Intrinsic DNA codes govern distinct modes of nucleosome-transcription factor interactions

Abstract

The interplay between transcription factors (TFs) and nucleosomes is central to gene regulation, yet most studies adopt a protein-centric perspective that largely overlooks DNA sequence contributions. Here, we identify four distinct nucleosomal DNA classes, including the canonical WW/SS pattern (W = A/T, S = G/C). All four nucleosome types are widespread across the human genome both in vivo and in vitro. Nucleosomes sharing the same pattern are rotationally in phase, whereas different patterns exhibit specific rotational offsets, revealing distinct DNA codes for nucleosome positioning. Analysis of 744 TFs across eukaryotes shows that the WW/SS and anti-WW/SS patterns are associated with TF binding in chromatin. Differences in the proportions of these patterns in a region, quantified as ΔNPS values, correlate closely with in vitro nucleosome occupancy. Integrating ΔNPS with in vivo nucleosome occupancy identifies four distinct modes of nucleosome–TF interaction and links nucleosomal DNA patterns to nucleosome-depleted regions around TF binding sites.

Here, the authors describe four distinct DNA sequence patterns in nucleosomes that shape how transcription factors interact with chromatin, revealing previously hidden DNA-encoded rules governing nucleosome positioning and gene regulation.

Introduction

Understanding the interaction modes between transcription factors (TFs) and nucleosomes is central to gene regulation^1–3. The traditional view holds that TFs compete with nucleosomes for access to their binding sites^4,5, a model supported by ChIP-seq analyses showing that TF-bound genomic regions are typically nucleosome-depleted regions (NDRs)⁶. However, numerous studies have demonstrated that a subset of TFs, including pioneer factors, can directly bind nucleosomal DNA^7,8.

Most studies of nucleosome-TF interactions have relied on protein-centric experimental approaches, such as electrophoretic mobility shift assays^9,10, single-molecule fluorescence methods^11,12, cryogenic electron microscopy^13–15, SELEX¹⁶, and protein microarray¹⁷. These experiments typically employed a limited number of nucleosomal DNA templates, including the synthetic “601” sequence¹⁸ or a small set of genomic loci with strong nucleosome positioning sequences (NPSs), such as yeast Hsp82¹¹, ribosomal protein L30 genes¹⁹, mouse ALBN1, NRCAM, and CX3CR1¹⁷, and the human lin28b locus²⁰. Consequently, prior work has focused primarily on TFs rather than their DNA-binding partners.

Nucleosome-TF interactions are influenced by the rotational phasing of TF binding motifs within nucleosomes^21–24. Early studies of the glucocorticoid receptors showed that motif accessibility depends on rotational positioning²¹, with similar effects observed for p53^22,23 and SRY-box 2²⁴. Rotational positioning is defined by which face of the DNA duplex contacts the histone core²⁵ and is strongly determined by DNA sequence^26–33, including the AA/TT³¹ and RR/YY³³ patterns. These motifs exhibit ~10-bp periodicity across eukaryotic genomes, such as yeast and nematodes^33,34.

One well-characterized DNA sequence motif is the WW/SS pattern (W = A/T, S = G/C), in which WW dinucleotides preferentially occupy minor-groove bending sites (minor-GBS), whereas SS dinucleotides are enriched at major-groove bending sites (major-GBS)³⁵. This pattern exhibits a ~10-bp periodicity and is structurally stabilized by conserved “sprocket” arginine residues that insert into minor grooves^36–38, favoring interactions with AT-rich DNA that narrows the minor groove^39–41. These periodic interactions form the structural basis for rotational positioning.

An opposite sequence motif, termed anti-WW/SS, places WW and SS dinucleotides in reversed groove contexts⁴², leading to less favorable histone–DNA interactions. We previously showed that anti-WW/SS nucleosomes are enriched in mammalian genic regions⁴³ and potentially stabilized by H4 and H2A histone N-tails⁴⁴, but their role in nucleosome-TF interactions remained unclear.

Here, we identify four distinct nucleosomal DNA patterns in the human genome (Types 1–4). Nucleosomes sharing the same pattern are rotationally in phase, whereas different types exhibit defined phasing relationships. We show that Type-1 (WW/SS) and Type-4 (anti-WW/SS) nucleosomes are preferentially associated with TF binding. The difference in their regional abundance correlates strongly with in vitro nucleosome occupancy (NO), and integration of this metric with in vivo NO reveals multiple modes of nucleosome-TF interaction.

Results

Four types of nucleosomes are widespread across the human genome

A major difference between nucleosome organization in vitro and in vivo is that nucleosomes are irregularly spaced in vitro but exhibit regular spacing in vivo^45–47. To assess whether nucleosomal DNA sequences also differ between these conditions, we extracted nucleosome core particle (NCP) DNA from a human in vitro nucleosome map⁴⁷ and an in vivo map from GM12878 cells⁴⁸, both generated using single-end sequencing (Supplementary Table 1). Aligned reads were extended to 147 bp to derive NCP DNA fragments.

Analysis of WW and SS dinucleotide frequencies across the 12 minor- and 12 major-groove bending sites (GBS) (Fig. 1a) revealed four distinct sequence patterns (Types 1–4; Fig. 1b and Supplementary Fig. 1). Type-1 nucleosomes display the canonical WW/SS pattern, with WW enriched in minor-GBS and SS enriched in major-GBS. Type-2 and Type-3 nucleosomes exhibit intermediate (“mixed”) patterns: Type-2 shows higher WW and SS frequencies in minor-GBS than in major-GBS, whereas Type-3 shows the opposite trend. Type-4 nucleosomes represent the inverse (anti-WW/SS) pattern, with WW enriched in major-GBS and SS enriched in minor-GBS.

