Sensitive Automated Measurement of Histone-DNA Affinities in Nucleosomes.

The DNA of eukaryotes is wrapped around histone octamers to form nucleosomes. Although it is well established that the DNA sequence significantly influences nucleosome formation, its precise contribution has remained controversial, partially owing to the lack of quantitative affinity data. Here, we present a method to measure DNA-histone binding free energies at medium throughput and with high sensitivity. Competitive nucleosome formation is achieved through automation, and a modified epifluorescence microscope is used to rapidly and accurately measure the fractions of bound/unbound DNA based on fluorescence anisotropy. The procedure allows us to obtain full titration curves with high reproducibility. We applied this technique to measure the histone-DNA affinities for 47 DNA sequences and analyzed how the affinities correlate with relevant DNA sequence features. We found that the GC content has a significant impact on nucleosome-forming preferences, but 10 bp dinucleotide periodicities and the presence of poly(dA:dT) stretches do not.

Introduction

Eukaryotes organize their genomes by wrapping their DNA around a complex of basic proteins called histones. Approximately 147 base pairs of DNA are wrapped 1.7 times around a histone octamer to form a nucleosome, which constitutes the basic unit of chromatin (Khorasanizadeh, 2004). Nucleosomes show a clear organization with respect to the DNA sequence especially around promoters, where a nucleosome-depleted region at the transcription start site is typically flanked by well-positioned nucleosomes on either side, referred to as the −1 and +1 nucleosomes, and by an array of regularly spaced nucleosomes downstream over the gene body (Jiang and Pugh, 2009, Lai and Pugh, 2017). It is commonly accepted that the in vivo positioning of nucleosomes is the result of the interplay of multiple determinants, which include the DNA sequence, but also the action of nucleosome remodelers, transcription factors, and other DNA-binding proteins/complexes such as the RNA polymerase II transcription machinery (Klemm et al., 2019). Among sequence parameters influencing nucleosome formation, the best studied are the GC content (Fenouil et al., 2012, Tillo and Hughes, 2009) and the base pair (bp) periodicity of flexible dinucleotides forming the contact sites of the nucleosomal DNA with the histone octamer (Drew and Calladine, 1987, Jin et al., 2016, Klug and Lutter, 1981, Shrader and Crothers, 1990, van der Heijden et al., 2012). Owing to the double helical nature of DNA, the same face contacts the histone complex every ten to eleven base pairs; nucleosome formation is thus facilitated by the periodical occurrence of alternatingly flexible (like AT, AA, or TT) and stiff (like GC) dinucleotide steps (kink-and-slide model) (Vasudevan et al., 2010). In addition, it is believed that the presence of short homopolymeric stretches of deoxyadenosine nucleotides referred to as poly(dA:dT) or dA:dT tracks, which are intrinsically stiff and are frequently found in nucleosome-depleted regions, impedes nucleosome formation (Jin et al., 2018, Raveh-Sadka et al., 2012, Segal and Widom, 2009). Overall, much effort has been devoted in recent years to characterize nucleosome sequence preferences in vitro (Krietenstein et al., 2012, Segal et al., 2006) and in vivo (Jin et al., 2018, Kaplan et al., 2010) and to predict nucleosome positions in the genome on this basis (Segal et al., 2006, Tillo et al., 2010). However, the precise degree to which the underlying sequence directs nucleosome positioning, and which sequence features are most important, is still a matter of debate (Jin et al., 2018, Kaplan et al., 2010, Struhl and Segal, 2013, Zhang et al., 2009). This is in part due to the fact that most studies have relied on MNase digestion of genomic DNA followed by deep sequencing and fragment counting to derive nucleosome occupancy values. This approach is subject to confounding effects of biological activity and/or the inherent biases of the MNase enzyme. It would thus be immensely useful to measure histone-DNA binding affinities directly, without other intervening features. However, few such data exist, mainly because of the lack of efficient experimental techniques.

