Design of a guidedNOMe-seq experiment
For guidedNOMe-seq we start out with a group of ROI, for example binding sites of a specific transcription factor (TF) that are subject to investigation in the respective study and, importantly, unrelated ROI, e.g. binding sites of an unrelated TF under the respective experimental conditions, that can be used as internal technical controls. The guidedNOMe-seq procedure entails the following steps (Fig. 2A): 1) Computational identification of ROI including filtering for the presence of GpCs in the three chromatin quantification windows. 2) Selection of suitable Cas9 cut sites up and downstream of the ROI chromatin quantification windows. It is important to select the Cas9 cut sites in such a way that all three chromatin quantification windows are sequenced from a single DNA molecule. The precise parameters depend on the sequencing platform and modality used. In this study, we made use of Illumina paired-end 300bp sequencing and thus were able to choose the ROI spanning 600 bp surrounding the TF motif center, with Cas9 targeting a window of 80bp at the 5’ and 3’ end of the ROI. This resulted in library insert sizes ranging from 440bp to 600bp, covering 3 nucleosomes on average. 3) Execution of the standard NOMe-seq protocol (Grand et al. 2021) until the isolation of genomic DNA. 4) Performing custom Cas9 digest using the in silico designed sgRNA pool specific to the ROI. 5) Ligation of methylated adapters (compatible with the respective sequencing technology used) containing unique molecular identifiers (UMIs) to the NOMe-treated Cas9-digested genomic DNA fragments. 6) Bisulfite conversion of the library followed by PCR amplification and sequencing on the platform of choice. 7) Alignment to bisulfite-aware reference genome and quantification of chromatin states.
Of note, this protocol results in a library containing a custom set of ROI with 5’ and 3’ ends defined by the Cas9 cut sites and thus allows simple estimation of ROI enrichment and hence quality of the acquired data.
Technical aspects of guidedNOMe-seq
To benchmark guidedNOMe-seq, we generated two biological replicate guidedNOMe-seq libraries from wild-type mouse embryonic stem cells (mESC) targeting 1226 ROI with binding sites for the well-studied transcription factor CTCF and 274 ROI entailing binding sites for the transcription factor REST.
Visual inspection and computational analysis of the guidedNOMe-seq data showed that the ROI enrichment works efficiently and that the sequencing reads start at the expected Cas9 cut sites (Fig. 2B). Notably, only 1.3% of ROI (20 out of 1500) showed no read coverage in both replicates, while for the vast majority fragment coverage over ROI is relatively homogenous and comparable between replicates (Fig. 2C and S1A). Moreover, the high correlation between replicates shows that the Cas9-sgRNA RNPs reproducibly enrich hundreds of ROI from a complex genome and that either ROI intrinsic or experimental parameters (e.g. Cas9 cutting efficiency) dictate fragment coverage (see below). Of the 4 and 5.5 million paired-end reads (which equals genomic coverage of ~ 1) sequenced in the two replicates, we found 197k and 186k reads spanning the selected ROI which corresponds to a 140- and 100-fold enrichment, respectively (see Methods). Importantly, due to the defined 5’ and 3’ cut sites in guidedNOMe-seq, 90% of reads that map to the ROI span all three chromatin quantification windows and are therefore usable to quantify chromatin states (Fig. 2B and 2D). The unique molecular identifiers (UMIs) present in the ligated adapters allow for the identification and removal of PCR duplicates. Plotting deduplicated fragment counts versus duplicated fragment counts shows that there are indeed PCR duplicates, indicating that the libraries are sequenced to saturation and demonstrating that the addition of UMIs is essential for such quantitative experimental approaches (Figure S1B). This analysis also confirms that UMIs are distributed equally over all ROI and scale proportionally to the coverage of the ROI, suggesting that there are no major locus-specific ligation or PCR biases for the over thousand different genomic regions.
