Detecting Protein-DNA binding in single molecules using antibody guided methylation

Abstract

Characterization of DNA binding sites for specific proteins is of fundamental importance in molecular biology. It is commonly addressed experimentally by chromatin immunoprecipitation and sequencing (ChIP-seq) of bulk samples (10³-10⁷ cells). We have developed an alternative method that uses a Chromatin Antibody-mediated Methylating Protein (ChAMP) composed of a GpC methyltransferase fused to protein G. By tethering ChAMP to a primary antibody directed against the DNA-binding protein of interest, and selectively switching on its enzymatic activity in situ, we generated distinct and identifiable methylation patterns adjacent to the protein binding sites. This method is compatible with methods of single-cell methylation-detection and single molecule methylation identification. Indeed, as every binding event generates multiple nearby methylations, we were able to confidently detect protein binding in long single molecules.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13072-025-00602-9.

Introduction

Characterization of protein-DNA binding sites is a fundamental aim in molecular biology. It is commonly addressed by chromatin immunoprecipitation (IP) and sequencing assay (ChIP-seq) in which the protein is immunoprecipitated, native or cross-linked to DNA, and then the bound DNA is sequenced [1]. Challenges associated with this method include the need for an available high quality antibody to pull down the DNA-protein complex, and the loss of primary substance throughout the (IP) procedure. Hence this procedure requires large sample size (typically 10³-10⁷ cells) and limits the ability to analyze small samples including single cells, rare cell types and states, and limited clinical samples.

Complementary and alternative methods to ChIP-seq include DNA adenine methyltransferase identification (DamID) [2]Targeted Gene Methylation (TAGM) [3] and Chromatin Immuno/Endogenous Cleavage (ChIC/ChEC/Cut&Run) [4, 5]. DamID works by expressing an adenine methyltransferase (MTase) fused to a protein of interest, transgenically expressed in the target cells. When the fused protein binds the DNA, an adenosine methylation (which does not occur naturally in eukaryotes) will label DNA sites proximal to the binding site of the target protein. With no currently available chemistry to directly distinguish methylated adenines through clonal-amplification based sequencing methods, it typically requires specific digestion or immunoprecipitation of the methylated DNA. TAGM works in a similar way using cytosine DNA methyltransferases, overcoming the chemistry limitation. ChIC and ChEC both rely on tethering a nuclease, either by antibodies or by fusion, to a protein of interest. This results in a specific cleavage pattern, allowing for the isolation and detection of double stranded fragments from the vicinity of the binding site. Cut&Run works similarly, is performed in situ and without crosslinking, and is widely used to probe protein-DNA interactions. More recently, directed methylation with long-read sequencing (DiMeLo-seq) was demonstrated by targeting Hia5, a non-specific DNA adenine methyltransferase, to antibodies of interest [6]. This enzyme was also used to observe the regulatory architectures of chromatin fibers [7]. In a non-sequencing approach, DNA accessibility of single molecules was probed by enzymatic attachment of fluorescent cofactors, followed by optical genome mapping in nanochannel arrays [8].

Though adenine methylation is difficult to detect without immunoprecipitation or methylation-sensitive restriction enzymes, methylated cytosines can be detected by bisulfite conversion and sequencing. Advances in single cell methylation analysis enable reliable detection of methylation in over 50% of the genome of a single cell [9]. But bisulfite sequencing introduces a significant reduction in mappability and causes DNA breakage resulting in short molecules. Methylated DNA can also be detected directly from single molecules using nanopore sequencing [10].

We reasoned that antibody-mediated tethering of an MTase to a protein of interest will enable us to use antibodies capable only of binding a target of interest, while avoiding losses and inefficiencies associated with immunoprecipitation. Furthermore, an MTase with a short recognition sequence can be used to accumulate signals over a relatively short stretch of DNA, while maintaining a high mapping resolution. Specificity is provided by the fact that NpCpH methylation does not generally occur in the human genome. To achieve our goal, we developed Chromatin Antibody-mediated Methylating Protein (ChAMP) - a GpC MTase [11] fused to protein G. By tethering ChAMP to primary antibodies, we were able to produce distinct methylation patterns adjacent to the protein binding sites, thus facilitating their identification.