The most widely used method for determining histone-DNA binding free energies in vitro was pioneered by Schrader, Crothers, and Widom (Lowary and Widom, 1998, Shrader and Crothers, 1990). In this approach, nucleosomes are typically reconstituted from purified core histones and DNA of mononucleosomal length by dialysis, or alternatively by stepwise dilution. A competition experiment is conducted using a mixture of the DNA of interest and low amounts of (usually radio- or fluorescently) labeled DNA, which serves as a reference to compare the nucleosome-forming capacity of different DNA sequences. Reconstituted samples are analyzed on polyacrylamide or agarose gels by an electrophoretic mobility shift assay (EMSA) (Thastrom et al., 1999) to calculate the fraction of reference DNA that reconstitutes into nucleosomes in a given DNA mixture. In this fashion, relative affinities (free energies) of histone octamers to differing DNA fragments can be determined. The assay works reliably but suffers from significant limitations: both the nucleosome reconstitution and the EMSA readout steps are time consuming and difficult to parallelize. This entails that histone-DNA free energies are usually determined using only a single concentration per sequence, raising issues regarding accuracy and reproducibility. As a result, affinity data are currently only available for a relatively limited number of sequences. Moreover, most studies have focused on artificially designed nucleosomal DNA sequences with strong, non-physiological binding properties (Eslami-Mossallam et al., 2016), whereas native genomic sequences have not been investigated extensively. Thus, there is a clear need for more accurate and comprehensive measurements of histone-DNA free energies.

In the current study, we developed a method to measure histone-DNA binding free energies in nucleosomes with high reproducibility and at medium throughput. Our technique is based on the classical approach but offers substantial improvements: (1) we carry out the competitive reconstitution of nucleosomes by small dilution steps using an automated liquid handling system and (2) we determine the fraction of bound DNA by measuring fluorescence anisotropy (FA) with an adapted epifluorescence microscope, following the approach we recently reported for measuring transcription factor-DNA binding energies (Jung, 2018, Jung, 2019). The high parallelization of the nucleosome reconstitution and the fast and sensitive fluorescence readout allowed us to obtain full titration curves for each individual histone-DNA interaction instead of single concentration measurements, resulting in a more accurate determination of binding free energies. We demonstrate the utility of this approach by measuring histone-DNA binding free energies for 47 different DNA sequences, including Drosophila melanogaster (D. mel) genomic nucleosomal sequences, synthetic DNA sequences derived from D. mel enhancers, and additional nucleosomal DNA sequences tested in previous studies (Cao et al., 1998, Filesi et al., 2000, Shrader and Crothers, 1990, Thastrom et al., 1999). We show that the free energies of nucleosome formation can be measured accurately and cover a wide dynamical range. Furthermore, we explored how the free energies correlate with DNA features such as the GC content, the 10 bp periodicity of flexible and stiff dinuleotides, and the number of short poly(dA:dT) stretches. We found GC content to be the most predictive feature in our data, explaining ∼30% of the variation of the free energies.

Results

Automated Assay to Determine Free Energies of Nucleosome Formation

We determined the free energies of nucleosome formation by titration of unlabeled DNA sequences (147–300 bp in length) competing for nucleosome reconstitution with a fluorescently labeled DNA reference in low amount (Figure 1A). Histones and competing DNAs are initially mixed at high salt concentration; in a slow dilution process, buffer is added in small steps, thus gradually increasing the interaction strength. The use of an automated liquid handling system permits carrying out this nucleosome reconstitution over a long period of time (12 h, typically overnight), helping to approximate thermodynamic equilibrium, and greatly improves the reproducibility and throughput of the assay. To ensure reproducibility of the reconstitution reaction, we had to limit evaporation (which typically occurs at borders and edges of well plate containers) and provide for a stable temperature. To this end, we designed and fabricated a metal block (Figure S5) accommodating up to 42 individual low protein binding tubes with a heated lid (Figure 1B and Transparent Methods). The metal block improves temperature stability and uniformity during the nucleosome formation process, while the heated lid reduces evaporation by preventing condensation at the lid.

Figure 1

Competitive Nucleosome Reconstitution and Binding Assay

(A) Schematic representation of the assay workflow.