We next generated single ROI plots to assess reproducibility between biological replicates. Reassuringly, both replicates produce very similar nucleosome and TF footprints (Fig. 2E). Further, when globally comparing the chromatin states called based on the footprint patterns (TF, nucleosome occupied, or open, see Fig. 1B) in all suitable ROI with a coverage of at least 30 reads, we found that guidedNOMe-seq results in highly reproducible chromatin state classifications (Fig. 2F). To further validate the chromatin state classifications, we compared the TF chromatin state quantification of all ROI with CTCF ChIP-seq signal. This has previously been reported to correlate, indicating that CTCF ChIP-seq enrichment is a reasonable proxy for TF occupancy at individual binding sites (Sonmezer et al. 2021). As expected, the control group consisting of REST ROI shows no correlation with CTCF ChIP-seq enrichment, whereas the CTCF ROI show a linear correlation between the TF state quantified by guidedNOMe-seq and CTCF ChIP-seq (r = 0.72) (Fig. 2G).
From these analyses we conclude that guidedNOMe-seq is a robust, specific, and versatile extension of existing NOMe-seq protocols that allows unique customization at a medium throughput.
Comparison with other ROI enrichment methods
We next set out to compare guidedNOMe-seq to other locus specific or genome-scale approaches. NOMe followed by locus-specific PCR-based analysis achieves high specificity and coverage and is therefore considered the gold standard (Figure S1D). It also results in defined 5’ and 3’ ends and thus a high percentage of informative reads, as observed for guidedNOMe-seq (Fig. 2D). However, it suffers from two major drawbacks: First, it is not easily scalable to hundreds or thousands of ROI. Secondly, it is prone to PCR biases and artifacts as primer design and PCR amplification using bisulfite treated DNA is inherently difficult (Lu et al. 2017; Wojdacz et al. 2008).
Another strategy to reduce complexity and thus sequencing costs is to combine NOMe-seq with commercial array capture-based enrichment as introduced by Sönmezer et al (Sonmezer et al. 2021). This strategy is comparable to guidedNOMe-seq but offers less flexibility and customizability. For example, the commercial array capture probes used in the Sönmezer et al study target ~ 2% of the mouse genome with a focus on regulatory regions such as transcription start sites (TSSs) and cis-regulatory elements (CREs) but were not specifically designed to enrich TSS distal ROI or TF binding sites. The ROI coverage between replicates is comparable with guidedNOMe-seq (Figure S1C), but the percentage of mapped reads overlapping with ROI is considerably higher when using array capture (90%) as compared to Cas9-mediated enrichment (23%; Figure S1D). However, many of those on-target sequencing reads from the array capture approach cannot be used for inferring chromatin states of individual loci because they do not span the three quantification windows. This is due to a simple technical reason: The array capture protocol relies on shearing of genomic DNA prior to hybridization, which is commonly performed by sonication. This leads to randomly sheared genomic fragments without defined 5’ and 3’ ends. Because of this, ~ 80% of the on-target reads do not span all three chromatin quantification windows. Consequently, these reads cannot be used for chromatin state quantification (Figures S1E). In addition, the inflexible commercial array design results in the enrichment of regions that are not necessarily of value to address a particular research question.
Taken together, whereas the guidedNOMe-seq protocol generates more off-target reads, array capture NOMe-seq yields more non-usable reads on target, which ultimately results in comparable percentages of reads that can be used for chromatin state quantification between techniques. To put this in numbers for our specific set of ROI and when we only look at the 465 ROI shared between the guidedNOMe-seq generated here and the published array capture NOMe-seq dataset (Sonmezer et al. 2021). If we wanted to achieve that ~ 60 percent of the ROI shared between guided and array capture libraries have at least 30x informative read coverage, we would need to sequence ~ thirty times more reads (125–150 million vs 4-5.5 million reads) in an array capture approach in comparison to guidedNOMe-seq due to the high number of non-informative ROI when studying a specific TF like CTCF.
