dimeric species of full-length EEPD1 and EEP domain were covalently captured by bis(sulfosuccinimidyl)suberate (BS3) crosslinking and visualized on denaturing PAGE (Fig. 1F). In contrast to the reported heterodimerization mediated by (HhH)₂ domains in other DNA repair proteins like the XPF–ERCC1 and Mus81–Eme1 dimers [71, 72], EEPD1’s N-terminal region containing both (HhH)₂ domains showed negligible self-association, even at concentration as high as 10 μM (∼0.6 mg/ml). Together, these results indicate that the EEP domain harbors the primary dimerization interface for EEPD1, with possible secondary transient association among (HhH)₂ domains.

EEP domain X-ray crystal structures reveal conserved αββα sandwich fold

Crystallization of full-length EEPD1 was precluded by its predicted and experimentally determined flexibility (Fig. 1A, D, and E, and Supplementary Table S3). We therefore targeted the hydrolase EEP domain. Our SAXS data collected from the EEP domain constructs revealed a Porod coefficient of 4, which is characteristic of a globular structure, a Dimensionless Kratky consistent with a folded structure, and thus suggests suitability for crystallization studies (Supplementary Fig. S5 and Supplementary Table S3) [70, 73]. We determined independent crystal structures of two constructs, human EEPD1^254–543 (EEP²⁵⁴) and EEPD1^245–543 (EEP²⁴⁵), crystallized under different conditions for data collection and refined to 1.6 and 2.0 Å, respectively (Fig. 2A and Supplementary Table S2). The subunits of the two crystal structures are nearly identical, with a superposition RMSD of 0.232 Å (for aligned 260 Cα atoms).

Figure 2.

X-ray crystal structures of the EEP domain revealed a conserved αββα-sandwich fold with a unique “handshake” dimer interface. (A) Domain schematic of EEPD1 and two constructs of EEP domain used for crystallization. (B) The 1.6-Å crystal structure shows EEP²⁵⁴ (PDB:9YSF) forms a dimer resulting in two EEP active sites facing each other. Zoomed-in view shows the dimer interface formed by an extensive hydrogen-bonding network emanating from Lys275 and Asn278, as well as Trp522 on loop β11–β12 from opposing subunits. In subunit A: β-strands in orange/yellow, α-helices in blue. Dimerizing elements: α1–α2 in green, β11–β12 loop in purple. His531 and Glu301 are shown as spheres to mark the putative active site. Two metal ions (green spheres) are shown in each subunit. (C) B-factor putty representation of the EEP dimer structure shows the dimer interface has an unexpectedly low B-factor (blue, thin tube) similar to the EEP core, in contrast to other loops with high B-factors (red, thick tube). (D) Overlay of the two independent EEP domain crystal structures, based on subunit A, highlights open and closed conformations. Distances are given between Asp448 Cα in each subunit. Zoomed-in view suggests binding of a metal–water cluster (from EEP²⁵⁴’s crystallization condition) coincided with Glu283 flipping its side chain, creating additional hydrogen bonds (gray dash) with Asn515 (dimer loop β11–β12) and Asn322 (α4 helix). This bends the β11–β12 loop relative to core structure around residues 513–514 (“wrist”), effectively causing one subunit to open up ∼12° relative to the other.

The overall structure of the EEP domain preserves the characteristic two-layered αββα sandwich of the EEP fold (InterPro: IPR036691) [74] (Fig. 2B and Supplementary Fig. S6A). Two mostly antiparallel six-strand sheets are nested on the interior with approximate two-fold internal pseudosymmetry. The last two strands β11 and β12 form one sheet with the first four strands β1–β4, while β5–β10 form the second sheet (Fig. 2B and Supplementary Fig. S6B). Two β-sheets are flanked by an α-helical layer on each side. Helices and loops connect the β-strands. Each subunit forms a rectangular prism of about 43 Å × 48 Å × 30 Å.

DALI search with our crystal structures quantified the high structural homology between the C-terminal domain and members of the EEP-fold phosphodiesterase/phosphatase superfamily, with RMSD 2.4–3.9, despite low sequence identity (<20%) (Supplementary Table S5) [75]. EEP enzymes hydrolyze phosphoester and phosphodiester bonds in a wide array of substrates, from nucleic acids to phospholipids [76]. This superfamily includes deoxyribonucleases like APE1 and DNase I, CCR4 family of deadenylases, and enzymes acting on phospholipids, including human neutral sphingomyelinase (nSMase) and phosphatidylinositol 5′-phosphatases (INPP5). The substantial substrate diversity recognized across this fold highlights that while structural similarity is useful, functional inference based on fold may be misleading [77–79].

Unique tryptophan-handshake dimer interface between EEP domains

Consistent with the biophysical and biochemical data, both independent crystal structures determined from different pH and crystallization conditions revealed that the EEP domain forms a dimer (Fig. 2B and D). One crystal form (EEP²⁵⁴) had a crystallographic symmetric dimer, while the other crystal form (EEP²⁴⁵) had pseudocrystallographic symmetry. The unusual dimer interface resembled a handshake: at the end of the β11–β12 loop that forms a β-hairpin, the apical Trp522 reaches into a hydrophobic surface on the opposing subunit formed in part by the alkyl chain of Lys275 and several aliphatic residues (Fig. 2B, and Supplementary Fig. S7C and D). The interface is highly conserved (Supplementary Fig. S8) and marked by an extensive internal hydrogen bond network that bridges distant sequence regions and would reduce flexibility of each β-hairpin. Most critically, Lys275 clips the main chain carbonyl of residues 268–269 (helix α1) together with residue 528 (loop β11–β12) (Fig. 2B). Asn278 is located at a right-angle turn connecting helices α1 and α2. Its carboxamide side chain “N-caps” the helix α2, resulting in a fully protonated amine moiety that forms bridging hydrogen bonds with the main chain carbonyl of Trp517 in cis and of opposing Trp522 in trans to reinforce the subunit interactions. This precise structural arrangement, which relies on a conserved and intricate interaction network, ensures optimal positioning of Trp522 to form a stable dimer interface, indicated by markedly low crystallographic temperature factors in the dimer interface (Fig. 2C and Supplementary Fig. S9A).

The complex network of interactions expands beyond the hydrophobic core. The β11–β12 loop is further secured by mainchain-to-mainchain hydrogen bonds with the juxtaposing β9–β10 loop and by Arg510 acting as another unique structural clip between these two loops (Supplementary Fig. S7A). Additionally, in the EEP²⁵⁴ structure, a magnesium–water cluster further extends bridging interaction. Glu283 on helix α2 is flipped to bind the Mg–water cluster, which creates new bridging hydrogen bonds with Asn515 on β11–β12 loop and Asn322 on helix α4 (Fig. 2D and Supplementary Fig. S7B).

