Mapping the sequence logic of DNA repair enzyme binding reveals mechanistic principles and evolutionary links

Mapping the sequence logic of DNA repair enzyme binding reveals mechanistic principles and evolutionary links | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Mapping the sequence logic of DNA repair enzyme binding reveals mechanistic principles and evolutionary links Ariel Afek, Noga Levy, Vered Levin Salomon, Sharon Greenwood, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7972250/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Mutations compromise genome stability, promote disease, and yet drive genetic diversity and evolution. DNA repair of mutagenic lesions acts to maintain genome integrity but is inherently imperfect, allowing mutations to emerge, persist and accumulate unevenly across the genome. Understanding when and where such mutations arise requires a deep understanding of the molecular factors that govern repair enzyme recognition. In base excision repair (BER), glycosylases must locate rare damaged bases that appear within diverse sequence and structural contexts across the genome, yet how these contexts modulate recognition and influence mutational outcomes remains unresolved. Here we introduce a high-throughput approach that quantifies BER-glycosylase binding across thousands of lesion-containing sequence contexts. Focusing on the cytosine-deamination pathway, we mapped the recognition landscapes of the human enzymes UDG and TDG, which act on the modified base uracil and on T:G mismatches derived from cytosine and 5-methylcytosine deamination, respectively. Our results reveal widespread sequence- and structure-dependent influences on binding, extending several bases from the lesion site and including non-additive interactions between flanking positions. Structural analyses implicate DNA-shape features such as base-step Slide and minor-groove width as determinants of recognition. Notably, sequence preferences influenced by flanks beyond the immediate neighbors reflect the cytosine–thymine (C/T) balance in the human genome, revealing a striking connection between long-range binding specificity and localized mutation patterns. Together, these findings and framework establish a generalizable approach for dissecting how DNA repair enzymes recognize lesions in sequence- and structure-specific contexts, providing a foundation for predictive models of how repair fidelity shapes genome evolution and human disease. Biological sciences/Biochemistry/DNA Biological sciences/Biochemistry/Enzymes/DNA repair enzymes Biological sciences/Molecular biology/DNA damage and repair/Base excision repair Biological sciences/Biochemistry/Proteins/DNA-binding proteins Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Genomic DNA is continuously challenged by spontaneous chemical reactions that generate thousands of lesions per cell per day 1 , including uracil from cytosine deamination and T:G mismatches from 5-methylcytosine deamination (Fig. 1 a). Cellular DNA repair pathways protect genomic integrity, with base excision repair (BER) providing the first line of defense against these frequent yet subtle forms of damage 2 , 3 . If left unrepaired 4 , these lesions are mutagenic, leading to genomic alterations that can be either creative or destructive. While mutations contribute to cancer, aging, and genetic disease, they also drive evolution and immunity development 5 – 9 . The efficiency with which BER enzymes recognize and excise lesions is therefore a key determinant of mutational outcomes 10 – 12 . Although the catalytic mechanisms of DNA glycosylases are well established 13 , 14 , a critical gap remains: how these enzymes discriminate among the vast diversity of sequence environments in the genome. Repair involves base flipping and local DNA distortion 15 , 16 , and crystallographic and mechanistic studies have revealed highly bent DNA conformations in the bound state 17 – 19 . However, these insights have largely come from a limited number of substrate examples, and a comprehensive, systematic framework for understanding how DNA sequence and structure influence lesion recognition remains lacking. Evidence indicates that flanking bases can modulate repair efficiency 15 , 20 – 22 . Biochemical studies have shown context effects for selected substrates, and structural analyses reveal that glycosylases often contact nucleotides beyond the lesion itself 23 , 24 (see example in Fig. 1 b). Mutational signatures in cancer genomes further suggest that context-dependent repair contributes to mutation biases 25 – 28 . Still, in contrast to transcription factors—where high-throughput assays have defined sequence rules, DNA-shape preferences, and cooperative interactions 29 – 32 —no comparable maps exist for DNA repair enzymes. The general principles by which BER glycosylases interpret sequence context therefore remain unresolved. The evolutionary consequences of cytosine deamination further reinforce the need to better understand the repair of this particular damage. Deamination of cytosine, if unrepaired, results in C→T transitions—one of the most common point mutations in the genome 9 , 38 (Fig. 1 b). In vertebrates, CpG dinucleotides are especially prone to methylation and subsequent deamination, contributing to their marked depletion over evolutionary time through C→T transitions 33 – 35 . Beyond CpGs, more subtle, context-dependent biases in C→T mutation rates contribute to fine-scale sequence composition 36 – 38 . The genomic C/T ratio thus captures the long-term balance between cytosine loss through deamination and its preservation by repair 39 , 40 (Fig. 1 a,b). Yet, whether such compositional patterns directly reflect intrinsic recognition rules of BER enzymes remains unknown. Here, we systematically define how the local DNA environment guides damage recognition by BER enzymes. We developed a high-throughput on-chip approach extending the principles of the protein-binding microarray 41 and saturation mismatch-binding assay 42 to measure the binding of UDG and TDG to uracil- and T:G-containing duplexes across thousands of sequence contexts, and to monitor catalytic activity through detection of the resulting apurinic/apyrimidinic (AP) sites (also known as abasic sites) 43 . UDG and TDG were selected as representative BER enzymes because of their complementary roles in repairing cytosine deamination products, thereby preventing the accumulation of C→T mutations. Importantly, our approach does not rely on polymerase-based methods or enzymatic incorporation of damage which limit the context scope of damage incorporation in classical methods. Instead, we directly synthesize the damaged substrates on-chip using a custom hairpin design that enables precise placement of lesions, including T:G mismatches for TDG and site-specific uracils for UDG in a double stranded region, in any desired sequence context (Fig. 1 c). The resulting measurements of binding and repair specificity afforded by our platform revealed that repair enzymes interpret information extending well beyond the damaged base, integrating flanking sequence and DNA-structural cues to determine recognition. When projected onto the human genome, these biochemical preferences coincide with local patterns of cytosine conservation, linking sequence-encoded recognition rules to the evolutionary imprint of DNA repair. Together, these results bridge molecular recognition and genome evolution, establishing a general framework for how DNA repair enzymes read the sequence- and structure-encoded signals of the genome. Results A high-throughput approach for systematic mapping of base excision repair recognition To exhaustively probe the influence of local sequence context on DNA damage recognition, we developed BER-Map (Base Excision Repair Enzyme Recognition Mapping), a high-throughput approach that embeds lesions into thousands of custom-designed DNA sequences with single-nucleotide precision (Fig. 1 c). We previously developed the saturation mismatch-binding assay (SaMBA) 42 to measure the effects of mismatches on protein–DNA interactions; here, we extend this approach to include chemically modified nucleotides. In all BER-Map libraries a lesion is fixed at a defined position while the surrounding sequence is systematically varied (e.g., 5′-NNNXNNN-3′, where N represents any nucleotide and X denotes the fixed damaged base). Multiple replicates per sequence context are included to ensure statistically robust measurements. Lesion-free sequences are always included as negative controls. All oligonucleotides were designed with a self-complementary segment that folds into a stem–loop hairpin conformation upon hybridization, ensuring that the lesion is presented in a double-stranded context (see Methods). Protein-binding measurements were performed directly on the microarrays by incubating the relevant repair enzyme—human UNG2 (the nuclear isoform, hereafter referred to as UDG) on uracil-containing libraries, in which uracil was paired with its physiological partners guanine (U:G) or adenine (U:A), and His-tagged human TDG (hereafter referred to as TDG) on T:G libraries. Binding levels were quantified from fluorescence signals, either directly from fluorophore-labeled protein or indirectly via fluorescent antibody detection (Fig. 1 c, Methods). Subsequently, we assayed enzymatic cleavage by detecting the excised product, the AP site generated upon base excision, using Aldehyde Reactive Probe (ARP) labeling (see Methods and Supplementary Fig. 1 for the AP detection workflow). This platform provides a versatile framework for systematically dissecting how sequence environments on the 5′ side, the 3′ side, or both, modulate repair enzyme binding, with binding measurements showing high reproducibility (see Supplementary Fig. 2 ). Comprehensive mapping of flank preferences reveals that damage recognition extends beyond adjacent positions BER-Map libraries provide a comprehensive framework for interrogating how sequence environments shape repair recognition. Using this platform, we first sought to elicit how bases flanking the lesion influence repair enzyme binding. We began by systematically evaluating the effects of single-nucleotide substitutions at each position by comparing binding to otherwise identical sequences differing only at the base and position of interest. For example, we selected a mid-affinity TDG target sequence and systematically substituted each flanking base with the three alternative nucleotides (Fig. 2 a). The resulting profile revealed that substitutions at multiple sites substantially altered binding, with clear preferences for specific bases at defined positions. For example, TDG showed a strong preference for G and reduced binding to C and T in position + 1 relative to the reference sequence. A similar analysis for UDG (Fig. 2 a, bottom) also revealed substantial flanking sequence effects, though with distinct positional preferences. In both cases, binding to damaged substrates was clearly greater than to non-damaged controls. While these illustrative profiles highlight the strong effects of flanking regions on binding, they each capture only a single sequence context. To move beyond individual examples, we leveraged the full BER-Map dataset to quantify the influence of each flanking position across all contexts. Specifically, we computed, for each position, the fraction of single-base substitutions that produced a statistically significant change in binding (statistical test and FDR correction in figure legend and in Methods). This analysis (Fig. 2 b) demonstrated that flanking substitutions were frequently significant. The strongest effects were observed at lesion-adjacent positions, particularly + 1, but distal flanks up to ± 3—which are typically not considered in current studies—also exerted a measurable and significant influence on recognition. We corroborated the effects of selected sequence contexts on UDG in solution using gel-based activity assays, obtaining steady-state rate constants ( k obs ) that closely tracked the array-derived trends (Fig. 2 c and Supplementary Fig. 3a-c ; see Methods). Moreover, TDG binding intensities also correlated well with published biochemical measurements 15 obtained for a smaller subset of sequences ( R ² = 0.79 for correlation with k max single-turnover rate constants; Supplementary Fig. 3d ). On top of the comparison between binding to other biochemical assays, we also examined how binding is related to product formation within our own assay. Remarkably, TDG binding correlated strongly with excised product formation on the array, whereas UDG displayed distinct sequence preferences and no detectable correlation with binding ( Supplementary Fig. 4 ), suggesting additional sequence-dependent influences beyond the binding step. Both enzymes were sensitive to bases up to position ± 3, the outermost range of our initial library. To test whether this influence extends further, we designed an additional library varying positions + 4 and + 5. Even these distal bases measurably affected binding ( Supplementary Fig. 5 ), suggesting that flanking sequence elements beyond the immediate lesion can modulate repair enzyme recognition, under certain contexts. While single-base (1-mer) analyses provide a useful approximation of positional effects, our platform enables systematic quantification of how combinations of flanking bases jointly influence binding. To capture multi-base contributions, we constructed a global representation of preferences across all 3-mer contexts. A circular recognition map (Fig. 2 d for TDG and Supplementary Fig. 6 for binding and ARP of UDG) summarizes these effects where within each half, corresponding to 5′ or 3′ flank, the inner rings encode the first two bases of each 3-mer, while the outer ring reports the average contribution of the third base, providing a compact visualization of thousands of measurements. Inspection of the circular map reveals 3-mer contexts with pronounced effects on damage binding. For example, TDG shows strong binding when the damage is flanked on the 3′ side by GCA (+ 1:+3), whereas binding is weak with CCT at the same positions (Fig. 2 d). Interestingly, other distinct 3′ 3-mers such as AGC can also support strong binding, similar to GCA. This highlights that completely different sequence compositions can yield similar or opposing binding outcomes, even though they flank the same damaged base. Together, these results demonstrate that base-excision repair recognition extends well beyond the lesion itself, with distal flanks exerting measurable influence, and motivated us to develop a global framework to capture these broader context-dependent effects in a systematic manner. Specificity is defined by base identity and dependencies To directly quantify how individual bases and their combinations influence damage recognition, we developed an analysis framework that captures both independent effects—how each base at each position contributes on its own—and conditional effects, which reflect dependencies between bases at different positions (i.e., how the effect of one base may depend on the identity of another). As a first step, we generated position weight matrix (PWM) representations of our dataset. PWMs are the standard framework for describing DNA-binding specificity in other DNA-binding protein classes and offer a first-order approximation of binding specificity based on additive positional contributions of each base. Moreover, PWMs may be represented in the form of compact sequence logos that provide an intuitive visual summary of binding preferences. To derive PWMs for UDG and TDG, we applied linear regression models where each base at each position was modeled as an independent feature (see Methods). These models explained most of the observed binding variation (R² = 0.78 for UDG and R² = 0.88 for TDG on the test set; see Supplementary Fig. 7a,b for performance and 7c,d for model coefficients), confirming that PWMs provide a reliable baseline description of repair enzyme specificity. The resulting PWM logos (Fig. 3 a) reveal that both enzymes are influenced by distal flanks and that the strongest contributions occur at position + 1: G for TDG and A for UDG, consistent with trends observed above (Fig. 2 a, b and Supplementary Fig. 6 ). They further highlight marked differences in relative contributions—TDG displays an exceptionally sharp preference for G at + 1, consistent with its role in recognizing T:G mismatches that predominantly occur in CpG contexts 15 , 40 , whereas UDG’s preference for A at + 1 is less pronounced, reflecting its broader role in removing uracil lesions that arise more uniformly across genomic contexts 45 . A limitation of traditional PWMs is that they assume positional independence. Whilst this provides a good first-order approximation, the influence of neighboring bases on each other’s geometry and orientation may create cooperative effects on binding specificity that should not be overlooked. Indeed, our circular recognition map (Fig. 2 d) revealed hints of such conditional effects between positions: for example, on the 3′ side, base C is favored at position + 3 when + 2 is A/G/T, but this preference shifts to G when + 2 is C. Moreover, similar dependencies were demonstrated to modulate transcription factor binding 31 , 32 , motivating us to test whether such context-dependent interactions likewise contribute to repair enzyme specificity. To systematically uncover these dependencies, we extended our prediction model by incorporating interaction terms between adjacent positions (see Methods). Incorporating these interactions significantly enhanced model performance (R² = 0.83 for UDG; R² = 0.94 for TDG on the test set; see Supplementary Fig. 7e for performance and Supplementary Table 1 for coefficient values). The major contributing coefficients are visualized in Fig. 3 b. For instance, the black arrow in Fig. 3 b (right) points to a blue arc that represents a strong positive interaction between C at position + 2 and G at position + 3, which is reflected by a large positive coefficient in the model. We next examined key model-identified interactions by comparing binding preferences across neighboring-base contexts. This analysis revealed context-dependent shifts in preferred bases across multiple interactions (Fig. 3 c as an example and Supplementary Fig. 8–9 for the other interactions visualized in Fig. 3 b). For example, while the PWM for TDG suggests a general preference for cytosine at + 3 (Fig. 3 a right, black arrow), the dependency model predicted—and direct comparisons confirmed—that this preference switches to guanine when + 2 is cytosine (Fig. 3 c, bottom). More concretely, Fig. 3 c (top panel) shows that C is preferred over G at the + 3 position when + 2 is A, T, or G, but this preference is reversed when + 2 is C, illustrating a context-dependent shift in base preference at + 3. Thus, extending beyond PWMs to include dependencies reveals hidden conditional effects, providing a more detailed picture of how repair enzymes interpret sequence context. DNA structure influences repair enzyme recognition Our results reveal that repair enzyme specificity is shaped not only by adjacent bases but also by distal flanks, with dependencies emerging between positions. Such dependencies often arise when proteins read the long-range structural properties of DNA, rather than simply the chemical signature of the individual bases alone. These observations raised the key question of whether DNA shape directs recognition of damage by repair enzymes, in an analogous manner to that demonstrated for several transcription factors 46 , 47 . Indeed, our comprehensive BER-Map data now provides the breadth and resolution needed for this systematic dissection of DNA shape contributions to repair enzyme recognition, removing the limitation of previous low-throughput studies which focus on a small number of specific sequence contexts. We therefore examined whether computationally calculated DNA shape features may account for the observed binding patterns in our analysis. To this end, we correlated predicted base step parameters (Tilt, Slide, Shift, Roll, Rise and Helical Twist) from flanking sequences (excluding the lesion site, for which shape values are undefined) 48 , 49 , with UDG and TDG binding measurements from our BER-Map libraries, averaged over dinucleotide steps (see Methods). This analysis revealed that multiple structural features across distinct positions contribute to binding (Fig. 4 a) with either strong positive (red) or negative (blue) correlation observed. Strikingly, some shape features exhibited strong standalone correlations with binding, as indicated by the more intense colors in Fig. 4 a—an unexpected result, as many structural features are normally intertwined. This suggests that some structural features may contribute to repair enzyme recognition more strongly than others. One such example is the strong significant relationship of the structural parameter Slide with UDG binding at the + 2/+3 base step