Fig. 1. Minor- and major-groove bending sites and four WW/SS-based sequence patterns along nucleosomal DNA.

a Schematic representation of the nucleosome core particle (PDB: 1KX5) containing 147 bp of DNA. The DNA duplex is divided into two halves separated by the dyad. In the “ventral” half of the particle, base-pair centers are shown as large spheres, and the sugar–phosphate backbone is rendered as a yellow ribbon. In the “dorsal” half, sticks connect base-pair centers (only small segments are visible). minor-GBS and major-GBS are highlighted in blue and red, respectively, and are labeled according to their superhelical locations (SHLs). b Four nucleosomal DNA sequence patterns identified in GM12878 cells (in vivo). At each nucleosomal position, the frequencies of AA, TT, AT, and TA dinucleotides (WW, shown in blue) and GG, CC, GC, and CG dinucleotides (SS, shown in red) were computed and symmetrized relative to the dyad (dashed line). Three–base pair running averages for WW and SS frequencies are plotted for Type-1, Type-2, Type-3, and Type-4 nucleosomes. Blue and red shaded regions denote the minor- and major-GBS, respectively, corresponding to their SHL designations ±1, ±1.5, ±2, ±2.5,… etc. Type-1 nucleosomes correspond to the canonical WW/SS pattern, whereas Type-4 nucleosomes correspond to the inverted anti-WW/SS pattern. In vivo nucleosome data were obtained from GM12878 cells (GSM920558).

To validate the four sequence-type classification, we performed two additional analyses. First, we analyzed seven independent in vivo nucleosome datasets from diverse human cell lines (Supplementary Table 1). Second, we examined 147-bp NCP DNA fragments derived from a paired-end MNase-seq dataset in human lymphoblastoid cells⁴⁹. Both analyses consistently reproduced the four nucleosome sequence patterns (Supplementary Figs. 2–9). These results confirm the robustness of the four nucleosomal DNA sequence types across multiple independent datasets. Notably, the observed patterns are independent of the MNase-seq protocol (single-end versus paired-end) and cellular context (in vivo versus in vitro).

In both in vivo (GM12878) and in vitro datasets, Type-1 nucleosomes are the most abundant class, followed by Type-4, Type-2, and Type-3 nucleosomes (Supplementary Tables 2–3). In GM12878 cells, these four types account for 34.2%, 31.0%, 20.2%, and 14.6% of nucleosomes, respectively (Supplementary Table 4).

To assess the effect of sequencing depth, we calculated nucleosome-type fractions across eight in vivo datasets with varying coverage (Supplementary Table 5). The average fractions closely matched those observed in GM12878 cells, with Type-1 nucleosomes being most abundant (34.9%), followed by Type-4 (31.0%), Type-2 (20.0%), and Type-3 (14.2%). These results indicate that nucleosome-type distributions are robust across datasets with varying sequencing depths.

Comparison of nucleosome-type fractions reveals notable differences between in vitro and in vivo (GM12878) landscapes (Supplementary Table 4). The in vitro map contains ~3% more Type-1 nucleosomes and ~3% fewer Type-4 nucleosomes than the in vivo map, whereas differences for Type-2 and Type-3 nucleosomes are minimal (~1%). These results suggest that Type-1 nucleosomes are energetically favored in vitro, consistent with nucleosome positioning in the absence of chromatin-associated proteins, reflecting intrinsic histone–DNA preferences. Supporting this interpretation, we found that Widom’s “601”, “603”, and “605” sequences⁵⁰ all conform to the Type-1 pattern, which are known to exhibit high affinities for histone octamers relative to other sequences.

These observations imply that chromatin-associated proteins, such as TFs, chromatin remodelers, and RNA polymerase, redistribute nucleosomes with different DNA-encoded patterns. Notably, this redistribution exerts stronger effects on Type-1 and Type-4 nucleosomes than on Type-2 and Type-3. Consistently, the difference in abundances between Type-1 and Type-4 nucleosomes is 9.8% in vitro, nearly three-fold higher than the corresponding difference in vivo (3.3%). In contrast, the shift between Type-2 and Type-3 nucleosomes is more modest (7.6% in vitro versus 5.6% in vivo, Supplementary Table 4).

In summary, two key conclusions emerge. First, all four nucleosome types are broadly distributed across the human genome in both in vitro and in vivo conditions. Second, Type-1 and Type-4 nucleosomes exhibit substantially larger differences between in vitro and in vivo contexts than Type-2 and Type-3, indicating that chromatin proteins exert stronger regulatory influence on nucleosomes with these two sequence patterns.

