Chemical Biology Approaches to Identify and Profile Interactors of Chromatin Modifications

Abstract

In eukaryotes, DNA is packaged with histone proteins in a complex known as chromatin. Both the DNA and histone components of chromatin can be chemically modified in a wide variety of ways, resulting in a complex landscape often referred to as the ‘epigenetic code’. These modifications are recognized by effector proteins that remodel chromatin and modulate transcription, translation, and repair of the underlying DNA. In this review, we examine the development of methods for characterizing proteins that interact with these histone and DNA modifications. ‘Mark first’ approaches utilize chemical, peptide, nucleosome, or oligonucleotide probes to discover interactors of a specific modification. ‘Reader first’ approaches employ arrays of peptides, nucleosomes, or oligonucleotides to profile the binding preferences of interactors. These complementary strategies have greatly enhanced our understanding of how chromatin modifications effect changes in genomic regulation, bringing us ever closer to deciphering this complex language.

Introduction

In eukaryotes, DNA is packaged in the form of chromatin.¹ The principal subunit of chromatin is the nucleosome, a complex consisting of DNA wrapped around an octameric protein core composed of two copies each of the canonical histones H2A, H2B, H3, and H4. Chromatin can take one of two general forms: a transcriptionally inactive state called heterochromatin in which nucleosomes are packed tightly together, restricting access to the DNA, and a transcriptionally active state called euchromatin in which nucleosomes are less tightly packed, and DNA is more accessible. The state of chromatin is regulated through several mechanisms, including post-translational modifications (PTMs) of the histone proteins or modifications to the DNA (Figure 1). These modifications regulate chromatin packaging and accessibility, typically by recruiting chromatin remodeling complexes or by directly perturbing the local charge and structure of chromatin.^1-3 Local chromatin structure, in turn, impacts the availability of DNA for the core processes of transcription, replication, and repair. Histone PTMs and DNA modifications are installed by enzymes known as ‘writers’, recognized by effector proteins known as ‘readers’, and removed by enzymes known as ‘erasers.’^{4, 5} Because of their role in regulating transcription, histone writers, readers, and erasers are attractive therapeutic targets.^{6, 7} For example, DNA demethylating agents such as azacitidine are FDA-approved to treat myelodysplastic syndromes.^{8, 9} Additionally, drugs targeting histone deacetylases have been approved for treatment of cutaneous and peripheral T-cell lymphomas, and other inhibitors in development show promise in treating cancer, immune diseases, and neurological disorders.¹⁰ In this review, we discuss chemical biology methods to identify what proteins bind to a specific chromatin modification and to characterize the binding behavior of modification interactor proteins.

Figure 1.

The epigenetic code consists of post-translational modifications (PTMs) to histone proteins and modification to cytosine bases in DNA. a) Structures of a subset of known histone PTMs b) A subset of modification sites on the N-terminal tail of human histone H3 for the PTMs in part a. All four canonical human histones contain many PTM sites in the N-terminal tails and core domains. c) The structure of modified cytosine bases, including 5-methylcytosine and its oxidized derivatives.

Histone Post-translational Modifications

Histones are extensively post-translationally modified on multiple residues and with many different chemotypes (Figure 1a and b).¹ Perhaps the most well-characterized histone PTMs are lysine acetylation (Kac) and lysine methylation, which can occur as a monomethyl, dimethyl, or trimethyl mark (Kme1, Kme2, and Kme3, respectively). Histone acetylation, which is generally associated with transcriptional activation, recruits chromatin modifying proteins and transcription factors (TFs), many of which contain an acetyllysine reader domain.¹¹ The most common Kac reader is the bromodomain (BRD), though other domains such as the YEATS and the Double PHD Finger (DPF) domains also recognize Kac and other lysine acylations.^11-14 BRD-containing proteins have diverse roles including as TFs, chromatin remodelers, helicases, and methyltransferases.¹² Of particular interest is BRD4, which drives expression of the MYC oncogene in several cancer types.¹⁵ Inhibitors of BRD4 such as the diazepines JQ1 and iBET show promise in treating hematological and tumorous cancers.^{16, 17}

Histone methylation can have transcriptionally activating or repressive functions depending on the specific site and state of methylation.^{18, 19} Some readers of histone methylation include the plant homeodomain (PHD) fingers, chromodomains, Tudor domains, PWWP domains, and bromo-adjacent homologous (BAH) domains.^{20, 21} Readers for histone phosphorylation include the BRCT domain and 14-3-3 family of proteins. ¹¹ Readers for some histone PTMs such as SUMOylation, ubiquitylation, and ADP-ribosylation have yet to be identified.¹

In addition to canonical lysine acetylation, mass spectrometric analyses have identified histone modifications derived from diverse acyl-CoA metabolites (Figure 1a), providing possible links between the metabolic state of the cell and transcriptional regulation.^22-26 Since these acyl marks are chemically distinct from lysine acetylation, an important question in the field is whether they each serve distinct signaling functions. For example, histone crotonylation (Kcr) plays a unique role in activating genes in post-meiotic male germline cells, and its function is mediated by the YEATS domain-containing protein Af9.^{22, 27} Apart from crotonylation, specific readers for other acylations (e.g., Klac, Khib, Kbhb, Kmal, and Ksuc) have yet to be determined.²⁵ Characterizing these metabolite-derived acyl PTMs and the mechanisms by which they exert their effects will provide valuable insight into the changes in transcriptional regulation that result from alterations in metabolite levels.^28-30