relative to the damage site (Fig. 4 b). Slide describes the lateral displacement of adjacent base pairs, a feature that can reshape groove geometry and promote local DNA deformation. Analysis of four UDG–DNA complexes from the Protein Data Bank (PDB), selected for having matched sequence flanks, using ProDB 50 revealed the consistent presence of marked deformation at this site, with narrowing of the minor groove width (MGW) by up to 2.6 Å compared to the + 1 base, reaching levels below those defined by the B-DNA envelope (Fig. 4 c as a representative example for narrowing of MGW, and Supplementary Fig. 10 ; see Methods). Notably, an arginine residue, frequently implicated in recognizing narrow minor grooves 30 , is positioned adjacent to this site (Arg276 in the hUNG-DNA complex, PDB: 1SSP 19 ; corresponding to residue 195 in structure coordinates). These observations pointed to a role for groove narrowing in damage recognition. Given prior evidence that transcription factors often exploit narrow minor grooves for recognition 30 , we investigated whether the UDG repair enzyme shows a similar trend. Indeed, analysis of the narrowest and widest 10% MGW sequences in our library revealed stronger average binding to narrow grooves at this position (Fig. 4 d; see Methods). Together, these results show that repair enzyme specificity arises from an interplay of base readout and local DNA structure, and that high-throughput approaches such as BER-Map enable these correlations to be systematically revealed. Repair enzyme recognition as a potential force in genome evolution Having shown that flanking bases and DNA shape modulate repair enzyme recognition in vitro , we next asked whether these mechanisms leave a detectable imprint on the human genome. Cytosine deamination and inefficient repair have long been linked to broad compositional biases in the genome, such as CG depletion and AT enrichment 33 , 34 . A useful way to capture this process is the current genomic C/T ratio: when cytosines are deaminated and not repaired, they are converted to thymines, reducing the relative abundance of cytosine over evolutionary time (as illustrated in Fig. 5 a). Thus, the C/T ratio provides an indirect readout of the long-term balance between cytosine loss and repair across sequence contexts. While these biases are well established, it remains unclear whether they reflect intrinsic recognition preferences of repair enzymes. We reasoned that if repair efficiency varies with flanking sequence—as reflected in the binding preferences of enzymes like TDG and UDG—these preferences might be mirrored in the current sequence composition of the genome, specifically in the pattern of genomic C/T ratios. To test this, we analyzed each of the 16 central triplets (NXN, with X denoting the lesion site) separately (see Fig. 5 b and Methods). For each triplet, we binned extended flanking sequences according to their repair enzyme binding strength measured on the array, then calculated the median genomic C/T ratio in each bin and tested for correlation between binding and C/T ratio. This approach controls for the immediate lesion context and isolates the influence of distal flanking bases, while minimizing noise from confounding features such as chromatin environment or replication timing. In the case of TDG, we observed a striking correlation between repair enzyme binding and genomic composition. For example, in the GXC triplet—defined by fixed G and C at the 5′ and 3′ positions—contexts with strong TDG binding showed markedly higher C/T ratios than weakly bound ones (Fig. 5 c), consistent with a model in which stronger TDG recognition reduces the accumulation of C→T mutations. Extending this analysis across all triplets revealed a similar trend in most cases, with 12 out of 16 triplets showing a positive correlation between TDG binding levels and genomic C/T ratio (Fig. 5 d). The four TDG triplets that did not show this correlation all shared an NXG context, corresponding to CpG sites—the predominant targets of cytosine methylation. This exception may reflect the unique mutational dynamics of 5-methylcytosine deamination, which yields T:G mismatches, together with the involvement of alternative repair enzymes such as MBD4, which specifically targets CpG deamination products 40 , 51 . We confirmed that the 12 out of 16 significant correlations are non-random by simulating 10,000 random control genomes matched to the human genome in overall base composition. For each simulated genome, we repeated the same correlation analysis across all 16 NXN triplets. While the human genome showed 12 significant correlations, none of the simulated genomes exhibited more than 4 (empirical P < 1 × 10⁻⁴; see Methods), confirming that the observed correlations are unlikely to have arisen by chance. In contrast, UDG binding did not show a comparable relationship, showing a weak, non-significant trend in the opposite direction ( Supplementary Fig. 11a-b for both binding and ARP measurements respectively). This lack of a comparable relationship may reflect a genuine biological difference but could also result from the smaller number of flank variants available for UDG in the current array design. Taken together, these findings are consistent with a model in which context-dependent TDG recognition contributes to fine-scale variation in mutability across the genome. Discussion We present BER-Map, a high-throughput framework that maps BER enzyme binding specificity across thousands of lesion contexts. This resource, the largest of its kind to our knowledge, uncovers how deeply flanking sequence modulates repair recognition and offers a scalable platform for integrative mechanistic and genomic analyses. By mapping the recognition rules of TDG and UDG at unprecedented resolution, we show that DNA recognition is profoundly shaped by its sequence environment. Flanking positions, including four or more nucleotides away from the lesion, exert significant influence on binding, indicating that BER enzymes interpret damage within a broad contextual framework rather than relying solely on the lesion base itself. Beyond individual positions, our combinatorial dataset reveals conditional dependencies between flanking bases, exposing higher-order recognition logic that extends past simple nearest-neighbor effects. Integrating sequence data with DNA structural features further demonstrates that enzymes are sensitive to DNA shape as well as base identity. Minor-groove width, base-step geometry, and other local deformations correlate with binding outcomes, suggesting that BER enzymes exploit structural cues to enhance lesion discrimination. Finally, the context-dependent binding rules uncovered here align with genomic composition, suggesting that the principles guiding repair recognition may also contribute to mutational biases and compositional asymmetries in the human genome. These results raise the possibility that the evolutionary landscape of genome stability is shaped not only by the chemistry of the formation of deaminated products but also by the recognition logic of repair enzymes themselves. The consistency across 12 of 16 triplets, together with the absence of such patterns in simulated genomes, strongly supports a model in which repair enzyme specificity leaves a measurable imprint on genome composition. While a detailed understanding of every step in repair enzyme activity is important, our study focuses primarily on binding rather than catalysis, since binding affinity is a key determinant of damage recognition and sequence context selectivity 52 , 53 . Our ARP assays, which monitor the excision step, confirm that for TDG, binding strength correlates closely with cleavage activity, whereas for UDG, the sequence context preferences of binding diverge from those of catalysis. This distinction highlights the value of measuring binding directly, as it isolates the recognition process from downstream enzymatic steps and demonstrates the potential of our approach to probe additional stages that collectively govern mutagenesis. Notably, recent studies have shown that impaired or redirected binding is often what drives mutation accumulation in vivo 54 . These findings emphasize the need to understand the rules governing damage recognition, and position damage binding as a central control point in genome stability. TDG and UDG were chosen as representative glycosylases because they play central and complementary roles in counteracting cytosine deamination—one of the most frequent sources of mutation in the genome—and thereby help prevent C→T substitutions, which are among the most common mutations observed 7 , 33 , 34 . Focusing on these two enzymes allowed us to capture distinct recognition strategies within the same pathway. The framework, however, is readily generalizable to other BER enzymes and lesion types. Expanding BER-Map to additional repair pathways could generate comprehensive catalogs of sequence- and structure-dependent recognition rules, offering a unified view of lesion recognition and its impact on genome stability and evolution. While our platform does not currently capture in vivo complexities such as transcription factor binding, nucleosome interactions, or broader chromatin context 10 , 11 , 55 , 56 , these factors have been shown to play critical roles in shaping damage recognition and mutagenesis. Notably, most such insights have come from studies on mismatch repair (MMR) and nucleotide excision repair (NER), highlighting the need for parallel investigations in the context of base excision repair (BER). Our platform begins to fill this gap by providing high-resolution, sequence-specific characterization of BER enzyme binding, and lays the groundwork for integrating in vivo regulatory layers in future studies. In particular, it enables systematic exploration of how transcription factors might interact with damaged DNA to modulate repair accessibility. By coupling high-throughput lesion mapping with sequence, structural, and genomic analyses, BER-Map provides an integrative framework for uncovering the rules of repair enzyme recognition. This approach bridges biochemical specificity with genome-scale mutational patterns, establishing the foundation for predictive models that link DNA sequence, structure, repair efficiency, and mutational outcome in human cells. Beyond mechanistic insight, such frameworks may ultimately help interpret mutational signatures in cancer and guide therapeutic strategies targeting repair pathways. Methods Protein expression and purification Human TDG (hTDG) was expressed using the pET28-hTDG plasmid (#70758, Addgene), following the protocol described in Schuermann et al. 2016 57 . Freshly transformed E. coli BL21(DE3) cells were grown in 1.5L LB medium supplemented with kanamycin to an OD₆₀₀ of 0.5, induced with 0.25 mM IPTG, and incubated at 15°C for 48 hours. Cells were harvested by centrifugation, resuspended in lysis buffer (20 mM Tris-HCl pH 7.5, 500 mM NaCl, 5 mM imidazole, 5% glycerol, 5 mM β-mercaptoethanol), and stored at °C. A tablet of protease inhibitor (Roche Diagnostics, catalog no. 11836170001) was added to the thawed cells and then cells were lysed by French press and sonication. Following centrifugation (35k rpm, for 1h), the supernatant was incubated with Ni-NTA resin for 1 hour at 4°C. hTDG was eluted with lysis buffer + 200 mM imidazole and dialyzed into 50 mM Tris-HCl pH 7.5, 100mM NaCl, 5% glycerol, 5mM β-mercaptoethanol. Further purification was performed using heparin and Q Sepharose chromatography as done in Schuermann et al. 2016. Final preparations were supplemented with 1 mM DTT and stored at − 80°C. Fluorescein (Fluor) Human UNG2 is a recombinant human UNG2 protein with an N-terminal fluorescein label that was produced and purified as described previously 58 . Briefly, an N-terminal cysteine was inserted before the first residue of recombinant full-length UNG2 to covalently react with fluorescein-5-maleimide. The N-terminal cysteine was possible because the UNG2 variant was expressed with a 8xHis-SUMO tag upstream of the cysteine, and the tag was removed with the SUMO protease during protein purification. The three endogenous cysteines of UNG2 were also mutated to alanines in UNG2(Fluor) to isolate labeling to the N-terminal cysteine. Previous studies indicated that UNG2(Fluor) had similar activity to wild-type UNG2 58 . Damaged sequences library design Uracil-containing libraries were synthesized by LC Sciences, using dUTP phosphoramidites. In contrast, T:G mismatch libraries, composed solely of canonical bases, were synthesized by Agilent Technologies using standard DNA synthesis protocols. To address the challenge posed by non-complementary pairing (T:G and U:G), we employed a hairpin-based design. Our oligonucleotides are self-annealing, forming an intramolecular hairpin structure. Each oligo includes a short loop sequence (5′-GGAA-3′) bridging two flanking arms, which hybridize to form a stem. The lesion site (uracil or thymine) is usually embedded within a 6-nucleotide variable region (3 bases on each side), positioned in the stem segment to mimic a double-stranded DNA context. Depending on the design, the library includes 1,024 to 16,384 unique sequences that systematically cover all possible 5- to 7-mer flanking contexts around the lesion (See Supplementary data 1–2 for all library designs used in this study). For the UDG libraries containing uracils, the design most commonly used in the Results section included fewer variable flanking positions (5′-NNUNNN-3′). As a result, the corresponding analyses and statistical comparisons in those cases were based on a smaller number of unique sequences. Each sequence probe is printed in 7–10 replicates to allow robust signal averaging, and additional sequences lacking any lesion serve as internal controls. Altogether, the design comprises tens of thousands of sequence probes, enabling high-resolution, quantitative mapping of sequence-context effects on glycosylase binding. A schematic of the single-stranded hairpin oligo design is shown in Fig. 1 c. BER glycosylase measurements To create the double-stranded hairpin structure, the microarray was incubated in annealing buffer (IDT catalog no. 11-05-01-03) at 85°C for 5 min. Then, gradually it was cooled down to room temperature. The binding step was performed as described previously 41 . Briefly, hybridized microarrays were blocked with 2% (w/v) nonfat milk for 1 h at room temperature. Following the blocking step and washes, the microarray was incubated with the repair enzyme of interest. Uracil libraries were incubated with 400 nM or 800 nM UDG for 0.5 h or 1 h respectively at room temperature in a reaction buffer containing 20 mM Tris-HCl pH 8, 1 mM DTT, 1 mM EDTA. T:G libraries were incubated with 1 µM TDG in a reaction buffer containing 50 mM Tris-HCl pH 8, 1 mM EDTA, 1 mM DTT, 5% Glycerol for 2 h at room temperature. To detect the binding of TDG, the relevant microarray was incubated with either Anti-His Alexa Fluor 647 (Qiagen, catalog no. 35370) or Anti-His Alexa Fluor 488 (Qiagen catalog no. 35310) for 1 h at room temperature. The antibody stock solutions as obtained from the supplier were diluted 1:18 in the respective reaction buffer. Fluorescein-labeled UDG was imaged directly after the binding step. The fluorescent signals were measured using a GenePix 4400A microarray scanner and extracted by GenePix Pro analysis software. Activity measurements using Aldehyde Reactive Probe (ARP) reagent Following scanning of protein binding, bound protein was removed from the microarray using the stripping protocol, as described previously 41 . To detect the formation of abasic (AP) sites, the microarray was incubated with 2 mg/mL ARP-biotin (GERBU Biotechnik GmbH, catalog No. A305-10) in 2% (w/v) nonfat milk for 45 min at room temperature. The slide was then washed for 3 min in 0.5% Tween-20 in PBS (1×), followed by 2 min in 0.01% Triton X-100 in PBS (1×). Subsequently, the microarray was incubated with 2 µg/mL streptavidin–Alexa Fluor 488 (Thermo Fisher, catalog No. S11223) in 2% nonfat milk for 45 min at room temperature. After incubation, the slide was washed twice for 3 min each in 0.05% Tween-20 in PBS (1×), and once for 2 min in PBS (1×). The arrays were scanned using a GenePix 4400A microarray scanner at an excitation wavelength of 488 nm. Microarray data preprocessing Binding data for all probes in each experiment were normalized using MASLINER and spatial-normalization functions based on Berger et al. (2009) 41 . Each sequence was represented by multiple replicate probes, allowing calculation of a reliable median fluorescence intensity across replicates, yielding one value per unique sequence. To correct for non-specific binding, we then subtracted the signal of the highest-intensity damage-free probe from all sequences, which defines the upper limit of non-specific interactions for the repair enzyme. This correction yields an intensity value of zero for non-specific binding. Because logarithmic transformation was applied, a constant of + 1 was added to all values to avoid undefined logarithms of zero. The resulting log-transformed values were used as normalized binding signals. Statistical analysis After preprocessing, all analyses were conducted using R software (version 4.2.2). Unless otherwise specified, sequence-level comparisons were performed using two-sided Wilcoxon tests (Mann–Whitney U test), and replicate-level comparisons using two-sided t -tests. When applicable, multiple comparison correction was applied using the method indicated in the Results section. P -values are reported either explicitly or using the specified convention in the figure legend or the following one: **** P < 0.0001; *** P < 0.001; ** P < 0.01; * P < 0.05; ns, not significant. Positional sensitivity analysis of flanking bases To assess how much each flanking position contributes to repair enzyme binding, we measured positional sensitivity by systematically testing whether changing a single base at a given position significantly alters binding. For each base position flanking the damage (− 3, − 2, −1, + 1, +2, + 3), we performed the following steps. First, we grouped sequences that were identical at all other flanking positions except the one being tested. Then, for each group, we compared the replicate-level fluorescence values (normalized as described above) between sequences carrying different bases at that position, and performed exhaustive pairwise t -tests between all base pairs (A vs C, A vs G, etc.), resulting in six comparisons per group. For example, when testing sensitivity at position − 3, we grouped all sequences that shared the same bases at positions − 2 through + 3, and compared binding signals between sequences differing only at position − 3. Per group of six comparisons, we adjusted P -values using the Benjamini-Hochberg procedure and recorded significance: if any pairwise comparison at a given position and context was significant (FDR < 0.05), that comparison was added to the sum of significant comparisons per position. Finally, we calculated the percentage of such significant comparisons at each position. These percentages are reported in Fig. 2 b as a metric for position-specific statistical variability in binding. Validation of UDG microarray data using gel-based uracil excision assays DNA oligonucleotides were purchased from IDT Technologies (see Supplementary Fig. 3a for the sequences). Uracil-containing oligonucleotides also had a 3′ fluorescein label (3′ 6-FAM modification). The oligonucleotides were purified by denaturing urea-TBE PAGE and then the buffer was exchanged into an assay buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 0.1 mM EDTA). To ensure that the signal that represents activity is coming only from dsDNA substrates, we mixed 8% molar excess of the unlabeled oligo with its complementary labeled oligo. The mixed oligonucleotides were then heated to > 95°C and slowly cooled to room temperature 59 , 60 . The uracil excision assay was performed at 22°C for 3.5 min in an assay buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 0.1 mM EDTA). We used 3 nM UDG and 0.5 µM DNA. In all cases, we intended for the enzyme to process less than ~ half of the DNA substrate for its most favorable sequences while detecting low turnover for its least favorable sequences. Because of the broad range of uracil excision activities for UDG and its catalytic domain on different DNA substrates 59 , 61 we could not always maintain strict steady-state conditions (less than ~ 10–15% of the substrate processed). Enzymatic reactions were quenched with 200 mM NaOH and heat. This step also cleaves the abasic sites produced by uracil excision. The substrate and product from quenched reactions were separated by denaturing urea-TBE PAGE, and the fluorescein end-labels on the oligonucleotides were visualized with in-gel fluorescence using an Azure c400 imager. The intensity of the substrate and product bands were quantified with Fiji/ImageJ 62 . The percent of substrate processed was calculated