(B) Image of the robotic system used, with enlarged image of the custom metal block (see also Figure S5) with its heated lid.

(C) Schematic depiction of the fluorescence microscopy setup used for the FA readout.

(D) Histone-DNA affinity single titration curves for three different competitor sequences, together with their corresponding fits (dashed lines), for a weak, medium, and strong binder as indicated. Error bars refer to the standard deviation of the FA measurements, which were used to weight the individual points in the fitting procedure (Transparent Methods).

(E) Assay validation, using DNA sequences measured in previous studies (Cao et al., 1998, Filesi et al., 2000, Shrader and Crothers, 1990, Thastrom et al., 1999). The relative free energies determined in previous studies are plotted against the corresponding values obtained in this study; dotted line shows linear regression; Pearson correlation coefficient R = 0.99.

The readout of the fraction of bound versus unbound fluorescently labeled reference DNA was carried out using FA (Roehrl et al., 2004) (Figure 1C) instead of the typical EMSA, thus offering the advantages of a fast and sensitive fluorescence readout. In brief, FA measures the rotational speed of a fluorescently labeled molecule: high FA indicates the presence of larger, and therefore slowly rotating, molecular complexes in the solution, in this case nucleosomes with incorporated fluorescently labeled reference DNA. If an unlabeled competitor DNA outcompetes this labeled reference for histone binding, the FA will decrease, as a higher proportion of the fluorescently labeled reference DNA molecules are unbound and thus rotating faster. Different FA levels (Figure S1) can therefore be used to calculate the fractions of bound versus unbound labeled reference DNA.

After nucleosome reconstitution, the samples are transferred to 96-well microscopy plates and FA is measured in each well using the microscopy setup described for HiP-FA in Jung, 2018, Jung, 2019 (Figure 1C). By performing the reconstitution with different unlabeled competitor concentrations, we obtain a full titration curve for each DNA sequence. The data can be fitted using the Hill equation, as shown for a weak (ΔΔG = 9.7 kJ·mol−1; DmeI08), a medium (ΔΔG = 7.2 kJ·mol−1; DmeI28), and a strong (ΔΔG = −2.2 kJ·mol−1; 601) competitor sequence (Figure 1D).

To validate our assay, we measured seven nucleosomal sequences that had been tested in other studies (601, TGGA-2, TAND-1, TG, Bombyx, 5S, and TG-T [Cao et al., 1998, Filesi et al., 2000, Shrader and Crothers, 1990, Thastrom et al., 1999]). As shown in Figure 1E, our results are in excellent agreement with the previously measured free energies, calculated relative to their respective reference sequences. To evaluate our FA readout, we also measured affinities for selected sequences by EMSA. EMSA with fluorescent labels is prone to quenching effects; nevertheless, we find a reasonable agreement between these measurements and the affinities obtained by FA (Figure S2).

Applying the Method to Genomic and Synthetic Nucleosomal DNA Sequences

Most studies measuring free energies of nucleosome formation in vitro focused on DNA sequences that were selected or designed to cover a large range of affinities. In fact, the strongest known binders, like the well-known 601 sequence, are the result of heavy selection and are not found in native genomic DNA. A few naturally occurring sequences have been tested, showing lower affinities and a smaller dynamical range than synthetic sequences (Thastrom et al., 1999), but to date no comprehensive direct measurement of histone affinities of genomic nucleosomal sequences has been conducted. Taking advantage of the throughput and accuracy of our technique, we therefore decided to test 29 endogenous nucleosomal DNA sequences from D. mel (denoted Dmel01 - Dmel29); the sequences were selected from −1 nucleosomes as determined by MNase-Seq (unpublished data), which are well positioned but typically contain fewer cis-regulatory sequence elements than the +1 nucleosomes and whose positioning is presumably less affected by biological activity (Mavrich et al., 2008a, Mavrich et al., 2008b). The sequences were chosen randomly, but such that a range of GC contents from ∼20% to 60% was represented. A second group of tested sequences (denoted Synt01 - Synt11) was derived from synthetic enhancers driving expression in D. mel embryos.

