Native nucleosomes intrinsically encode genome organization principles

Native mononucleosome purification

We used the hydroxyapatite (HAP) based protocol with minor modifications⁵¹ (see Supplementary Note 1 for full details). In brief, we cultured mammalian cell lines, including human embryonic stem cells H1-hESC (WiCell), GM12878 (Coriell Institute) and ES-E14TG2a (a gift from Ian Chambers, University of Edinburgh), and collected approximately 100 million cells. Next, we purified the nuclei with 0.3% NP-40 buffer and performed MNase digestion at 37 °C for 10 min in the presence of protease-inhibitor cocktails and other deacetylation and dephosphorylation inhibitors. The soluble mononucleosomes were saved after centrifugation of the insoluble nuclei debris in a cold room. The nucleosome samples were incubated with hydroxyapatite slurry for 10 min, and then unbound proteins were removed by repetitive washing with intermediate salt buffers. Finally, the nucleosomes were eluted with phosphate buffer from the hydroxyapatite slurry. The eluted fraction was checked by extracting DNA from the nucleosome through phenol-chloroform extraction and running a 2% agarose gel. The HAP elution contained mononucleosomes, naked DNA and oligonucleosomes. We applied further size selection of mononucleosomes using Mini Prep Cell (Biorad) gel-based size-selection purification. The quality of the final mononucleosome sample was checked by running a 2% agarose gel and a 20% SDS–PAGE gel. The purified mononucleosomes were stored on ice in a cold room for less than a week before the condensation reaction, or they were frozen in liquid nitrogen with 20% glycerol for long-term storage at −80 °C. All cell lines used in this study were routinely tested for mycoplasma contamination and confirmed to be negative throughout the duration of the study. 

Nucleosome condensation assay

The purified native mononucleosome sample was extensively dialysed into 10 mM Tris pH 7.5 buffer through several buffer exchanges using an Amicon Ultra 10-kDa filter (MilliporeSigma). In each condensation reaction, the final concentration of nucleosome or DNA was 50 ng µl⁻¹ as DNA weight, and BSA was added to the final 0.2 mg ml⁻¹ to stabilize the nucleosome core particle. The condensation buffer condition was 10 mM Tris pH 7.5 with more salt depending on the condensing agents (50 mM NaCl for spermine and 250 mM NaCl for PEG (molecular weight, 8 kDa)). We prepared 8–16 samples with different concentrations of condensing agents simultaneously. They were incubated at room temperature for 10 min and centrifuged at 16,000g for 10 min, and the supernatant was saved. The soluble-nucleosome concentration was measured using a Nanodrop UV spectrometer, and the nucleosome sample integrity was checked by running the 2% agarose gel (Supplementary Fig. 1). The rest of the nucleosomes in the supernatant were saved for use in high-throughput sequencing.

Next-generation sequencing and library preparation

Using phenol-chloroform extraction, genomic DNA was extracted from the nucleosome, which was either the input control sample or the supernatant saved from the nucleosome condensation assay. The extracted DNA sample was then washed several times with distilled water using an Amicon Ultra 10-kDa filter (MilliporeSigma). Using the NEBNext Ultra II DNA library preparation kit (NEB), the DNA was adapter-ligated and indexed for Illumina next-generation sequencing (NGS). The final indexing PCR was conducted in 5–7 cycles. We used a HiSeq 2500 or a NovaSeq 6000 platform (Illumina) for 50 bp-by-50 bp pair-end sequencing. In each experimental condition, we sequenced the samples over multiple titration points to get data with 10-kb resolution but deeply sequenced a few selected titration points to achieved approximately 20× coverage of the entire human genome at single-nucleosome resolution. In this paper, we focused mainly on the titration points near complete depletion of the solution fraction, in which we could observe the highest contrast of nucleosome condensabilities with strong selection power (for example, [spermine] = 0.79 mM in Fig. 1b and [HP1α] = 6.25 µM in Extended Data Fig. 8a).