Although the dimeric interface buries a relatively small surface area (∼490.6 Ų) [80], analogous to a handshake, the dimer’s K_d of 19 nM indicates stability. While Trp522 is critical for dimerization, the nearby conserved Trp524, exposed to the solvent, would be available to coordinate with an external protein or nucleic acid binding partner. Of potential functional relevance, overlay of our two experimental structures reveal differences in dimer “openness”—when both dimer structures were overlaid by their subunit A, the subunit B differed in their relative position by ∼12°. The more open EEP²⁵⁴ dimer may partly reflect the presence of the Mg–water cluster inducing the Glu283 flipping and Asn515–Glu283–Asn322 network, bending the mainchain ∼5° (up to 9° at apex) around Leu513–Thr514 (the “wrist”) (Fig. 2D, and Supplementary Fig. S9B and Supplementary Movie S1). This difference between crystal structures suggests that the otherwise stable interface is “malleable” allowing switching between open and closed states in response to a potential allosteric regulation from DNA, metal ion, or protein–protein interaction.

Open and closed states are sampled in solution

We collected SAXS data to determine EEPD1 structural dynamics in solution (Fig. 3A-F). To create atomic models of EEP domain dimer consistent with the SAXS sample, we modeled missing regions into the two crystal structures. The SAXS data generally matched the models, indicating that dimer interface observed in the crystal structures occurred in solution. Yet, neither of the models alone delivered a satisfactory fit with high χ² = 2.78 (open) and 1.93 (closed). We therefore input either the open or the closed form as a starting model into a molecular dynamics (MD)-SAXS program, BilboMD [45] (Fig. 3D). We set the entire dimer as a single rigid body or allowed relative motion between each EEP core domain and the β-hairpin dimer interface (defined as separate rigid bodies) around the pliable “wrist.” In either strategy, loops poorly defined in the crystallographic electron density were allowed to move. BilboMD will score model(s) based on their fit to the experimental SAXS data.

Figure 3.

SAXS and cryo-EM data support EEPD1 as an open-close dimer with four (HhH)₂ domains flexibly tethered. (A) Experimental reciprocal-space SAXS data for full-length EEPD1 (gray) and EEP domain (black) are overlaid with curves calculated from best-fitting ensembles of atomic models generated by BilboMD. (B) EEPD1’s real-space P(r) function from SAXS experiment (gray) is overlaid with the calculated function from the best-fitting ensemble (blue dash). (C) Full-length EEPD1 models are consistent with dimeric EEP domains with flexibly tethered (HhH)₂ domains. Model color: EEP in orange, (HhH)₂ in blue and green (D) χ² plot of different modeling strategies to obtain models most consistent with experimental SAXS data of the EEP²⁴⁵ construct. Starting models are based on either open (EEP²⁵⁴) or closed (EEP²⁴⁵) crystal structure. The two domains were either rigid or allowed to move relative to each other. (E) P(r) function of EEP²⁴⁵’s experimental SAXS data is overlaid with the calculated function from best-fitting ensembles from (D). (F) Best-fitting ensembles are consistent with sampling of open/close conformations, as indicated by the distance of the two opposing Asp448 Cα from each subunit. (G) Cryo-EM 2D classification, density map of full-length, dimeric EEPD1 (PDB: 9YI2). The closed form based on EEP²⁴⁵ crystal structure fits the map. Only the core dimer EEP is visible, with strong density of the handshake interface. Crystallographically undetected loop β7–α5 is also partly observed in cryo-EM map.

The BilboMD strategy that started with the model based on the closed crystal structure while allowing “wrist” motion yielded the best-fitting ensemble with χ² = 0.97 (Fig. 3A, D, and E). The two models had Asp448–Asp448 distances consistent with the open or closed crystal structures and were equivalently represented in the population (56% and 44%, respectively), indicating that the two independent crystal structures represent the two major states in solution even without divalent metal in the buffer (Fig. 3F). These conformations first observed in the crystals are not just a consequence of the crystallographic lattice or excessive metal in the crystallization condition (Fig. 3D–F). Despite the AlphaFold server successfully predicting the EEP domain dimer geometry and handshake interface [43] (Fig. 3C), the output models failed to predict an open state, even when Mg²⁺ was included in the prediction.

To determine if the (HhH)₂ domains interacted with the core EEP dimer in solution, we applied BilboMD with AlphaFold dimer models of full-length EEPD1 against experimental SAXS data of the full-length protein (Fig. 3A and B). The high degree of flexibility in the full-length protein prevented open and closed conformational analysis. The best models were consistent with the (HhH)₂ domains independent from the core EEP dimer domains, like beads on a string (Fig. 3C). Notably, full-length EEPD1’s Porod coefficient of 2.4 (Supplementary Table S3) indicates a much higher flexibility than another beads-on-string protein RPA (Porod coefficient: 3.3) [70, 81]

Our efforts to further characterize full-length dimer EEPD1 in complex with dsDNA using cryo-EM did not detect (HhH)₂ domains or bound nucleic acid in the 2D classifications and reconstructed models due to their flexibility; however, the density of the EEP domain dimer was distinctly evident. We successfully fit the closed-form model into the density map and refined the structure to an overall 3.6-Å resolution (Supplementary Figs S3 and S4, and Supplementary Table S4). Notably because of its potentially functional location, the density of the loop β7–α5 is partially visible, consistent with flexibility (Fig. 3G).

EEPD1 lacks the essential Asn residue for phosphodiester cleavage

Although EEPD1 uniquely has two N-terminal (HhH)₂ domains, its EEP domain shares a similar fold with other EEP superfamily hydrolases (Fig. 4A and B). Superfamily members share six highly conserved motifs (Fig. 4A, and Supplementary Figs S6A and S8, and Supplementary Table S5): (i) WN on β1, (ii) QE on β2, (iii) H on β7, (iv) DXN on β8, (v) D before β10, and (vi) DH before β12 (β-strand numbering from N-terminus to C-terminus, Supplementary Fig. S6B) [82]. For clarity when comparing with other EEP proteins, they will be footnoted with motif numbers: Asn_(i), Glu_(ii), His_(iii), Asp_(iv)–Asn_(iv), Asp_(v), and Asp_(vi)–His_(vi). All active site residues cluster on one edge of the β-sheets, within a negatively charged pocket in a cleft formed by several loops reaching out from the same edge (Supplementary Figs S6A and S10). Five out of six EEP active site motifs are conserved in EEPD1: (i) Asn267, (ii) Glu301, (iii) His404, (v) Asp490, and (vi) Glu530–His531. However, the striking exception is seen in motif iv: unlike all other EEP hydrolases (Fig. 4A and C, and Supplementary Table S5) that have an invariant catalytic Asp–Asn pair (DXN), mammalian EEPD1 has replaced Asn with Gly (Gly450 in human EEPD1) (DXG), disrupting the hydrolase-critical motif, suggesting EEPD1 alone lacks nuclease activity.

Figure 4.

EEPD1 EEP domain lacks key catalytic residues essential for phosphodiester bond hydrolysis and for abasic site specific cleavage. (A) Multiple sequence alignment of human EEP nucleases. Residues specific to either EEPD1 or APE1 are highlighted in yellow and marked with asterisks. EEPD1 secondary structures and features are mapped on the sequence. (B) Domain schematics for EEPD1 and exemplary human EEP nucleases (core EEP domain in light orange). (C) Superposition of EEPD1’s active site to APE1 (with substrate, PDB:5DFI). Attacking water (w) is accurately positioned and activated by both APE1 Asp210 and Asn212. Tyr171, Asn174, and Asn212 bind the scissile phosphate and are essential for APE1’s highly selective activity at AP-sites. All three are replaced in EEPD1 (marked by asterisks in A and C). (D) THF- or AP-site-containing dsDNA substrates (S) and nucleolytic products (P) were separated by denaturing TBE-Urea gel. 10 nM 5′-³²P-labeled substrate was incubated with 105, 210, and 315 nM EEPD1 or 180 nM APE1. The reaction buffer contains 100 mM KCl and 4 mM MgCl₂. Reactions were incubated at 37°C for 30 min before termination. Gel is a representative of three technical repeats.