The 465 ROI that are shared between our guidedNOMe-seq data and the array capture NOMe-seq have a similar amount of reads at the different sequencing depths used (~ 50k for guided, ~ 50k-150k for array). When comparing average protection profiles over CTCF and REST binding sites, the two methods reveal very similar average protection patterns over all ROI (Fig. 2H). Compared to methods like ChIP-, MNase- or ATAC-seq, an advantage of NOMe-seq is the ability to quantify chromatin states with single DNA molecule resolution. To determine if guidedNOMe-seq also compares well with the array capture NOMe-seq at this level, we quantified chromatin states with single DNA molecule and single ROI resolution. Reassuringly, both enrichment strategies again agree very well (Fig. 2I). Pearson correlation coefficients between experimental approaches (Open r = 0.59 (REST) and 0.78 (CTCF), nucleosome r = 0.79 (REST) and 0.83 (CTCF) and TF r = 0.85 (REST) and 0.92 (CTCF)) further corroborate the visual inspection of the data. In conclusion, NOMe-seq results in highly reproducible classification of chromatin states irrespective of the selected enrichment protocol, the respective experimenters, or even laboratories. Notably, when solely considering oligo synthesis costs (sgRNA oligos versus capture oligos), guidedNOMe-seq is more cost effective up to ~ 5000 ROI. For larger ROI numbers, designing a custom capture array would be more economical (Figure S1F). NOMe profiling, with and without Cas9-mediated enrichment of ROI has also been employed in combination with long-read Oxford Nanopore sequencing (Battaglia et al. 2022; Lee et al. 2020; Wang et al. 2019). This method referred to as nanoNOMe-seq, produces much longer reads at the expense of throughput. Due to the long-read length, nanoNOMe is perfectly suited to study nucleosome positioning and TF binding spanning hundreds of nucleosomes for a handful of ROI while array-capture and guided-NOMe can only query 2–4 nucleosomes in a single sequencing read but for hundreds to thousands of ROI. Due to difference in studied organism (human/yeast vs mouse), we could not directly compare nanoNOMe with guidedNOMe inferred chromatin states.
We conclude that guidedNOMe-seq, array capture NOMe-seq and nanoNOMe-seq are nicely complementary both with their own strengths and weaknesses (see also Discussion).
sgRNA design principles for guidedNOMe-seq
As outlined above, the usability of ROI depends on the presence of at least one GpC in all three chromatin quantification windows (Fig. 1B). Therefore, we determined the fraction of ROI that fulfil this criterium. When bootstrapping random genomic regions, on average 20 percent of ROI in the mouse genome are suitable for quantification by guidedNOMe-seq. Focusing on ROI containing a REST or CTCF ChIP-seq peak and binding motif we found that approximately 50 percent of these regions are suitable (Fig. 3A). The observed higher percentage of suitable REST and CTCF associated ROI can be explained by the presence of GpCs in their binding motifs (Johnson et al. 2007; Kim et al. 2007). The non-uniform distribution of GpCs throughout the genome and relative to a specific set of ROI is an inherent caveat of NOMe-seq that cannot easily be changed and hence needs to be taken into account when designing an experiment.
Further, the fraction of ROI that can be targeted depends on the stringency of sgRNA design parameters. Namely, the target space allowed for Cas9 surrounding the ROI and the chosen cutoffs regarding on- and off-target scores of the identified sgRNAs. In this study we used an 80bp target space at 5’ and 3’ ends of each ROI, did not select for GpCs in the chromatin quantification windows because of the GC-rich binding motifs of the TFs of interest, neglected on-target scores and omitted sgRNAs with a very poor off-target score (see methods).
To assess the impact of the above parameters, we performed in silico analyses with different settings. First, we investigated the relationship between Cas9 target space size and the number of targetable ROI for REST, CTCF and random genomic regions. Already with the smallest target space (50bp) Cas9 can, depending on the analyzed ROI group, liberate between 1/5 − 1/3 of the ROI. As expected, the fraction of targetable ROI increases when Cas9 target space increases. Notably, with a target space of 100bp, Cas9 can target half or more of the ROI analyzed (Fig. 3B). Using these numbers, in combination with the GpC distribution analysis described above, we estimated that we can target between 9k and 26k out of the 70k CTCF ChIP-seq peaks depending on the chosen cut offs.