In total, we obtained, after quality control (Transparent Methods), 147 titrations of 47 different DNA sequences (three measurements per sequence on average; Figure 2). All free energies (ΔΔG) were determined relative to our reference sequence, a weakened version of the 601 sequence (601_dpl), which has the same GC content as the original 601 but shows less pronounced 10 bp dinucleotide periodicity patterns and is thus well suited to measure weaker competitor sequences (Transparent Methods for details). Overall, our measured free energies range over 12 kJmol·mol−1 (Figure 2), which is similar to the dynamical range reported by Thastrom et al. (1999); reproducibility is high, with a mean coefficient of variation (CV) of 24%. Interestingly, both the 601 sequence and its weakened derivative 601_dpl are strong outliers, and the free energies of all the other sequences are distributed over a much smaller range of 5.3 kJ·mol−1.

Figure 2

Overview of All Measured Histone-DNA Affinities

All affinities are shown as free energy of nucleosome formation relative to the 601_dpl reference sequence; mean ± SEM over, on average, three replicates. Names are taken from the original publications or indicate the different groups of sequences: Dmelxx: selected −1 nucleosomes from D. mel, Syntxx: synthetic enhancer constructs driving expression during D. mel embryo development.

Dependency on GC Content and Other DNA Features

To gain insight into the parameters contributing to histone-DNA affinities, we correlated our data with several known sequence determinants of nucleosome formation, namely, the GC content and sequence features affecting bending: the 10 bp periodicities of flexible (WW where W is A or T) and stiff (SS where S is G or C) dinucleotides and the presence of homopolymeric sequences poly(dA:dT).

We first plotted the free energies of nucleosome formation as a function of the GC content for all investigated sequences (Figure 3). We observe a decrease of ΔΔG (i.e., increased affinity) with GC contents for GC content values <∼0.5 and an inverted trend with increasing ΔΔGs at higher GC content values. Thus, our data indicate that binding free energies are strongly influenced by the GC content, but in a non-monotonous fashion: there appears to be an optimal GC content value of ∼0.5. Interestingly, sequences from all three groups (Figure 3) follow this falling and rising pattern. The simplest model to describe the behavior of ΔΔG with respect to the GC content is given by a segmented linear regression with two segments, intersecting at the optimal GC content. We fitted the data using this model and found a relatively good correlation with a Pearson correlation coefficient (R) of 0.54 (p = 0.0001) and an optimal GC content value of 0.49, corresponding to a minimum ΔΔG value of 6.4 kJ·mol−1. Thus, the GC content alone explains ∼30% (R2) of the variance in the data. Note that for this analysis the extreme values of the two 601 variants were excluded, although both have a near-optimal GC content.

Figure 3

Correlation between Free Energy of Nucleosome Formation and GC Content

Scatterplot showing the relative free energies against the GC content for all investigated nucleosomal sequences. Data points are color coded according to their provenance; note the extreme values of the two 601 variants (yellow). Linear regression was performed over two sections for sequences with GC contents < or >0.49, respectively, resulting in a Pearson correlation coefficient between predicted and observed free energies of 0.54.