Material & methods

ChAMP design and synthesis

Protein DNA sequence was designed in-silico, ordered as a gblock (IDT), PCR amplified and SLICEd [18] into the pUC57-kan plasmid (See Supplementary file Champ_v1.0.gb). The design includes a GpC methyltransferase [11] fused to a protein G lacking the albumin binding domain. Linkers, a TEV cleavage site, a FLAG-tag and a His-tag were also included. Due to its methylation activity, only strains that do not restrict methylated DNA resulted in colonies after transformation. Moreover, bacteria expressing the plasmid need to be frozen or kept at the exponential growth phase, otherwise residual protein expression results in DNA methylation and bacterial death. The plasmid was maintained in 10G elite Escherichia coli bacteria (Lucigen).

ChAMP purification

The full protocol is available as a supplementary file and online at link. ChAMP was expressed in T7 Express lysY E. coli (C3010, NEB). Bacteria were grown overnight at 37 °C, 250 rpm, in 50 ml of LB supplemented with 50 µg/ml of kanamycin. Bacteria were moved into 200 ml of fresh LB supplemented with 1 mM of IPTG (no kanamycin added) and grown for additional 4 h. We note that ChAMP was readily detected without induction in both C3010 and 10G bacteria. Cells were cooled on ice for 15 min and centrifuged at 10,000G in a cooled centrifuge. The pellet was resuspended in GpC buffer (50 mM NaCl; 50 mM Tris-HCl; 10 mM DTT; pH 8.5) and moved to a 50 ml tube. The tube was centrifuged and the pellet resuspended in 10 ml of lysis buffer (GpC buffer + 0.5% triton X100 + protease inhibitor + 0.1 mM EDTA). This lysis buffer was supplemented with nothing (method 1 in Fig. 1E), 10% glycerol, 0.1% sodium deoxycholate and 0.1% SDS (method 2 in Fig. 1E), or 0.2% sodium deoxycholate (method 3 in Fig. 1E), and incubated at 4 °C rotating for 1 h. Samples were centrifuged for 10 min at 5,500G and the supernatant (~ 10 ml) was incubated with 0.5 ml his-tag beads for 1 h at 4 °C while rotating. Beads were loaded on a column and washed with 5 ml of GpC buffer followed by washes with 1.5 ml of GpC buffer with increasing concentration of Imidazole. ChAMP typically eluted between 200 and 1000 mM of Imidazole, with the earlier fractions containing more protein, and the later fractions being more pure. The eluant was aliquoted to single use tubes and stored at -80 °C. As all 3 methods yielded identical results, method 1 was used for all subsequent experiments.

Fig. 1.

ChAMP specifically methylates GpC near the proteins of interest. (A) diagram of ChAMP design. (B) Western blot of lysed T7 Express lysY E. coli expressing ChAMP show direct binding of the secondary antibody. No primary antibody was used (see Material & Methods) (C) Protein staining of HIS-tag purified ChAMP. (D) Detecting in-situ antibody binding. Immunofluorescence of HeLa cells stained with DAPI, a primary rabbit anti lamin B antibody and (i) positive control - AF555 anti-rabbit (first panel) or (ii) negative control - Cy3 anti mouse (no binding expected due to mismatch between primary and secondary species - second panel; inset at higher exposure shows no weak lamina staining) or (iii) ChAMP plus Cy3 anti mouse (bridged binding - third panel). As ChAMP has multiple IgG binding domains, it can bridge the rabbit and anti-mouse antibodies in panel 3, visualizing the nuclear envelope and proving correct localization. Scale bar − 10 μm (E) Detecting GpC methylation. GpC methylation inhibits HaeIII digestion. A PCR product containing multiple HaeIII cut sites was treated with either a lysate or purified ChAMP, in the presence of SAM, and digested with HaeIII. 3 purification methods were tested (see Material & Methods). Digested product was resolved on a gel. (F) Validation of GpC methylation in-vitro by ChAMP. Text shows genomic sequence, arrows point to GpC. Top diagram - Purified DNA was incubated with the ChAMP, bisulfite converted and Sanger sequenced. Bottom diagram - Validation of GpC methylation in-situ by ChAMP. Fixed cells were target methylated with ChAMP, DNA extracted, bisulfite converted and amplified by PCR. Significant methylation only seen upon mild fixation conditions. (G) Left - Restriction-PCR design overview. Genome track showing the targeted CTCF peak, HaeIII cut sites (estr enzymes - black lines) and PCR primers (red arrows). Right - ChAMP can specifically methylate DNA. Fixed cells treated with the ChAMP protocol showed a strong band following restriction-PCR, however no band at the correct size was seen when ChAMP was omitted and only a weak band when the CTCF antibody was omitted. Following the methylation reaction, DNA was digested with HaeIII and PCR reaction with limited number of cycles amplified the DNA products. Schematic representation of the method, applied to CTCF