The four types of nucleosomes exhibit distinctive genomic distribution

To investigate how the four nucleosome types are distributed across the genome in vivo and in vitro, we first computed normalized NO at every base pair and then calculated pairwise correlations between nucleosome maps (Fig. 2 and Supplementary Figs. 10, 11). Both in vivo and in vitro datasets revealed a consistent pattern. The full nucleosome map showed the highest correlation with the Type-1 map (0.74 in vivo, 0.76 in vitro; Fig. 2a and Supplementary Fig. 11a), consistent with the fact that Type-1 nucleosomes are the most abundant class (Supplementary Table 4). The second-highest correlation was observed between the full map and the Type-4 map (0.72 in vivo, 0.71 in vitro; Fig. 2d and Supplementary Fig. 11d). Lower correlations were observed for Type-2 (0.60 in vivo, 0.63 in vitro; Fig. 2b and Supplementary Fig. 11b) and Type-3 (0.45 in vivo, 0.54 in vitro; Fig. 2c and Supplementary Fig. 11c), reflecting their lower abundance. Notably, the ranking of these correlations mirrors the relative frequencies of the four nucleosome types (Supplementary Table 4).

Fig. 2. Comparison of in vivo nucleosome occupancy maps across the four nucleosome types.

Density dot plots show normalized nucleosome occupancy per base pair in the complete nucleosome dataset plotted against occupancy in each nucleosome type: Type-1 (a), Type-2 (b), Type-3 (c), and Type-4 (d). The color scale reflects the number of data points within each region of the plot. Pearson correlation coefficients were computed for each pairwise comparison to quantify the similarity between the full nucleosome map and individual nucleosome-type maps. In vivo nucleosome data were obtained from GM12878 cells (GSM920558).

Comparisons among the four nucleosome-type maps revealed several additional insights. First, Type-1 and Type-4 nucleosome maps exhibited the strongest cross-correlation in NO (0.51 in vivo, 0.47 in vitro; Supplementary Figs. 10c and 11g). Second, Type-2 and Type-3 maps showed the next highest correlation values (0.41 in vivo, 0.39 in vitro; Supplementary Figs. 10d and 11h). In contrast, all other pairwise comparisons displayed relatively low correlations, ranging from 0.13 to 0.24 in vivo (Supplementary Fig. 10a, b, e, f) and 0.27 to 0.32 in vitro (Supplementary Fig. 11e, f, i, j).

These results indicate that nucleosomes with specific DNA sequence patterns tend to co-localize across the genome. In particular, Type-1 and Type-4 nucleosomes frequently occur together, whereas Type-2 and Type-3 nucleosomes form a distinct grouping. This suggests that the four nucleosome types exhibit non-random spatial organization along chromatin, likely reflecting DNA-encoded rotational preferences and their interactions with chromatin-associated factors.

Nucleosomes with different sequence patterns have specific rotational phasing

To examine the relative positioning of the four nucleosome types across the genome, we applied the distance autocorrelation (DAC) and distance cross-correlation (DCC) functions described previously⁵¹. The DAC function measures the positional relationships between pairs of NCPs from the same set (e.g., all Type-1 nucleosomes)^33,47,52, whereas the DCC function measures positional relationships between NCPs from two different sets (e.g., Type-1 versus Type-4 nucleosomes).

We first computed the DAC function for the complete nucleosome set, which contains all four sequence patterns. In the in vitro dataset, weak peaks are observed at ~10 bp intervals (10, 20, 30, and 40 bp; Fig. 3a), consistent with previous analyses of the same dataset⁴⁷. These periodicities reflect overlapping nucleosome positions with similar rotational orientations, indicating that DNA is wrapped such that the helical face contacting the histone octamer repeats every full turn of the DNA helix. In contrast, the in vivo dataset showed no such periodicity (Fig. 3a), suggesting that the intrinsic rotational pattern is disrupted by chromatin-associated factors.

Fig. 3. Distance auto-correlation profiles and a model for the relative positioning of the four nucleosome types.

Distance auto-correlation (DAC) functions for all (a), Type-1 (b), Type-2 (c), Type-3 (d), and Type-4 (e) nucleosomes in in vivo (red) and in vitro (green). Peaks occur at ~10 × n bp (where n = 1, 2, 3, …), indicating ~10 bp periodicity and in-phase rotational alignment within each nucleosome type. f–k Distance cross-correlation (DCC) functions between nucleosome types. DCC functions were computed for all pairwise combinations of nucleosome types: Type-1 vs. Type-2 (f), Type-1 vs. Type-3 (g), Type-1 vs. Type-4 (h), Type-2 vs. Type-3 (i), Type-2 vs. Type-4 (j), and Type-3 vs. Type-4 (k). Peak positions define characteristic rotational offsets between nucleosome types. l Model of relative positioning of the four nucleosome types. Dashed lines indicate nucleosome start positions. Spacing relationships are inferred from DAC (b–e) and DCC (f–k) peak locations. In vivo nucleosome data were obtained from GM12878 cells (GSM920558). In vitro data were generated by reconstitution of human genomic DNA with histone octamers (GSE25133).

We then computed the DAC function for each nucleosome type separately. Strikingly, all four nucleosome types exhibited strong ~10 bp periodicity in both in vivo and in vitro datasets (Fig. 3b–e). Type-1 nucleosomes showed clear peaks at ~10 bp intervals (Fig. 3b), as did Type-2 (Fig. 3c), Type-3 (Fig. 3d), and Type-4 (Fig. 3e) nucleosomes. These results indicate that nucleosomes sharing the same sequence pattern are rotationally in phase, irrespective of cellular context.