as the product intensity divided by the sum of the substrate and product intensities. Using this value, we determined the rate of activity for the enzyme by (% of substrate processed × µM substrate in the reaction) / reaction time. The rate was divided by the enzyme concentration used in the assay to calculate k obs values. In all cases, we conducted three to four independent assays. Data modeling by multiple linear regression To map the binding preferences of repair enzymes across different sequence contexts, we used multiple linear regression models with one-hot encoded DNA sequences as features for enzyme binding prediction. Each 6-mer (TDG) or 5-mer (UDG) sequence context (5′-NNNXNNN-3′ or 5′-NNXNNN-3′) was one-hot encoded across all positions, yielding a set of 24 or 20 mononucleotide features respectively. The data was split into 80% train and 20% test. We first fitted simple regression models to capture additive effects—i.e., the independent contributions of each base (fitting included intercept fitting). Model performance and coefficients are shown in Supplementary Fig. 7a-d . We next incorporated dinucleotide interaction features between adjacent positions, also using one-hot encoding for adjacent dinucleotides in every available base-step. Eventually, including the mononucleotide features, for TDG there were 104 binary features, and for UDG 84 binary features. The data were split into training (60%), validation (20%), and test (20%) sets. We first used LassoCV with 5-fold cross-validation on the training set to identify an initial optimal value of the regularization parameter (alpha). Based on this result, we manually explored a range of nearby alphas (e.g., ×1, ×1.5, ×2, ×5, ×10) to evaluate the tradeoff between model performance and sparsity, as measured on the validation set by R² and the number of non-zero coefficients. From this sweep, we selected a “sweet spot” alpha that retained high predictive performance on the validation set while reducing the number of redundant features (alpha = 0.001 for TDG and alpha = 0.002 for UDG). A final model was then trained using this alpha and evaluated on the independent test set. The final selected (non-zeroed out) coefficients were used as a measure of feature importance, allowing us to identify the most informative mono and di sequence features for enzyme binding (as seen in Fig. 3 b). Model performances are shown in Supplementary Fig. 7e and coefficients for each interaction (non-additive) model are reported in Supplementary Table 1 . Base step DNA shape features and MGW analysis To investigate whether local DNA structural features contribute to differential repair enzyme binding, we systematically analyzed predicted DNA shape parameters of the sequence flanks in our microarray library. We focused on dinucleotide steps within 3-mer flanks (NNN) located upstream and downstream of the central damage (i.e., positions − 3 to − 1 and + 1 to + 3 relative to the damaged base at position 0), due to the lack of structural information on damaged DNA. Structural values were obtained from the DeepShape webserver 48 , 49 , which can provide predictions for each possible 3-mer sequence across multiple shape features (HelT, Roll, Rise, Slide, Shift, and Tilt) based on high-throughput simulations. Each probe in our dataset (5′-NNNXNNN-3′) contains a unique 3-mer sequence on either side of the damage site; we assigned the corresponding structural values to each flank and aligned them by position relative to the lesion. Because multiple probes may share the same 3-mer flank (e.g., both AAAXAAA and CAAXAAA share the 3′ flank AAA), the same structural values were assigned redundantly across all probes in which that 3-mer appeared, consistent with their representation in the experimental design. Subsequently, for each base-step position (e.g., base 1→2, 2→3), we extracted the dinucleotide identity (e.g., AA, AC, etc.) and computed the average structural value for that dinucleotide across all contexts. In parallel, we computed the average normalized binding signal for the same dinucleotide at the same position, based on all probes containing it. This resulted in a position-wise matched dataset of mean structural values and mean binding intensities per dinucleotide, which we used to assess correlations between structure and binding preferences. For cases where library design included only 2-mer flanks (e.g. NNUNNN), we extracted structural data the same way only per dinucleotides and continued similarly. To assess the influence of minor groove width (MGW) on UDG binding, we used an additional microarray library design and a different database. MGW is typically more reliable when computed over longer DNA contexts, such as pentamers, due to its dependence on surrounding base stacking and groove geometry. Therefore, we utilized an asymmetric design where the uracil was fixed at the 5′ end, allowing systematic variation of the 3′-flanking region (5′-UNNNNN-3′). MGW values were downloaded from DNAShape 63 and assigned to the central base of each 5-mer flank (i.e., the third base, + 3, in the UNN N NN sequence), enabling further analysis with our binding measurements per corresponding flank. Structural analysis of UDG–DNA complex crystal structures To assess local DNA distortions around the lesion site, we analyzed four available crystal structures of uracil-bound UDG–DNA complexes: 1EMH, 1EMJ 64 , 1SSP, and 2SSP 19 . These structures were selected based on their shared sequence flanks surrounding the lesion site, allowing for a controlled comparison of structural features. DNA structural parameters, MGW and Slide, were extracted using the ProDB pipeline 50 , which integrates output from CURVES+. To visualize the distortions, DNA sequences from the structures were aligned according to the uracil position. Genomic analysis (C/T ratio analysis) To calculate the C/T ratio per context in the genome, we used the Homo sapiens (human) genome assembly GRCh38.p14 (hg38) from UCSC, as our reference genome. For that, we extracted reference genome sequences using the BSgenome.Hsapiens.UCSC.hg38 package from Bioconductor 65 . We scanned the genome (excluding sex chromosomes) for all 6-mer sequences with either a cytosine (C) or thymine (T) at the third position (i.e., NNCNNNN and NNTNNNN, respectively). For each unique 6-mer, we counted the number of occurrences on the forward strand of the genome. For the C/T ratio for each unique 6-mer, the number of occurrences in the reference genome was then divided as follows: NNCNNNN counts / NNTNNNN counts = C/T ratio per context. Since it is known that adjacent nucleotides greatly affect mutation accumulation and generally the evolution of a genome 66 , we decided to group our ratio results into triplets. We consider a ‘Triplet’ as the group of 6-mers sharing the same immediate 3′ and 5′ adjacent nucleotides (e.g., CXA, GXC). Afterward, for each triplet, we binned the data by the repair enzyme binding measurement. It is important to note that the number of samples per bin may change between libraries and affect the observed significance. Specifically, the unique 6-mers were ordered by binding signal levels and divided into groups of 16 bins, each containing 16 observations. Then, for each bin, the median was calculated for the C/T ratio, and the average was calculated for the normalized binding levels. Pearson correlation tests were done for each triplet on the binned data and corrected for multiple comparisons by the Benjamini-Hochberg procedure. The analysis shown corresponds to data averaged over the − 3 base position from the asymmetric library design (5′-NNNTNNNN-3′), which yielded the strongest and most consistent correlations with genomic C/T ratios. To test whether the observed C/T correlations with TDG binding levels could arise by chance, we generated a null distribution using simulated genomes. Random genomes of 10 Mb in length were generated by sampling nucleotides according to their frequencies in the human hg38 reference genome (A = 29.5%, C = 20.4%, G = 20.5%, T = 29.6%), thereby matching its overall composition. For each of 10,000 simulations, the C/T ratio was calculated per sequence context and correlated with the measured TDG binding levels across triplets following the same exact procedure described above. We then compared the number of significant triplet-level correlations in the real genome (12) to this null distribution. The empirical P -value was calculated as the proportion of simulations with as many or more significant triplets than observed. Declarations Data availability All raw and processed microarray data are provided as Supplementary Data 1–2 for TDG and UDG, respectively (Excel .xlsx format). Each sheet within these files corresponds to an individual library design and measurement, as indicated in its title and in the notes sheet. Relevant source code for data modeling supporting the findings of this study are available at: https://github.com/afeklab/LR-models-for-repair-enzyme-specificity-predictions . Acknowledgements We thank Yuval Bayer and Ruben Solozabal for insightful discussions on specificity modeling. We also thank Michael O’Hagan for carefully reviewing the manuscript, and Deborah Fass for valuable comments and suggestions. This research was supported by Israel Science Foundation (ISF) Grant No. 1174/22 and by a research grant from the Shimon and Golde Picker – Weizmann Annual Grant. B.P.W. and S.N.G. were supported by an award to B.P.W. from the National Institutes of Health (R01GM135152). Author information Authors and Affiliations Department of Chemical and Structural Biology, Weizmann Institute of Science, Rehovot, Israel Noga Levy, Vered Levin Salomon, Naama Kessler & Ariel Afek Department of Molecular genetics, Weizmann Institute of Science, Rehovot, Israel Omer Erez Department of Cell & Molecular Biology, Rowan-Virtua School of Osteopathic Medicine and Rowan-Virtua School of Translational Biomedical Engineering and Sciences, Rowan University, Stratford, New Jersey, United States of America Sharon N. Greenwood & Brian P. Weiser Contributions N.L. and A.A. designed the study. A.A. and V.L.S. supervised the research. N.L. developed the assay, generated high throughput enzyme-DNA binding data, and analyzed the data. S.N.G and B.P.W purified the human fluorescent UDG, performed the gel-based activity experiment and analyzed its data. O.E. helped with the genomic analysis. N.K. purified human TDG. N.L., V.L.S. and A.A wrote the manuscript, with inputs from all authors. All the authors critically reviewed the manuscript and approved the final version. Competing interests The authors declare no competing interests. Materials & Correspondence Correspondence to Ariel Afek. References Lindahl, T. Instability and decay of the primary structure of DNA. Nature 362 , 709–715 (1993). Krokan, H. E. & Bjørås, M. Base Excision Repair. Cold Spring Harb Perspect Biol 5 , a012583 (2013). Zharkov, D. O. Base excision DNA repair. Cellular and Molecular Life Sciences 65 , 1544–1565 (2008). Aitken, S. J. et al. Pervasive lesion segregation shapes cancer genome evolution. Nature 583 , 265–270 (2020). Liston, A., Humblet-Baron, S., Duffy, D. & Goris, A. Human immune diversity: from evolution to modernity. Nat Immunol 22 , 1479–1489 (2021). Quintana-Murci, L. Human Immunology through the Lens of Evolutionary Genetics. Cell 177 , 184–199 (2019). Wallace, S. S., Murphy, D. L. & Sweasy, J. B. Base excision repair and cancer. Cancer Lett 327 , 73–89 (2012). López-Otín, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. The Hallmarks of Aging. Cell 153 , 1194–1217 (2013). Alexandrov, L. B. et al. Signatures of mutational processes in human cancer. Nature 500 , 415–421 (2013). Gonzalez-Perez, A., Sabarinathan, R. & Lopez-Bigas, N. Local Determinants of the Mutational Landscape of the Human Genome. Cell 177 , 101–114 (2019). Anderson, C. J. et al. Strand-resolved mutagenicity of DNA damage and repair. Nature 630 , 744–751 (2024). Robinson, P. S. et al. Inherited MUTYH mutations cause elevated somatic mutation rates and distinctive mutational signatures in normal human cells. Nat Commun 13 , 1–12 (2022). Parikh, S. S., Putnam, C. D. & Tainer, J. A. Lessons learned from structural results on uracil-DNA glycosylase. Mutation Research/DNA Repair 460 , 183–199 (2000). Stivers, J. T. Extrahelical Damaged Base Recognition by DNA Glycosylase Enzymes. Chemistry – A European Journal 14 , 786–793 (2008). Dow, B. J., Malik, S. S. & Drohat, A. C. Defining the Role of Nucleotide Flipping in Enzyme Specificity Using 19F NMR. J Am Chem Soc 141 , 4952–4962 (2019). Slupphaug, G. et al. A nucleotide-flipping mechanism from the structure of human uracil-DNA glycosylase bound to DNA. Nature 384 , 87–92 (1996). Buechner, C. N., Maiti, A., Drohat, A. C. & Tessmer, I. Lesion search and recognition by thymine DNA glycosylase revealed by single molecule imaging. Nucleic Acids Res 43 , 2716–2729 (2015). Zharkov, D. O., Mechetin, G. V. & Nevinsky, G. A. Uracil-DNA glycosylase: Structural, thermodynamic and kinetic aspects of lesion search and recognition. Mutation Research/Fundamental and Molecular Mechanisms of Mutagenesis 685 , 11–20 (2010). Parikh, S. S. et al. Base excision repair initiation revealed by crystal structures and binding kinetics of human uracil‐DNA glycosylase with DNA. EMBO J 17 , 5214–5226 (1998). Donigan, K. A. & Sweasy, J. B. Sequence context-specific mutagenesis and base excision repair. Mol Carcinog 48 , 362–368 (2009). Nilsen, H., Yazdankhah, S. P., Eftedal, I. & Krokan, H. E. Sequence specificity for removal of uracil from U·A pairs and U·G mismatches by uracil-DNA glycosylase from Escherichia coli, and correlation with mutational hotspots. FEBS Lett 362 , 205–209 (1995). Eftedal, I., Guddal, P. H., Slupphaug, G., Volden, G. & Krokan, H. E. Consensus sequences for good and poor removal of uracil from double stranded DNA by uracil-DNA glycosylase. Nucleic Acids Res 21 , 2095–2101 (1993). Orndorff, P. B. et al. Uracil-DNA glycosylase efficiency is modulated by substrate rigidity. Sci Rep 13 , 1–14 (2023). Schärer, O. D., Kawate, T., Gallinari, P., Jiricny, J. & Verdine, G. L. Investigation of the mechanisms of DNA binding of the human G/T glycosylase using designed inhibitors. Proceedings of the National Academy of Sciences 94 , 4878–4883 (1997). Poetsch, A. R. The genomics of oxidative DNA damage, repair, and resulting mutagenesis. Comput Struct Biotechnol J 18 , 207–219 (2020). Rogozin, I. B. et al. Mutational signatures and mutable motifs in cancer genomes. Brief Bioinform 19 , 1085–1101 (2018). Alexandrov, L. B. et al. The repertoire of mutational signatures in human cancer. Nature 578 , 94–101 (2020). Kaplanis, J. et al. Genetic and chemotherapeutic influences on germline hypermutation. Nature 605 , 503–508 (2022). Jolma, A. et al. DNA-binding specificities of human transcription factors. Cell 152 , 327–339 (2013). Rohs, R. et al. The role of DNA shape in protein-DNA recognition. Nature 461 , 1248–1253 (2009). Rohs, R. et al. Origins of specificity in protein-DNA recognition. Annu Rev Biochem 79 , 233–269 (2010). Stormo, G. D. & Zhao, Y. Determining the specificity of protein-DNA interactions. Nat Rev Genet 11 , 751–760 (2010). Schorderet, D. F. & Gartler, S. M. Analysis of CpG suppression in methylated and nonmethylated species. Proceedings of the National Academy of Sciences 89 , 957–961 (1992). Bird, A. P. DNA methylation and the frequency of CpG in animal DNA. Nucleic Acids Res 8 , 1499–1504 (1980). Hess, S. T., Blake, J. D. & Blake, R. D. Wide variations in neighbor-dependent substitution rates. J Mol Biol 236 , 1022–1033 (1994). Aggarwala, V. & Voight, B. F. An expanded sequence context model broadly explains variability in polymorphism levels across the human genome. Nat Genet 48 , 349–355 (2016). Oman, M., Alam, A. & Ness, R. W. How Sequence Context-Dependent Mutability Drives Mutation Rate Variation in the Genome. Genome Biol Evol 14 , (2022). Youk, J., An, Y., Park, S., Lee, J. K. & Ju, Y. S. The genome-wide landscape of C:G > T:A polymorphism at the CpG contexts in the human population. BMC Genomics 21 , 1–11 (2020). Fryxell, K. J. & Zuckerkandl, E. Cytosine Deamination Plays a Primary Role in the Evolution of Mammalian Isochores. Mol Biol Evol 17 , 1371–1383 (2000). Bellacosa, A. & Drohat, A. C. Role of base excision repair in maintaining the genetic and epigenetic integrity of CpG sites. DNA Repair (Amst) 32 , 33–42 (2015). Berger, M. F. & Bulyk, M. L. Universal protein-binding microarrays for the comprehensive characterization of the DNA-binding specificities of transcription factors. Nat Protoc 4 , 393–411 (2009). Afek, A. et al. DNA mismatches reveal conformational penalties in protein–DNA recognition. Nature 587 , 291-296 (2020). Ide, H. et al. Synthesis and Damage Specificity of a Novel Probe for the Detection of Abasic Sites in DNA. Biochemistry 32 , 8276–8283 (1993). Coey, C. T. et al. Structural basis of damage recognition by thymine DNA glycosylase: Key roles for N-terminal residues. Nucleic Acids Res 44 , 10248–10258 (2016). Krokan, H. E., Drabløs, F. & Slupphaug, G. Uracil in DNA - Occurrence, consequences and repair. Oncogene 21 , 8935–8948 (2002). Gordân, R. et al. Genomic Regions Flanking E-Box Binding Sites Influence DNA Binding Specificity of bHLH Transcription Factors through DNA Shape. Cell Rep 3 , 1093–1104 (2013). Samee, M. A. H., Bruneau, B. G. & Pollard, K. S. A De Novo Shape Motif Discovery Algorithm Reveals Preferences of Transcription Factors for DNA Shape Beyond Sequence Motifs. Cell Syst 8 , 27-42.e6 (2019). Li, J., Chiu, T. P. & Rohs, R. Predicting DNA structure using a deep learning method. Nat Commun 15 , 1–12 (2024). Li, J. & Rohs, R. Deep DNAshape webserver: prediction and real-time visualization of DNA shape considering extended k-mers. Nucleic Acids Res 52 , W7–W12 (2024). Mitra, R., Cohen, A. S., Sagendorf, J. M., Berman, H. M. & Rohs, R. DNAproDB: an updated database for the automated and interactive analysis of protein–DNA complexes. Nucleic Acids Res 53 , D396–D402 (2025). Silveira, A. B. et al. Base-excision repair pathway shapes 5-methylcytosine deamination signatures in pan-cancer genomes. Nature Communications 15 , 1–17 (2024). Hardeland, U., Bentele, M., Jiricny, J. & Schar, P. Separating substrate recognition from base hydrolysis in human thymine DNA glycosylase by mutational analysis. Journal of Biological Chemistry 275 , 33449–33456 (2000). Bellamy, S. R. W. & Baldwin, G. S. A kinetic analysis of substrate recognition by uracil-DNA glycosylase from herpes simplex virus type 1. Nucleic Acids Res 29 , 3857–3863 (2001). Zhu, W. et al. DNA mutagenesis driven by transcription factor competition with mismatch repair. Cell 188 , 5735-5747.e15 (2025). Schnable, B. L. et al. Thymine DNA glycosylase combines sliding, hopping, and nucleosome interactions to efficiently search for 5-formylcytosine. Nature Communications 15 , 1–11 (2024). Sabarinathan, R., Mularoni, L., Deu-Pons, J., Gonzalez-Perez, A. & Lopez-Bigas, N. Nucleotide excision repair is impaired by binding of transcription factors to DNA. Nature 532 , 264–267 (2016). Schuermann, D. et al. 3CAPS – a structural AP–site analogue as a tool to investigate DNA base excision repair. Nucleic Acids Res 44 , 2187–2198 (2016). Weiser, B. P., Stivers, J. T. & Cole, P. A. Investigation of N-Terminal Phospho-Regulation of Uracil DNA Glycosylase Using Protein Semisynthesis. Biophys J 113 , 393–401 (2017). Weiser, B. P., Rodriguez, G., Cole, P. A. & Stivers, J. T. N-terminal domain of human uracil DNA glycosylase (hUNG2) promotes targeting to uracil sites adjacent to ssDNA–dsDNA junctions. Nucleic Acids Res 46 , 7169–7178 (2018). Weiser, B. P. Analysis of uracil DNA glycosylase (UNG2) stimulation by replication protein A (RPA) at ssDNA-dsDNA junctions. Biochimica et Biophysica Acta (BBA) - Proteins and Proteomics 1868 , 140347 (2020). Greenwood, S. N., Kulkarni, R. S., Mikhail, M. & Weiser, B. P. Replication Protein A Enhances Kinetics of Uracil DNA Glycosylase on ssDNA and Across DNA Junctions: Explored with a DNA Repair Complex Produced with SpyCatcher/SpyTag Ligation. ChemBioChem 24 , e202200765 (2023). Schindelin, J. et al. Fiji: An open-source platform for biological-image analysis. Nat Methods 9 , 676–682 (2012). Zhou, T. et al. DNAshape: a method for the high-throughput prediction of DNA structural features on a genomic scale. Nucleic Acids Res 41 , W56–W62 (2013). Parikh, S. S. et al. Uracil-DNA glycosylase–DNA substrate and product structures: Conformational strain promotes catalytic efficiency by coupled stereoelectronic effects. Proceedings of the National Academy of Sciences 97 , 5083–5088 (2000). Team TBD. BSgenome.Hsapiens.UCSC.hg38: Full genomic sequences for Homo sapiens (UCSC genome hg38) . R package version 1.4.5 (2023). Hodgkinson, A. & Eyre-Walker, A. Variation in the mutation rate across mammalian genomes. Nat Rev Genet 12 , 756–766 (2011). Additional Declarations There is NO Competing Interest. Supplementary Files Suptable1monoanddicoefficientsfromLRindependentadjacentposmodel.xlsx Table 1 Supplementaryfigures.docx Supplementary Figures Supdata1TDGbindingARPcombined.xlsx Supplementary data 1 Supdata2UDGbindingARPcombined.xlsx Supplementary data 2 Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7972250","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":541238322,"identity":"f7e61347-d8d0-4065-9b9b-43cf21f91977","order_by":0,"name":"Ariel Afek","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABC0lEQVRIie3QMUsDMRTA8RcOziVn1ndQ9Cu8Ti5iP0iXSNcGHAXLeVLQ5W4vFNqvoEvng4NzCe1aOIcWwcnhJjmhQgMdpEIyO+S/JfDjvQTA5/uXsRQ2AAJ4UADQ7z0/dREJEKc8lMckdA06EE7HtzYipuXDRsIVCq6/sLlJ+mKaF9COoGMjuLwek4QBxnm+iCcUqsnbUrKsciym2SNKCBJaRYuaE1fpekgQpXZyrtlTK+Eeeyv+Ue8I1dwQ9uMgZKaY55dIURbWQKSeDQlcU7qajVHSK6KuLr4zkurFkLJToZWc6ZNt09zeocgG79TuEjVbD7vbz9FlT9ief1jvz7kwX+kEPp/P53O3Bx8sUg5CNW/EAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0001-8584-9879","institution":"Weizmann Institute of Science","correspondingAuthor":true,"prefix":"","firstName":"Ariel","middleName":"","lastName":"Afek","suffix":""},{"id":541238323,"identity":"886cc991-14b9-4625-ae24-b7551cc67dec","order_by":1,"name":"Noga Levy","email":"","orcid":"https://orcid.org/0009-0005-1405-889X","institution":"Weizmann Institute of Science","correspondingAuthor":false,"prefix":"","firstName":"Noga","middleName":"","lastName":"Levy","suffix":""},{"id":541238324,"identity":"5afe905a-4d5e-4a76-8ece-71e71bb832b6","order_by":2,"name":"Vered Levin Salomon","email":"","orcid":"","institution":"Weizmann Institute of Science","correspondingAuthor":false,"prefix":"","firstName":"Vered","middleName":"Levin","lastName":"Salomon","suffix":""},{"id":541238325,"identity":"8807d940-968f-4939-95df-69eb31a2b92e","order_by":3,"name":"Sharon Greenwood","email":"","orcid":"","institution":"Rowan University","correspondingAuthor":false,"prefix":"","firstName":"Sharon","middleName":"","lastName":"Greenwood","suffix":""},{"id":541238326,"identity":"6f355e16-490b-4f9a-9e3e-9c93e855ba0f","order_by":4,"name":"Naama Kessler","email":"","orcid":"","institution":"Weizmann Institute of Science","correspondingAuthor":false,"prefix":"","firstName":"Naama","middleName":"","lastName":"Kessler","suffix":""},{"id":541238327,"identity":"b88de98d-3fe9-4e57-bae3-59aec50c29de","order_by":5,"name":"Omer Erez","email":"","orcid":"","institution":"Weizmann Institute of Science","correspondingAuthor":false,"prefix":"","firstName":"Omer","middleName":"","lastName":"Erez","suffix":""},{"id":541238328,"identity":"7355365d-8c07-4b82-ac64-1e62bce0e8e0","order_by":6,"name":"Brian Weiser","email":"","orcid":"","institution":"Rowan University","correspondingAuthor":false,"prefix":"","firstName":"Brian","middleName":"","lastName":"Weiser","suffix":""}],"badges":[],"createdAt":"2025-10-28 17:28:45","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7972250/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7972250/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95580334,"identity":"2b2fb4da-a232-4819-8a91-5756099321a6","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7656894,"visible":true,"origin":"","legend":"","description":"","filename":"DNAdamagerepairUDGTDGmutagenesisfinal.