Mechanistically, Asp_(iv)-Asn_(iv) are catalytically critical for the EEP superfamily members to precisely position and activate the attacking water that starts the hydrolytic reaction (Fig. 4C) [10, 83, 84]. Additionally, Asn_(iv) binds the scissile phosphate oxygen in the product. Mutation of either residue (D210N or N212A) in APE1 severely reduces activity by at least four orders of magnitude [7, 85, 86]. The position between these two catalytic residues heavily favors hydrophobic side chains (Supplementary Table S5), which anchor and ensure precise positioning of the flanking catalytic side chains. The other active site motifs, which are conserved in EEPD1, act in metal ion or phosphodiester backbone binding but not initiation of hydrolysis: they improve EEP hydrolase catalytic efficiency in the context of a catalytically intact DXN motif iv. Altogether the conservation pattern within the active site supports EEPD1's ability to bind metal ion and nucleic acid phosphate similarly to other EEP domains but not its DNA nuclease ability.

EEPD1 lacks two added critical residues for AP-site cleavage

Although biological data has implicated EEPD1 in endonucleolytic cleavage of AP-site-containing dsDNA [17], EEPD1 is missing additional APE1 critical residues. Notably, AP-site nucleolytic activity typified by APE1 relies on a key Tyr171 in place of His_(iii) (Fig. 4A and C, and Supplementary Table S5). Y171H mutation diminished APE1’s nucleolytic activity by 50 000-fold [87]. Similarly, APE1-specific residue Asn174 (Fig. 4A) is reported as essential for catalysis, adding to the growing evidence that AP-endonuclease activity by APE1 requires a modified catalytic landscape which EEPD1 and other EEP hydrolases lack [88].

As there was a discrepancy between our structural analysis and previous reports of EEPD1 as an active nuclease, we experimentally tested and compared EEPD1 activities with known EEP nucleases. We tested affinity-tagged EEPD1, expressed recombinantly in both bacteria and HEK293 cells. With EEPD1 expressed in mammalian cells, we did not see detectable hydrolytic activity against dsDNA substrates with AP lesion or its analog THF even when we incubated substrates with EEPD1 at excessive concentrations (Fig. 4D and Supplementary Fig. S11A). The kinetics of a potential cleavage reaction thus cannot be further determined. With bacteria-expressed EEPD1, we initially detected trace AP-endonuclease activity. However, such activity was eliminated when EEPD1 was produced in RPC501(Xth-/Nfo-) strain that is devoid of bacterial AP-endonuclease homologs (Supplementary Fig. S11B and C). Similarly, despite substantial binding affinity at lower ionic strength (see Fig. 9), we did not detect hydrolase activity against any lesion-mimicking DNA structures even at high enzyme-to-substrate ratios and extensive reaction times, in agreement with our comparative structural analysis. These data suggest that while a partner protein might provide the activity or the missing catalytic moiety, as seen for TFIIS [89], our combined structural and biochemical analyses do not support EEPD1 alone as an active nuclease.

Figure 9.

EEPD1 binding to dynamic dsDNA intermediates requires dimeric full-length protein. (A) EMSA shows binding to 5′-Cy5-labeled 45-nt ssDNA or 45-bp dsDNA fragment (20 nM) at 10 mM KCl, with increasing EEPD1 titer. (B) MST measurements of EEPD1 binding affinity for various dsDNA structures at 50-mM KCl. The labeled strand is the same 5′-Cy5-labeled 45-nt oligo as in (A). (C) MST measurements of DNA bubble binding by truncated EEPD1 constructs. Domains are colored according to schematics in (A): (HhH)₂ in blue and green, EEP in orange. Losing one or two (HhH)₂ increasingly impairs binding. The N-terminal fragment (aa 31–201, purple trace) alone doesn’t bind DNA. (D) MST measurements of EEPD1 mutants’ binding to DNA bubble. H404A is pseudo active site mutation, W522A is dimer-disrupting mutation. (HhH)₂-truncated construct Δ1–129 (from C, blue curve) is copied here for comparison. Indicated titers of StrepII-EEPD1 (also applying to truncations and mutants) assume a dimeric state. All MST dissociation constants K_d are mean values with standard deviation from three independent experiments. Experiments of binding-defective groups were done at least twice. ND: not determined. (E) Model for how dsDNA (cartoon) could be encircled by EEPD1 dimer (surface and cartoon), based on overlays with APE1–dsDNA complex (PDB: 1DEW) onto either subunit of closed EEPD1 dimer (PDB: 9YXY). A “di-Trp-Pro” pocket, at the bottom of the DNA-binding channel, would lie next to the major groove of the DNA (zoom). A potential extra-helical base flipping is indicated by arrow. (F) The “di-Trp-Pro” pocket is 12-Å wide, formed by symmetrically related tryptophans and prolines in the dimer interface, could theoretically accommodate two bases. Residues that line this pocket have significant ET scores (red dots). (G) Experimental simulated annealing omit map shows a density (*) sandwiched between the tryptophans, which could be a docking site for aromatic DNA bases.

Displaced active site metal ion

EEP-fold enzymes coordinate divalent metals (most often Mg²⁺) that bind scissile phosphate in enzymatic reactions (Supplementary Fig. S6A and C, and Supplementary Table S5) [9, 74, 83, 90]. There was no significant electron density to indicate metal coordination near the conventional metal-binding residue Glu301 in the EEP²⁵⁴ structure, albeit crystallized with 0.3 M MgCl₂. However, a stronger density, likely a Mg²⁺ with five coordinated water molecules [91], is seen near His531’s Nε. The partial (∼60%) metal occupancy displaced from the catalytic position is consistent with the disorder seen in other apo EEP structures: metal occupancy is expected to increase upon substrate binding [8, 11]. Notably, in other EEP domains, the metal ion is also coordinated by two ordered water molecules placed by Asp_(vi). In EEPD1, the Asp_(vi) is replaced by Glu530. Although Asp and Glu are often considered functional equivalents, in our apo structures Glu530 is pushed away from the catalytic core by Gln269 (Supplementary Fig. S6C), which, along with the observed low metal occupancy, suggests that concerted events—such as substrate binding and side chain reorientation—are necessary for metal ion positioning.

Evolution preserves its unique dimer interface but not key nuclease residues

Our structural and biochemical analyses of human EEPD1 revealed two dominant features: dimerization and missing catalytic residues for nuclease activity. If these standout features are functionally important, they could have evolved by selection over time. To test this, we curated and aligned the orthologous sequences for EEPD1-like proteins from nearly 200 species, defined as comprising at least one N-terminal (HhH)₂ domain and a C-terminal EEP domain [49, 92]. EEPD1 was only found in eukaryotes, with most in Metazoans. Unlike APE1 orthologs that occur in multiple domains of life [8, 93], no EEPD1 ortholog was identified in eubacteria or archaea as well as in plants or lower eukaryotes. We find that the common ancestor with an EEPD1-like protein likely dates to ∼590 million years ago during the Neoproterozoic Oxygenation Event in which earth oxygen concentrations rose significantly and after which, living organisms required enhanced oxidative defenses [48, 94].