Western blots

Samples were mixed with a protein loading buffer (Li-Cor 928-40004) and heated to 99 °C for 5 min. Proteins were resolved on a Bis-Tris denaturing protein gel (NuPAGE, Thermo Fisher) and transferred to a nitrocellulose membrane (iBlot mini, Thermo Fisher). The membrane was blocked for 1 h with 5% BSA at room temperature and incubated with secondary antibody (Li-Cor IRDye 925-68021 or 925-32210) for half an hour. While ChAMP contains both FLAG and His tags, no primary antibody was used, as the protein G IgG-binding domains are capable of binding most secondary antibodies directly, even after heating and membrane transfer.

Antibody-guided methylation

The full protocol is available as a supplementary file (ChAMP_protocol_v1.0_Freeze.pdf) and online at link. Briefly, cells were grown on poly-lysine coated 96-well plates or T25 flasks for 24–48 h. Samples were fixed with 0.1-1% formaldehyde for 5 min. Cells were permeabilized with 0.5% Triton X-100 in PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, and 1.8 mM KH2PO4) for 7 min. Samples were heated to 65 °C for 10 min and blocked with PBST (PBS supplemented with 0.1% Tween 20) with 5% BSA and 0.1 M glycine for 30 min. Samples were incubated with a primary antibody, washed, incubated with ChAMP, washed with detergents and incubated with SAM (S-adenosyl-methionine) for 1 h at 37 °C. Formaldehyde fixed samples were reverse-crosslinked at 65 °C overnight. Proteins were digested by proteinase K and DNA used for library preparation.

PCR validation

ChAMP methylation was validated by PCR using the primers AAGGAGAAGACAGAGTAGAGACTGC and ATGAGGAGGCTGGATGAGGT. PCR was performed using MyTaq HS Red Mix (Bioline) with 66 °C annealing and 15s elongation. Real-time PCR was performed using SYBR green (Invitrogen) and annealing at 64 °C. Ct values of cut were subtracted from uncut and used to calculate the fraction of uncut available for amplification.

Not CTCF 1	AGGACAGGAGCAGGAGACAG
Not CTCF 1	CTGGACAAGGCGCAGAGT
Not CTCF 2	ATCTGATGGGAGGTGGAGTG
Not CTCF 2	CAGACTTCAGGGAGGCTGAC
CTCF 1	ATGAGGAGGCTGGATGAGGT
CTCF 1	GCCAGGACAGCAATTCTAGG
CTCF 2	CTACTCCAGCACAGCCACCT
CTCF 2	CTGTCCCAGGCTGACAGG

Low input antibody-guided methylation

Resuspended HeLa cells were counted and serially diluted in medium. A single drop was placed in the middle of multiple wells in a 96 well plate, at a calculated cell number aiming for the desired range. Cells were allowed to settle in an incubator for an hour, to maximize cell adherence in the center of the well. Medium was added and the cells were incubated overnight. Two researchers independently counted the number of cells in each well. Wells with the right number of cells (2-1000 cells) were processed as in “Antibody-guided methylation”, followed by post-amplification enrichment with 5 cycles of random priming.

Library Preparation

The library preparation protocol is as a supplementary file and online at link.

Nanopore libraries

Modified DNA was purified by phenol-chloroform extraction and libraries were prepared using the Rapid Sequencing (SQK-RAD004; Oxford Nanopore) kit, according to manufacturer’s instructions.

Data Availability

Sequencing data has been deposited to https://www.ncbi.nlm.nih.gov/bioproject?term=PRJNA1041053.