Genetic and epigenetic datasets

All the genome references and epigenetic data used in this work, including DNA methylation, histone ChIP-seq and Hi-C, are shown in Supplementary Tables 1–11.

Computation of genome-wide nucleosome condensability

First, we obtained coverage profiles along the genome for input control and for the supernatant sample of each titration after the alignment of pair-end reads on the hg38 human genome assembly using Bowtie2 software⁵². On the basis of the coverage profile of the input control data, the position of each mononucleosome was localized by calling the peaks or finding the local maxima of the coverage profile. Beginning by randomly choosing a peak, the algorithm searched for all peaks in both directions, not allowing overlaps of more than 40 bp between 147-bp peak windows. For each nucleosome peak, the area of coverage in a window (we picked 171 bp as the window size) was computed for both the control and supernatant samples. The ratio of supernatant versus input read coverage area was combined with the titration curve measured by a UV–VIS spectrometer during the nucleosome condensation assay to estimate the survival probability of nucleosomes in the supernatant after condensation. Then, the negative natural log of this survival probability was used as a condensability metric for each mononucleosome peak. For the finer regular sampling used in plotting metagene profiles, the genome was binned into a 171-bp window with 25-bp sliding steps to compute the coverage area and the condensability scores. For a larger scale, we binned the genome into 1 kb or 10 kb and counted the reads aligned onto each bin to compute the condensability scores as the negative natural log of the ratio of supernatant to input read counts or the estimated survival probability inferred from the titration data. To avoid taking log of zero values, we added one pseudo-count to each input and supernatant read counts during the condensability calculation.

Computation of a condensation point, c
_1/2

The condensation point, c_1/2, was computed by using the survival probabilities of nucleosomes in multiple spermine concentrations. For each 10-kb genomic bin, we estimated the nucleosome counts in the input and supernatants after condensation in different spermine concentrations. We obtained the data points of spermine concentrations versus the soluble fraction of nucleosomes and fitted them with a logistic function. We then defined c_1/2 as the spermine concentration when the soluble fraction was half of the input.

Using z-score computations as an enrichment metric

We used the z-score as the enrichment metric for genetic and epigenetic features. For example, we counted the number of CpG dinucleotides in each mononucleosome and standardized their distribution by subtracting the mean across all nucleosomes and dividing it by the standard deviation. Thus, each mononucleosome was assigned with a z-score of the CpG dinucleotide counts as the metric of how enriched or depleted the CpG was compared with the average in the unit of standard deviation. For the partitioned or grouped dataset of the quantile analysis, we used the averaged z-score for each partition as the enrichment metric.

Data stratification and conditional correlation

To minimize the confounding effects between the genetic and epigenetic features of nucleosome condensation, the data were divided into subgroups that had one varying test variable, but all other variables were constant. For example, to evaluate whether AT content was correlated with condensability, the data were divided into smaller groups with the same genetic and epigenetic features, such as H3K4me3 and CpG methylations, except for AT content. In each stratified subgroup, we checked the correlation between AT content and condensability. We then defined the conditional correlation between AT content and condensability as the weighted average of all correlations over the stratified subgroups, weighted according to the data size of each subgroup. In practice, it was difficult to obtain enough data for each stratified subgroup when the feature set is high dimensional. In this case, we discretized each genetic–epigenetic feature into a specific number. All histone ChIP-seq scores were discretized into 10 numbers, and other scores were discretized into 100 numbers.

NMF decomposition

The genetic–epigenetic features of all mononucleosomes in chromosome 1 were linearly decomposed into ten basis property classes using a Scikit-learn NMF Python package. The nucleosomes were clustered into each property class with the highest component value in linear decomposition.