We then grouped the ortholog sequences based on the Tree of Life (https://tree.opentreeoflife.org) and compared their consensus sequences. Strikingly, compared to the unique dimerizing elements that are always present since the emergence of the gene, the canonical active site motif iv in EEPD1 became more tolerant to variations (Fig. 5): while early metazoans retained the DXN_(iv) signature, the Asn_(iv) went under deselection. In chordates, Asn_(iv) is only kept in a subclass of Reptilia, namely Archelosauria clade (birds, crocodilians, and turtles), whereas in mammals it is always replaced by Gly, which completely removes all side chain atoms.

Figure 5.

Conservation of EEPD1 sequences along Tree of Life shows early emergence and stabilization of dimerization elements but not key catalytic motifs. EEPD1 ortholog sequences were only identified in Metazoa based on the characteristic 2x(HhH)₂ + EEP domain configuration. Distant orthologs in oomycetes and Gracilariopsis (red algae) lack dimerizing motifs and thus are isolated (bottom, gray box). Dimerizing motifs are well conserved through all branches. In contrast, the key DXN signature of the active site motif iv is only found in invertebrates representing early animal ancestors and has been variable. Height of the residue symbols is proportional to the relative frequency of the residue at the position. Created in BioRender. Shen, R. (2026) https://BioRender.com/kq2ijo6

To better quantitate evolutionary significance, we used evolutionary trace (ET) analysis to identify the most critical features of the protein [54, 55, 95]. In EEPD1, the most evolutionarily important residues (top 2%), representing the first subset (Fig. 6A, red circle; Fig. 6B, red sphere), are predominantly located at the dimer interface plus the myristoylation site Gly2 [21, 27, 96, 97]. The second ET analysis subset, representing the next tier of importance (2%–5%, orange circles), consists mostly of residues in the two (HhH)₂ domains. Unexpectedly for active site residues, the ones in EEPD1 (blue dots) were scored as being not exceptionally critical (11%–78% range). In retrospect, this scoring is consistent with a prominent role in binding, where the larger interface is more tolerant of single mutations. As a control, we performed the ET analysis on APE1. In stark contrast, the corresponding active site residues (red and orange dots) in APE1 were scored as the most evolutionarily critical throughout its much longer phylogenetic history (top 5%, except for Asp308 and His309), in line with their role in abasic site hydrolysis (Fig. 6C).

Figure 6.

EEPD1 dimer-mediating elements are more evolutionarily conserved than its nuclease active site residues. (A) ET scores of each amino acid (middle) mapped relative to the domain schematic (top). Lower coverage means higher evolutionary significance. The most evolutionarily conserved residues map to two dimerization elements in the EEP domain (green and purple zones), plus the myristoylation site Gly2. The active site residues (blue) are not well conserved evolutionarily. EA analysis (bottom) predicts fitness impact of the recorded clinical VUSs in tumor samples. EA score > 80 is likely to be the most severe [95]. (B) EEPD1’s evolutionarily conserved residues in ET (red and orange spheres) and most impactful VUSs in EA (black dots) are mapped onto the full-length dimeric EEPD1 model (top) and on dimeric EEP²⁵⁴ crystal structure (bottom, zoomed-in view of interface). Subunit B (dark gray) is partially shown for reference. (C) ET scoring for APE1, shown for comparison. Well-defined catalytic residues are solid circles colored by ET ranking.

By comparing EEPD1’s EEP domain with reported structures of EEP domains from all other major subclasses under the superfamily in the Conserved Domain Database [76], we found that EEPD1 has a significantly longer, more protruding β11–β12 loop (Supplementary Fig. S12). While bacterial sphingomyelinases have a similar outreaching loop for membrane insertion, they do not dimerize [98, 99]. This distinct loop modification, along with the conserved “lysine clip” Lys275 that is also absent in other EEP domains (Fig. 4A), suggests these structural distinctions are hallmarks of EEPD1’s evolutionary history and are critical for its dimerization mode.

EEPD1 expression in tumors correlates with oxidative replication stress and patient outcome

Our evolutionary analysis suggests EEPD1 co-evolved with metazoan development during environmental oxygenation, leading us to examine its relationship with oxidative stress response pathways in tumors. Transcriptome analysis across 33 TCGA cancer types strikingly showed that EEPD1 expression consistently correlated positively with the oxidative DNA glycosylases NEIL1 and OGG1, which operate at stalled replication forks, but not with other Base Excision Repair (BER) components (Fig. 7A). This is in line with a functional link between EEPD1 and the processing of oxidative DNA lesions during replication stress.

Figure 7.

EEPD1 expression in tumors correlates with replication stress and its dysregulation may predict patient outcome. (A) Hierarchical clustering of ten oxidative stress-related nuclear genes across 33 TCGA tumor types identifies a stress/transcriptional-response arm (GADD45A, CDKN1A, NFE2L2, and FOXO3) and an oxidative DNA lesion/replication-fork arm (NEIL1, OGG1, XRCC1, PARP1, APEX1, and POLB). EEPD1 correlates preferentially NEIL1 and OGG1 within the latter, consistent with its recruitment to oxidatively stressed reversed forks. (B) Box plot of EEPD1 gene mRNA expression in tumors and matched normal tissues from TCGA. P-values by Wilcoxon tests are notated on top. (C) EEPD1 gene expression in normal brain tissue from GTEx, gliomas from TCGA. P-values were derived from one-way ANOVA. (D) Kaplan–Meier (K-M) survival plot for LGG separated into three equal-size groups based on EEPD1 expression. P-value on the left corner is derived from log-rank test for trend. P-values from pairwise multiple comparisons are also given. (E) K-M survival plot for UVM separated into two equal-size groups based on median EEPD1 expression. Hazard ratio (with 95% confidence interval) and log-rank test P-value is shown. (F) EEPD1 expression in LGG patients separated by IDH status. P-value derived from unpaired t-test. (G) K-M survival plot of all glioma patients. GBM is individually grouped, while LGG are grouped by molecular subtype. IDH-mutant, 1p/19q-codeleted subtype (magenta dash) is further split into groups A and B by median EEPD1 expression. Survival of either group is compared against IDH-mutant, 1p/19q-intact group C. A Bonferroni-corrected α threshold for P-value comparison is set to 0.005.

Differential steady-state mRNA level analysis of ∼7000 tumor samples and the matched controls available in TCGA revealed EEPD1 is significantly upregulated in kidney, colon, prostate, and liver cancers, but downregulated in thyroid, breast, and lung cancers (Fig. 7B). Notably, EEPD1 showed the highest mean expression in LGG. Although matched controls for glioma were scarce in TCGA, integrated TCGA/GTEx analysis confirmed strong upregulation in both LGG and glioblastoma multiforme (GBM) relative to normal brain tissue (Fig. 7C).