Thus, not surprisingly, increasing target space for sgRNA selection improves the number of targetable loci in the genome. Yet, this comes at the expense of correspondingly shortening the length of quantifiable DNA around the TF motif and must thus be considered carefully and adjusted to the read-length of the available NGS sequencing platform.
Next, we determined if in silico predicted sgRNA on- and off-target scores can be used to optimize sgRNA selection and read coverage uniformity of ROI. Surprisingly, this analysis making use of the 1500 ROI in our experimental set up revealed a very poor correlation between sgRNA on-target scores and observed read coverage over ROI. Towards this end, we employed seven different sgRNA prediction algorithms but none of them could predict cutting efficiency, respectively read coverage, satisfyingly (Figs. 3C and S2A)(Hoberecht et al. 2022). Nevertheless, we would still recommend in silico prediction of sgRNA on-target scores because our analyses revealed that sgRNAs with very poor predicted on-target scores and the presence of simple repeats in ROI should be avoided when possible (Figure S2B).
Lastly, we analyzed potential adverse effects of retaining sgRNAs with predicted off-targets. We first determined the percentage of reads associated with the on-target Cas9 cleavage activity. In both replicates we found ~ 10% of reads precisely originating from the predicted on-target sgRNA-Cas9 cleavage sites, but also noticed that another ~ 4% of reads were within 600bp of the intended cleavage site (Fig. 3D and S2C). We attribute this to the combined effect of targeted digestion by Cas9 in combination with random breakage of genomic DNA on the other side, which stochastically results in DNA fragments of suitable sizes for NGS library generation. In addition, ~ 7 percent of reads are associated with predicted off-target sites, which is three times more than expected by chance (Fig. 3D). Consistent with literature, we found that predicted off-targets without PAM sequence contribute minimally to off-target reads (Figure S2D)(Mojica et al. 2009).
In summary, the sgRNA analysis revealed that on-target score predictions vary considerably between used algorithms and correlate poorly with the observed ROI read count distribution. We therefore assume that ROI sequence features (e.g. simple repeats (Figure S2B)) and potential biases in the library preparation outweigh the selection of sgRNA on their predicted on-target score except for sgRNAs with a very low score. The sgRNA off-target analysis showed that using only perfect sgRNAs without off-targets will most probably decrease off-target sequencing reads. However, the gain will likely be minimal, as the majority (> 65%) of the reads not associated with a ROI are not linked to predicted sgRNA off-target sites but arise due to random genome breakage during the NOMe-seq protocol. Thus, careful handling of nuclei and isolated DNA, as recently optimized by Battaglia et al., is an effective precaution to reduce off-target reads (Battaglia et al. 2022).
Strengths and limitations of guidedNOME-seq
Despite the steadily decreasing costs for next-generation sequencing, financial constraints are still a major bottleneck for large experiments such as time-courses or comparative studies with dozens of samples. To determine the minimal sequencing depth that is required for reliable quantification of chromatin states by guidedNOMe-seq, we systematically down-sampled the guidedNOMe-seq libraries and subsequently performed analysis at different resolutions.
We first generated average profiles plotting the fraction of GpCs that are protected from methylation by M.CviPI per nucleotide, while keeping the CTCF and REST ROI separate. As expected, using 4 million guidedNOMe-seq reads, we observed M.CviPI protection footprints for the TF bound to its motif and for positioned nucleosomes up and downstream of the TF binding site (Fig. 3E). At this sequencing depth, almost all individual GpC protection values (grey dots) nicely align with the LOESS smoothed curve. When serially down sampling total read numbers, the average protection pattern indicated by the LOESS line persists throughout while the noise level of individual data points increases. However, even with as little as 50,000 reads, TF and nucleosome footprints were still detectable (Fig. 3E). To quantify the increase in single nucleotide noise we computed the Pearson correlation between the fraction of protection per nucleotide position with decreasing read numbers versus the full 4 million total reads sample. This revealed correlation coefficients ranging from 0.92 − 0.64, indicating that with as little as 50k reads, reproducibility is still very high (Figure S2E).