The first peak corresponds to nucleosomes shifted by 10 bp relative to the reference position (starting at 0 bp). Subsequent peaks show slight deviations from exact 10 bp spacing (Fig. 3b–e), consistent with a helical periodicity of ~10.25 bp observed in yeast nucleosomes⁵¹. This value closely matches previous estimates, including 10.18 ± 0.05 bp per turn measured by hydroxyl radical footprinting of 5S DNA in reconstituted nucleosomes⁵³, as well as structural estimates of 10.30 bp per turn from the 147 bp NCP crystal structure⁵⁴ and 10.23 and 10.15 bp per turn from 146 bp NCP structures⁵⁵.

Analysis of the DCC function between different nucleosome types yielded several additional insights. First, DCC peaks were generally stronger for in vivo datasets than in vitro (Fig. 3f–k), likely to reflect the larger number of in vivo NCPs (Supplementary Tables 2 and 3). Second, Type-1 and Type-4 nucleosomes displayed strong out-of-phase peaks at ~5, ~15, ~25, ~35, and ~45 bp in both datasets (Fig. 3h), indicating a consistent half-turn rotational offset. Third, the same out-of-phase relationship was observed between Type-2 and Type-3 nucleosomes (Fig. 3i). Finally, all other pairwise comparisons produced two sets of peaks, one at ~13, ~23, ~33 bp and another at ~18, ~28, ~38 bp, across multiple type pairs (Fig. 3f, g, j, k), further demonstrating that nucleosomes with distinct sequence patterns have characteristic rotational spacing.

Because the sum of all individual DAC and DCC functions recapitulates the DAC function of the complete nucleosome population (Supplementary Tables 6 and 7), these patterns reveal that nucleosomes with different sequence motifs are arranged in a highly regular but previously hidden rotational organization.

Based on these results, we propose a rotational phasing model for the four nucleosome types, using Type-1 nucleosomes as the reference (Fig. 3l). Type-1 nucleosomes are spaced at ~10 bp intervals, reflecting their in-phase rotational alignment. Type-4 nucleosomes, which are out of phase with Type-1, are offset by ~5 bp (Fig. 3h). For Type-2 and Type-3 nucleosomes, relative positioning with respect to Type-1 follows distances of 10n + 3 bp or 10n + 8 bp (where n = 0, 1, 2, …), consistent with the two peak groups observed in the DCC analyses (Fig. 3f, g, j, k). Type-2 and Type-3 nucleosomes are themselves offset by ~5 bp, indicating an out-of-phase relationship (Fig. 3i). Notably, these positional relationships do not imply physical overlap of nucleosomes; rather, the correlations shown in the model (Fig. 3l) reflect ensemble patterns derived from different cells.

In summary, we identify four distinct nucleosomal DNA sequence patterns associated with nucleosome rotational positioning. Nucleosomes with the same sequence pattern align rotationally in phase, whereas nucleosomes with different patterns adopt specific, predictable rotational offsets relative to one another.

Type-1 and Type-4 nucleosomes are associated with TF binding

Because Type-1 and Type-4 nucleosomes show the largest differences between in vitro and in vivo maps (Supplementary Table 4), we hypothesized that these two nucleosome types influence TF binding in vivo, and that such effects should be reduced at motifs not bound by TFs. We used CTCF as a model system because its nucleosome organization has been extensively characterized^52,56–60. Recently, it has been shown that CTCF can readily displace a well-positioned nucleosome covering its motif, provided that the nucleosome is not marked by CpG methylation or repressive histone modifications⁵⁶.

We began by examining NO around CTCF motifs. As expected, in vivo NO shows a characteristic pattern: nucleosomes are depleted directly over CTCF-bound motifs (GSM935611) and form strongly ordered arrays in the flanking regions (solid line, Fig. 4a), consistent with previous studies^52,58,60. In contrast, unbound motifs exhibit a prominent occupancy peak with a normalized value of ~1.25 (dashed lines, Fig. 4a). The in vitro NO data reveal an occupancy peak of similar height for both bound and unbound motifs (Fig. 4b), indicating that CTCF motifs intrinsically favor nucleosome formation in the absence of CTCF. Together, these results suggest that CTCF binding promotes local nucleosome depletion, whereas, in its absence, the motif is preferentially occupied by a nucleosome.

Fig. 4. Differences in nucleosome occupancy and ΔNPS between CTCF-bound and unbound motifs.

In vivo (a) and in vitro (b) nucleosome occupancy profiles for CTCF-bound (solid lines) and unbound (dashed lines) motifs. Nucleosome occupancy at each nucleotide was normalized by dividing the number of nucleosome sequences covering that position by the genome-wide average number of nucleosome sequences per base pair. The dynamic range of in vivo occupancy (0.6–1.7) is substantially larger than that of in vitro occupancy (0.8–1.3). Dashed guides are provided to facilitate comparison between profiles. Fractions of Type-1 (c) and Type-4 (d) nucleosomes surrounding CTCF-bound (solid lines) and unbound (dashed lines) motifs. Fractions were computed in 10 bp bins across the motif-centered window. e ΔNPS profiles for CTCF-bound (solid line) and unbound (dashed line) motifs. Two-tailed Student’s t tests were used to compare mean nucleosome-type fractions (c, d) and ΔNPS values (e) across all human chromosomes between bound and unbound motif sets, as well as between bound motifs and their ±1 kb flanking regions. In vivo nucleosome data were obtained from GM12878 cells (GSM920558). CTCF ChIP–seq data (GSM935611) were also derived from GM12878 cells. The in vitro nucleosome dataset (GSE25133) was generated by reconstituting human genomic DNA with histone octamers.