Machine-learning models

First, we randomly selected 0.1 million nucleosomes from chromosome 1 for machine learning. For this dataset, the ridge regressor, supported vector regressor, gradient-boosting regressor, random-forest regressor and multilayer perception regressor were trained and validated using tenfold cross-validations. All machine-learning training and predictions were done using the Scikit-learn Python package. All analysis details are available and documented as IPython notebooks in our Github repository (https://github.com/spark159/condense-seq).

Predicting the condensability of mononucleosomes

The condensability scores of mononucleosomes, as measured in H1-hESC cell lines using a spermine concentration of 0.79 mM, were predicted as a linear combination of the condensability scores of each PTM library member nucleosome measured at the same spermine concentration. For each PTM, the ChIP-seq signals on mononucleosomes were normalized by dividing them by the average ChIP-seq signal of the nucleosomes on chromosome 1, enabling comparison of different histone modifications at the same magnitude. The average of three measurements was used as the condensability score for each PTM. We restricted our analysis to mononucleosomes with at least six different types of PTM to prevent condensability from being influenced predominantly by PTMs not analysed in this study. The linear model was constructed as follows:

where C_mono represents the predicted condensability of a mononucleosome, ChIP_PTM indicates the normalized ChIP-seq signal and C_PTM denotes the condensability of PTM-library nucleosomes. For further analysis, mononucleosomes were stratified using ChromHMM, and the predicted condensability of each chromatin state was compared with its measured counterpart (Extended Data Fig. 6d–f).

Nucleosome reconstitution with canonical human octamers

Individual human histones H2A, H2B, H3.1 and H4 were purchased from the Histone Source (Colorado State University) and the octamers were reconstituted and purified following the standard protocol⁵³. Then nucleosomes were reconstituted using Widom 601 DNA or purified genomic DNA by following the standard gradient salt-dialysis protocol⁵⁴. Nucleosomes were further purified using Mini Prep Cell (Bio-Rad) to eliminate naked DNA or other by-product contaminants. For the PTM-library condense-seq experiment, the background reconstituted nucleosomes were made of Widom 601 DNA designed to have the same length and sequence as in the PTM library but with different primer-binding sequences, so it could not be amplified along with the library members. For the reconstitution of genomic DNA from GM12878, the genomic nucleosomal DNA was carefully purified at a size of 150 bp by 6% PAGE purification (Bio-rad Mini Prep Cell) following the phenol-chloroform extraction of DNA from HAP-purified mononucleosomes. A histone octamer titration was required for each DNA batch because very small increments of octamer can induce aggregation and loss of mononucleosome yield. Reconstituted nucleosomes were further purified using a 6% polyacrylamide 29:1 Native PAGE column (Bio-Rad Mini Prep Cell). To increase the stability of mononucleosomes during PAGE separation, 0.02% NP40 was added to the column, running and elution buffers. Nucleosomes containing fractions were concentrated and stored on ice at 4 °C for immediate use.

Purification of the HP1α and HP1β tSUV39H1 complex

We expressed and purified HP1α following the previous protocol⁶. In brief, we expressed HP1α with a His₆ affinity tag in Escherichia coli Rosetta (DE3) strains (MilliporeSigma) at 18 °C overnight. After cell lysis, the protein was first purified by cobalt-NTA affinity purification. The His tag was then cleaved by TEV protease, which was removed by anion-exchange purification using a HiTrap Q HP column (GE Healthcare). The HP1α was further purified by size selection using a Superdex-75 16/60 size-exclusion column (GE Healthcare). The HP1β with a truncated SUV39H1 complex (HP1β tSUV39H1) was similarly purified following a previous protocol²⁰.

Nucleosome condensation assay of the PTM library

The PTM library was prepared as previously described³³. The nucleosome condensation reaction of the PTM library was performed similarly, as described for the native mononucleosomes. However, because of the limited amount of the PTM-library sample, we spiked only a 1% (v/v) sample amount of the library into 99% (v/v) of reconstituted human nucleosomes as background for the condensation reaction. For condensation experiments using HP1α, a final concentration of 50 ng µl⁻¹ of DNA or nucleosome (DNA weight) was used in the reaction buffer (10 mM Tris-HCl pH 7.5, 100 mM NaCl, 0.2 mg ml⁻¹ BSA) with 5% (v/v) PEG 8000 as a crowding agent. Various amounts of HP1α were added to start the condensation.