Another sequence feature that is thought to impact histone-DNA affinities is the periodic occurrence of flexible and rigid dinucleotides, aligning with the periodic changes in orientation of the DNA double strand facing the histones. In fact, the 601 sequence, which lies outside the general trend we observed with respect of GC content (Figure 3), contains very strong dinucleotide periodicities (Lowary and Widom, 1998), as do other specifically engineered sequences. Thus, we sought to determine whether dinucleotide periodicities influence the free energies among the pool of our measured sequences (owing to their extreme values, the 601 variants were again excluded from this analysis). We started by computing the 10 bp periodicity of WW (A or T) and SS (C or G) dinucleotides for all sequences using an autocorrelation function (Figures 4 and S3A and Transparent Methods) (Cui and Zhurkin, 2010). For both dinucleotide groups, we found relatively low 10 bp autocorrelations; the same result was obtained using Fourier transform, the alternative commonly used method (data not shown). A closer inspection of the individual autocorrelation values (Figure 4) revealed that the small average 10 bp autocorrelation observed is driven by a few literature sequences that in fact had been selected for their strong 10 bp dinucleotide periodicities (besides 601: Bombyx, TG, or TG-T). By contrast, most of the genomic −1 nucleosome and synthetic enhancer sequences exhibit very low 10 bp autocorrelation values. Note that this is not simply due to our sequence selection: the dinucleotide autocorrelations for all −1 nucleosomes and for all nucleosomes in D. mel show the same distribution of values (Figures S3B and S3C). When we then analyzed the influence of this sequence feature on the free energy of nucleosome formation by plotting the autocorrelation values versus ΔΔG (Figure 4), we found only a very weak correlation (R = −0.15 and −0.16 for WW and SS, respectively), which is driven by the artificial (=literature) sequences (without them R = 0.06 [SS] and 0.07 [WW]). This suggests that, at least among native nucleosomal sequences, dinucleotide periodicities are not a prominent feature and have no discernible impact on histone affinities.

Figure 4

Relationship between the Free Energy of Nucleosome Formation and the Corresponding Autocorrelation at a Shift of 10 bp

No significant correlation can be observed for our sequences (R = −0.15 and −0.16 for WW and SS, respectively). The extreme points (autocorrelation>15) originate from literature-derived sequences selected for their strong periodicities (Bombyx, TG, TG-T).

Finally, we evaluated the influence of poly (dA:dT) tracks, which are thought to disfavor nucleosome formation, on our measured histone binding energies (Figure S4). We do not find any significant effect, perhaps due to the low number of sequences containing long tracks (Figures S4A and S4B). However, for some of the sequences from the group of synthetic enhancers, in which this feature was systematically varied, we observe a weak effect of very short dA:dT tracks (n = 2 and 3) on ΔΔG, with changes of maximally 2 kJ·mol−1 (Figure S4C).

Discussion

We developed an assay to determine histone-DNA affinities using salt titration and fluorescence anisotropy microscopy. Aided by automated liquid handling, we achieve slow gradual changes in the salt concentration by dilution in small steps over a long period of time and thus ensure that nucleosome formation takes place at conditions close to equilibrium. The readout is performed in an automated fashion using fluorescence anisotropy and allows the reproducible measurement of relatively small effects. This medium-throughput pipeline permits carrying out a full titration series, which is typically not possible in other competitive nucleosome formation assays. Since the analysis thus relies on a fitting procedure over a larger dataset, the determination of the free energies is more robust and it allows introducing a quality control step for the detection of experimental errors (see Transparent Methods).

Most recent efforts to systematically determine nucleosome sequence preferences have relied on genome-scale assays, where genomic DNA is digested by MNase into (typically mono-nucleosomal) fragments, which are then analyzed by high-throughput sequencing. For any given DNA region, the number of fragments covering it provides a measure of nucleosome occupancy at that position; these occupancy values in turn can then be used to derive nucleosome sequence preferences. However, when conducted using in vivo chromatin, the results of this procedure will reflect not only the inherent sequence preferences of the histone octamer, but also the action of the plethora of other biological factors influencing both nucleosome occupancy and positioning. Even if conducted on nucleosomes reconstituted in vitro, the imprecision and biases in cutting by the MNase will have a significant impact (Jin et al., 2018). Moreover, the result for a given stretch of genomic sequence will always also be influenced by the nucleosome binding properties of the surrounding sequence and corresponding steric hindrance effects.

By contrast, our assay can measure the “pure” histone binding affinity of any given piece of DNA precisely, without influence of other biological factors, chromatin features, or neighboring sequences, and it can do so with a throughput that is sufficient to comprehensively investigate the sequence determinants of nucleosome formation. Our approach cannot match the scale of the genomic sequencing-based assays and does rely on prior knowledge to select the sequences to be tested, but in permitting to isolate and measure the purely sequence-dependent aspect of nucleosome formation, it provides a valuable complement - similar to the way in vitro measurements of transcription factor-DNA affinities complement the in vivo tracking of transcription factor occupancies using ChIP-Seq.