We next examined the distribution of the four nucleosome types around CTCF-bound motifs in vivo. Type-1 and Type-4 nucleosomes show clear changes at the motif relative to flanking regions (~2–3%; Supplementary Fig. 12a), whereas Type-2 and Type-3 nucleosomes exhibit minimal variation (<1%; Supplementary Fig. 12b), suggesting that Type-1 and Type-4 nucleosomes play a dominant role in shaping local nucleosome organization.

To assess whether the observed ~2–3% differences in Type-1 and Type-4 fractions in vivo are statistically significant, we performed two comparisons: between CTCF-bound and unbound motifs, and between motif centers and positions ±1 kb from the center. The ±1 kb regions were chosen because (i) they are minimally influenced by the sequence specificity of CTCF binding motifs, and (ii) they approximate the genomic background levels of Type-1 and Type-4 nucleosomes. CTCF-bound motifs exhibit a significantly higher fraction of Type-1 nucleosomes than unbound motifs (Student’s t-test, P < 10⁻⁷) and the ±1 kb flanking regions (P < 10⁻¹⁰) (Fig. 4c). Conversely, CTCF-bound motifs show a significantly lower fraction of Type-4 nucleosomes than unbound motifs (P < 10⁻³) and the ±1 kb regions (P < 10⁻¹⁰) (Fig. 4d).

To capture the opposing trends between Type-1 and Type-4 nucleosomes in vivo, we used the ΔNPS metric, defined as the percentage difference between Type-1 and Type-4 nucleosome positioning sequences (NPS) within a given region⁴³. CTCF-bound motifs exhibit significantly higher ΔNPS values than unbound motifs (P < 10⁻⁷) and their ±1 kb flanking regions (P < 10⁻¹¹) (Fig. 4e). Together, these results demonstrate that the differences in Type-1 and Type-4 nucleosome fractions are statistically significant. Notably, nucleosomes and CTCF cannot occupy the same DNA molecule simultaneously at bound motifs. Accordingly, the patterns shown in Fig. 4c–e reflect ensemble averages across cell populations rather than co-occupancy within individual molecules.

Plotting the in vitro NO profile revealed a pronounced peak centered on CTCF motifs, closely resembling the ΔNPS profile (green and blue lines in Supplementary Fig. 13a). This concordance suggests that ΔNPS captures the intrinsic propensity for nucleosome formation around CTCF-binding sites and reflects, at least in part, the contribution of DNA sequence to nucleosome positioning in chromatin. These findings further imply that, in the absence of CTCF binding, the motif region is occupied by a nucleosome, consistent with previous reports^56–59.

To exclude the possibility that nucleosomal signals at bound CTCF sites arise from MNase protection by CTCF rather than bona fide nucleosomes, we validated our results using 147-bp nucleosomal DNA fragments derived from paired-end MNase-seq datasets⁴⁹ (Supplementary Fig. 14a, c–e). In addition, NO profiles were recalculated while accounting for CTCF motif directionality (Supplementary Fig. 14b). These analyses revealed only minor asymmetry in nucleosome organization around CTCF-bound motifs, consistent with previous studies⁶⁰.

In vivo nucleosome occupancy and ΔNPS exhibit distinct patterns around TF binding sites

To investigate how TFs influence nucleosome organization at their genomic binding sites, we compared, for each TF, its in vivo NO profile with its ΔNPS profile. For CTCF, we observed a clear DIP/PEAK pattern: in vivo NO shows a pronounced dip over bound motifs (GSM935611), while the ΔNPS profile shows a corresponding peak (red and blue lines in Supplementary Fig. 13a). Because GC content has previously been linked to in vitro NO at TF sites⁶¹, we also examined GC content around CTCF motifs. Indeed, GC content peaks at the motif center, mirroring the in vitro NO profile (Supplementary Fig. 13c).

A second pattern, DIP/DIP, was observed for SPI1: both in vivo NO and ΔNPS exhibited a dip at bound motifs (Supplementary Fig. 13b). The nearest ΔNPS peak is located ~70–80 bp away from the center of the motifs. The in vitro NO profile also shows a dip, correlating well with ΔNPS. However, GC content behaves differently here: it shows a peak at the motif (Supplementary Fig. 13d), indicating that GC content does not correlate with intrinsic NO in this case.

Two additional patterns were identified in other TFs. NFIC displayed a PEAK/PEAK pattern, where both in vivo NO and ΔNPS profiles show peaks at bound motifs (Supplementary Fig. 13e). NFYA showed a PEAK/DIP pattern, with in vivo NO showing a peak but ΔNPS showing a dip (Supplementary Fig. 13f). For both TFs, the in vitro NO profile matches the ΔNPS profile. GC content correlates with intrinsic NO for NFIC (Supplementary Fig. 13g), but not for NFYA (Supplementary Fig. 13h).

Using these four patterns, we classified 512 mammalian TFs and 232 yeast TFs into four groups, DIP/PEAK, DIP/DIP, PEAK/PEAK, and PEAK/DIP, according to their in vivo NO and ΔNPS profiles (Supplementary Tables 8–10). Representative average profiles for each group are shown in Fig. 5a–d. We also identified an AMBIGUOUS category for TFs in GM12878 cells lacking either a clear peak or dip in their in vivo NO profiles (Supplementary Fig. 15).

Fig. 5. Representative nucleosome occupancy and ΔNPS profiles for the four TF pattern groups.