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/d147d423d615066648254613.docx"},{"id":95580319,"identity":"16b9f9c0-2bf2-4d7a-bfc9-debb399d5342","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9036,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS2580992.json","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/955c0091f3b2ab3ffb459e53.json"},{"id":95655347,"identity":"24f8e84d-7869-4856-9fa2-1b3e3f83c098","added_by":"auto","created_at":"2025-11-11 16:15:30","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16928592,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfigures.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/9a978ecad2bdc8affb932852.docx"},{"id":95580322,"identity":"2768cc6e-2631-4b3f-997a-8f3e095d59c4","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":154818,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25809920enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/332d0180a0d24b407e05a693.xml"},{"id":95580325,"identity":"da42f2eb-e204-4068-a79a-6be3da256e83","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"pdf","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1903762,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/86024a4c600897bb9c425f3a.pdf"},{"id":95656123,"identity":"9401a6e5-b3c3-43a6-b439-8209579c38a8","added_by":"auto","created_at":"2025-11-11 16:17:48","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1418019,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/1b84647f243191c3e90512fe.pdf"},{"id":95655993,"identity":"657cc789-834f-47dd-8b06-21d8dcee467c","added_by":"auto","created_at":"2025-11-11 16:17:30","extension":"pdf","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2251175,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/1271c96a2c3f5622c89489d1.pdf"},{"id":95580339,"identity":"acbe8424-c255-48ce-a2a6-fa16f5d97a64","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"pdf","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1172637,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/fc5333010e38a3bf438f1473.pdf"},{"id":95655954,"identity":"27370a91-d61a-46f3-a0fb-643b2c05c432","added_by":"auto","created_at":"2025-11-11 16:17:24","extension":"pdf","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":888781,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/6fb6aca7dac442b1590dfe00.pdf"},{"id":95656183,"identity":"0f2fcb60-5ef6-424c-b19e-1a699ef38bdf","added_by":"auto","created_at":"2025-11-11 16:17:57","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1758736,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/ef9db2c6a59a6ef9a07b7f69.png"},{"id":95656063,"identity":"76cc687a-9d04-4a7a-8e4c-271cc1cf8fe4","added_by":"auto","created_at":"2025-11-11 16:17:43","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1435224,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/6bd69282ef21ba930773bdc4.png"},{"id":95580321,"identity":"1b3511a3-3ae5-4d19-ae47-c32e78632f2d","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1083087,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/806809126f46554af34d05d4.png"},{"id":95656092,"identity":"d3c420a9-5669-48bf-9c9f-c5e27946aef0","added_by":"auto","created_at":"2025-11-11 16:17:45","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":707427,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/8f4e4994ec2ae752e126793d.png"},{"id":95580338,"identity":"06a00ea8-38a1-411c-aef7-00951c9becc1","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1073175,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/f811625c7bab8f08ff13ad86.png"},{"id":95655803,"identity":"444f1b9d-6377-4e0f-9161-714bebbeeb21","added_by":"auto","created_at":"2025-11-11 16:16:59","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":409405,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/ae916e7d9f24160b9ddb0c78.png"},{"id":95655222,"identity":"de899e23-a6f4-45fc-8267-ed89991d92c8","added_by":"auto","created_at":"2025-11-11 16:14:46","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":419839,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/2e063110abe59f3c0dedcae4.png"},{"id":95580333,"identity":"3c098e05-94dd-402d-9c79-2a10204fe051","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":318035,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/04c5b921d2b9a1a05139b73d.png"},{"id":95580342,"identity":"6a70b014-4e6e-4bb6-9567-3a2558a93ba4","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":144625,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/69d0512bdb0a1864d9d63f25.png"},{"id":95580332,"identity":"556be66c-1819-4c91-9add-cabcf5bf33cf","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":332384,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/d841de24f61c2ad0f6d1d239.png"},{"id":95580337,"identity":"a12cc9ae-dca1-42e5-9c7d-b16d87fdba42","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"xml","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153251,"visible":true,"origin":"","legend":"","description":"","filename":"NCOMMS25809920structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/6e5d0e1831fc3f69f57d253d.xml"},{"id":95580340,"identity":"ca3a5dfe-2686-4957-87d2-ac6ac14fe7a6","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"html","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":165230,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/0279008764afa9bd3b6278c8.html"},{"id":95580313,"identity":"78764365-b348-46b6-96df-88efffdbc91a","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":699371,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHigh-throughput mapping of lesion recognition across sequence contexts. a,\u003c/strong\u003e Cytosine and methylcytosine deamination generate uracil and T:G mismatches, which arise in diverse sequence contexts across the genome. \u003cstrong\u003eb,\u003c/strong\u003e Base-excision repair enzymes such as human UDG and TDG identify and excise lesions within duplex DNA; protein–DNA proximity and possible interactions extend a few base pairs beyond the damaged base, as illustrated by the TDG–DNA crystal structure (PDBID: 5HF7\u003csup\u003e44\u003c/sup\u003e). Recognition is essential to prevent mutagenic C:G→T:A transitions. \u003cstrong\u003ec,\u003c/strong\u003e Schematic representation of the BER-Map platform. Thousands of oligonucleotides carrying a defined lesion (red) in systematically varied sequence contexts are synthesized on a microarray and then folded into hairpins that present the lesion in a double-stranded region. Binding by fluorescently labelled UDG or His-tagged TDG is quantified either directly or via fluorescent antibody detection. Example design (right) shows all flank positions varied (–3 to +3). This approach enables systematic dissection of how local and distal flanking sequences modulate repair enzyme recognition of DNA damage.\u003c/p\u003e","description":"","filename":"Binder11.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/e5bc20e9f2448f4b1322620a.png"},{"id":95580314,"identity":"574c063e-ad86-4296-a59e-fc7706e6615f","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":191294,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePositional sensitivities and sequence effects on lesion recognition. a, \u003c/strong\u003eEffect of single-nucleotide substitutions on repair enzyme binding. Shown are changes in binding signal relative to a reference sequence. Top: TDG, GTCXAAA with X = T:G mismatch. Bottom: UDG, GCXGTG with X = U. Bars represent the substitution effect at each flanking position and are colored by the identity of the substituted base. The lesion site is fixed at position 0 (dashed line). Significant substitutions are outlined with bold lines, based on two-tailed \u003cem\u003et\u003c/em\u003e-tests comparing replicate fluorescence values of the reference sequence and the corresponding mutant. \u003cstrong\u003eb,\u003c/strong\u003eGlobal analysis of positional sensitivities across all BER-Map contexts (top: TDG, bottom: UDG). For each flanking position across all contexts (for TDG n=6,144 and for UDG n=1,536), we calculated the fraction of single-base substitutions that significantly altered binding (see Methods). Bars represent the proportion of statistically significant comparisons per position (FDR \u0026lt; 0.05) with an example in the right top panel for one such comparison with the corresponding adjusted \u003cem\u003eP\u003c/em\u003e-value. \u003cstrong\u003ec,\u003c/strong\u003e Example of binding effects from bases beyond the immediate flank (with A fixed at the +1 position) observed in BER-Map (left) and validated by gel-based activity assays (right). Reported \u003cem\u003eP\u003c/em\u003e-value was calculated using replicate-level two-tailed \u003cem\u003et\u003c/em\u003e-test. \u003cstrong\u003ed,\u003c/strong\u003e Global view of TDG binding preferences displayed as a circular recognition map (see Supplementary Fig. 6 for UDG). The right half summarizes mean signals for all 3-mer contexts on the 3′ side of the lesion (+1 to +3), and the left half summarizes the 5′ side (−1 to −3). Colors indicate average binding intensity. Examples highlight strong binding in the GCA context and weak binding in CCT of 3′ side.\u003c/p\u003e","description":"","filename":"Binder12.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/8e8aebdfa9de769459bead52.png"},{"id":95656059,"identity":"3324573a-b167-4c3d-8cb5-55c4f5df73c2","added_by":"auto","created_at":"2025-11-11 16:17:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":173085,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eVisualization of TDG and UDG binding preferences and context-dependent base interactions.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003ePWM logos for UDG (left) and TDG (right). Letters represent regression coefficient values from linear models, showing the additive contribution of each base at each position (relative to the damage site) to the normalized binding signal. Coefficients were mean-centered per position so that values reflect relative base preferences. The arrow in the TDG PWM highlights a specific base preference at position +3, which is further explored in subsequent panels. \u003cstrong\u003eb, \u003c/strong\u003eSequence context contributions to UDG and TDG binding, derived from a linear regression model with interaction terms. Vertical bars represent single-base coefficients from this model at each position, with height indicating coefficient magnitude and color denoting base identity. Curved arcs represent pairwise interaction coefficients from the same model: blue for positive and orange for negative effects. Arc width reflects interaction strength, and arc height corresponds to the predicted binding signal based on both the individual base effects (bars) and their interaction (arc). Interactions below |0.2| Δln(binding signal) are rendered faintly for clarity. The arrow in the TDG panel shows the interaction demonstrated in (c). \u003cstrong\u003ec,\u003c/strong\u003e Positional interaction effects between position +2 and +3 for TDG damage binding. We focused on 6-mer contexts of the form 5′-NNNTCNN-3′, fixing the base at position +1 as C to reduce sequence diversity. Within this subset, we grouped sequences by the base at position +2 (A, T, G, or C) and compared binding signals between matched pairs that differed only at position +3, containing either C or G (i.e., a C→G substitution at +3). The top panel shows TDG binding signals across matched pairs, with colored lines connecting sequences that differ only at position +3. Line color indicates whether the substitution increased (green) or decreased (red) binding. Statistical comparisons were performed using paired \u003cem\u003et\u003c/em\u003e‑tests within each group, with significance shown above (***\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001; **\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01; *\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05; ns, not significant). The bottom panel summarizes these differences as Δln(binding signal) = G − C.\u003c/p\u003e","description":"","filename":"Binder13.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/8c83baf625f2fb45539f44ba.png"},{"id":95655195,"identity":"0f41b1cc-a70c-41e7-b0bf-26f7e0161fe0","added_by":"auto","created_at":"2025-11-11 16:14:34","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":195695,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eLocal DNA shape features in the 3′ flank of uracil correlate with UDG binding.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Correlation heatmap between UDG and TDG binding levels (averaged across all sequences containing the same adjacent dinucleotide step when available, see Methods) and predicted DNA structural parameters from DeepShape\u003csup\u003e48,49\u003c/sup\u003e. Colors correspond to Pearson’s R and significant correlations (Benjamini-Hochberg FDR-adjusted \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05) are outlined in bold black. Multiple‑comparison correction was applied separately for each flank (3′ or 5′) and each enzyme dataset. \u003cstrong\u003eb,\u003c/strong\u003e Scatter plot showing the relationship between mean Slide values at step +2 to +3 and mean UDG binding signal. Slide values represent the average predicted structural parameter for each dinucleotide across all sequence contexts in which that dinucleotide occurs, and binding signals are likewise averaged over all such sequences. The reported \u003cem\u003eP\u003c/em\u003e‑value corresponds to the Benjamini-Hochberg-adjusted \u003cem\u003eP\u003c/em\u003e from the multiple‑comparison analysis in (a). \u003cstrong\u003ec, \u003c/strong\u003eAn example of UDG-DNA structural analysis showing the narrowing of MGW in positions 3′ relative to the damaged site (PDBID: 2SSP\u003csup\u003e19\u003c/sup\u003e). \u003cstrong\u003ed,\u003c/strong\u003e Boxplot comparing UDG binding signals for contexts with narrow (n=104) vs. wide (n=108) minor groove width (MGW) at position +3 (taken from Pentamer datasets, see Methods). Contexts with narrower grooves show significantly higher binding. Wilcoxon \u003cem\u003eP\u003c/em\u003e-value is presented.\u003c/p\u003e","description":"","filename":"Binder14.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/e545e5516e717a11b6bf244a.png"},{"id":95580318,"identity":"cd162449-3a30-465f-b4c6-4397bd0aa948","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":180700,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTDG specificity for non-adjacent flanks correlates with mutation protection index (C/T ratio).\u003c/strong\u003e \u003cstrong\u003ea, \u003c/strong\u003eSpontaneous deamination of cytosine leads to the formation of promutagenic lesions that, if left unrepaired, results in C→T transition mutations. Over multiple generations, such events can accumulate at many loci across the genome. Because cytosine bases are embedded in distinct flanking sequences, C→T mutations arise in diverse sequence contexts. This results in variable C/T genomic ratios fordifferent sequence contexts, which we use as a proxy for historical mutation rates in our comparative analysis with TDG binding levels (see panels b–d). \u003cstrong\u003eb, \u003c/strong\u003eSchematic illustrating triplet-based grouping for C/T ratio and binding analysis. For every flanking sequence context tested in the BER-Map assay, a corresponding genomic C/T ratio was calculated. Shown here are sequence pairs belonging to a specific NXN triplet (where the central damaged base and its immediate neighbors are fixed), illustrating how distal flanks (−2, +2 to +4) vary within each group. These variations are the key drivers of the observed correlations between enzyme binding and genomic mutation patterns shown in panels c–d.\u003cstrong\u003e c,\u003c/strong\u003e Example correlation for the GXC triplet: each point represents a bin of sequences that share the same immediate triplet (GXC) but differ in distal flanking context (as shown in panel b). Bins with lower TDG binding (x-axis) tend to show lower genomic C/T ratios (y-axis), indicating greater susceptibility to C→T substitutions. Pearson’s R, confidence interval and \u003cem\u003eP\u003c/em\u003e-value are reported. \u003cstrong\u003ed, \u003c/strong\u003eSummary of correlation coefficients after Pearson correlation test between TDG context preferences to C/T ratio of these contexts per triplet, as exemplified in panel c.X in the NXN triplets context depicted on the X-axis is the T:G mismatch position. All correlations \u003cem\u003eP\u003c/em\u003e-values are indicated after Benjamini-Hochberg correction for multiple comparisons.\u003c/p\u003e","description":"","filename":"Binder15.png","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/c3a432ee0a8987382d8c75c0.png"},{"id":95660128,"identity":"f4e630aa-932f-43eb-80d9-cf36b4ef4a21","added_by":"auto","created_at":"2025-11-11 16:30:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3115766,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/7d12133d-b2d3-4ae5-b10b-3900fc51867b.pdf"},{"id":95580315,"identity":"2171cd5d-5c7b-410b-8c56-425dbc5a917a","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":13106,"visible":true,"origin":"","legend":"Table 1","description":"","filename":"Suptable1monoanddicoefficientsfromLRindependentadjacentposmodel.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/5a82348cfa7c84c6cfa7d86c.xlsx"},{"id":95655553,"identity":"848e9593-11b8-489a-8967-65829a2811cc","added_by":"auto","created_at":"2025-11-11 16:16:27","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":16928592,"visible":true,"origin":"","legend":"Supplementary Figures","description":"","filename":"Supplementaryfigures.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/cd253d5b713e4abdbfd8c65a.docx"},{"id":95656002,"identity":"6a5b0059-1e1f-4929-8c77-ad16915a1fd1","added_by":"auto","created_at":"2025-11-11 16:17:31","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":16709970,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary data 1\u003c/p\u003e","description":"","filename":"Supdata1TDGbindingARPcombined.