Results

We envisioned that a protein capable of binding primary antibodies, switching-on its enzymatic activity and distinctly modifying adjacent DNA, can provide an alternative method to ChIP-seq in the characterization of protein binding sites (Graphical abstract and Fig. 1A). The advantages of this approach include (i) avoidance of the sample loss inherent in the immunoprecipitation methods (ii) multiple and bi-modal adjacent DNA modifications from a single enzyme-molecule can allow for a confident detection from a single sequenced molecule (iii) the primary antibody only needs to bind to its target, and not immunoprecipitate it, enabling the use of antibodies with weaker binding capacities (iv) modifications can be directly identified from single long molecules, enabling the exploration of 3D folding and the relations between adjacent sites (v) development of new MTases with a different recognition sequence can allow the multiplexing of several DNA-binding proteins on a single molecule. This is similar to DiMeLo-seq, however, the use of GpC MTase keeps this method compatible with bisulfite sequencing.

Method development

We designed (Fig. 1A and supplementary data), expressed (Fig. 1B) and purified (Fig. 1C) a fusion protein that (a) binds to antibodies and (b) methylates GpC in the presence of S-Adenosyl methionine (SAM). We validated that the fusion protein binds to antibodies in-vitro and in situ (Fig. 1B and D) and is able to GpC methylate purified DNA in-vitro (Fig. 1E, F).

Next, we developed (Supp. Figure 1A-C; Supp. Figure 2; Supplementary protocol) and optimized a protocol for antibody-guided methylation of fixed samples. As a proof-of-concept, we focused on the well characterized CCCTC-binding factor (CTCF) in HeLa cells. Standard formaldehyde fixation significantly inhibited DNA methylation in situ. However, short fixation time with low formaldehyde concentration enabled enzyme mediated methylation (Supp. Figure 1B). Methylation was further enhanced by a brief heating of the fixed samples prior to adding the enzyme (Supp. Figure 1C).

We then validated our ability to bind to primary antibodies of interest (Fig. 1D), to methylated CpG in-vitro (Fig. 1E, F), and to in situ methylate near a CTCF binding site using restriction-PCR and Sanger bisulfite sequencing (Fig. 1G). While methylation was enriched near the target of interest, nonspecific binding created background methylation (Supp. Figure 3A, lower left panel). Washes with detergents minimized nonspecific binding without affecting enzymatic activity (Supp. Figure 3A, B).

Methylation patterns

Natural GpCpH (H being any nucleotide but G) methylation rarely occurs in mammalian cells, and is thereby easily distinguishable from existing methylation patterns. To assess data quality, we sequenced samples to low-coverage and mapped the GCH methylation distribution relative to the closest known CTCF binding site, selecting for conditions that minimize off-target methylation. Incomplete conversion of cytosine into uracil resulted in background (typically 0.5-1%), however, as unconverted cytosines distributed randomly, requiring multiple adjacent methylation events (henceforth dense methylation) removed much of the background. To map the methylation patterns from CHAMP, we sequenced the HeLa CTCF samples to a higher coverage (122 M mapped reads). We aligned all the called methylation events relative to known CTCF peak centers, then applied a sliding window threshold, requiring 7 events within 200 bases, resulting in a strong peak overlapping the center of CTCF peaks. This pattern was not observed when either ChAMP or the primary antibody were omitted (Fig. 2A). Zooming in without the sliding window, revealed a double peak ~ 30 bp on either side of the CTCF ChIP-seq peak center (Fig. 2B), presumably because CTCF and the bound antibody-ChAMP complex limit enzyme accessibility to adjacent DNA, as observed by in similar conditions [12]. Weak secondary peaks, approximately 220 bp from peak center, may be the result of nucleosome occupancy (see below).

Fig. 2.