Average profiles of in vivo nucleosome occupancy (red), in vitro nucleosome occupancy (green), and ΔNPS (blue) are shown for TFs classified into the DIP/PEAK (a), DIP/DIP (b), PEAK/PEAK (c), and PEAK/DIP (d) groups. For GM12878 cells, the DIP/PEAK, DIP/DIP, and PEAK/PEAK groups include 70, 39, and 14 TFs, respectively. Because GM12878 contains no TFs in the PEAK/DIP group, 15 TFs from HeLa cells belonging to this category were used for (d). All profiles are symmetrized relative to the motif center (position 0). In the DIP/DIP group (b), the ΔNPS peak closest to the motif occurs approximately 70–80 bp from the center. In vivo nucleosome data were obtained from GM12878 cells (GSM920558). In vitro data were generated by reconstitution of human genomic DNA with histone octamers (GSE25133). Corresponding ChIP–seq datasets are listed in Supplementary Table 8.

We classified the 512 mammalian TFs into two datasets: a stringent set containing 410 TFs and a non-stringent set containing 512 TFs (Supplementary Table 9). The stringent set includes only TFs with a single dominant profile pattern across all examined cell lines. In contrast, the non-stringent set includes all TFs, including those with tied dominant patterns, which are assigned to each corresponding profile group.

Importantly, the distinct profile groups found in human TFs are conserved across species: the same categories are observed in mouse TFs (Supplementary Fig. 16) and yeast TFs (Supplementary Fig. 17). This indicates that these patterns likely represent evolutionarily conserved modes of nucleosome-TF interactions.

Discussion

Our analysis of 12 in vivo and in vitro nucleosome maps and 744 TFs across multiple species yields several key findings. First, we identify four distinct nucleosomal DNA patterns that are widespread across the human genome in both in vivo and in vitro contexts. Second, nucleosomes with these patterns exhibit specific rotational offsets, revealing previously unrecognized DNA-encoded rules for nucleosome positioning. Third, two patterns, Type-1 and Type-4, are strongly associated with TF binding, with greater enrichment or depletion at bound than unbound motifs. Fourth, the difference in their relative abundance, quantified as ΔNPS, correlates closely with in vitro NO around TF-binding sites. Finally, integration of ΔNPS with in vivo occupancy profiles reveals four distinct modes of nucleosome–TF interaction.

Based on our data, we propose a unified model for the modes of nucleosome-TF interactions in vivo (Fig. 6), derived from the stringent dataset. For simplicity, the model focuses on direct interactions between TFs and nucleosomes, although additional factors, such as ATP-dependent chromatin remodelers, are likely to contribute.

1. DIP/PEAK group (41% of TFs). These TFs have strong intrinsic nucleosomes located directly over their binding motifs. The in vivo depletion of nucleosomes suggests direct competition between TFs and these intrinsically positioned nucleosomes. In other words, TF binding displaces strong DNA-encoded nucleosomes.

2. DIP/DIP group (32% of TFs). These TFs have strong intrinsic nucleosomes positioned ~70–80 bp away from the motif center. Thus, motifs are situated either in linker DNA or near the nucleosome edge, making them easily accessible. The absence of a strong DNA-encoded nucleosome at the motif itself is likely to contribute to the NDRs observed in vivo.

3. PEAK/PEAK group (11% of TFs). These TFs have strong intrinsic nucleosomes centered on their motifs. The enrichment of nucleosomal DNA in vivo suggests that these TFs bind DNA while it remains wrapped around a nucleosome, binding “at the center” relative to the nucleosome ends, though not necessarily at the dyad. These TFs appear to recognize or stabilize a well-positioned nucleosome.

4. PEAK/DIP group (3% of TFs). These TFs lack strong DNA-encoded nucleosomes near their motifs (ΔNPS shows a dip), yet nucleosomes are enriched in vivo. This suggests that nucleosome positioning is imposed not by DNA sequence but by protein factors such as chromatin remodelers or even the TFs themselves.

Fig. 6. A model illustrating how DNA-encoded nucleosome patterns determine modes of nucleosome-TF interactions.

Schematic representations of the four major TF profile groups, DIP/PEAK, DIP/DIP, PEAK/PEAK, and PEAK/DIP, highlight how intrinsic DNA sequence patterns shape nucleosome positioning and influence transcription factor binding. Percentages reflect the proportion of TFs in the stringent dataset. The AMBIGUOUS group (13% of TFs) is not shown because it lacks a clear peak or dip in in vivo nucleosome occupancy at motif centers. Their modes of nucleosome interaction appear heterogeneous or context-dependent and require further investigation.

Our data indicate that, for DIP/DIP TFs, NDRs may arise because the local DNA sequence disfavors the formation of strong, stable nucleosomes. In this case, the genomic region is intrinsically nucleosome-poor, and TF binding does not require competition with nucleosomes. In contrast, TFs in the DIP/PEAK group are associated with strong intrinsic nucleosomes positioned over their motifs, as reflected by high ΔNPS values. For these TFs, NDRs likely result from direct competition, in which DNA-encoded nucleosomes are displaced or excluded upon TF binding. Together, these results show that NDRs can arise through at least two distinct sequence-driven mechanisms: (i) the absence of strong intrinsic nucleosomes (DIP/DIP) and (ii) TF–nucleosome competition at sites with strong intrinsic nucleosomes (DIP/PEAK).