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/9aff15e22260ebe156519779.xlsx"},{"id":95580320,"identity":"2bd1c3ca-3760-4907-a9b4-e60b7010306a","added_by":"auto","created_at":"2025-11-10 19:54:38","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":1539856,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary data 2\u003c/p\u003e","description":"","filename":"Supdata2UDGbindingARPcombined.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-7972250/v1/a33da2ebc814a1cce319d27b.xlsx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Mapping the sequence logic of DNA repair enzyme binding reveals mechanistic principles and evolutionary links","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGenomic DNA is continuously challenged by spontaneous chemical reactions that generate thousands of lesions per cell per day\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e, including uracil from cytosine deamination and T:G mismatches from 5-methylcytosine deamination (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). Cellular DNA repair pathways protect genomic integrity, with base excision repair (BER) providing the first line of defense against these frequent yet subtle forms of damage\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. If left unrepaired\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e, these lesions are mutagenic, leading to genomic alterations that can be either creative or destructive. While mutations contribute to cancer, aging, and genetic disease, they also drive evolution and immunity development\u003csup\u003e\u003cspan additionalcitationids=\"CR6 CR7 CR8\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. The efficiency with which BER enzymes recognize and excise lesions is therefore a key determinant of mutational outcomes\u003csup\u003e\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAlthough the catalytic mechanisms of DNA glycosylases are well established\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, a critical gap remains: how these enzymes discriminate among the vast diversity of sequence environments in the genome. Repair involves base flipping and local DNA distortion\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, and crystallographic and mechanistic studies have revealed highly bent DNA conformations in the bound state\u003csup\u003e\u003cspan additionalcitationids=\"CR18\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. However, these insights have largely come from a limited number of substrate examples, and a comprehensive, systematic framework for understanding how DNA sequence and structure influence lesion recognition remains lacking.\u003c/p\u003e\u003cp\u003eEvidence indicates that flanking bases can modulate repair efficiency\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Biochemical studies have shown context effects for selected substrates, and structural analyses reveal that glycosylases often contact nucleotides beyond the lesion itself\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e (see example in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb). Mutational signatures in cancer genomes further suggest that context-dependent repair contributes to mutation biases\u003csup\u003e\u003cspan additionalcitationids=\"CR26 CR27\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Still, in contrast to transcription factors\u0026mdash;where high-throughput assays have defined sequence rules, DNA-shape preferences, and cooperative interactions\u003csup\u003e\u003cspan additionalcitationids=\"CR30 CR31\" citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e\u0026mdash;no comparable maps exist for DNA repair enzymes. The general principles by which BER glycosylases interpret sequence context therefore remain unresolved.\u003c/p\u003e\u003cp\u003eThe evolutionary consequences of cytosine deamination further reinforce the need to better understand the repair of this particular damage. Deamination of cytosine, if unrepaired, results in C\u0026rarr;T transitions\u0026mdash;one of the most common point mutations in the genome\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb). In vertebrates, CpG dinucleotides are especially prone to methylation and subsequent deamination, contributing to their marked depletion over evolutionary time through C\u0026rarr;T transitions\u003csup\u003e\u003cspan additionalcitationids=\"CR34\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. Beyond CpGs, more subtle, context-dependent biases in C\u0026rarr;T mutation rates contribute to fine-scale sequence composition\u003csup\u003e\u003cspan additionalcitationids=\"CR37\" citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e. The genomic C/T ratio thus captures the long-term balance between cytosine loss through deamination and its preservation by repair\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e,\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea,b). Yet, whether such compositional patterns directly reflect intrinsic recognition rules of BER enzymes remains unknown.\u003c/p\u003e\u003cp\u003eHere, we systematically define how the local DNA environment guides damage recognition by BER enzymes. We developed a high-throughput on-chip approach extending the principles of the protein-binding microarray\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e and saturation mismatch-binding assay\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e to measure the binding of UDG and TDG to uracil- and T:G-containing duplexes across thousands of sequence contexts, and to monitor catalytic activity through detection of the resulting apurinic/apyrimidinic (AP) sites (also known as abasic sites)\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. UDG and TDG were selected as representative BER enzymes because of their complementary roles in repairing cytosine deamination products, thereby preventing the accumulation of C\u0026rarr;T mutations. Importantly, our approach does not rely on polymerase-based methods or enzymatic incorporation of damage which limit the context scope of damage incorporation in classical methods. Instead, we directly synthesize the damaged substrates on-chip using a custom hairpin design that enables precise placement of lesions, including T:G mismatches for TDG and site-specific uracils for UDG in a double stranded region, in any desired sequence context (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec).\u003c/p\u003e\u003cp\u003eThe resulting measurements of binding and repair specificity afforded by our platform revealed that repair enzymes interpret information extending well beyond the damaged base, integrating flanking sequence and DNA-structural cues to determine recognition. When projected onto the human genome, these biochemical preferences coincide with local patterns of cytosine conservation, linking sequence-encoded recognition rules to the evolutionary imprint of DNA repair. Together, these results bridge molecular recognition and genome evolution, establishing a general framework for how DNA repair enzymes read the sequence- and structure-encoded signals of the genome.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eA high-throughput approach for systematic mapping of base excision repair recognition\u003c/h2\u003e\u003cp\u003eTo exhaustively probe the influence of local sequence context on DNA damage recognition, we developed BER-Map (Base Excision Repair Enzyme Recognition Mapping), a high-throughput approach that embeds lesions into thousands of custom-designed DNA sequences with single-nucleotide precision (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec). We previously developed the saturation mismatch-binding assay (SaMBA)\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e to measure the effects of mismatches on protein\u0026ndash;DNA interactions; here, we extend this approach to include chemically modified nucleotides. In all BER-Map libraries a lesion is fixed at a defined position while the surrounding sequence is systematically varied (e.g., 5\u0026prime;-NNNXNNN-3\u0026prime;, where N represents any nucleotide and X denotes the fixed damaged base). Multiple replicates per sequence context are included to ensure statistically robust measurements. Lesion-free sequences are always included as negative controls. All oligonucleotides were designed with a self-complementary segment that folds into a stem\u0026ndash;loop hairpin conformation upon hybridization, ensuring that the lesion is presented in a double-stranded context (see Methods).\u003c/p\u003e\u003cp\u003eProtein-binding measurements were performed directly on the microarrays by incubating the relevant repair enzyme\u0026mdash;human UNG2 (the nuclear isoform, hereafter referred to as UDG) on uracil-containing libraries, in which uracil was paired with its physiological partners guanine (U:G) or adenine (U:A), and His-tagged human TDG (hereafter referred to as TDG) on T:G libraries. Binding levels were quantified from fluorescence signals, either directly from fluorophore-labeled protein or indirectly via fluorescent antibody detection (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec, Methods). Subsequently, we assayed enzymatic cleavage by detecting the excised product, the AP site generated upon base excision, using Aldehyde Reactive Probe (ARP) labeling (see Methods and \u003cb\u003eSupplementary Fig.\u0026nbsp;1\u003c/b\u003e for the AP detection workflow). This platform provides a versatile framework for systematically dissecting how sequence environments on the 5\u0026prime; side, the 3\u0026prime; side, or both, modulate repair enzyme binding, with binding measurements showing high reproducibility (see \u003cb\u003eSupplementary Fig.\u0026nbsp;2\u003c/b\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eComprehensive mapping of flank preferences reveals that damage recognition extends beyond adjacent positions\u003c/h3\u003e\n\u003cp\u003eBER-Map libraries provide a comprehensive framework for interrogating how sequence environments shape repair recognition. Using this platform, we first sought to elicit how bases flanking the lesion influence repair enzyme binding. We began by systematically evaluating the effects of single-nucleotide substitutions at each position by comparing binding to otherwise identical sequences differing only at the base and position of interest. For example, we selected a mid-affinity TDG target sequence and systematically substituted each flanking base with the three alternative nucleotides (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). The resulting profile revealed that substitutions at multiple sites substantially altered binding, with clear preferences for specific bases at defined positions. For example, TDG showed a strong preference for G and reduced binding to C and T in position\u0026thinsp;+\u0026thinsp;1 relative to the reference sequence. A similar analysis for UDG (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, bottom) also revealed substantial flanking sequence effects, though with distinct positional preferences. In both cases, binding to damaged substrates was clearly greater than to non-damaged controls.\u003c/p\u003e\u003cp\u003eWhile these illustrative profiles highlight the strong effects of flanking regions on binding, they each capture only a single sequence context. To move beyond individual examples, we leveraged the full BER-Map dataset to quantify the influence of each flanking position across all contexts. Specifically, we computed, for each position, the fraction of single-base substitutions that produced a statistically significant change in binding (statistical test and FDR correction in figure legend and in Methods). This analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb) demonstrated that flanking substitutions were frequently significant. The strongest effects were observed at lesion-adjacent positions, particularly\u0026thinsp;+\u0026thinsp;1, but distal flanks up to \u0026plusmn;\u0026thinsp;3\u0026mdash;which are typically not considered in current studies\u0026mdash;also exerted a measurable and significant influence on recognition.\u003c/p\u003e\u003cp\u003eWe corroborated the effects of selected sequence contexts on UDG in solution using gel-based activity assays, obtaining steady-state rate constants (\u003cem\u003ek\u003c/em\u003e\u003csub\u003eobs\u003c/sub\u003e) that closely tracked the array-derived trends (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec and \u003cb\u003eSupplementary Fig.\u0026nbsp;3a-c\u003c/b\u003e; see Methods). Moreover, TDG binding intensities also correlated well with published biochemical measurements\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e obtained for a smaller subset of sequences (\u003cem\u003eR\u003c/em\u003e\u0026sup2; = 0.79 for correlation with \u003cem\u003ek\u003c/em\u003e\u003csub\u003emax\u003c/sub\u003e single-turnover rate constants; \u003cb\u003eSupplementary Fig.\u0026nbsp;3d\u003c/b\u003e). On top of the comparison between binding to other biochemical assays, we also examined how binding is related to product formation within our own assay. Remarkably, TDG binding correlated strongly with excised product formation on the array, whereas UDG displayed distinct sequence preferences and no detectable correlation with binding (\u003cb\u003eSupplementary Fig.\u0026nbsp;4\u003c/b\u003e), suggesting additional sequence-dependent influences beyond the binding step.\u003c/p\u003e\u003cp\u003eBoth enzymes were sensitive to bases up to position\u0026thinsp;\u0026plusmn;\u0026thinsp;3, the outermost range of our initial library. To test whether this influence extends further, we designed an additional library varying positions\u0026thinsp;+\u0026thinsp;4 and +\u0026thinsp;5. Even these distal bases measurably affected binding (\u003cb\u003eSupplementary Fig.\u0026nbsp;5\u003c/b\u003e), suggesting that flanking sequence elements beyond the immediate lesion can modulate repair enzyme recognition, under certain contexts.\u003c/p\u003e\u003cp\u003eWhile single-base (1-mer) analyses provide a useful approximation of positional effects, our platform enables systematic quantification of how combinations of flanking bases jointly influence binding. To capture multi-base contributions, we constructed a global representation of preferences across all 3-mer contexts. A circular recognition map (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed for TDG and \u003cb\u003eSupplementary Fig.\u0026nbsp;6\u003c/b\u003e for binding and ARP of UDG) summarizes these effects where within each half, corresponding to 5\u0026prime; or 3\u0026prime; flank, the inner rings encode the first two bases of each 3-mer, while the outer ring reports the average contribution of the third base, providing a compact visualization of thousands of measurements. Inspection of the circular map reveals 3-mer contexts with pronounced effects on damage binding. For example, TDG shows strong binding when the damage is flanked on the 3\u0026prime; side by GCA (+\u0026thinsp;1:+3), whereas binding is weak with CCT at the same positions (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed). Interestingly, other distinct 3\u0026prime; 3-mers such as AGC can also support strong binding, similar to GCA. This highlights that completely different sequence compositions can yield similar or opposing binding outcomes, even though they flank the same damaged base.\u003c/p\u003e\u003cp\u003eTogether, these results demonstrate that base-excision repair recognition extends well beyond the lesion itself, with distal flanks exerting measurable influence, and motivated us to develop a global framework to capture these broader context-dependent effects in a systematic manner.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eSpecificity is defined by base identity and dependencies\u003c/h3\u003e\n\u003cp\u003eTo directly quantify how individual bases and their combinations influence damage recognition, we developed an analysis framework that captures both independent effects\u0026mdash;how each base at each position contributes on its own\u0026mdash;and conditional effects, which reflect dependencies between bases at different positions (i.e., how the effect of one base may depend on the identity of another).\u003c/p\u003e\u003cp\u003eAs a first step, we generated position weight matrix (PWM) representations of our dataset. PWMs are the standard framework for describing DNA-binding specificity in other DNA-binding protein classes and offer a first-order approximation of binding specificity based on additive positional contributions of each base. Moreover, PWMs may be represented in the form of compact sequence logos that provide an intuitive visual summary of binding preferences.\u003c/p\u003e\u003cp\u003eTo derive PWMs for UDG and TDG, we applied linear regression models where each base at each position was modeled as an independent feature (see Methods). These models explained most of the observed binding variation (R\u0026sup2; = 0.78 for UDG and R\u0026sup2; = 0.88 for TDG on the test set; see \u003cb\u003eSupplementary Fig.\u0026nbsp;7a,b\u003c/b\u003e for performance and \u003cb\u003e7c,d\u003c/b\u003e for model coefficients), confirming that PWMs provide a reliable baseline description of repair enzyme specificity.\u003c/p\u003e\u003cp\u003eThe resulting PWM logos (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea) reveal that both enzymes are influenced by distal flanks and that the strongest contributions occur at position\u0026thinsp;+\u0026thinsp;1: G for TDG and A for UDG, consistent with trends observed above (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, b \u003cb\u003eand Supplementary Fig.\u0026nbsp;6\u003c/b\u003e). They further highlight marked differences in relative contributions\u0026mdash;TDG displays an exceptionally sharp preference for G at +\u0026thinsp;1, consistent with its role in recognizing T:G mismatches that predominantly occur in CpG contexts\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e, whereas UDG\u0026rsquo;s preference for A at +\u0026thinsp;1 is less pronounced, reflecting its broader role in removing uracil lesions that arise more uniformly across genomic contexts\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eA limitation of traditional PWMs is that they assume positional independence. Whilst this provides a good first-order approximation, the influence of neighboring bases on each other\u0026rsquo;s geometry and orientation may create cooperative effects on binding specificity that should not be overlooked. Indeed, our circular recognition map (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed) revealed hints of such conditional effects between positions: for example, on the 3\u0026prime; side, base C is favored at position\u0026thinsp;+\u0026thinsp;3 when +\u0026thinsp;2 is A/G/T, but this preference shifts to G when +\u0026thinsp;2 is C. Moreover, similar dependencies were demonstrated to modulate transcription factor binding\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e, motivating us to test whether such context-dependent interactions likewise contribute to repair enzyme specificity.\u003c/p\u003e\u003cp\u003eTo systematically uncover these dependencies, we extended our prediction model by incorporating interaction terms between adjacent positions (see Methods). Incorporating these interactions significantly enhanced model performance (R\u0026sup2; = 0.83 for UDG; R\u0026sup2; = 0.94 for TDG on the test set; see \u003cb\u003eSupplementary Fig.\u0026nbsp;7e\u003c/b\u003e for performance and \u003cb\u003eSupplementary Table\u0026nbsp;1\u003c/b\u003e for coefficient values). The major contributing coefficients are visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb. For instance, the black arrow in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb (right) points to a blue arc that represents a strong positive interaction between C at position\u0026thinsp;+\u0026thinsp;2 and G at position\u0026thinsp;+\u0026thinsp;3, which is reflected by a large positive coefficient in the model.\u003c/p\u003e\u003cp\u003eWe next examined key model-identified interactions by comparing binding preferences across neighboring-base contexts. This analysis revealed context-dependent shifts in preferred bases across multiple interactions (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec as an example and \u003cb\u003eSupplementary Fig.\u0026nbsp;8\u0026ndash;9\u003c/b\u003e for the other interactions visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). For example, while the PWM for TDG suggests a general preference for cytosine at +\u0026thinsp;3 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea right, black arrow), the dependency model predicted\u0026mdash;and direct comparisons confirmed\u0026mdash;that this preference switches to guanine when +\u0026thinsp;2 is cytosine (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec, bottom). More concretely, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec (top panel) shows that C is preferred over G at the +\u0026thinsp;3 position when +\u0026thinsp;2 is A, T, or G, but this preference is reversed when +\u0026thinsp;2 is C, illustrating a context-dependent shift in base preference at +\u0026thinsp;3.