We therefore assessed overall survival (OS) of cancer patients divided evenly based on EEPD1 expression. High EEPD1 expression was associated with a better OS in LGG (Fig. 7D), but conversely, with poorer OS in uveal melanoma (UVM) (Fig. 7E). In LGG, stratification by EEPD1 expression revealed a substantial survival difference, with median OS differing by approximately two- to three-fold across expression strata, consistent with a clinically meaningful effect size rather than a marginal statistical association. As molecular characterization and stratification of LGG is established with superior outcome prediction, we further analyzed EEPD1 differential expression in LGG subtypes. EEPD1 expression is elevated in isocitrate dehydrogenase (IDH)-mutant tumors (Fig. 7F) associated with metabolic reprogramming, increased oxidative stress, and epigenetic dysregulation [100]. Crucially, in the most favorable prognostic group (IDH-mutant, chromosome 1p/19q-codeleted [101]), higher EEPD1 expression remained a statistically significant low-risk factor. In fact, patients with lower-than-median EEPD1 expression from this group showed no better OS than the IDH-mutant, 1p/19q-intact group (Fig. 7G, group B and C, P > 0.005). These findings indicate EEPD1 expression aligns with oxidative stress pathways and suggest it may hold significant prognostic value, particularly for refining patient stratification in gliomas.

Dimer disrupting mutations are predicted as most impactful for fitness

Given the correlation between EEPD1 status and cancer patient survival (Fig. 7), we sought to further test the implications of EEPD1 cancer mutations by applying a structure-independent evolutionary action (EA) scoring which assesses the functional impact of a coding variant [95, 102]. Using EA, we predicted the negative impact of EEPD1’s variants of unknown significance (VUS) from all tumor samples in clinical databases and mapped the most severe predictions onto our structure. Notably, the highest-scoring (most impactful) mutations clustered at the dimer interface, suggesting that these recorded cancer VUS may impact dimerization (Fig. 6A and B, black dot, and Fig. 8A; Supplementary Table S6). To test and validate the importance of the dimer interface residues, we introduced single amino acid substitutions (W522A, K275A/E) and clinical variants (G526W/V, S523P) into the EEP domain. We expressed and purified these mutants and performed BS3 crosslinking studies. All mutants showed a varying degree of dimerization perturbations (Fig. 8B). A titration of EEPD1^W522A to the wild-type ATTO488–EEPD1 failed to produce a measurable dimerization K_d in MST (Fig. 8C). The selected mutation affecting EEPD1 dimerization was further confirmed by SEC–SAXS–MALS analyses, where they showed smaller size and mass compared to wild-type (Fig. 8D and Supplementary Table S3). Interestingly, the K275A displayed a more extended conformation than other monomeric mutants, consistent with a dual role for Lys275 in both interfacing dimer and clipping multiple intramolecular structural elements together (Fig. 8D and E).

Figure 8.

Interface point mutations disrupt dimerization. (A) List of VUSs with highest EA score (likely to have the most severe impact). Probable impacts are categorized and annotated. Three mutations (bold, colored) were selected for follow-up characterization. (B) BS3 crosslinking reaction of EEP domain and dimer interface mutants, separated on SDS–PAGE. (C) MST measurement of EEPD1 dimerization. Titration of unlabeled full-length W522A mutant failed to saturate wild-type ATTO488–EEPD1. K_d of wild-type EEPD1 is an average of three measurements with standard deviation. (D) SEC–SAXS chromatogram (integrated intensity from selected q-range versus frame) with solution volume (radius of gyration, R_g) data points across peak. Color of the curves matches panel (B). (E) Real space paired-distance function P(r) derived from SAXS experimental data of W522A or K275 mutants. Disruption of K275A on stability of the dimer interface is manifested as tailing of an otherwise gaussian peak, increased distribution in higher distance r, and higher D_max. See Supplementary Table S3.

EEPD1 preferentially binds dsDNA components

Though inactive as a nuclease, the presence of four (HhH)₂ domains and two EEP domains in the EEPD1 dimer, its localization to the nucleus [2, 103], and its role in genome stability suggests that EEPD1 could function in DNA binding. We therefore tested full-length EEPD1 for binding to various structured DNA substrates by EMSA and more quantitative measurement by MST.

In the EMSA, EEPD1 shifted both single-stranded (ss) and dsDNA to the gel wells. The gel band characteristics suggested qualitatively cooperative binding of the DNA (Fig. 9A). By MST, EEPD1 displayed highest affinity binding, 30–35 nM K_d, to substrates with dynamic dsDNA intermediate structures (Fig. 9B), which was comparable to APE1 binding affinity to its substrates [83, 104]. EEPD1 had more than two-fold higher affinity for dsDNA structures than ssDNA. Moreover, its Hill coefficient > 1 (1.3–1.6) in association to dsDNA structures suggests cooperative binding, whereas no positive cooperativity was observed for ssDNA (Supplementary Table S7). The MST experiments were performed with low ionic strength (up to 50 mM KCl), as higher ionic strength significantly reduced DNA binding (Supplementary Fig. S13A).

Contrary to the previous interpretation that EEPD1 acted as an endonuclease on AP-sites [17] but consistent with our own results of no hydrolytic activity, the presence of THF did not enhance EEPD1 binding compared to intact dsDNA (Fig. 9B). Furthermore, in an EMSA experiment conducted at 100 mM KCl, EEPD1 did not bind THF-containing dsDNA, while APE1 retained sub-micromolar-K_d binding that was stable in gel electrophoresis (Supplementary Fig. S13B). Along with the relatively shallow surface topology surrounding EEPD1’s active site, key contacting residues on APE1’s loops including the unusual double-insertion by Met270 and Arg177 enabling tight dsDNA binding at near-physiological ionic strength [7], are also absent in EEPD1. Therefore, the EEPD1 active site environment not only lacks key catalytic side chains but also other structural elements and surface geometry for efficient AP recognition and endonucleolytic cleavage [83].

Full-length dimeric EEPD1 is required for optimal DNA binding

The (HhH)₂ domains and EEP domain are both known to engage DNA in other enzymes [15, 105]. However, EEPD1 is exceptional in tethering two (HhH)₂ domains per EEP domain, making four (HhH)₂ domains and two EEP domains per dimer. To determine which domains are responsible for DNA binding, we created a series of split constructs (Fig. 9C). Interestingly by MST, we found that neither the (HhH)₂ domains nor the EEP domain alone could stably engage DNA. Deleting even a single (HhH)₂ domain from each full-length subunit severely diminished binding. Only full-length EEPD1 binding to substrate has Hill coefficient > 1, suggesting full-length EEPD1 integrity is necessary for stable nucleic acid engagement.

Furthermore, we tested if the protein’s dimeric configuration is relevant for binding. Strikingly, dimer-disrupting mutations (W522A and K275E) diminished nucleic acid binding, confirming that the dimer form is essential for EEPD1’s DNA engagement (Fig. 9D). Notably, either disruption of dimer (W522A) or truncation of one (HhH)₂ domain (Δ1–129) effectively cuts the number of (HhH)₂ at a functional unit from four to two. On MST, they had similarly negative impacts on binding. Increasing numbers of (HhH)₂ domain enhancing DNA binding is documented in topoisomerase V from Methanopyrus genus [106]. An additional contribution of the C-terminal EEP domain to DNA binding is suggested by the observation that the dimer mutant retained residual binding capacity while the N-terminal fragment did not with the same two (HhH)₂ domains. Overall, our collective observations demonstrate that EEPD1 binds DNA through a cooperative mechanism that requires the full-length protein in its obligate dimeric state.