Our analyses validate the existence of four nucleosomal DNA types across multiple independent in vivo and in vitro MNase-seq datasets, with Type-1 and Type-4 being the most prevalent patterns. This distribution is further supported by nucleosome maps generated using a chemical cleavage method based on hydroxyl radicals⁴³, indicating that Type-4 nucleosomes are unlikely to arise from artifacts such as asymmetric MNase trimming.

The mechanistic basis underlying the formation of Type-4 nucleosomes remains to be fully elucidated. However, our previous work⁴³ showed that these nucleosomes are enriched in mammalian genic regions, suggesting a potential role in facilitating transcription elongation. In the present study, we further observe that Type-4 nucleosomes are enriched around binding sites of TFs exhibiting the DIP/DIP pattern, consistent with a role in facilitating TF access. Together, these findings suggest that Type-1 and Type-4 DNA patterns may act as intrinsic sequence features that modulate interactions between nucleosomes and chromatin-associated proteins.

Methods

Genome-wide nucleosome positioning (MNase-seq) datasets

Both in vitro and in vivo nucleosome maps from yeast and higher eukaryotes were used in this study, comprising ten in vivo datasets and two in vitro datasets (Supplementary Table 1). The ten in vivo nucleosome maps included eight human datasets, one mouse dataset, and one yeast dataset. The human datasets were derived from eight cell lines: GM12878 (GSM920558), K562 (GSM920557), MCF-7 (GSE51097), H1-hESC (GSM1194220), HepG2 (GSM3718063), HeLa (GSE100401), HCT116 (GSM3889219), and IMR90 (GSM1095279). The mouse dataset was generated from mouse embryonic stem cells (mESC; GSM2183911), and the yeast dataset was obtained from cells grown to late-log phase in synthetic complete medium (GSM651368). Two in vitro nucleosome positioning datasets were also analyzed—one from humans (GSE25133) and one from yeast (GSE13622).

Nucleosome occupancy maps were obtained either directly from ENCODE or generated by aligning MNase-seq short reads to reference genomes using Bowtie2 (Supplementary Table 1). For example, occupancy tracks (bigWig format) for GM12878 and K562 were downloaded from ENCODE. For datasets without pre-computed occupancy files, MNase-seq reads in FASTQ format were aligned using Bowtie2 with parameters: --sensitive -I 0 -X 500 --no-discordant.

For single-end datasets, mapped reads were extended from their 5′ ends to a canonical fragment length of 147 bp. For paired-end datasets, read pairs with insert sizes between 120 and 180 bp were retained. These fragments accounted for 75%, 73%, 46%, and 31% of all paired reads in the H1-hESC, MCF-7, HepG2, and HCT116 MNase-seq datasets, respectively, and were used for downstream analysis. Nucleosomes on repetitive elements are included. For one paired-end dataset (GSE36979), positions of 147-bp nucleosomal DNA fragments were obtained from published data⁴⁹.

To generate occupancy profiles for each nucleosome type, nucleosomes were classified as Type-1, Type-2, Type-3, or Type-4 based on their DNA sequence patterns. For each human cell line, BAM files were obtained from UCSC or generated using Bowtie2, and corresponding SAM files were processed.

For single-end datasets, dyad positions were defined by adding 73 bp to the start position of mapped reads. For paired-end datasets, dyad positions were defined as the midpoints of mapped read pairs. In all datasets, nucleosomal DNA fragments derived directly from MNase-seq reads were used to capture alternative nucleosome positioning within a cell population. Nucleosomal DNA fragments (147 bp) centered on dyads were then extracted from the reference genome and assigned to one of the four sequence types.

Genome-wide NO for each type was computed from dyad counts using a center-weighted scheme⁵⁸. This approach provides improved boundary resolution compared to alternative methods⁵⁸ and produces smoothing of dyad counts consistent with previously generated NO maps (e.g., GM12878; GSE35586)⁴⁸.

The mESC dataset was provided as center-weighted scores, which were converted to BedGraph format using the same center-weighting scheme⁵⁸ prior to downstream analysis. For yeast datasets, NO profiles were generated directly from dyad counts using the same center-weighting procedure⁵⁸.

Classification of nucleosomes with different NPS patterns

Nucleosomal DNA sequences were classified into four types, e.g., Type-1, Type-2, Type-3, and Type-4, based on the relative abundance of WW and SS dinucleotides in minor- and major-GBS⁴³. Type-1 nucleosomes exhibit the canonical WW/SS pattern. For a 147-bp NCP fragment, if (1) the total number of WW dinucleotides in the 12 minor-GBS is greater than that in the 12 major-GBS, and (2) the total number of SS dinucleotides in the 12 major-GBS is greater than that in the 12 minor-GBS, the fragment is classified as Type-1. Type-4 nucleosomes display the anti-WW/SS pattern, the inverse of Type-1. A 147-bp fragment is assigned to Type-4 if (1) WW dinucleotides are more abundant in the 12 major-GBS than in the minor-GBS, and (2) SS dinucleotides are more abundant in the 12 minor-GBS than in the major-GBS. Type-2 and Type-3 nucleosomes represent “mixed” patterns. Type-2 fragments contain more WW and more SS dinucleotides in minor-GBS than in major-GBS. Type-3 fragments contain more WW and more SS dinucleotides in major-GBS than in minor-GBS.