\u003c/p\u003e\u003cp\u003eThus, extending beyond PWMs to include dependencies reveals hidden conditional effects, providing a more detailed picture of how repair enzymes interpret sequence context.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eDNA structure influences repair enzyme recognition\u003c/h3\u003e\n\u003cp\u003eOur results reveal that repair enzyme specificity is shaped not only by adjacent bases but also by distal flanks, with dependencies emerging between positions. Such dependencies often arise when proteins read the long-range structural properties of DNA, rather than simply the chemical signature of the individual bases alone. These observations raised the key question of whether DNA shape directs recognition of damage by repair enzymes, in an analogous manner to that demonstrated for several transcription factors\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e. Indeed, our comprehensive BER-Map data now provides the breadth and resolution needed for this systematic dissection of DNA shape contributions to repair enzyme recognition, removing the limitation of previous low-throughput studies which focus on a small number of specific sequence contexts.\u003c/p\u003e\u003cp\u003eWe therefore examined whether computationally calculated DNA shape features may account for the observed binding patterns in our analysis. To this end, we correlated predicted base step parameters (Tilt, Slide, Shift, Roll, Rise and Helical Twist) from flanking sequences (excluding the lesion site, for which shape values are undefined)\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e,\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e, with UDG and TDG binding measurements from our BER-Map libraries, averaged over dinucleotide steps (see Methods). This analysis revealed that multiple structural features across distinct positions contribute to binding (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea) with either strong positive (red) or negative (blue) correlation observed. Strikingly, some shape features exhibited strong standalone correlations with binding, as indicated by the more intense colors in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea\u0026mdash;an unexpected result, as many structural features are normally intertwined. This suggests that some structural features may contribute to repair enzyme recognition more strongly than others.\u003c/p\u003e\u003cp\u003eOne such example is the strong significant relationship of the structural parameter Slide with UDG binding at the +\u0026thinsp;2/+3 base step relative to the damage site (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). Slide describes the lateral displacement of adjacent base pairs, a feature that can reshape groove geometry and promote local DNA deformation.\u003c/p\u003e\u003cp\u003eAnalysis of four UDG\u0026ndash;DNA complexes from the Protein Data Bank (PDB), selected for having matched sequence flanks, using ProDB\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e revealed the consistent presence of marked deformation at this site, with narrowing of the minor groove width (MGW) by up to 2.6 \u0026Aring; compared to the +\u0026thinsp;1 base, reaching levels below those defined by the B-DNA envelope (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec as a representative example for narrowing of MGW, and \u003cb\u003eSupplementary Fig.\u0026nbsp;10\u003c/b\u003e; see Methods). Notably, an arginine residue, frequently implicated in recognizing narrow minor grooves\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, is positioned adjacent to this site (Arg276 in the hUNG-DNA complex, PDB: 1SSP\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e; corresponding to residue 195 in structure coordinates).\u003c/p\u003e\u003cp\u003eThese observations pointed to a role for groove narrowing in damage recognition. Given prior evidence that transcription factors often exploit narrow minor grooves for recognition\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, we investigated whether the UDG repair enzyme shows a similar trend. Indeed, analysis of the narrowest and widest 10% MGW sequences in our library revealed stronger average binding to narrow grooves at this position (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed; see Methods).\u003c/p\u003e\u003cp\u003eTogether, these results show that repair enzyme specificity arises from an interplay of base readout and local DNA structure, and that high-throughput approaches such as BER-Map enable these correlations to be systematically revealed.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eRepair enzyme recognition as a potential force in genome evolution\u003c/h3\u003e\n\u003cp\u003eHaving shown that flanking bases and DNA shape modulate repair enzyme recognition \u003cem\u003ein vitro\u003c/em\u003e, we next asked whether these mechanisms leave a detectable imprint on the human genome.\u003c/p\u003e\u003cp\u003eCytosine deamination and inefficient repair have long been linked to broad compositional biases in the genome, such as CG depletion and AT enrichment\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. A useful way to capture this process is the current genomic C/T ratio: when cytosines are deaminated and not repaired, they are converted to thymines, reducing the relative abundance of cytosine over evolutionary time (as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea). Thus, the C/T ratio provides an indirect readout of the long-term balance between cytosine loss and repair across sequence contexts.\u003c/p\u003e\u003cp\u003eWhile these biases are well established, it remains unclear whether they reflect intrinsic recognition preferences of repair enzymes. We reasoned that if repair efficiency varies with flanking sequence\u0026mdash;as reflected in the binding preferences of enzymes like TDG and UDG\u0026mdash;these preferences might be mirrored in the current sequence composition of the genome, specifically in the pattern of genomic C/T ratios.\u003c/p\u003e\u003cp\u003eTo test this, we analyzed each of the 16 central triplets (NXN, with X denoting the lesion site) separately (see Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb and Methods). For each triplet, we binned extended flanking sequences according to their repair enzyme binding strength measured on the array, then calculated the median genomic C/T ratio in each bin and tested for correlation between binding and C/T ratio. This approach controls for the immediate lesion context and isolates the influence of distal flanking bases, while minimizing noise from confounding features such as chromatin environment or replication timing.\u003c/p\u003e\u003cp\u003eIn the case of TDG, we observed a striking correlation between repair enzyme binding and genomic composition. For example, in the GXC triplet\u0026mdash;defined by fixed G and C at the 5\u0026prime; and 3\u0026prime; positions\u0026mdash;contexts with strong TDG binding showed markedly higher C/T ratios than weakly bound ones (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec), consistent with a model in which stronger TDG recognition reduces the accumulation of C\u0026rarr;T mutations. Extending this analysis across all triplets revealed a similar trend in most cases, with 12 out of 16 triplets showing a positive correlation between TDG binding levels and genomic C/T ratio (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed). The four TDG triplets that did not show this correlation all shared an NXG context, corresponding to CpG sites\u0026mdash;the predominant targets of cytosine methylation. This exception may reflect the unique mutational dynamics of 5-methylcytosine deamination, which yields T:G mismatches, together with the involvement of alternative repair enzymes such as MBD4, which specifically targets CpG deamination products\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e,\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eWe confirmed that the 12 out of 16 significant correlations are non-random by simulating 10,000 random control genomes matched to the human genome in overall base composition. For each simulated genome, we repeated the same correlation analysis across all 16 NXN triplets. While the human genome showed 12 significant correlations, none of the simulated genomes exhibited more than 4 (empirical \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1 \u0026times; 10⁻⁴; see Methods), confirming that the observed correlations are unlikely to have arisen by chance.\u003c/p\u003e\u003cp\u003eIn contrast, UDG binding did not show a comparable relationship, showing a weak, non-significant trend in the opposite direction (\u003cb\u003eSupplementary Fig.\u0026nbsp;11a-b\u003c/b\u003e for both binding and ARP measurements respectively). This lack of a comparable relationship may reflect a genuine biological difference but could also result from the smaller number of flank variants available for UDG in the current array design.\u003c/p\u003e\u003cp\u003eTaken together, these findings are consistent with a model in which context-dependent TDG recognition contributes to fine-scale variation in mutability across the genome.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe present BER-Map, a high-throughput framework that maps BER enzyme binding specificity across thousands of lesion contexts. This resource, the largest of its kind to our knowledge, uncovers how deeply flanking sequence modulates repair recognition and offers a scalable platform for integrative mechanistic and genomic analyses.\u003c/p\u003e\u003cp\u003eBy mapping the recognition rules of TDG and UDG at unprecedented resolution, we show that DNA recognition is profoundly shaped by its sequence environment. Flanking positions, including four or more nucleotides away from the lesion, exert significant influence on binding, indicating that BER enzymes interpret damage within a broad contextual framework rather than relying solely on the lesion base itself.\u003c/p\u003e\u003cp\u003eBeyond individual positions, our combinatorial dataset reveals conditional dependencies between flanking bases, exposing higher-order recognition logic that extends past simple nearest-neighbor effects.\u003c/p\u003e\u003cp\u003eIntegrating sequence data with DNA structural features further demonstrates that enzymes are sensitive to DNA shape as well as base identity. Minor-groove width, base-step geometry, and other local deformations correlate with binding outcomes, suggesting that BER enzymes exploit structural cues to enhance lesion discrimination.\u003c/p\u003e\u003cp\u003eFinally, the context-dependent binding rules uncovered here align with genomic composition, suggesting that the principles guiding repair recognition may also contribute to mutational biases and compositional asymmetries in the human genome. These results raise the possibility that the evolutionary landscape of genome stability is shaped not only by the chemistry of the formation of deaminated products but also by the recognition logic of repair enzymes themselves. The consistency across 12 of 16 triplets, together with the absence of such patterns in simulated genomes, strongly supports a model in which repair enzyme specificity leaves a measurable imprint on genome composition.\u003c/p\u003e\u003cp\u003eWhile a detailed understanding of every step in repair enzyme activity is important, our study focuses primarily on binding rather than catalysis, since binding affinity is a key determinant of damage recognition and sequence context selectivity\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e,\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. Our ARP assays, which monitor the excision step, confirm that for TDG, binding strength correlates closely with cleavage activity, whereas for UDG, the sequence context preferences of binding diverge from those of catalysis. This distinction highlights the value of measuring binding directly, as it isolates the recognition process from downstream enzymatic steps and demonstrates the potential of our approach to probe additional stages that collectively govern mutagenesis. Notably, recent studies have shown that impaired or redirected binding is often what drives mutation accumulation \u003cem\u003ein vivo\u003c/em\u003e\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e. These findings emphasize the need to understand the rules governing damage recognition, and position damage binding as a central control point in genome stability.\u003c/p\u003e\u003cp\u003eTDG and UDG were chosen as representative glycosylases because they play central and complementary roles in counteracting cytosine deamination\u0026mdash;one of the most frequent sources of mutation in the genome\u0026mdash;and thereby help prevent C\u0026rarr;T substitutions, which are among the most common mutations observed\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. Focusing on these two enzymes allowed us to capture distinct recognition strategies within the same pathway. The framework, however, is readily generalizable to other BER enzymes and lesion types. Expanding BER-Map to additional repair pathways could generate comprehensive catalogs of sequence- and structure-dependent recognition rules, offering a unified view of lesion recognition and its impact on genome stability and evolution.\u003c/p\u003e\u003cp\u003eWhile our platform does not currently capture \u003cem\u003ein vivo\u003c/em\u003e complexities such as transcription factor binding, nucleosome interactions, or broader chromatin context\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e,\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e, these factors have been shown to play critical roles in shaping damage recognition and mutagenesis. Notably, most such insights have come from studies on mismatch repair (MMR) and nucleotide excision repair (NER), highlighting the need for parallel investigations in the context of base excision repair (BER). Our platform begins to fill this gap by providing high-resolution, sequence-specific characterization of BER enzyme binding, and lays the groundwork for integrating \u003cem\u003ein vivo\u003c/em\u003e regulatory layers in future studies. In particular, it enables systematic exploration of how transcription factors might interact with damaged DNA to modulate repair accessibility.\u003c/p\u003e\u003cp\u003eBy coupling high-throughput lesion mapping with sequence, structural, and genomic analyses, BER-Map provides an integrative framework for uncovering the rules of repair enzyme recognition. This approach bridges biochemical specificity with genome-scale mutational patterns, establishing the foundation for predictive models that link DNA sequence, structure, repair efficiency, and mutational outcome in human cells. Beyond mechanistic insight, such frameworks may ultimately help interpret mutational signatures in cancer and guide therapeutic strategies targeting repair pathways.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003eProtein expression and purification\u003c/h2\u003e\n \u003cp\u003eHuman TDG (hTDG) was expressed using the pET28-hTDG plasmid (#70758, Addgene), following the protocol described in Schuermann et al. 2016\u003csup\u003e57\u003c/sup\u003e. Freshly transformed \u003cem\u003eE. coli\u003c/em\u003e BL21(DE3) cells were grown in 1.5L LB medium supplemented with kanamycin to an OD₆₀₀ of 0.5, induced with 0.25 mM IPTG, and incubated at 15\u0026deg;C for 48 hours. Cells were harvested by centrifugation, resuspended in lysis buffer (20 mM Tris-HCl pH 7.5, 500 mM NaCl, 5 mM imidazole, 5% glycerol, 5 mM \u0026beta;-mercaptoethanol), and stored at \u0026deg;C. A tablet of protease inhibitor (Roche Diagnostics, catalog no. 11836170001) was added to the thawed cells and then cells were lysed by French press and sonication. Following centrifugation (35k rpm, for 1h), the supernatant was incubated with Ni-NTA resin for 1 hour at 4\u0026deg;C. hTDG was eluted with lysis buffer\u0026thinsp;+\u0026thinsp;200 mM imidazole and dialyzed into 50 mM Tris-HCl pH 7.5, 100mM NaCl, 5% glycerol, 5mM \u0026beta;-mercaptoethanol. Further purification was performed using heparin and Q Sepharose chromatography as done in Schuermann et al. 2016. Final preparations were supplemented with 1 mM DTT and stored at \u0026minus;\u0026thinsp;80\u0026deg;C.\u003c/p\u003e\n \u003cp\u003eFluorescein (Fluor) Human UNG2 is a recombinant human UNG2 protein with an N-terminal fluorescein label that was produced and purified as described previously\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e. Briefly, an N-terminal cysteine was inserted before the first residue of recombinant full-length UNG2 to covalently react with fluorescein-5-maleimide. The N-terminal cysteine was possible because the UNG2 variant was expressed with a 8xHis-SUMO tag upstream of the cysteine, and the tag was removed with the SUMO protease during protein purification. The three endogenous cysteines of UNG2 were also mutated to alanines in UNG2(Fluor) to isolate labeling to the N-terminal cysteine. Previous studies indicated that UNG2(Fluor) had similar activity to wild-type UNG2\u003csup\u003e58\u003c/sup\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eDamaged sequences library design\u003c/h2\u003e\n \u003cp\u003eUracil-containing libraries were synthesized by LC Sciences, using dUTP phosphoramidites. In contrast, T:G mismatch libraries, composed solely of canonical bases, were synthesized by Agilent Technologies using standard DNA synthesis protocols. To address the challenge posed by non-complementary pairing (T:G and U:G), we employed a hairpin-based design. Our oligonucleotides are self-annealing, forming an intramolecular hairpin structure. Each oligo includes a short loop sequence (5\u0026prime;-GGAA-3\u0026prime;) bridging two flanking arms, which hybridize to form a stem. The lesion site (uracil or thymine) is usually embedded within a 6-nucleotide variable region (3 bases on each side), positioned in the stem segment to mimic a double-stranded DNA context. Depending on the design, the library includes 1,024 to 16,384 unique sequences that systematically cover all possible 5- to 7-mer flanking contexts around the lesion (See \u003cstrong\u003eSupplementary data 1\u0026ndash;2\u003c/strong\u003e for all library designs used in this study). For the UDG libraries containing uracils, the design most commonly used in the Results section included fewer variable flanking positions (5\u0026prime;-NNUNNN-3\u0026prime;). As a result, the corresponding analyses and statistical comparisons in those cases were based on a smaller number of unique sequences. Each sequence probe is printed in 7\u0026ndash;10 replicates to allow robust signal averaging, and additional sequences lacking any lesion serve as internal controls. Altogether, the design comprises tens of thousands of sequence probes, enabling high-resolution, quantitative mapping of sequence-context effects on glycosylase binding. A schematic of the single-stranded hairpin oligo design is shown in Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ec.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eBER glycosylase measurements\u003c/h2\u003e\n \u003cp\u003eTo create the double-stranded hairpin structure, the microarray was incubated in annealing buffer (IDT catalog no. 11-05-01-03) at 85\u0026deg;C for 5 min. Then, gradually it was cooled down to room temperature. The binding step was performed as described previously\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Briefly, hybridized microarrays were blocked with 2% (w/v) nonfat milk for 1 h at room temperature. Following the blocking step and washes, the microarray was incubated with the repair enzyme of interest. Uracil libraries were incubated with 400 nM or 800 nM UDG for 0.5 h or 1 h respectively at room temperature in a reaction buffer containing 20 mM Tris-HCl pH 8, 1 mM DTT, 1 mM EDTA. T:G libraries were incubated with 1 \u0026micro;M TDG in a reaction buffer containing 50 mM Tris-HCl pH 8, 1 mM EDTA, 1 mM DTT, 5% Glycerol for 2 h at room temperature. To detect the binding of TDG, the relevant microarray was incubated with either Anti-His Alexa Fluor 647 (Qiagen, catalog no. 35370) or Anti-His Alexa Fluor 488 (Qiagen catalog no. 35310) for 1 h at room temperature. The antibody stock solutions as obtained from the supplier were diluted 1:18 in the respective reaction buffer. Fluorescein-labeled UDG was imaged directly after the binding step. The fluorescent signals were measured using a GenePix 4400A microarray scanner and extracted by GenePix Pro analysis software.