Dimeric EEP domain encircles the implicated DNA binding channel

To query the structural environment for DNA binding, we examined the topological and electrostatic surface around the active site and compared it to other EEP proteins [107]. EEP-fold proteins generally have a relatively narrow cleft near the active site, for insertion of nucleotides or lipids (Supplementary Fig. S14). In contrast, EEPD1 has a broader surface, which lacks the extensive positively charged surface seen in APE1 and LINE-1 that contacts a stretch of phosphodiesterase backbone (Supplementary Fig. S10A and C). However, overlaying DNA from an APE1 complex, we find dsDNA fits within a channel formed at the dimer interface and that EEPD1 Lys347, Lys481, and Lys487, besides the bound metal, would provide phosphodiester path contacts (Supplementary Fig. S10B).

Importantly, the architectural orientation of the two subunits places the pseudo active sites facing each other. Notably, based on an overlay of APE1, dsDNA fits surprisingly well between the two EEPD1 subunits (Fig. 9E). In the overlay, the phosphodiester backbone modeled onto each subunit lines up, indicating that the relative orientation of the two EEPD1 subunits may be strategic for placing the DNA phosphate backbone next to the dimer interface. Underneath the DNA backbone path, the two subunits form a striking bowl-shaped “di-Trp-Pro pocket” in the tryptophan handshake. From each subunit, two Trp522’s are positioned ∼12 Å apart on opposite sides, and two Pro519’s floor the pocket (Fig. 9F). Indeed, density in the pocket is noticed in the EEP²⁴⁵ structure, where we fit one imidazole (Fig. 9G). These experimentally observed densities in DNA-free crystal structure point to potential binding sites for aromatic bases, sandwiched between the two tryptophans.

Notably, three loops are positioned over the inferred DNA binding channel: loop β3–β4 and the two flexible loops β5–β6 and β7–α5 that are crystallographically undetermined due to low density (Fig. 2D, and Supplementary Fig. S6A and B). In the open form EEP²⁵⁴ (PDB: 9YSF), the narrowest gaps measure 24–26 Å, as suitable for accessibility to dsDNA (diameter 20 Å). In the closed form EEP²⁴⁵ structure (PDB: 9YXY), the narrowed gap measures ∼13 Å and would enable these loops to clamp into the major grooves (Supplementary Fig. S9B and C, and Supplementary Movie S2), supporting a DNA binding functionality for the handshake dimer interface and its adjustable “wrist.”

EEPD1 binds and protects reversed replication forks from nucleolytic degradation

Considering compelling evidence for EEPD1’s preferential binding to complex dsDNA structures, we hypothesized that this reflects it being targeted to dynamically stalled and reversed DNA replication forks, which are enriched during oxidative stress [67, 108, 109]. To test for this in cells, we used SIRF for the quantitative assessment of protein binding to the nascent DNA replication fork [65–68]. By leveraging a PLA between the EEPD1 protein target and biotinylated nascent DNA, SIRF captures proteins associated with the newly synthesized DNA in situ. Importantly, the total amount of nascent DNA is normalized to the PLA signals in SIRF to distinguish signal changes that are caused by protein–DNA interaction changes from those caused indirectly by changes in amounts of nascent DNA substrate. We applied low concentrations of CPT or hydrogen peroxide (H₂O₂), which are conditions validated to promote fork reversal in U2OS cells both by electron microscope [108, 110, 111], and by RF-SIRF [67], which is a PLA-based assay that we recently developed to detect reversed forks in single cells. Under these conditions, we observed a significant increase in EEPD1–SIRF signals with CPT induced fork reversal (0.067 without and 0.429 with CPT, P < 0.0001), and H₂O₂-induced fork reversal (0.001 without and 0.289 with H₂O₂, P < 0.0001) (Fig. 10A and B, and Supplementary Fig. S15). The SIRF signal is specific to EEPD1 as knock-down of EEPD1 by siRNA strongly reduces the EEPD1–SIRF signals (Fig. 10A and B, P < 0.0001, and Supplementary Fig. S15). These results suggest EEPD1 binds reversed DNA replication forks.

Figure 10.

EEPD1 targets and protects reversed replication forks. (A) Representative images of EEPD1–SIRF with 15 min of camptothecin (CPT, 50 nM) or hydrogen peroxide (H₂O₂, 20 μM) treatments in U2OS with EEPD1 knockdown (siEEPD1) or mock transfection (CTRL); NT, nontreatment; scale bar: 20 μm. (B) Scatter plot of EEPD1–SIRF. The bar represents the median (n = 99–129 for CTRL and n = 37–118 for siEEPD1). Two independent biological repeats. Top, conceptual schematic of EEPD1–SIRF [EEP domains in orange and light orange, (HhH)₂ domains in blue and green]. (C) Representative images of EEPD1–SIRF with 15 min of CPT (50 nM) or H₂O₂ (20 μM) treatments in U2OS and 3KO cells (U2OS cells with SMARCAL1, ZRANB3, and HLTF knock out). NT (non-treatment). Scale bar, 20 μm. (D) Scatter plot of EEPD1–SIRF. The bar represents the median (n = 99–129 for CTRL and n = 272–416), four independent biological repeats. (E) Representative images of DNA fibers in U2OS with EEPD1 knockdown (siEEPD1) or mock transfection (CTRL). Top, experimental sketch; scale bar: 20 μm. (F) Scatter plot of CldU fiber tract length divided by IdU DNA fiber spreads; hydroxyurea (HU, 4 mM, 4 h), MRE11 inhibitor PFM39 (50 μM). The bar represents the median (n = 63–82), three independent biological repeats, P values were derived from one-way ANOVA; ns P > 0.05, *P < 0.05, **P < 0.01, ****P < 0.0001.

To genetically verify this, we performed EEPD1–SIRF under the same conditions in U2OS 3KO cells [112] (herein 3KO cells), which have limited reversed replication forks due to CRISPR–Cas9 knock-out of the three primary translocases SMARCAL1, ZRANB3, and HLTF responsible for most of cellular fork reversal [109]. EEPD1–SIRF signals are significantly reduced in 3KO cells compared to U2OS cells with either CPT (0.265 in U2OS and 0.174 in 3KO, P < 0.0001) or H₂O₂ (0.358 in U2OS and 0.201 in 3KO, P < 0.0001), suggesting a preferential association of EEPD1 to reversed replication forks (Fig. 10C and D). Taken together, the data show that EEPD1 preferentially localizes to reversed replication forks in cells, which are complex dsDNA structures at stressed DNA replication forks.