NGS library preparation and sequencing of the PTM library

The DNA sample was purified by phenol-chloroform extraction followed by several washes with distilled water using an Amicon Ultra filter (MilliporeSigma). The DNA library was then prepared for Illumina NGS sequencing by PCR using Phusion HF master mix (NEB) and custom indexed primers for the PTM library³³. During amplification, the background nucleosome DNA was not amplified because it has different primer-binding sequences. We used MiSeq (Illumina) for sequencing libraries with custom primers, following previous protocols³³.

Condensability calculation for the PTM library

The PTM library was de-multiplexed on the basis of the DNA hexamer barcodes by using a custom Python script and Bowtie2 aligner⁵². Then we approximated the nucleosome counts using information about the total soluble fraction, which was measured by a UV–VIS spectrometer, and the fraction of the individual members in the library, which was measured by Illumina sequencing. Finally, we computed the survival probability of each member in the library, which is the number of the remaining nucleosomes in the solution after condensation over input control. A negative log of survival probability was used for the condensability metric. For the PTM library, condensability averaged over many titration points was used as a condensability score for further analysis.

Nucleosome–nucleosome interaction-energy calculations

Coarse-grained molecular-dynamics simulations of chromatin were done using OpenMM software⁵⁵. Chromatin was modelled as beads-on-a-string polymers with each bead representing a genomic segment 25 kb long. Energy terms for bonds, excluded volume, spherical confinement and sequence-dependent contacts were defined. Sequence-dependent contact energies were parameterized using read counts from condense-seq experiments. Contact probability matrixes were computed from these simulation trajectories and compared with experimental Hi-C contact maps. Full simulation details are provided in the Supplementary Note 2.

Mouse CD8⁺ T cell culture and in vitro activation

Wild-type C57BL/6 mice and mice expressing Cre recombinase (CD4Cre) under the control of the CD4 promoter and Rosa^26eYFP were purchased from Jackson Laboratories, and Odc^flox/flox mice were purchased from the KOMP repository. For experiments involving epigenetic marks, the spleen of Odc^flox/flox or Odc^+/+ Rosa^26eYFP mice were used to isolate and transduce T cells in vitro. All mice were bred and maintained in specific pathogen-free conditions under protocols approved by the Animal Care and Use Committee of Johns Hopkins University, in accordance with the Guide for the Care and Use of Animals. Mice used for all experiments were littermates and were matched for age and sex (both male and female mice were used). Mice of all strains were typically 8–12 weeks of age. Naive CD8⁺ T cells were isolated from the spleens of mice 8–12 weeks old using a negative-selection CD8 T cell kit (MojoSort Mouse CD8 T Cell Isolation Kit) according to the manufacturer’s protocol. Isolated T cells (1 × 10⁶ per ml) were activated using plate-bound anti-CD3 (5 μg ml⁻¹) and soluble anti-CD28 (0.5 μg ml⁻¹) in T cell media (1640 Roswell Park Memorial Institute medium with 10% fatal calf serum, 4 mM l-glutamine, 1% penicillin/streptomycin and 55 μM β-mercaptoethanol) supplemented with 100 U ml⁻¹ rhIL-2 (Peprotech). Cells were cultured at 37 °C in humidified incubators with 5% CO₂ and atmospheric oxygen for 24 h after activation. After 48 h, T cells were removed from anti-CD3 and anti-CD28 and cultured at a density of 1 × 10⁶ per ml in rhIL-2 (100 U ml⁻¹) at 37 °C for 7 days, with a change of media and fresh rhIL-2 every 24 h. To inhibit ODC, cells were incubated with 2.5 mM DFMO for 24 h at day 6 of culture. Odc^−/−, wild-type and DFMO-treated cells were collected at day 7 for chromatin isolation and sequencing.