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eActivity measurements using Aldehyde Reactive Probe (ARP) reagent\u003c/h2\u003e\n \u003cp\u003eFollowing scanning of protein binding, bound protein was removed from the microarray using the stripping protocol, as described previously\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. To detect the formation of abasic (AP) sites, the microarray was incubated with 2 mg/mL ARP-biotin (GERBU Biotechnik GmbH, catalog No. A305-10) in 2% (w/v) nonfat milk for 45 min at room temperature. The slide was then washed for 3 min in 0.5% Tween-20 in PBS (1\u0026times;), followed by 2 min in 0.01% Triton X-100 in PBS (1\u0026times;). Subsequently, the microarray was incubated with 2 \u0026micro;g/mL streptavidin\u0026ndash;Alexa Fluor 488 (Thermo Fisher, catalog No. S11223) in 2% nonfat milk for 45 min at room temperature. After incubation, the slide was washed twice for 3 min each in 0.05% Tween-20 in PBS (1\u0026times;), and once for 2 min in PBS (1\u0026times;). The arrays were scanned using a GenePix 4400A microarray scanner at an excitation wavelength of 488 nm.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003eMicroarray data preprocessing\u003c/h2\u003e\n \u003cp\u003eBinding data for all probes in each experiment were normalized using MASLINER and spatial-normalization functions based on Berger et al. (2009)\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Each sequence was represented by multiple replicate probes, allowing calculation of a reliable median fluorescence intensity across replicates, yielding one value per unique sequence.\u003c/p\u003e\n \u003cp\u003eTo correct for non-specific binding, we then subtracted the signal of the highest-intensity damage-free probe from all sequences, which defines the upper limit of non-specific interactions for the repair enzyme. This correction yields an intensity value of zero for non-specific binding. Because logarithmic transformation was applied, a constant of +\u0026thinsp;1 was added to all values to avoid undefined logarithms of zero. The resulting log-transformed values were used as normalized binding signals.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical analysis\u003c/h2\u003e\n \u003cp\u003eAfter preprocessing, all analyses were conducted using R software (version 4.2.2). Unless otherwise specified, sequence-level comparisons were performed using two-sided Wilcoxon tests (Mann\u0026ndash;Whitney U test), and replicate-level comparisons using two-sided \u003cem\u003et\u003c/em\u003e-tests. When applicable, multiple comparison correction was applied using the method indicated in the Results section. \u003cem\u003eP\u003c/em\u003e-values are reported either explicitly or using the specified convention in the figure legend or the following one: **** \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001; *** \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001; ** \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.01; * \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; ns, not significant.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n \u003ch2\u003ePositional sensitivity analysis of flanking bases\u003c/h2\u003e\n \u003cp\u003eTo assess how much each flanking position contributes to repair enzyme binding, we measured positional sensitivity by systematically testing whether changing a single base at a given position significantly alters binding. For each base position flanking the damage (\u0026minus;\u0026thinsp;3, \u0026minus;\u0026thinsp;2, \u0026minus;1, +\u0026thinsp;1, +2, +\u0026thinsp;3), we performed the following steps. First, we grouped sequences that were identical at all other flanking positions except the one being tested. Then, for each group, we compared the replicate-level fluorescence values (normalized as described above) between sequences carrying different bases at that position, and performed exhaustive pairwise \u003cem\u003et\u003c/em\u003e-tests between all base pairs (A vs C, A vs G, etc.), resulting in six comparisons per group. For example, when testing sensitivity at position \u0026minus;\u0026thinsp;3, we grouped all sequences that shared the same bases at positions \u0026minus;\u0026thinsp;2 through +\u0026thinsp;3, and compared binding signals between sequences differing only at position \u0026minus;\u0026thinsp;3. Per group of six comparisons, we adjusted \u003cem\u003eP\u003c/em\u003e-values using the Benjamini-Hochberg procedure and recorded significance: if any pairwise comparison at a given position and context was significant (FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05), that comparison was added to the sum of significant comparisons per position. Finally, we calculated the percentage of such significant comparisons at each position. These percentages are reported in Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb as a metric for position-specific statistical variability in binding.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\n \u003ch2\u003eValidation of UDG microarray data using gel-based uracil excision assays\u003c/h2\u003e\n \u003cp\u003eDNA oligonucleotides were purchased from IDT Technologies (see \u003cstrong\u003eSupplementary Fig.\u0026nbsp;3a\u003c/strong\u003e for the sequences). Uracil-containing oligonucleotides also had a 3\u0026prime; fluorescein label (3\u0026prime; 6-FAM modification). The oligonucleotides were purified by denaturing urea-TBE PAGE and then the buffer was exchanged into an assay buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 0.1 mM EDTA). To ensure that the signal that represents activity is coming only from dsDNA substrates, we mixed 8% molar excess of the unlabeled oligo with its complementary labeled oligo. The mixed oligonucleotides were then heated to \u0026gt;\u0026thinsp;95\u0026deg;C and slowly cooled to room temperature\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e59\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n \u003cp\u003eThe uracil excision assay was performed at 22\u0026deg;C for 3.5 min in an assay buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 0.1 mM EDTA). We used 3 nM UDG and 0.5 \u0026micro;M DNA. In all cases, we intended for the enzyme to process less than ~\u0026thinsp;half of the DNA substrate for its most favorable sequences while detecting low turnover for its least favorable sequences. Because of the broad range of uracil excision activities for UDG and its catalytic domain on different DNA substrates\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e59\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e we could not always maintain strict steady-state conditions (less than ~\u0026thinsp;10\u0026ndash;15% of the substrate processed). Enzymatic reactions were quenched with 200 mM NaOH and heat. This step also cleaves the abasic sites produced by uracil excision. The substrate and product from quenched reactions were separated by denaturing urea-TBE PAGE, and the fluorescein end-labels on the oligonucleotides were visualized with in-gel fluorescence using an Azure c400 imager. The intensity of the substrate and product bands were quantified with Fiji/ImageJ\u003csup\u003e62\u003c/sup\u003e. The percent of substrate processed was calculated as the product intensity divided by the sum of the substrate and product intensities. Using this value, we determined the rate of activity for the enzyme by (% of substrate processed\u0026thinsp;\u0026times;\u0026thinsp;\u0026micro;M substrate in the reaction) / reaction time. The rate was divided by the enzyme concentration used in the assay to calculate \u003cem\u003ek\u003c/em\u003e\u003csub\u003eobs\u003c/sub\u003e values. In all cases, we conducted three to four independent assays.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n \u003ch2\u003eData modeling by multiple linear regression\u003c/h2\u003e\n \u003cp\u003eTo map the binding preferences of repair enzymes across different sequence contexts, we used multiple linear regression models with one-hot encoded DNA sequences as features for enzyme binding prediction. Each 6-mer (TDG) or 5-mer (UDG) sequence context (5\u0026prime;-NNNXNNN-3\u0026prime; or 5\u0026prime;-NNXNNN-3\u0026prime;) was one-hot encoded across all positions, yielding a set of 24 or 20 mononucleotide features respectively. The data was split into 80% train and 20% test. We first fitted simple regression models to capture additive effects\u0026mdash;i.e., the independent contributions of each base (fitting included intercept fitting). Model performance and coefficients are shown in \u003cstrong\u003eSupplementary Fig.\u0026nbsp;7a-d\u003c/strong\u003e. We next incorporated dinucleotide interaction features between adjacent positions, also using one-hot encoding for adjacent dinucleotides in every available base-step. Eventually, including the mononucleotide features, for TDG there were 104 binary features, and for UDG 84 binary features. The data were split into training (60%), validation (20%), and test (20%) sets. We first used LassoCV with 5-fold cross-validation on the training set to identify an initial optimal value of the regularization parameter (alpha). Based on this result, we manually explored a range of nearby alphas (e.g., \u0026times;1, \u0026times;1.5, \u0026times;2, \u0026times;5, \u0026times;10) to evaluate the tradeoff between model performance and sparsity, as measured on the validation set by R\u0026sup2; and the number of non-zero coefficients. From this sweep, we selected a \u0026ldquo;sweet spot\u0026rdquo; alpha that retained high predictive performance on the validation set while reducing the number of redundant features (alpha\u0026thinsp;=\u0026thinsp;0.001 for TDG and alpha\u0026thinsp;=\u0026thinsp;0.002 for UDG). A final model was then trained using this alpha and evaluated on the independent test set. The final selected (non-zeroed out) coefficients were used as a measure of feature importance, allowing us to identify the most informative mono and di sequence features for enzyme binding (as seen in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb). Model performances are shown in \u003cstrong\u003eSupplementary Fig.\u0026nbsp;7e\u003c/strong\u003e and coefficients for each interaction (non-additive) model are reported in \u003cstrong\u003eSupplementary Table\u0026nbsp;1\u003c/strong\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\n \u003ch2\u003eBase step DNA shape features and MGW analysis\u003c/h2\u003e\n \u003cp\u003eTo investigate whether local DNA structural features contribute to differential repair enzyme binding, we systematically analyzed predicted DNA shape parameters of the sequence flanks in our microarray library. We focused on dinucleotide steps within 3-mer flanks (NNN) located upstream and downstream of the central damage (i.e., positions \u0026minus;\u0026thinsp;3 to \u0026minus;\u0026thinsp;1 and +\u0026thinsp;1 to +\u0026thinsp;3 relative to the damaged base at position 0), due to the lack of structural information on damaged DNA. Structural values were obtained from the DeepShape webserver\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e48\u003c/span\u003e,\u003cspan class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e, which can provide predictions for each possible 3-mer sequence across multiple shape features (HelT, Roll, Rise, Slide, Shift, and Tilt) based on high-throughput simulations. Each probe in our dataset (5\u0026prime;-NNNXNNN-3\u0026prime;) contains a unique 3-mer sequence on either side of the damage site; we assigned the corresponding structural values to each flank and aligned them by position relative to the lesion. Because multiple probes may share the same 3-mer flank (e.g., both AAAXAAA and CAAXAAA share the 3\u0026prime; flank AAA), the same structural values were assigned redundantly across all probes in which that 3-mer appeared, consistent with their representation in the experimental design. Subsequently, for each base-step position (e.g., base 1\u0026rarr;2, 2\u0026rarr;3), we extracted the dinucleotide identity (e.g., AA, AC, etc.) and computed the average structural value for that dinucleotide across all contexts. In parallel, we computed the average normalized binding signal for the same dinucleotide at the same position, based on all probes containing it. This resulted in a position-wise matched dataset of mean structural values and mean binding intensities per dinucleotide, which we used to assess correlations between structure and binding preferences. For cases where library design included only 2-mer flanks (e.g. NNUNNN), we extracted structural data the same way only per dinucleotides and continued similarly.\u003c/p\u003e\n \u003cp\u003eTo assess the influence of minor groove width (MGW) on UDG binding, we used an additional microarray library design and a different database. MGW is typically more reliable when computed over longer DNA contexts, such as pentamers, due to its dependence on surrounding base stacking and groove geometry. Therefore, we utilized an asymmetric design where the uracil was fixed at the 5\u0026prime; end, allowing systematic variation of the 3\u0026prime;-flanking region (5\u0026prime;-UNNNNN-3\u0026prime;). MGW values were downloaded from DNAShape\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e and assigned to the central base of each 5-mer flank (i.e., the third base, +\u0026thinsp;3, in the UNN\u003cspan type=\"BoldUnderline\" class=\"BoldUnderline\" name=\"Emphasis\"\u003eN\u003c/span\u003eNN sequence), enabling further analysis with our binding measurements per corresponding flank.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\n \u003ch2\u003eStructural analysis of UDG\u0026ndash;DNA complex crystal structures\u003c/h2\u003e\n \u003cp\u003eTo assess local DNA distortions around the lesion site, we analyzed four available crystal structures of uracil-bound UDG\u0026ndash;DNA complexes: 1EMH, 1EMJ\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e, 1SSP, and 2SSP\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. These structures were selected based on their shared sequence flanks surrounding the lesion site, allowing for a controlled comparison of structural features. DNA structural parameters, MGW and Slide, were extracted using the ProDB pipeline\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e, which integrates output from CURVES+. To visualize the distortions, DNA sequences from the structures were aligned according to the uracil position.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\n \u003ch2\u003eGenomic analysis (C/T ratio analysis)\u003c/h2\u003e\n \u003cp\u003eTo calculate the C/T ratio per context in the genome, we used the \u003cem\u003eHomo sapiens\u003c/em\u003e (human) genome assembly GRCh38.p14 (hg38) from UCSC, as our reference genome. For that, we extracted reference genome sequences using the BSgenome.Hsapiens.UCSC.hg38 package from Bioconductor\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e. We scanned the genome (excluding sex chromosomes) for all 6-mer sequences with either a cytosine (C) or thymine (T) at the third position (i.e., NNCNNNN and NNTNNNN, respectively). For each unique 6-mer, we counted the number of occurrences on the forward strand of the genome. For the C/T ratio for each unique 6-mer, the number of occurrences in the reference genome was then divided as follows: NNCNNNN counts / NNTNNNN counts\u0026thinsp;=\u0026thinsp;C/T ratio per context. Since it is known that adjacent nucleotides greatly affect mutation accumulation and generally the evolution of a genome\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e, we decided to group our ratio results into triplets. We consider a \u0026lsquo;Triplet\u0026rsquo; as the group of 6-mers sharing the same immediate 3\u0026prime; and 5\u0026prime; adjacent nucleotides (e.g., CXA, GXC). Afterward, for each triplet, we binned the data by the repair enzyme binding measurement. It is important to note that the number of samples per bin may change between libraries and affect the observed significance. Specifically, the unique 6-mers were ordered by binding signal levels and divided into groups of 16 bins, each containing 16 observations. Then, for each bin, the median was calculated for the C/T ratio, and the average was calculated for the normalized binding levels. Pearson correlation tests were done for each triplet on the binned data and corrected for multiple comparisons by the Benjamini-Hochberg procedure. The analysis shown corresponds to data averaged over the \u0026minus;\u0026thinsp;3 base position from the asymmetric library design (5\u0026prime;-NNNTNNNN-3\u0026prime;), which yielded the strongest and most consistent correlations with genomic C/T ratios.\u003c/p\u003e\n \u003cp\u003eTo test whether the observed C/T correlations with TDG binding levels could arise by chance, we generated a null distribution using simulated genomes. Random genomes of 10 Mb in length were generated by sampling nucleotides according to their frequencies in the human hg38 reference genome (A\u0026thinsp;=\u0026thinsp;29.5%, C\u0026thinsp;=\u0026thinsp;20.4%, G\u0026thinsp;=\u0026thinsp;20.5%, T\u0026thinsp;=\u0026thinsp;29.6%), thereby matching its overall composition. For each of 10,000 simulations, the C/T ratio was calculated per sequence context and correlated with the measured TDG binding levels across triplets following the same exact procedure described above.\u003c/p\u003e\n \u003cp\u003eWe then compared the number of significant triplet-level correlations in the real genome (12) to this null distribution. The empirical \u003cem\u003eP\u003c/em\u003e-value was calculated as the proportion of simulations with as many or more significant triplets than observed.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e\u003cu\u003eData availability\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll raw and processed microarray data are provided as \u003cstrong\u003eSupplementary Data 1\u0026ndash;2\u003c/strong\u003e for TDG and UDG, respectively (Excel .xlsx format). Each sheet within these files corresponds to an individual library design and measurement, as indicated in its title and in the notes sheet.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRelevant source code for data modeling supporting the findings of this study are available at:\u0026nbsp;\u003cstrong\u003ehttps://github.com/afeklab/LR-models-for-repair-enzyme-specificity-predictions\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cu\u003eAcknowledgements\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Yuval Bayer and Ruben Solozabal for insightful discussions on specificity modeling. We also thank Michael O\u0026rsquo;Hagan for carefully reviewing the manuscript, and Deborah Fass for valuable comments and suggestions.\u003c/p\u003e\n\u003cp\u003eThis research was supported by Israel Science Foundation (ISF) Grant \u0026nbsp; No. 1174/22 and by a research grant from the Shimon and Golde Picker \u0026ndash; Weizmann Annual Grant.\u003c/p\u003e\n\u003cp\u003eB.P.W. and S.N.G. were supported by an award to B.P.W. from the National Institutes of Health (R01GM135152).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cu\u003eAuthor information\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAuthors and Affiliations\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDepartment of Chemical and Structural Biology, Weizmann Institute of Science, Rehovot, Israel\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNoga Levy, Vered Levin Salomon, Naama Kessler \u0026amp; Ariel Afek\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDepartment of Molecular genetics, Weizmann Institute of Science, Rehovot, Israel\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOmer Erez\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDepartment of Cell \u0026amp; Molecular Biology, Rowan-Virtua School of Osteopathic Medicine and Rowan-Virtua School of Translational Biomedical Engineering and Sciences, Rowan University, Stratford, New Jersey, United States of America\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSharon N. Greenwood \u0026amp; Brian P. Weiser\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cu\u003eContributions\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eN.L. and A.A. designed the study. A.A. and V.L.S. supervised the research. N.L. developed the assay, generated high throughput enzyme-DNA binding data, and analyzed the data. S.N.G and B.P.W purified the human fluorescent UDG, performed the gel-based activity experiment and analyzed its data. O.E. helped with the genomic analysis. N.K. purified human TDG. N.L., V.L.S. and A.A wrote the manuscript, with inputs from all authors. All the authors critically reviewed the manuscript and approved the final version.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cu\u003eCompeting interests\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cu\u003eMaterials \u0026amp; Correspondence\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrespondence to Ariel Afek.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eLindahl, T. Instability and decay of the primary structure of DNA. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e362\u003c/strong\u003e, 709\u0026ndash;715 (1993).