(A) Frequency of GpC methylation peaks detected by ChAMP, aligned to the center of CTCF ChIP-seq peaks (position 0; filtered by minimum of 7 events in 200 bp of sliding window moving average). The results show strong enrichment of methylation near known CTCF binding sites, indicating successful detection of CTCF-DNA interactions. In contrast, control conditions in which either ChAMP or the CTCF antibody was omitted show no significant enrichment, confirming the specificity of the method. (B) Zoom-in on ChAMP without the density filter showing a double peak. (C) Post-amplification enrichment - Schematic overview of the post-enrichment protocol. After bisulfite conversion, DNA is amplified by PCR for 1–5 rounds with random primers that do not introduce GpC. The purified product is further amplified, introducing adapters that lack GpC. The resulting product is re-methylated with the ChAMP protein. As this is amplified bisulfite converted DNA, any GpC found represents a methylated GpC (including GpCpG) in the original sample. (D) Distance of GpCpH methylation relative to the center of the closest CTCF ChIP-seq peak using post-amplification enrichment

To evaluate ChAMP’s ability to identify protein binding sites, we developed a basic peak calling strategy for initial comparisons with established methods. Methylation events were expanded using bedtools slop and called using callpeaks2.pl (https://github.com/Henikoff/Cut-and-Run/). However, more sophisticated algorithms that account for the unique properties of antibody-guided methylation will be needed for optimal peak identification, as discussed below.

Methylation enrichment

Without target enrichment, ChAMP requires genome-wide coverage. While this provides the additional information of bound/unbound ratios, it requires deeper sequencing than standard ChIP-seq experiments. Enrichment for methylated DNA can reduce the sequencing requirements. However, target enrichment may involve primary substance losses, thus inhibiting our ability to apply it to small samples. We circumvented this limitation by creating a post-amplification library enrichment protocol that enriches for GpC methylation, but not CpG (Fig. 2C). In our bisulfite converted library-preparation protocol we have three consecutive PCRs, used for random priming, amplification and indexing, respectively [9]. We modified the primers used in the first two PCR reactions to preclude the introduction of GpCs. Following the bisulfite conversion and amplification PCR, the only remaining GpCs present in the sample originated from a methylated GpC. We then re-methylated the sample (see Supplementary protocol) and immunoprecipitated the methylated DNA, effectively enriching for molecules that were originally labeled. We then performed a final amplification with indexing PCR prior to sequencing. Following the enrichment protocol, a clear structure matching nucleosome-free DNA surrounding the CTCF binding sites was seen in the cumulative data (Fig. 2D). Mapping methylation sites revealed that these predominantly overlap the CTCF ChIP-seq and Cut&Run signal (Fig. 3A, B). To validate the generality of ChAMP we tested it with an antibody directed to the histone modification H3K27ac. Dense methylated sites yielded methylation patterns that overlapped with a ChIP-seq signal (Fig. 3C). Additional methylation sites often overlapped DNase hypersensitive regions or were adjacent to exons, suggesting the possibility of 3D folding.

Fig. 3.

(A) Genome browser track showing dense methylation events for enrichment ChAMP using a CTCF antibody. These methylations partially co-localize with CTCF Cut&Run and ChIP-seq. (B) Zoom-in on a ChIP-seq peak. (C) Genome browser track showing dense methylation events for ChAMP using a H3K27ac antibody. These methylations partially co-localize with H3K27ac ChIP-seq

Single molecule mapping of binding sites

Various single molecule sequencing technologies allow direct detection of DNA modifications [14, 15]. We used the modified version of nanopolish that is trained to call GpC methylation from Lee et al. [10, 12]. Nanopolish uses a hidden Markov model trained on enzymatically methylated DNA to distinguish between cytosine and 5-methylcytosine. While single molecule sequencing and methylation detection have higher error rates than clonal amplification and sequence by synthesis, it increases the maximal read length by 3 orders of magnitude, to over 100 kb. At these scales, the extended methylation pattern resulting from a single bound site can be seen on a single molecule. Moreover, the relationship between the occupancy of neighboring sites and how distance affects it can be analyzed. Despite using fixed cells, we were able to obtain multi-kilobase sequences with good quality scores (Supp. Figure 4; N50: 8327 bp; [16]). For each read, we calculate the GpC methylation probabilities for k-mers along the sequence [12]. Methylation was greatly enriched near known H3K27ac binding sites, with multiple adjacent k-mers showing high probability of methylation (Fig. 4A, B).

Fig. 4.