Lentiviral production and cell transduction

HEK293T cells were transfected using Lipofectamine 3000 (Thermo Fisher Scientific) with the lentiviral packaging vectors pCAG-eco and psPAX.2 plus Cre-expressing vector pLV-EF1-Cre-PGK-Puro (all obtained from Addgene). The produced lentivirus was collected from the supernatant of the cells. CD8⁺ naive T lymphocytes isolated from Odc^+/+ Rosa^26eYFP mice or Odc^flox/flox Rosa^26eYFP were transduced by centrifugation in the presence of polybrene (8 mg ml⁻¹) in a plate treated with anti-CD3 (5 μg ml⁻¹), soluble anti-CD28 (0.5 μg ml⁻¹) and 100 U ml⁻¹ rhIL-2. The virus was removed after 6 h and fresh media containing anti-CD28⁺ IL-2 was added again. After two days, the transduced cells were selected by flow cytometry and sorted by expression of YFP (Cre⁺ cells) in the CD8⁺ live-cell population and cultured in the presence of 100 U ml⁻¹ rhIL-2 for two more days.

Assessment of epigenetic marks by flow cytometry

Transduced CD8⁺ YFP⁺ sorted T cells from Odc^+/+ and Odc^flox/flox were fixed and stained for intracellular immunostaining. The measurement of the histone methylation and acetylation marks enrichment was done using flow cytometry for sorted CD8⁺ eYFP⁺ T cells from Odc^+/+ and Odc^flox/flox (wild type and KO, respectively) mice, and they were fixed for 60 min at room temperature using a FOXP3 permeabilization kit (eBioscience) and stained for 90 min with primary antibodies against H3K36me3 (Polyclonal, from Abcam), H3K4me3 (clone C42D8), H3K27ac (clone D5E4), H3K27me3 (clone C36B11), H3K9ac (clone C5B11) and rabbit monoclonal antibody IgG isotype control (DA1E) (all from Cell Signaling Technology unless stated otherwise) and stained for 30 min with donkey anti-rabbit IgG (H + L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor Plus 647 (Thermo) at room temperature. Cells were gated on diploid cells with ‘single’ DNA content based on FxCycle staining (Thermo Fisher) in the live-cell gate.

Histone PTM enrichment measurement

For the mass-spectrometry measurement, native mononucleosomes were purified from the GM12878 cell line and a nucleosome condensation assay was similarly performed using spermine (250 ng µl⁻¹ nucleosome, 0.079 mM spermine in 10 mM Tris-HCl pH 7.5 buffer at room temperature). The input/soluble/pellet nucleosome sample was washed several times in 10 mM Tris-HCl pH 7.5 buffer using an Amicon Ultra filter (10-kDa cut-off) to remove spermine and kept at 70 °C for 20 min to dissociate DNA from the histones. The free DNA was further removed in the desalting step of the mass-spectrometry process. About 20 µg of purified histone was derivatized using propionic anhydride⁵⁶ followed by digestion with 1 µg trypsin for bottom-up mass spectrometry. The desalted peptides were then separated in a Thermo Scientific Acclaim PepMap 100 C18 HPLC Column (250 mm length, 0.075 mm internal diameter, reversed-phase, 3 µm particle size) fitted on a Vanquish Neo UHPLC system (Thermo Scientific) using an HPLC gradient as follows: 2% to 35% solvent B (A = 0.1% formic acid; B = 95% MeCN, 0.1% formic acid) over 50 min, to 99% solvent B in 10 min, all at a flow rate of 300 nl min⁻¹. About 5 µl of a 1 µg µl⁻¹ sample was injected into a QExactive-Orbitrap mass spectrometer (Thermo Scientific) and a data-independent acquisition was carried on, as described previously⁵⁶. In brief, full-scan mass spectrometry (m/z 295–1,100) was acquired in an Orbitrap with a resolution of 70,000 and an AGC target of 1 × 10⁶. Tandem mass spectrometry was set in centroid mode in the ion trap using sequential isolation windows of 24 m/z with an AGC target of 2 × 10⁵, a CID collision energy of 30 and a maximum injection time of 50 ms. The raw data were analysed using in-house software, EpiProfile⁵⁷. The chromatographic profile and isobaric forms of peptides were determined using precursor and fragment-extracted ions. The data were output as peptide relative ratios (percentages) of the total area under the extracted ion chromatogram of a particular peptide form to the sum of unmodified and modified forms belonging to the same peptide with the same amino acid sequence. The log₂-transformed fold change in the peptide relative ratio in the soluble/pellet fraction versus the input was computed as the enrichment metric. Using the unmodified peptide as the reference, the difference in fold change between the PTM modified peptide and the unmodified peptide was computed and plotted as a heatmap.