We next focused our analysis on the chromatin state quantification of single ROI by assessing the GpC protection state of individual DNA molecules (according to Figs. 1B and 2F). We followed a similar down-sampling approach as described above to determine how chromatin state classifications perform under limited coverage conditions. In contrast to the average plots, the accuracy of the individual ROI chromatin state classification rapidly deteriorates with decreasing read numbers (Figure S2F). This is likely caused by an increase in sampling noise due to the lower coverage per ROI at lower sequencing depths (Fig. 3F). To determine the required sequencing depth, we compared Pearson correlation coefficients of inferred ROI chromatin states (TF-bound and Nucleosome-bound) between replicates with increasing subsampled read counts. In line with the replicate scatterplots for the TF chromatin state, this showed a steep increase in correlation coefficient with increasing read numbers, which starts to flatten out between 1 and 4 million reads. That is, sequencing more than 4 million reads for 1500 ROI can increase reproducibility, albeit only relatively small improvements are likely (Fig. 3G).
In summary, we establish the experimental boundaries of guidedNOMe-seq and find that biological reproducibility of the protocol is excellent, without applying any normalization steps. Hence, guidedNOMe-seq is an important extension to existing NOMe-seq protocols, enabling researchers to quantify how perturbations influence chromatin states on hundreds of loci in parallel.
guidedNOMe-seq reveals asymmetric nucleosome patterns at ChAHP-bound loci
To exemplify the power of guidedNOMe-seq to quantitatively measure how chromatin states change upon perturbations, we used genetic inactivation of the ChAHP complex as a case study. We have previously shown that the ChAHP complex, consisting of the chromatin remodeler CHD4, the zinc-finger protein ADNP, and heterochromatin protein HP1gamma, binds over 15,000 loci in mESCs (Ostapcuk et al. 2018). These ChAHP-bound loci often reside in a repetitive SINE B2 element and are also enriched for CTCF. When ChAHP function is perturbed (e.g. through the removal of ADNP), chromatin accessibility and CTCF binding increases specifically at ChAHP-bound sites, suggesting that ChAHP competes with CTCF for binding on chromatin and/or restricts chromatin accessibility at those sites (Kaaij et al. 2019; Ostapcuk et al. 2018). However, the molecular mechanisms behind ChAHP function and its interplay with CTCF are not well understood. We thus set out to quantify nucleosome occupancy and TF binding in Adnp−/− mESCs and compared this to the guidedNOMe-seq data set from wild type and CTCF-depleted mESCs. To address the interplay between ADNP/ChAHP and CTCF, we subdivided the CTCF ROI into two groups: group 1 contains ROI where both ADNP/ChAHP and CTCF are enriched and group 2 consists of ROI where only CTCF is bound based on ChIP-seq data (Fig. 4A) (Kaaij et al. 2019). Reassuringly, the protection profile in the REST-bound control group is very similar between all conditions, further validating the high reproducibility of guidedNOMe-seq between experiments and conditions (Fig. 4B). At loci where both CTCF and ADNP/ChAHP are enriched, we find a pronounced increase of positioned nucleosome upon ADNP loss. Loci only bound by CTCF do not show a different protection/nucleosome positioning pattern in the absence of ADNP/ChAHP. In contrast, the protection pattern of CTCF-associated groups is markedly changed following 24 hours of CTCF depletion. Here, the characteristic GpC protection pattern in wild type cells is lost. Of note, the GpC protection after CTCF depletion is at a similar level as the protection elicited by + 1/-1 nucleosomes in wild type cells, indicating that the ROI are not in an open, nucleosome free state when CTCF is depleted, but rather occupied by non-positioned nucleosomes (Fig. 4B). This random nucleosome positioning is supported when inspecting individual ROI where we observe large (> 100bp), seemingly randomly located stretches of protected GpCs in individual DNA molecules (Figure S3A). When quantified at the single ROI level, we find that around 20% of DNA molecules are TF bound in the two CTCF-associated groups at any given moment in wild type cells. Specifically, in the CTCF group associated with ADNP/ChAHP, this increases to ~ 50% in the absence of ADNP, whereas upon CTCF depletion TF binding decreases to nearly 0 percent in both CTCF-associated groups (Fig. 4C).