(A) Genome browser track showing methylation probability of k-mers along a single ~ 150 kb molecule from a ChAMP experiment using an H3K27ac antibody. Methylation probability was enriched, but not exclusive, to ChIP-seq peaks, open chromatin and ChromeHMM Active Promoters. (B) Zoom-in on a ChIP-seq peak. HeLaS3 H3K27ac - ChIP-seq signal and peaks; HeLaS3 ChromHMM - chromatin state track; HeLaS3 DS - DNaseI HS Density Signal

Discussion

In the current study we present a new method to map protein-DNA interactions, using low input and the possibility of mapping the binding sites on single long molecules. Importantly, the mapping is done in-situ, while preserving much of the DNA structure during methylation. Limitations of the system include (i) the need for a suitable antibody and motif (GpC) sites near the target of interest; (ii) the lower mappability of bisulfite-converted DNA requires more sequencing than alternative methods; (iii) ChAMP has only been tested with open chromatin near the target of interest.

But mapping protein binding on single DNA molecules can be used to answer fundamental questions about genome organization and functionality. For example, the distribution of CTCF binding sites in the genome is not random. Specifically, a ChIP-seq validated CTCF binding site has a high probability to be in the vicinity of another such site. This can be the result of sites being enriched in specific genomic regions [17]without any direct or indirect interactions (neutral model). Alternatively, adjacent CTCF binding sites can provide redundancy for a specific function, with only one of the sites typically being occupied (exclusive model). Finally, adjacent sites can facilitate cooperative binding and 3D organization (cooperative model). These possibilities can be distinguished by calculating how CTCF binding of one site affects the probability of CTCF binding on an adjacent site on the same molecule. This method can also allow the exploration of alleles on protein occupancy in adjacent sites.

Peak calling for ChAMP signals presents unique computational challenges compared to traditional ChIP-seq analysis. While this manuscript provides an initial peak calling approach, optimal analysis must account for the distinctive characteristics of antibody-guided methylation patterns. The density and distribution of GpC sites throughout the genome create an inherent bias that influences signal detection, as regions with higher GpC content naturally yield more methylation events when bound. Future peak calling algorithms should normalize for this variation in GpC density to prevent artifactual enrichment in GpC-rich regions. Additionally, nucleosome positioning plays a crucial role in accessibility of DNA to the ChAMP enzyme, as evidenced by the observed double-peak pattern flanking CTCF binding sites and secondary peaks at approximately 220 bp intervals. An advanced peak calling algorithm could integrate nucleosome occupancy maps to differentiate between true binding events and regions where accessibility, rather than binding, drives the methylation pattern.

Reducing the minimal required cell number can allow for the exploration of rare cell types. While ChIP-seq experiments typically require a large number of cells, Cut&Run experiments targeting CTCF have been successfully performed on 1000 cells [13]. By contrast, protocols for CpG methylation detection from single cells have been developed [9]. As the output of ChAMP labeling is compatible with the input of single-cell CpG methylation protocols, we hypothesize that in combination with post-amplification enrichment, informative experiments can be performed from low cell numbers.

Future studies will expand these capabilities by generating ChAMP variants targeting sequences other than GpC. A MTase lacking sequence specificity can generate more signal with improved mapping, while MTases targeting distinct sequences can be multiplexed to map the binding sites of multiple proteins on a single DNA molecule. These can even be extended to other (non 5-methylcytosine) methylations and base modifications, using single molecule technologies that can distinguish them. Moreover, direct conversion of DNA, for example by deaminases or by MTases and SAM analogs like 5’-amino-5’-deoxyadenosine, may also be used with improved genomic mapping. By directly editing DNA, all sequencing methods can be used with high accuracy and while maintaining high mappability. Development of such variants will enable the identification of multiple distinct protein-DNA interactions on a single molecule.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

DZB designed the experiments, with assistance from FSC and MRE. AT and DZB performed the experiments. IL, JTS and WT assisted with the nanopore analysis. All authors reviewed the manuscript.

Funding

This work was supported by the Israeli Science Foundation (grants 654/20 and 632/20 to DZB), the Center for Artificial Intelligence & Data Science in Tel Aviv University (TAD to DZB) and NIH intramural support for project HG-200305 (FSC).

Data availability

Sequencing data has been deposited to https://www.ncbi.nlm.nih.gov/bioproject? term=PRJNA1041053.

Declarations

Ethics and consent to participate declarations

Not applicable.

Consent to publish

Not applicable.

Competing interests

The authors declare no competing interests.