Reversed replication forks require protection from nucleolytic degradation to suppress genome instability, a function distinct from DNA repair and attributed to many notorious tumor suppressors including BRCA1/2 and Fanconi Anemia (FANC) pathway members [61–63, 111, 113–115]. Given that EEPD1 suppresses cancer, targets reversed replication forks but lacks nuclease activity, we sought to test if EEPD1 may have a protective function at forks. Using single-molecule DNA fiber spreading, we tested for fork protection in U2OS cells and those lacking EEPD1 by siRNA knock down (Fig. 10E and F). By sequentially labeling DNA replication tracts with IdU followed by CldU before replication stalling with hydroxyurea (HU, see experimental sketch, Fig. 10E, top), the ratio of CldU/IdU tract lengths can indicate the selective loss of CldU label during replication stalling [62]. U2OS cells maintain a ratio of CldU/IdU tract lengths of approximately 1 with and without HU, suggesting that the nascent DNA tracts with replication stalling remain intact (Fig. 10E and F, P = ns). In contrast, cells lacking EEPD1 show a significant reduction in nascent CldU/IdU tract ratio with HU, suggesting a selective loss of CldU tracts with replication stress (Fig. 10E and F, P < 0.0001). Evidently, MRE11 nuclease is responsible for CldU tract loss as inhibition with the selective MRE11 inhibitor PFM39 prevents nascent fork degradation (Fig. 10E and F, P < 0.0001), as seen for other BRCA1/2 and FANC pathway members [61, 62, 111, 113, 115–117]. Collectively, these data show that EEPD1 binds and protects reversed replication forks from nucleolytic degradation by MRE11 nuclease: this defines EEPD1 as an unexpected key replication fork protection protein rather than a nuclease.

Discussion

With the first structural and biophysical characterization alongside molecular and cellular tests, we here identify EEPD1 as a dimeric clamp that binds complex dynamic dsDNA structures typically formed at stalled and reversed replication forks to protect them from nucleolytic degradation. EEPD1 has been thought to act as both a 5′-flap nuclease analogous to FEN1, and an abasic site endonuclease similar to APE1 in response to oxidative base damage and stalled replication forks [2, 4, 17]. However, our findings necessitate a reinterpretation of these purported nuclease roles. First, we find that purified EEPD1 remains catalytically inactive across a diverse array of DNA substrates. Second, this surprising absence of activity is fundamentally explained by our structure-based evolutionary analyses revealing systematic removal of key active-site residues essential for initiating phosphodiester cleavage at both 5′-flap and abasic sites. Third, functional assay results show EEPD1 binds to complex DNA substrates, localizes to replication forks, and protects reversed replication forks degradation as evident from experiments with validated MRE11 inhibitor [64]. These combined observations support an EEPD1 function in protection of metazoan dsDNA at reversed forks from MRE11 nuclease activity in response to oxidative stress. With reports of EEPD1’s role on plasma membrane mediated by its conserved lipidation sites [21, 22, 27], future studies should test if cholesterol and lipid oxidation release EEPD1 from the plasma membrane for its nuclear replication fork protection function as part of an interorganelle oxidative stress response. As a strictly metazoan innovation, we propose that EEPD1’s primary function has evolved from an EEP nuclease into a protective dsDNA binder for the replication stress response.

EEPD1 achieves this through an unprecedented yet evolutionarily conserved mechanism in metazoans: forming a unique EEP domain obligate dimer via a striking tryptophan handshake interface. Although not examined here, this conserved dimer and its pseudonuclease site (with its phosphate placement) could also have a metabolic role [27] that now can be tested by structure-informed mutagenesis. Notably EEPD1 could act similarly to canonical DNA clamps found in DNA replication stress response complexes such as FANCD2/FANCI or mono-ubiquitinated PCNA that can protect stalled replication forks from nucleolytic degradation [113, 118–122]. Interestingly, the dimer creates dynamic open-closed states with a DNA-encircling channel between two subunits overlying a central di-Trp-Pro pocket suited to recognize and clamp dsDNA. As binding of a single-base flap into FEN1's distal binding pocket increases its binding affinity by ~100-fold [123, 124], we reason that EEPD1–DNA binding affinity could similarly increase for dynamic dsDNA intermediates. Sculpting these intermediates to insert bases into the di-Trp-Pro pocket would thereby favor and stabilize the closed-clamp conformation (see Supplementary Movie S2). Furthermore, interacting protein partners may enhance the cooperative binding of EEPD1 to reversed replication forks at physiological ionic strength, analogous to how PCNA stimulates FAN1 binding at forks [125]. Moreover, dimer EEPD1’s four flexibly tethered (HhH)₂domains can collectively hold two added turns of dsDNA, and it shows cooperative binding to dsDNA but not ssDNA, consistent with the EEP and (HhH)₂ domains functioning synergistically in their canonical dsDNA-binding modes. EEPD1's exceptionally high flexibility, as shown by a strikingly low Porod coefficient (2.4) in our SAXS experiments, supports and explains its potential to adapt to multiple dynamic dsDNA structures. Given its multimodal and flexible DNA binding features, we propose that EEPD1 is structured to bind and tame dynamic multistrand DNA structures that are most prevalent at stressed forks (Supplementary Fig. S16).

Surprisingly, our findings suggest EEPD1 has lost its nuclease activity during metazoan evolution despite retaining dsDNA binding. Given the canonical phosphodiesterase role for EEP superfamily proteins, the mismatch between EEPD1’s substantial biological importance in replication restart and genome stability [2, 6, 17] and its relatively negligible inferred in vitro nuclease activity seems paradoxical. Notably, EEPD1 absence in zebrafish embryos and human cancer cells leads to the strong biological phenotype of genome instability, anaphase bridges, and micronuclei [2, 6, 17]. Yet, its implied nuclease activities are orders of magnitude lower than recognized EEP hydrolases including APE1 [7, 17, 83] or flap endonuclease FEN1 [4, 124, 126], and our own biochemical assays with purified EEPD1 detected no nuclease activity. The very low inferred nuclease activity at biologically reasonable enzyme–substrate ratios precludes robust kinetic measurements for compelling biochemical characterization of its DNA substrate specificity or mechanism of enzymatic action, such as done for FEN1 [126] and for numerous EEP hydrolases [25, 83, 84, 127].

Instead, our combined evolutionary analyses show that mammalian EEPD1 has strikingly discarded the key catalytic Asn_(iv). Importantly, this key Asn positions and activates the attacking water in EEP superfamily members [10, 26, 83–85]. The evolutionarily traceable deselection of the hallmark catalytic residue that initiates the phosphodiesterase reaction for the EEP superfamily argues for selective removal of nuclease activity. Consistently, absence of two other key residues for hydrolase activity of the structurally related APE1 undermines support for EEPD1’s potential as a backup AP-endonuclease [17, 85, 87, 88]. However, we recognize that the combined structural and evolutionary conservation data do not exclude the possibility that EEPD1 has adapted to an as yet undetected hydrolytic reaction that does not require water as a nucleophile, such as with the RNA 2′ hydroxyl. Nevertheless, the high degree of variation among all six active site motifs in EEPD1 during metazoan evolution is in stark contrast to the strict conservation of its dimerizing residues and of the active site motifs for functional EEP nucleases like APE1 (Fig. 6 and Supplementary Table S6). Our combined results therefore support retention of robust DNA-binding capacity without nuclease activity as key to EEPD1’s function and its biological role in replication stress responses in metazoans.