\u003c/li\u003e\n \u003cli\u003eKrokan, H. E. \u0026amp; Bj\u0026oslash;r\u0026aring;s, M. Base Excision Repair. \u003cem\u003eCold Spring Harb Perspect Biol\u003c/em\u003e\u003cstrong\u003e5\u003c/strong\u003e, a012583 (2013).\u003c/li\u003e\n \u003cli\u003eZharkov, D. O. Base excision DNA repair. \u003cem\u003eCellular and Molecular Life Sciences\u003c/em\u003e\u003cstrong\u003e65\u003c/strong\u003e, 1544\u0026ndash;1565 (2008).\u003c/li\u003e\n \u003cli\u003eAitken, S. J. \u003cem\u003eet al.\u003c/em\u003e Pervasive lesion segregation shapes cancer genome evolution. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e583\u003c/strong\u003e, 265\u0026ndash;270 (2020).\u003c/li\u003e\n \u003cli\u003eListon, A., Humblet-Baron, S., Duffy, D. \u0026amp; Goris, A. Human immune diversity: from evolution to modernity. \u003cem\u003eNat Immunol\u003c/em\u003e\u003cstrong\u003e22\u003c/strong\u003e, 1479\u0026ndash;1489 (2021).\u003c/li\u003e\n \u003cli\u003eQuintana-Murci, L. Human Immunology through the Lens of Evolutionary Genetics. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e177\u003c/strong\u003e, 184\u0026ndash;199 (2019).\u003c/li\u003e\n \u003cli\u003eWallace, S. S., Murphy, D. L. \u0026amp; Sweasy, J. B. Base excision repair and cancer. \u003cem\u003eCancer Lett\u003c/em\u003e\u003cstrong\u003e327\u003c/strong\u003e, 73\u0026ndash;89 (2012).\u003c/li\u003e\n \u003cli\u003eL\u0026oacute;pez-Ot\u0026iacute;n, C., Blasco, M. A., Partridge, L., Serrano, M. \u0026amp; Kroemer, G. The Hallmarks of Aging. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e153\u003c/strong\u003e, 1194\u0026ndash;1217 (2013).\u003c/li\u003e\n \u003cli\u003eAlexandrov, L. B. \u003cem\u003eet al.\u003c/em\u003e Signatures of mutational processes in human cancer. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e500\u003c/strong\u003e, 415\u0026ndash;421 (2013).\u003c/li\u003e\n \u003cli\u003eGonzalez-Perez, A., Sabarinathan, R. \u0026amp; Lopez-Bigas, N. Local Determinants of the Mutational Landscape of the Human Genome. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e177\u003c/strong\u003e, 101\u0026ndash;114 (2019).\u003c/li\u003e\n \u003cli\u003eAnderson, C. J. \u003cem\u003eet al.\u003c/em\u003e Strand-resolved mutagenicity of DNA damage and repair. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e630\u003c/strong\u003e, 744\u0026ndash;751 (2024).\u003c/li\u003e\n \u003cli\u003eRobinson, P. S. \u003cem\u003eet al.\u003c/em\u003e Inherited MUTYH mutations cause elevated somatic mutation rates and distinctive mutational signatures in normal human cells. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, 1\u0026ndash;12 (2022).\u003c/li\u003e\n \u003cli\u003eParikh, S. S., Putnam, C. D. \u0026amp; Tainer, J. A. Lessons learned from structural results on uracil-DNA glycosylase. \u003cem\u003eMutation Research/DNA Repair\u003c/em\u003e\u003cstrong\u003e460\u003c/strong\u003e, 183\u0026ndash;199 (2000).\u003c/li\u003e\n \u003cli\u003eStivers, J. T. Extrahelical Damaged Base Recognition by DNA Glycosylase Enzymes. \u003cem\u003eChemistry \u0026ndash; A European Journal\u003c/em\u003e\u003cstrong\u003e14\u003c/strong\u003e, 786\u0026ndash;793 (2008).\u003c/li\u003e\n \u003cli\u003eDow, B. J., Malik, S. S. \u0026amp; Drohat, A. C. Defining the Role of Nucleotide Flipping in Enzyme Specificity Using 19F NMR. \u003cem\u003eJ Am Chem Soc\u003c/em\u003e\u003cstrong\u003e141\u003c/strong\u003e, 4952\u0026ndash;4962 (2019).\u003c/li\u003e\n \u003cli\u003eSlupphaug, G. \u003cem\u003eet al.\u003c/em\u003e A nucleotide-flipping mechanism from the structure of human uracil-DNA glycosylase bound to DNA. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e384\u003c/strong\u003e, 87\u0026ndash;92 (1996).\u003c/li\u003e\n \u003cli\u003eBuechner, C. N., Maiti, A., Drohat, A. C. \u0026amp; Tessmer, I. Lesion search and recognition by thymine DNA glycosylase revealed by single molecule imaging. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e43\u003c/strong\u003e, 2716\u0026ndash;2729 (2015).\u003c/li\u003e\n \u003cli\u003eZharkov, D. O., Mechetin, G. V. \u0026amp; Nevinsky, G. A. Uracil-DNA glycosylase: Structural, thermodynamic and kinetic aspects of lesion search and recognition. \u003cem\u003eMutation Research/Fundamental and Molecular Mechanisms of Mutagenesis\u003c/em\u003e\u003cstrong\u003e685\u003c/strong\u003e, 11\u0026ndash;20 (2010).\u003c/li\u003e\n \u003cli\u003eParikh, S. S. \u003cem\u003eet al.\u003c/em\u003e Base excision repair initiation revealed by crystal structures and binding kinetics of human uracil‐DNA glycosylase with DNA. \u003cem\u003eEMBO J\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 5214\u0026ndash;5226 (1998).\u003c/li\u003e\n \u003cli\u003eDonigan, K. A. \u0026amp; Sweasy, J. B. Sequence context-specific mutagenesis and base excision repair. \u003cem\u003eMol Carcinog\u003c/em\u003e\u003cstrong\u003e48\u003c/strong\u003e, 362\u0026ndash;368 (2009).\u003c/li\u003e\n \u003cli\u003eNilsen, H., Yazdankhah, S. P., Eftedal, I. \u0026amp; Krokan, H. E. Sequence specificity for removal of uracil from U\u0026middot;A pairs and U\u0026middot;G mismatches by uracil-DNA glycosylase from Escherichia coli, and correlation with mutational hotspots. \u003cem\u003eFEBS Lett\u003c/em\u003e\u003cstrong\u003e362\u003c/strong\u003e, 205\u0026ndash;209 (1995).\u003c/li\u003e\n \u003cli\u003eEftedal, I., Guddal, P. H., Slupphaug, G., Volden, G. \u0026amp; Krokan, H. E. Consensus sequences for good and poor removal of uracil from double stranded DNA by uracil-DNA glycosylase. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 2095\u0026ndash;2101 (1993).\u003c/li\u003e\n \u003cli\u003eOrndorff, P. B. \u003cem\u003eet al.\u003c/em\u003e Uracil-DNA glycosylase efficiency is modulated by substrate rigidity. \u003cem\u003eSci Rep\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, 1\u0026ndash;14 (2023).\u003c/li\u003e\n \u003cli\u003eSch\u0026auml;rer, O. D., Kawate, T., Gallinari, P., Jiricny, J. \u0026amp; Verdine, G. L. Investigation of the mechanisms of DNA binding of the human G/T glycosylase using designed\u0026thinsp;inhibitors. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e\u003cstrong\u003e94\u003c/strong\u003e, 4878\u0026ndash;4883 (1997).\u003c/li\u003e\n \u003cli\u003ePoetsch, A. R. The genomics of oxidative DNA damage, repair, and resulting mutagenesis. \u003cem\u003eComput Struct Biotechnol J\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, 207\u0026ndash;219 (2020).\u003c/li\u003e\n \u003cli\u003eRogozin, I. B. \u003cem\u003eet al.\u003c/em\u003e Mutational signatures and mutable motifs in cancer genomes. \u003cem\u003eBrief Bioinform\u003c/em\u003e\u003cstrong\u003e19\u003c/strong\u003e, 1085\u0026ndash;1101 (2018).\u003c/li\u003e\n \u003cli\u003eAlexandrov, L. B. \u003cem\u003eet al.\u003c/em\u003e The repertoire of mutational signatures in human cancer. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e578\u003c/strong\u003e, 94\u0026ndash;101 (2020).\u003c/li\u003e\n \u003cli\u003eKaplanis, J. \u003cem\u003eet al.\u003c/em\u003e Genetic and chemotherapeutic influences on germline hypermutation. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e605\u003c/strong\u003e, 503\u0026ndash;508 (2022).\u003c/li\u003e\n \u003cli\u003eJolma, A. \u003cem\u003eet al.\u003c/em\u003e DNA-binding specificities of human transcription factors. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e152\u003c/strong\u003e, 327\u0026ndash;339 (2013).\u003c/li\u003e\n \u003cli\u003eRohs, R. \u003cem\u003eet al.\u003c/em\u003e The role of DNA shape in protein-DNA recognition. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e461\u003c/strong\u003e, 1248\u0026ndash;1253 (2009).\u003c/li\u003e\n \u003cli\u003eRohs, R. \u003cem\u003eet al.\u003c/em\u003e Origins of specificity in protein-DNA recognition. \u003cem\u003eAnnu Rev Biochem\u003c/em\u003e\u003cstrong\u003e79\u003c/strong\u003e, 233\u0026ndash;269 (2010).\u003c/li\u003e\n \u003cli\u003eStormo, G. D. \u0026amp; Zhao, Y. Determining the specificity of protein-DNA interactions. \u003cem\u003eNat Rev Genet\u003c/em\u003e\u003cstrong\u003e11\u003c/strong\u003e, 751\u0026ndash;760 (2010).\u003c/li\u003e\n \u003cli\u003eSchorderet, D. F. \u0026amp; Gartler, S. M. Analysis of CpG suppression in methylated and nonmethylated species. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e\u003cstrong\u003e89\u003c/strong\u003e, 957\u0026ndash;961 (1992).\u003c/li\u003e\n \u003cli\u003eBird, A. P. DNA methylation and the frequency of CpG in animal DNA. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 1499\u0026ndash;1504 (1980).\u003c/li\u003e\n \u003cli\u003eHess, S. T., Blake, J. D. \u0026amp; Blake, R. D. Wide variations in neighbor-dependent substitution rates. \u003cem\u003eJ Mol Biol\u003c/em\u003e\u003cstrong\u003e236\u003c/strong\u003e, 1022\u0026ndash;1033 (1994).\u003c/li\u003e\n \u003cli\u003eAggarwala, V. \u0026amp; Voight, B. F. An expanded sequence context model broadly explains variability in polymorphism levels across the human genome. \u003cem\u003eNat Genet\u003c/em\u003e\u003cstrong\u003e48\u003c/strong\u003e, 349\u0026ndash;355 (2016).\u003c/li\u003e\n \u003cli\u003eOman, M., Alam, A. \u0026amp; Ness, R. W. How Sequence Context-Dependent Mutability Drives Mutation Rate Variation in the Genome. \u003cem\u003eGenome Biol Evol\u003c/em\u003e\u003cstrong\u003e14\u003c/strong\u003e, (2022).\u003c/li\u003e\n \u003cli\u003eYouk, J., An, Y., Park, S., Lee, J. K. \u0026amp; Ju, Y. S. The genome-wide landscape of C:G \u0026gt; T:A polymorphism at the CpG contexts in the human population. \u003cem\u003eBMC Genomics\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 1\u0026ndash;11 (2020).\u003c/li\u003e\n \u003cli\u003eFryxell, K. J. \u0026amp; Zuckerkandl, E. Cytosine Deamination Plays a Primary Role in the Evolution of Mammalian Isochores. \u003cem\u003eMol Biol Evol\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 1371\u0026ndash;1383 (2000).\u003c/li\u003e\n \u003cli\u003eBellacosa, A. \u0026amp; Drohat, A. C. Role of base excision repair in maintaining the genetic and epigenetic integrity of CpG sites. \u003cem\u003eDNA Repair (Amst)\u003c/em\u003e\u003cstrong\u003e32\u003c/strong\u003e, 33\u0026ndash;42 (2015).\u003c/li\u003e\n \u003cli\u003eBerger, M. F. \u0026amp; Bulyk, M. L. Universal protein-binding microarrays for the comprehensive characterization of the DNA-binding specificities of transcription factors. \u003cem\u003eNat Protoc\u003c/em\u003e\u003cstrong\u003e4\u003c/strong\u003e, 393\u0026ndash;411 (2009).\u003c/li\u003e\n \u003cli\u003eAfek, A. \u003cem\u003eet al.\u003c/em\u003e DNA mismatches reveal conformational penalties in protein\u0026ndash;DNA recognition. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e587\u003c/strong\u003e, 291-296 (2020).\u003c/li\u003e\n \u003cli\u003eIde, H. \u003cem\u003eet al.\u003c/em\u003e Synthesis and Damage Specificity of a Novel Probe for the Detection of Abasic Sites in DNA. \u003cem\u003eBiochemistry\u003c/em\u003e\u003cstrong\u003e32\u003c/strong\u003e, 8276\u0026ndash;8283 (1993).\u003c/li\u003e\n \u003cli\u003eCoey, C. T. \u003cem\u003eet al.\u003c/em\u003e Structural basis of damage recognition by thymine DNA glycosylase: Key roles for N-terminal residues. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e44\u003c/strong\u003e, 10248\u0026ndash;10258 (2016).\u003c/li\u003e\n \u003cli\u003eKrokan, H. E., Drabl\u0026oslash;s, F. \u0026amp; Slupphaug, G. Uracil in DNA - Occurrence, consequences and repair. \u003cem\u003eOncogene\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 8935\u0026ndash;8948 (2002).\u003c/li\u003e\n \u003cli\u003eGord\u0026acirc;n, R. \u003cem\u003eet al.\u003c/em\u003e Genomic Regions Flanking E-Box Binding Sites Influence DNA Binding Specificity of bHLH Transcription Factors through DNA Shape. \u003cem\u003eCell Rep\u003c/em\u003e\u003cstrong\u003e3\u003c/strong\u003e, 1093\u0026ndash;1104 (2013).\u003c/li\u003e\n \u003cli\u003eSamee, M. A. H., Bruneau, B. G. \u0026amp; Pollard, K. S. A De Novo Shape Motif Discovery Algorithm Reveals Preferences of Transcription Factors for DNA Shape Beyond Sequence Motifs. \u003cem\u003eCell Syst\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 27-42.e6 (2019).\u003c/li\u003e\n \u003cli\u003eLi, J., Chiu, T. P. \u0026amp; Rohs, R. Predicting DNA structure using a deep learning method. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 1\u0026ndash;12 (2024).\u003c/li\u003e\n \u003cli\u003eLi, J. \u0026amp; Rohs, R. Deep DNAshape webserver: prediction and real-time visualization of DNA shape considering extended k-mers. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e52\u003c/strong\u003e, W7\u0026ndash;W12 (2024).\u003c/li\u003e\n \u003cli\u003eMitra, R., Cohen, A. S., Sagendorf, J. M., Berman, H. M. \u0026amp; Rohs, R. DNAproDB: an updated database for the automated and interactive analysis of protein\u0026ndash;DNA complexes. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e53\u003c/strong\u003e, D396\u0026ndash;D402 (2025).\u003c/li\u003e\n \u003cli\u003eSilveira, A. B. \u003cem\u003eet al.\u003c/em\u003e Base-excision repair pathway shapes 5-methylcytosine deamination signatures in pan-cancer genomes. \u003cem\u003eNature Communications\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 1\u0026ndash;17 (2024).\u003c/li\u003e\n \u003cli\u003eHardeland, U., Bentele, M., Jiricny, J. \u0026amp; Schar, P. Separating substrate recognition from base hydrolysis in human thymine DNA glycosylase by mutational analysis. \u003cem\u003eJournal of Biological Chemistry\u003c/em\u003e\u003cstrong\u003e275\u003c/strong\u003e, 33449\u0026ndash;33456 (2000).\u003c/li\u003e\n \u003cli\u003eBellamy, S. R. W. \u0026amp; Baldwin, G. S. A kinetic analysis of substrate recognition by \u0026nbsp;uracil-DNA glycosylase from herpes simplex virus type 1. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e29\u003c/strong\u003e, 3857\u0026ndash;3863 (2001).\u003c/li\u003e\n \u003cli\u003eZhu, W. \u003cem\u003eet al.\u003c/em\u003e DNA mutagenesis driven by transcription factor competition with mismatch repair. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e188\u003c/strong\u003e, 5735-5747.e15 (2025).\u003c/li\u003e\n \u003cli\u003eSchnable, B. L. \u003cem\u003eet al.\u003c/em\u003e Thymine DNA glycosylase combines sliding, hopping, and nucleosome interactions to efficiently search for 5-formylcytosine. \u003cem\u003eNature Communications\u0026nbsp;\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 1\u0026ndash;11 (2024).\u003c/li\u003e\n \u003cli\u003eSabarinathan, R., Mularoni, L., Deu-Pons, J., Gonzalez-Perez, A. \u0026amp; Lopez-Bigas, N. Nucleotide excision repair is impaired by binding of transcription factors to DNA. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e532\u003c/strong\u003e, 264\u0026ndash;267 (2016).\u003c/li\u003e\n \u003cli\u003eSchuermann, D. \u003cem\u003eet al.\u003c/em\u003e 3CAPS \u0026ndash; a structural AP\u0026ndash;site analogue as a tool to investigate DNA base excision repair. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e44\u003c/strong\u003e, 2187\u0026ndash;2198 (2016).\u003c/li\u003e\n \u003cli\u003eWeiser, B. P., Stivers, J. T. \u0026amp; Cole, P. A. Investigation of N-Terminal Phospho-Regulation of Uracil DNA Glycosylase Using Protein Semisynthesis. \u003cem\u003eBiophys J\u003c/em\u003e\u003cstrong\u003e113\u003c/strong\u003e, 393\u0026ndash;401 (2017).\u003c/li\u003e\n \u003cli\u003eWeiser, B. P., Rodriguez, G., Cole, P. A. \u0026amp; Stivers, J. T. N-terminal domain of human uracil DNA glycosylase (hUNG2) promotes targeting to uracil sites adjacent to ssDNA\u0026ndash;dsDNA junctions. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e46\u003c/strong\u003e, 7169\u0026ndash;7178 (2018).\u003c/li\u003e\n \u003cli\u003eWeiser, B. P. Analysis of uracil DNA glycosylase (UNG2) stimulation by replication protein A (RPA) at ssDNA-dsDNA junctions. \u003cem\u003eBiochimica et Biophysica Acta (BBA) - Proteins and Proteomics\u003c/em\u003e\u003cstrong\u003e1868\u003c/strong\u003e, 140347 (2020).\u003c/li\u003e\n \u003cli\u003eGreenwood, S. N., Kulkarni, R. S., Mikhail, M. \u0026amp; Weiser, B. P. Replication Protein A Enhances Kinetics of Uracil DNA Glycosylase on ssDNA and Across DNA Junctions: Explored with a DNA Repair Complex Produced with SpyCatcher/SpyTag Ligation. \u003cem\u003eChemBioChem\u003c/em\u003e\u003cstrong\u003e24\u003c/strong\u003e, e202200765 (2023).\u003c/li\u003e\n \u003cli\u003eSchindelin, J. \u003cem\u003eet al.\u003c/em\u003e Fiji: An open-source platform for biological-image analysis. \u003cem\u003eNat Methods\u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e, 676\u0026ndash;682 (2012).\u003c/li\u003e\n \u003cli\u003eZhou, T. \u003cem\u003eet al.\u003c/em\u003e DNAshape: a method for the high-throughput prediction of DNA structural features on a genomic scale. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e41\u003c/strong\u003e, W56\u0026ndash;W62 (2013).\u003c/li\u003e\n \u003cli\u003eParikh, S. S. \u003cem\u003eet al.\u003c/em\u003e Uracil-DNA glycosylase\u0026ndash;DNA substrate and product structures: Conformational strain promotes catalytic efficiency by coupled stereoelectronic effects. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e\u003cstrong\u003e97\u003c/strong\u003e, 5083\u0026ndash;5088 (2000).\u003c/li\u003e\n \u003cli\u003eTeam TBD. \u003cem\u003eBSgenome.Hsapiens.UCSC.hg38: Full genomic sequences for\u003c/em\u003e\u003cem\u003eHomo sapiens\u003c/em\u003e\u003cem\u003e(UCSC genome hg38)\u003c/em\u003e. R package version 1.4.5 (2023).\u003c/li\u003e\n \u003cli\u003eHodgkinson, A. \u0026amp; Eyre-Walker, A. Variation in the mutation rate across mammalian genomes. \u003cem\u003eNat Rev Genet\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 756\u0026ndash;766 (2011).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7972250/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7972250/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMutations compromise genome stability, promote disease, and yet drive genetic diversity and evolution. DNA repair of mutagenic lesions acts to maintain genome integrity but is inherently imperfect, allowing mutations to emerge, persist and accumulate unevenly across the genome.\u003c/p\u003e\u003cp\u003eUnderstanding when and where such mutations arise requires a deep understanding of the molecular factors that govern repair enzyme recognition. In base excision repair (BER), glycosylases must locate rare damaged bases that appear within diverse sequence and structural contexts across the genome, yet how these contexts modulate recognition and influence mutational outcomes remains unresolved.\u003c/p\u003e\u003cp\u003eHere we introduce a high-throughput approach that quantifies BER-glycosylase binding across thousands of lesion-containing sequence contexts. Focusing on the cytosine-deamination pathway, we mapped the recognition landscapes of the human enzymes UDG and TDG, which act on the modified base uracil and on T:G mismatches derived from cytosine and 5-methylcytosine deamination, respectively. Our results reveal widespread sequence- and structure-dependent influences on binding, extending several bases from the lesion site and including non-additive interactions between flanking positions. Structural analyses implicate DNA-shape features such as base-step Slide and minor-groove width as determinants of recognition. Notably, sequence preferences influenced by flanks beyond the immediate neighbors reflect the cytosine\u0026ndash;thymine (C/T) balance in the human genome, revealing a striking connection between long-range binding specificity and localized mutation patterns.\u003c/p\u003e\u003cp\u003eTogether, these findings and framework establish a generalizable approach for dissecting how DNA repair enzymes recognize lesions in sequence- and structure-specific contexts, providing a foundation for predictive models of how repair fidelity shapes genome evolution and human disease.\u003c/p\u003e","manuscriptTitle":"Mapping the sequence logic of DNA repair enzyme binding reveals mechanistic principles and evolutionary links","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-10 19:54:33","doi":"10.21203/rs.3.rs-7972250/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"acb5b756-3dd7-46cd-a6f9-7516e88abc6e","owner":[],"postedDate":"November 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":57580947,"name":"Biological sciences/Biochemistry/DNA"},{"id":57580948,"name":"Biological sciences/Biochemistry/Enzymes/DNA repair enzymes"},{"id":57580949,"name":"Biological sciences/Molecular biology/DNA damage and repair/Base excision repair"},{"id":57580950,"name":"Biological sciences/Biochemistry/Proteins/DNA-binding proteins"}],"tags":[],"updatedAt":"2026-05-04T17:35:48+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-10 19:54:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7972250","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7972250","identity":"rs-7972250","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}