Calibrated ChIP-seq

We followed a published ChIP protocol⁵⁸ with minimal modifications. Antibody-conjugated beads were prepared by adding 50 µl of Protein A beads per ChIP reaction (Thermo Fisher) to a 2 ml tube, washing twice with 1 ml of blocking buffer (0.5% BSA in PBS) and resuspending in 100 µl blocking buffer per ChIP reaction. Antibody was then added to the beads (4 µl of H3K27ac antibody (Novus ab4729) and 4 µl of H3K27me3 (Novus ab192985) plus 2 µg of spike-in antibody (ActiveMotif) per reaction), and the mixture was incubated with rotation for 1–3 h. Crosslinked cell pellets were resuspended in 4 ml of lysis buffer LB1 (50 mM HEPES, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% Igepal CA-630, 0.25% Triton X-100, pH adjusted to 7.5, 1× protease inhibitors) and incubated in LB1 for 10 min at 4 °C with rotation. Cells were then spun down at 2,000g, at 4 °C for 3 min. The supernatant was discarded and pellets were resuspended in 4 ml of LB2 (10 mM Tris-HCl pH 8, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, pH 8.0, 1× protease inhibitors) and incubated at 4 °C with rotation for 5 min, then spun down (with the same settings). The supernatant was removed and cells were then resuspended in 1.5 ml of LB3 (10 mM Tris-HCl pH 8, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% Na-deoxycholate, 0.5% N-lauroylsarcosine, pH 8.0, 1× protease inhibitors) and transferred to 2-ml tubes. Sonication was performed using a Fisher 150E Sonic Dismembrator with the following settings: 50% amplitude, 30 s on, 30 s off for 12 min total time. The sonicated sample was spun down at 20,000g and 4 °C for 10 min, and the supernatant was transferred to a 5 ml tube. Then, 1.5 ml of LB3 (with no protease inhibitor), 300 µl of 10% Triton X-100, and 120 ng of Drosophila spike-in chromatin (ActiveMotif) per 25 µg of ChIP’ed chromatin were added to each sample. The entire solution was mixed by inversion. The 2-ml tubes containing antibody-conjugated beads were placed on a magnetic rack, washed three times with 1 ml of blocking buffer, and resuspended in 50 µl of blocking buffer per ChIP reaction. We then transferred 50 µl of antibody-conjugated beads to each ChIP reaction and incubated them overnight at 4 °C with rotation. ChIP samples were transferred to a 1.5 ml LoBind tube, placed on a magnetic stand and washed six times with 1 ml RIPA buffer (50 mM HEPES, 500 mM LiCl, 1 mM EDTA, 1% Igepal CA-630, 0.7% Na-deoxycholate, pH 7.5) and once with 1 ml TBE buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl). The supernatant was discarded, and the beads were eluted in 50 µl elution buffer EB (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) and incubated at 65 °C overnight with shaking at 1,000 rpm. We then added 40 µl TE buffer to the mixture to dilute the SDS, followed by 2 µl of 20 mg ml⁻¹ RNaseA (New England BioLabs), and samples were incubated for 15 min at 37 °C. Then, 4 µl of 20 mg ml⁻¹ Proteinase K (New England BioLabs) was added and the samples were incubated for 1 h at 55 °C. The genomic DNA was column purified and eluted in 41 µl of nuclease-free water. Sequencing libraries were prepared using the NEB Next Ultra II End Repair/dA-Tailing Module (New England BioLabs), using half volumes. Libraries were amplified with 10 (H3k27ac) or 13 (H3k27me3) cycles of PCR using single indexed primers. ChIP’ed DNA samples were then pooled, quantified with QuBit and qPCR (BioRad), and sequenced on a NextSeq 1000 Illumina machine using paired 2 × 50 bp reads. Reads were demultiplexed after sequencing using bcl2fastq and aligned to the mm10 genome using bowtie2. Samtools63 was used to filter for a mapping quality greater than or equal to 25, remove singleton reads, convert to BAM format and remove potential PCR duplicates and index reads.