For any genomic region, we quantify the relative contributions of the two complementary sequence patterns using ΔNPS, defined as ΔNPS = Type 1 (%) – Type 4 (%), which reflects the balance between WW/SS and anti-WW/SS nucleosome patterns.

Type-1 through Type-4 nucleosomes were identified in both in vivo (GM12878; Supplementary Table 2) and in vitro datasets (Supplementary Table 3). Overall, the number of nucleosomes detected in the in vitro dataset is approximately one-third of that detected in vivo (compare Supplementary Tables 2 and 3).

Genome-wide protein-DNA interaction (ChIP-seq) datasets

In total, 2235 human and mouse ChIP–seq datasets were obtained from ENCODE across eight human cell lines (Supplementary Table 1). For 1202 datasets (53.8%), position-specific scoring matrices (PSSMs) for TF binding motifs were retrieved by matching TF names to entries in the HOCOMOCO v11, JASPAR 2022, and FactorBook databases. These PSSMs were used to represent the sequence variability of TF binding motifs. For each of the 1202 datasets, processed BED files containing mapped ChIP–seq fragments were downloaded from ENCODE or literature for downstream analysis (Supplementary Tables 8 and 10).

TF binding sites were identified by scanning each ChIP fragment with the corresponding PSSM using MEME suite tools. When multiple motif hits were present within a fragment, the highest-scoring site was retained for downstream analysis. For yeast, we used the processed ChIP-seq dataset^62,63 in which the center of the binding motif was already provided for each ChIP fragment.

For CTCF, unbound motifs were defined as follows. The human genome was scanned using the CTCF position-specific scoring matrix (PSSM; CTCF_human_h11MO.meme) with FIMO at a significance threshold of 10⁻⁴. CTCF-bound sites were obtained from ChIP–seq data in GM12878 cells (GSM935611), with processed BED files downloaded from ENCODE. Predicted motifs overlapping ChIP-defined binding sites were excluded to generate a set of unbound motifs. From this set, an equal number of unbound motifs were randomly selected to match the number of ChIP-derived bound sites for downstream analysis.

Classification of TF patterns using in vivo nucleosome occupancy and ΔNPS profiles

Transcription factors were classified into five categories, DIP/PEAK, DIP/DIP, PEAK/PEAK, PEAK/DIP, and AMBIGUOUS, based on the patterns of in vivo NO and ΔNPS profiles centered on TF bound motifs.

Classification criteria were defined using the relative positions of global maxima and minima within each profile, normalized by the full dynamic range of the signal:

DIP/PEAK: The in vivo NO profile exhibits a global minimum (or a value below the lower one-third of its dynamic range), while the ΔNPS profile exhibits a global maximum (or a value above the upper one-third of its range) at the motif center.

DIP/DIP: Both the in vivo NO and ΔNPS profiles show a global minimum (or values within the lowest one-third of their respective ranges) at the motif center.

PEAK/PEAK: Both the in vivo NO and ΔNPS profiles display a global maximum (or values within the upper one-third of their ranges) at the motif center.

PEAK/DIP: The in vivo NO profile shows a global maximum (upper one-third of its range), while the ΔNPS profile shows a global minimum (lower one-third of its range) at the motif center.

AMBIGUOUS: TFs whose in vivo NO profiles exhibit no clear peak or dip, i.e., values at the motif center fall between the lower and upper one-third thresholds, were placed in the AMBIGUOUS group.

This classification scheme ensures that each TF is assigned based on robust, quantitative features of nucleosome organization and intrinsic DNA-encoded nucleosome strength around its binding motifs.

Pattern labeling

Some transcription factors exhibited more than one TF pattern across different cell lines. To derive a consensus label for each TF, we counted the occurrences of all patterns observed for that TF and assigned as its consensus classification the pattern(s) tied for the highest frequency. Thus, a TF could have a single consensus label (e.g., DIP/DIP) or multiple labels if ties occurred (e.g., DIP/DIP, DIP/PEAK, and AMBIGUOUS). In addition to this comprehensive dataset, we generated a stringent dataset in which TFs with more than one consensus label were removed. The stringent dataset, therefore, contains only TFs that exhibit a single, consistent pattern across all examined cell lines.

Inter-nucleosome distance correlation function

To quantify the relative spatial organization of nucleosomes, we applied DAC and DCC functions⁵¹. For each pair of nucleosomal DNA fragments, we computed the distance between their dyad positions. When identical nucleosome sequences occur multiple times at the same genomic location, their contributions are weighted multiplicatively; for example, if two sequences occur 5 and 10 times, respectively, their DAC contribution is counted as 5 × 10 = 50.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: F.C. Experiments: C.W.C. and S.U.N. Investigation: C.W.C., S.U.N., and F.C. Project administration: F.C. Software: C.W.C., S.U.N., and F.C. Writing—original draft: C.W.C and F.C. Writing—review & editing: C.W.C., S.U.N., I.A., J.M.F., J.R.P., P.P., and F.C. Project leads: F.C.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

F.C. discloses support for publication for this work from NIH [R15GM149587 and R21GM152740], and a Faculty Education and Development (FEAD) grant from the College of Science at RIT.

Data availability

All data used in the study are publicly available. The accession numbers of the data are provided in supplementary tables.

Code availability

All computational code and pipelines are available in GitHub https://github.com/rit-cui-lab/Hidden-DNA-codes-influence-the-interaction-between-nucleosomes-and-transcription-factors/tree/main

Competing interests

The authors declare no conflicts of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74448-4.