Although our assay can capture histone-DNA affinities ranging over more than 2.5 orders of magnitude (in a non-log scale), we obtain a more limited dynamical range of only one order of magnitude for the histone binding free energies of the native genomic and synthetic enhancer sequences, consistent with observations in previous studies (Thastrom et al., 1999). Although we cannot be certain that our selected sequences represent the entire range of natively occurring nucleosomes, this suggests that the biological impact of sequence preferences on nucleosome positioning in vivo (Kaplan et al., 2010) is caused by relatively small variations in absolute affinities.

We found the GC content to be the one feature contributing significantly to the free energies of nucleosome formation for the investigated DNA sequences, explaining ∼30% of the variance in the data (Figure 3). This finding is in agreement with studies of genome-wide nucleosome occupancy based on MNase digestion (Fenouil et al., 2012, Tillo and Hughes, 2009). Interestingly, we also observe that the relationship is non-monotonous, with an optimum GC content of ∼0.5. Although some genomic studies found a monotonous dependency (Tillo et al., 2010), others do support the idea of an optimum GC content (Fenouil et al., 2012, Wang et al., 2014): Fenouil et al. (2012) examined GC content preferences of +1 nucleosomes and found an occupancy maximum at GC contents of 0.44 (in vivo) and 0.58 (in vitro), respectively; Wang et al. (Wang et al., 2014) found the occupancy of nucleosomes in exons to peak around a GC content of 0.5. However, the GC dependency of nucleosome occupancy observed in these genomic studies might be attributable to either biological activity (Tillo et al., 2010) or to the cutting biases of the MNase (Jin et al., 2018). Our data now show, with an independent method that captures the pure sequence preference of the histone-octamer, that GC content is indeed an important driving force of nucleosome formation.

Numerous studies have focused on 10 bp dinucleotide periodicity, which affects the bending properties of the DNA, as a major determinant of nucleosome formation. Although this feature certainly appears dominant in many designed sequences, with the 601 clone the strongest known case, we find neither strong intrinsic periodicities in our random selection of genomic sequences and synthetic enhancers nor any significant correlation between the autocorrelation values for these sequences and our ΔΔG measurements (Figure 4). This suggests that, in naturally occurring (euchromatic) DNA, dinucleotide periodicity is not an important parameter driving differential histone-octamer binding. Finally, we investigated the influence on nucleosome formation of the occurrence of poly (dA:dT) stretches, which are thought to inhibit histone binding. Although globally we did not find an effect, perhaps due to the paucity of longer poly (dA:dT) tracks in our sample, we did observe, for a sub-group of synthetic enhancer sequences that are highly similar and expected to bind with the same strength based on GC content, a slight but detectable decrease of histone-DNA affinity with growing numbers of nucleotides in short dA:dT tracks (Figure S4C). Although more detailed investigation is needed, this suggests that even short dA:dT tracks can lead to a weakening of histone-DNA interactions. Interestingly, results from a genome-wide reconstitution study in yeast suggest that rather than directly affecting the histone-DNA binding interaction, poly(dA:dT) tracks might have a role in facilitating nucleosome displacement (Krietenstein et al., 2016), which our assay by design cannot track.

In conclusion, we have developed a method to measure the free energies of nucleosome formation reliably and at medium throughput, and we pinpoint the GC content of the DNA sequence as an important determinant of the histone-DNA interaction. We believe that our method will be highly valuable in investigating and comprehensively characterizing the sequence preferences of histone octamers. It can potentially be extended to investigate more complex interactions like the interplay between pioneer transcription factors and nucleosomes.

Limitations of the Study

Although the approach described here is much higher in throughput than existing in vitro methods, it cannot match the scale of genome-wide sequencing-based assays and thus relies on prior knowledge to select the sequences to be tested. To further substantiate our findings regarding the sequence determinants of histone-DNA affinities, more extensive studies with a larger selection of genomic and synthetic sequences will be needed.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.