Inactivating an enzymatic fold while retaining DNA recognition functions has precedence in DNA replication stress and in DNA repair. The SLFN nuclease families contains both nuclease active and inactive members, in both dimeric and monomeric forms: SLFN11 cuts both transfer RNA (tRNA) and DNA, but SLFN5 lost its hydrolase-essential residue allowing it to bind but not cut DNA and SLNF2 actively suppresses tRNA cutting [128–132]. As another example, alkyltransferase-like proteins (ATLs) lost the reactive cysteine and alkyltransferase activity of O⁶-alkylguanine DNA-alkyltransferases (AGT) but retained the ability to not only bind but flip out the damaged base [133–135]. ATLs evolved to recruit active nuclease partners to DNA damage sites rather than acting as a catalytic hydrolase itself. Indeed, EEPD1 is reported to interact with EXO1 [4]. However, replication fork reversal principally is a DNA protective reaction to avoid dsDNA breaks and protect from ssDNA gap formation [108, 109, 136]. Thus, the fork protection function identified here suggests that EEPD1 lost its nuclease activity to suppress nucleolytic processing at forks, while retaining its dsDNA binding specificity to antagonize and regulate nucleolytic attack.

Distinct DNA structure-specific assemblies can distinguish DNA–protein complexes acting in replication fork protection and DNA repair. Dynamic dsDNA–protein complex structures are increasingly recognized as tell-tale features of reversed replication forks [61, 137]. In contrast, dsDNA break repair by homology directed repair (HDR) principally requires ssDNA–protein complexes for templated repair [61, 138]. EEPD1 previously was implicated in replication restart and HDR [2, 4, 17]. In analogy to the canonical tumor suppressor, BRCA2, which has two distinct, genetically, physiologically and mechanistically separable functions in fork protection and HDR [61, 62], it is possible that EEPD1 too has functions in both HDR and fork protection. However, the dimeric clamping of dsDNA and lack of nucleolytic activity of EEPD1 revealed here is consistent with our observed replication fork protective function.

Fork protection is linked to cancer suppression [61, 116, 139]. Crucially, experimental data showing that tumor-associated VUSs in the dimer interface abolish dimerization and DNA binding provide mechanistic evidence linking cancer to dysfunctional EEPD1 (Fig. 8). Here the observed correlation between EEPD1 expression in tumors, oxidative replication stress, and patient outcome supports and extends its nuclear functions, linking the newly identified EEPD1 dimer and fork protection mechanism to oxidative replication stress and better patient survival in certain cancers (Fig. 7). Importantly, the association between EEPD1 expression and patient survival should be interpreted in the context of pathway-level replication stress responses rather than implying EEPD1 acts as a solitary factor. Consistent with this view, genome-wide dependency datasets indicate that EEPD1 is not broadly required for cancer cell viability (https://depmap.org/portal/gene/EEPD1?tab=dependency) [140], suggesting that reduced survival in LGG tumors with low EEPD1 expression likely reflects impaired coordination of replication fork protection pathways rather than loss of a single essential gene. In LGGs, where IDH mutations drive metabolic reprogramming, increased reactive oxygen species, and epigenetic dysregulation, EEPD1 expression may therefore support the integrity of oxidative replication stress response networks including its functional partners at reversed replication forks, which are more likely to be defective in tumors.

EEPD1 upregulation is linked to radiation resistance in esophageal cancer patients treated by chemoradiotherapy [23], so there is a need for timely molecular tools to modulate EEPD1 function and test its therapeutic potential. Our results provide the structural basis for a precision oncology strategy for resensitizing cells to radiation therapy. The discovery of EEPD1’s unique tryptophan handshake dimer and its adjustable “wrist” enables two distinct therapeutic avenues previously unattainable: (i) designing allosteric inhibitors that lock the EEPD1 complex out of its active DNA-clamping conformation to enhance the effect of radiation or other therapies, and (ii) utilizing the novel di-Trp-Pro pocket as a specific structural anchor, moving drug design away from the challenging extended and conserved central DNA binding channel. Ultimately, a clamping, rather than cleaving, mechanism redefines EEPD1’s function in the metazoan oxidative replication stress response, unveiling a non-canonical fork protection target that could be exploited to advance next-generation precision cancer therapies.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by National Institutes of Health (NIH) [P01 CA092584 to J.A.T., Z.A., S.E.T., R35 CA220430 to J.A.T., R01 AG074009 to O.L.]; National Institute on Aging [U01-AG068214 to O.L.]; National Science Foundation [DBI-2344149 to O.L.]; Department of Defense [BC220523 to O.L.]; The Central Research Institute of Fukuoka University [257302 to I.K.]; MD Anderson Cancer Center bridge funds to K.S., MD Anderson Cancer Center Department of Cancer Biology grant to K.S.; and Robert A. Welch Chemistry Chair [G-0010 to J.A.T.]. Cryo-EM experiments were facilitated by Baylor College of Medicine cryo-EM core and the cryo-EM core facility at UTHealth as supported by Cancer Prevention and Research Institute of Texas [RP190602], and by the Pacific Northwest Center for Cryo-EM (PNCC) as supported by Department of Health and Human Services [R24GM154185]. Diffraction data were collected at SSRL Beamline 12-1 and 12-2, supported by Department of Health and Human Services [P30GM124169] for beamline operations. X-ray scattering at SIBYLS is supported by the Integrated Diffraction Analysis Technologies program of the US Department of Energy Office of Biological and Environmental Research. The Advanced Light Source (Berkeley, CA) is a national user facility operated by Lawrence Berkeley National Laboratory on behalf of the US Department of Energy under contract number DE-AC02-05CH11231, Office of Basic Energy Sciences. Funding to pay the Open Access publication charges for this article was provided by R35 CA220430.

Data availability

The TCGA RNA-Seq normalized rsem and clinical data underlying this article is available at https://doi.org/10.5281/zenodo.7885656. Source codes used for the TCGA bioinformatic analyses are available at https://doi.org/10.5281/zenodo.17550102. C++ codes for integrating t-test and linear regression with other compute programs for large-scale analyses are available at https://doi.org/10.5281/zenodo.16810772 and https://doi.org/10.5281/zenodo.16810790.

X-ray diffraction data and atomic coordinates along with structure factors have been deposited with the Protein Data Bank as 9YSF (EEP²⁵⁴) and 9YXY (EEP²⁴⁵).

SEC–SAXS raw data, final merged SAXS curves, P(r) functions, together with SEC–MALS reports are deposited to the SIMPLE SCATTERING database (https://simplescattering.com/). The SAXS accession IDs are XSTRXN7L (EEPD1), XSZV3EBD (EEP²⁴⁵), XSVGQDFV (EEP^245,G526W), XSCNTM2R (EEP^245,G526V), XS2XTQWG (EEP^245,W522A), XSHBS5V7 (EEP^245,K275E), XSSCJRQV (EEP^245,K275A), and XSDOI2CP (EEP^245,S523P). Ensembles best fitting EEP²⁴⁵ experimental data are deposited and linked to entry XSZV3EBD.

Structural coordinate and cryo-EM density map of EEPD1 have been deposited with the Protein Data Bank as 9YI2 and with EMDB as EMD-72976, respectively.

The Alvinella pompejana EEPD1 gene underlying this article is available in the GenBank Nucleotide Database at https://www.ncbi.nlm.nih.gov/genbank/ and can be accessed with accession number PX436288.