Two-colour smFRET imaging for nucleosome unwrapping

Biotinylated Cy3/Cy5 20N20 mononucleosomes (25 mM Hepes-KOH pH 7.6, 5% glycerol, 0.017% NP-40, 70 mM KCl, 3.6 mM MgCl2 and 0.1 mg ml⁻¹ BSA) were incubated in surface-functionalized chambers for 2 min. Free nucleosomes were flushed out with dilution buffer containing imaging additives (oxygen-scavenging system: 0.8% w/v dextrose, 2 mM Trolox, 1 mg ml⁻¹ glucose oxidase (Sigma-Aldrich) and 500 U ml⁻¹ catalase (Sigma-Aldrich)). Basal nucleosome fluorescent emission was recorded to control density and FRET signal before the addition of spermine. A total of 10 short movies (100 ms exposure time) of 20 frames each were taken (10 frames using Cy3 excitation and 10 frames using Cy5 excitation). Spermine was introduced to the imaging chamber in dilution buffer containing imaging additives and incubated for 10 min. Short movies were taken using the settings explained above. FRET histograms were generated from donor and acceptor fluorescent intensities of single molecules. The details of the nucleosome construct and single-molecule imaging conditions can be found in Supplementary Note 3.

Single-molecule nucleosome pull-down assay

Biotinylated Cy3-H2A(K120C) 20N0 mononucleosomes were dialysed into 10 mM Tris pH 7.5 buffer through three buffer exchanges using an Amicon Ultra 10-kDa filter (MilliporeSigma). Nucleosomes were diluted to 7.5 nM and BSA was added to a concentration of 0.2 mg ml⁻¹. For condensation, 5 nM mononucleosomes were mixed with 0.4 mM spermine in 10 mM Tris pH 7.5 and 50 mM NaCl. The reaction was covered from light and incubated at room temperature for 10 min. Before immobilization, spermine-condensed nucleosomes were mixed and immediately diluted 50 times in 10 mM Tris pH 7.5, 50 mM NaCl and 0.4 mM spermine (pull-down buffer). Dilution flowed into neutravidin functionalized chambers and incubated for 10 min with the quartz slide facing down. The chamber was washed with pull-down buffer including imaging additives. Short movies of 20 frames (100 ms exposure time) were taken using Cy3 excitation. Laser intensity was regulated to control the intense fluorescent signal from large condensates immobilized on the single-molecule surface. A control experiment was done in which spermine was removed from the condensation reaction and pull-down buffers. Nucleosomes were diluted 500-fold for immobilization and only single nucleosome spots were observed. Detailed information of nucleosome constructs and single-molecule imaging conditions are in Supplementary Note 3 and Supplementary Table 12.

Reporting summary

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