To determine which ROI show significant changes in their chromatin state, we used an implementation of edgeR in our dinoR package. This revealed that indeed many ROI undergo a statistically significant chromatin state change from nucleosome bound to TF-bound between wild type and Adnp KO samples (Figure S3B-E). Accordingly, the increase in TF-bound chromatin states in Adnp−/− cells is coupled to a decrease in nucleosome bound DNA molecules and a small increase in open chromatin. Further, these results indicate that the nucleosome and TF protection patterns observed in wild type cells are largely dependent on CTCF, as positioning is nearly completely lost upon CTCF depletion.
Of note, when comparing the wild type protection patterns between CTCF bound and CTCF and ADNP/ChAHP co-bound loci, we noticed an asymmetry in the ADNP-bound group in some single ROI plots which we never observed in the CTCF or the REST single ROI plots (Fig. 4D). To further explore this, we split this group of ROI based on CTCF ChIP enrichment (high – medium – low) and generated average protection profiles (Fig. 4E). The asymmetry is primarily visible in the two groups with medium and low CTCF levels. Interestingly, the maximum GpC protection is similarly high on both sides of the binding motif, indicating that the fuzzier downstream region is occupied by nucleosomes, which however are not well-positioned in contrast to the upstream region. Upon Adnp loss the asymmetry is lost in all groups, indicating that the fuzzy nucleosome positioning downstream of the binding site might indeed be due to ChAHP. Hence we wanted to quantify the observed asymmetry with single ROI resolution. To do this we adapted the chromatin quantification window approach and introduced two novel states, namely “upstream positioned nucleosome” and “downstream positioned nucleosome” (Figure S4A). Analyzing these two additional chromatin states revealed that ~ 30% of the ADNP/CTCF co-bound ROI significantly lose DNA molecules classified as positioned nucleosome upstream in the absence of Adnp (Figure S4B and C). At present we do not know how the observed asymmetry is established and if it has a functional role in reducing CTCF binding. But this likely results from the chromatin remodeling activity of CHD4, which has previously been shown to reduce access to chromatin by increasing nucleosome densities in a non-positioned manner at its target loci (Bornelov et al. 2018; Clarkson et al. 2019; de Dieuleveult et al. 2016; Morris et al. 2014). These data confirm that CTCF only and REST-bound sites show symmetric nucleosome positioning, whereas CTCF and ADNP enriched ROI show asymmetry in their nucleosome positioning that correlates with the presence of ChAHP.
guidedNOMe-seq reveals insights into the interplay between ChAHP and CTCF with temporal resolution.
Most of our knowledge regarding the function of genes in biological systems is through the generation of constitutive gene knock-out models and subsequent observation of loss-of-function phenotypes. These studies have proven very powerful, but measure endpoint situations without temporal resolution and can therefore be prone to secondary effects. With the recent development of small-molecule inducible degradation tags that allow rapid and specific degradation of endogenously tagged proteins, high-resolution time course measurements of biological processes became feasible (Nabet et al. 2018; Nishimura et al. 2009). Such experiments have for example been performed to measure the temporal dependency of 3D genome organization on the presence of the cohesin subunit RAD21(Rao et al. 2017). Also, it proved to be a powerful approach to assess the relevance of chromatin remodelers like BRG1 for TF binding to DNA (Rao et al. 2017; Schick et al. 2021) .
Unfortunately, high resolution time course experiments coupled to genome-wide read-outs such as ChIP-seq, HiC, ATAC-seq, or RNA-seq are still very expensive and often require complex normalization procedures for data analysis due to sparsity of data-points. As described above, guidedNOMe-seq requires relatively low sequencing depth to achieve excellent reproducibility without any normalization and therefore appears to be ideally suited to measure chromatin state dynamics during time course experiments.
To test if guidedNOMe-seq can provide insight into chromatin state dynamics, we established an endogenous gene fusion of Adnp with FKBP12F36V and mNeon-Green (Fig. 4F; top). The FKBP12F36V tag allows for rapid degradation of ADNP upon addition of the dTAG-13 compound, whereas mNeon-Green allows for the direct quantification of ADNP protein levels by flow cytometry (Nabet et al. 2018). Furthermore, the FKBP12F36V degradation system is reversible. Upon dTAG-13 wash-out, ADNP levels should slowly restore, making it also possible to perform Adnp recovery time-course experiments as well. We first established the depletion and recovery kinetics of ADNP after dTAG-13 treatment and wash-out (Fig. 4F; bottom). In agreement with studies on other proteins, we found that the FKBP12F36V tag leads to rapid and near complete depletion of ADNP upon addition of dTAG-13 (Nabet et al. 2018; Schick et al. 2021). 5 minutes after dTAG-13 addition, a small decrease in mNeon-Green signal was already observable. After 1h of treatment, fluorescence intensity was close to background levels (untagged control). Recovery of protein levels after dTAG-13 wash-out is largely dependent on protein synthesis rates. In the case of Adnp recovery started 2h after wash-out and steadily increased to about 35% after 8 hours and recovery was complete 24 hours after wash-out. We next generated guidedNOMe-seq libraries for these 16 time points in biological duplicates throughout the ADNP depletion and recovery time course. Utilizing the same three ROI classes as before, the GpC protection pattern changed as early as 15 minutes after initiation of ADNP depletion specifically at ROI where ADNP is present (Fig. 4G). The methylation protection at the CTCF motif and the positioning of nucleosomes steadily increased during the first hour of ADNP depletion and was comparable to that of Adnp−/− cells at later time points (Fig. 4B and 4G). Upon reversal of the depletion (dTAG-13 wash-out), we observed first signs of recovery towards the untreated state after 8 hours. Reversion was completed within 24 hours. These results show that ChAHP bound ROI are converted into CTCF bound sites upon ADNP depletion within an hour. In the other direction, although the reversal of CTCF bound sites to ChAHP controlled loci is constrained by ADNP protein synthesis rates, we observe the first signs of the reversal 8 hours after dTAG-13 wash-out when ADNP protein levels are back to 35% compared to untreated cells (Fig. 4F).
We next set out to quantify chromatin state dynamics, as introduced in Fig. 1B, with single DNA molecule resolution throughout the time course. The ROI present in the ADNP/CTCF group showed a marked increase in TF-bound chromatin state already after 15–30 minutes of Adnp depletion while the number of nucleosome-bound molecules reduced correspondingly (Fig. 4H and I). On the other hand, CTCF only and the REST control group showed minimal dynamics throughout the time-course, in line with the above average analysis. To statistically stratify these observations dinoR was used. This revealed that after 15 minutes of ADNP depletion we already find the first ROI to show a significant increase in TF chromatin state. The number of significant changes increases during the first hour and then remains constant thereafter. Intriguingly, this increase in TF chromatin states is completely reverted 24H after dTAG wash-out (Figs. 4I).
Of note, the ADNP associated asymmetric nucleosome positioning that we identified in wild type vs Adnp−/− ESCs (Fig. 4D, E and S4B, C), follows ADNP levels throughout the depletion and recovery time course, which again supports the idea that the downstream loss of precise nucleosome positioning is due to the presence of functional ChAHP complex (Figure S4D). However, further experiments will be required to unequivocally attribute this effect to a specific activity of ChAHP and to investigate its consequence for CTCF binding.
These data demonstrate that the ChAHP complex is continuously required to counteract CTCF binding at its target sites. The ADNP recovery experiments show that there are no signs of long-term epigenetic memory that would reduce ChAHP activity at CTCF loci. This strongly suggests that the ChAHP-CTCF binding landscape is established cell autonomously in a rapid and reversible manner.
Taken together, we conclude that guidedNOMe-seq is a powerful technique to quantify how genetic or small-molecule induced perturbations influence the chromatin state at hundreds of individual ROI either at steady-state conditions or in time-course series with many samples requiring minimal amounts of NGS sequencing.