Affinity probes for identifying readers of histone PTMs

Affinity-based probes (AfBPs) have found use as tools to discover proteins that will bind to a specific histone PTM. Such AfBPs contain several components: 1) the histone peptide or protein containing the PTM of interest, 2) a photo-crosslinking moiety to aid in detection of low-affinity interactions (optional), and 3) a functional handle for probe visualization or retrieval (Figure 2).^31-33 These probes are typically used to pulldown proteins from cell lysates or nuclear extracts. The bound proteins are then identified and quantified using Stable Isotope Labelling by Amino Acids in Cell Culture (SILAC) mass spectrometry (MS).^34-36

Figure 2.

a) Chemical structures of typical diazirine (left) and benzophenone (right) crosslinkers before and after crosslinking to a serine residue on a target protein. The diazirine reaction proceeds by way of a carbene intermediate whereas the benzophenone reaction proceeds through a diradical intermediate. B) Schematic representation of a sample AfBP workflow for peptide (top) or mononucleosome (bottom) probes.

AfBPs typically take one of two forms: a peptide mimicking the tail region of a histone protein, or a whole nucleosome assembled with a semi-synthetically prepared histone containing the PTM of interest.^{31, 33} The photocrosslinking moiety used can vary or even be left out altogether. Older studies used a benzophenone moiety, which reacts to form a reactive radical species upon UV irradiation.³¹ Diazirine photoprobes have become the standard in the field due to their small steric footprint and other superior properties. A 2015 study comparing these two photocrosslinkers found that the diazirine probes have a much higher crosslinking efficiency, less position dependence, and lower off-target crosslinking than the benzophenone moiety.³⁷ The diazirine crosslinker can be installed on a variety of residues via leucine (‘photo-Leu’), methionine (‘photo-Met’), lysine (‘photo-Lys’), or cysteine (via thiol alkylation).^{32, 38, 39, 40}

Peptide Probes

The use of peptide AfBPs to identify histone PTM interactors was first described in 2006 using biotinylated H3K4me3 and H3K9me3 probes with no photocrosslinker.⁴¹ The following year, Timmers, Mann, and coworkers used a similar biotinylated AfBP to identify several H3K4me3 binding proteins using SILAC-MS. ⁴² Their ‘hits’ included several known readers of H3K4me3 as well as TAF subunits of the TFIID complex. This was an important finding, because it demonstrated direct recruitment of a TF by a histone PTM. The authors followed-up their proteomics findings with experiments to validate the TFIID/H3K4me3 interaction in cells. Interestingly, the authors also found that nearby marks could enhance (e.g., H3K9ac/H3K14ac) or reduce (e.g., H3R2me) the TFIID-H3K4me3 interaction, providing evidence that multidentate interactions between functional complexes and chromatin occur.⁴²

In a subsequent study, the Mann lab used a series of biotinylated peptide AfBPs to identify interactors of H3K4me3, H3K36me3, H3K9me3, H3K27me3, and H4K20me3.³⁵ With the H3K4me3 probe, they again found TFIID subunits and also found the SAGA complex, a transcriptional co-activator. The authors determined that the Sgf29 subunit mediates recruitment of human SAGA to H3K4me3 via its double Tudor domain. They further identified several uncharacterized proteins, including BAP18, which they characterized as a subunit of the NuRF/BPTF complex.³⁵ With the H3K36me3 probe, they identified several proteins containing a PWWP domain (e.g., N-PAC and NSD2) among their most significant hits. For both H3K9me3 and H3K27me3, they identified many members of the ORC and PRC1 complexes as well as the chromodomain-containing proteins CDYL and CDYL2. H4K20me3 also interacted with members of the ORC complex but showed interactions with many members of the TFIID complex in a similar fashion to H3K4me3. ³⁵ Overall, their work showcased the power of AfBPs and quantitative proteomics to identify PTM readers and reader complexes based on clustering behavior in the data.³⁵ Similarly, Dillon and coworkers demonstrated that peptide probes can be used for identifying complexes that bind to specific PTMs because members of complexes tend to cluster together in the proteomics data.⁴³ They investigated the H3K9me3/S10phos combination and found a novel complex that binds this set of marks which includes the proteins ATRX, DAXX, and several members of the FACT complex.⁴³ The Reinberg lab employed an AfBP to investigate the H4K20me1 mark and found 53BP1 to be a reader.⁴⁴ Since 53BP1 is involved in the DNA damage response and cell cycle progression, the authors hypothesized that H4K20me1 could be a DNA damage signal. Indeed, they found that laser-induced DNA damage resulted in recruitment of 53BP1 to DNA, while knockdown of PR-Set7 (the methylase which installs H4K20me1) abolished the effect.⁴⁴

The Kapoor lab was one of the first to use a photocrosslinker in an AfBP to conduct pulldowns from whole cell lysate.³⁴ They synthesized an H3K4me3 peptide with a benzophenone moiety and an alkyne handle for copper-catalyzed azide-alkyne cycloaddition (CuAAC) click chemistry.⁴⁵ With this probe, they pulled down known binders of H3K4me3 from cell lysates and identified a few new binders such as MORC3.³⁴ In 2014, the Xiang lab developed a peptide probe containing H3K4cr and a benzophenone photocrosslinker. Using this probe, they found that the deacetylases SIRT1/2/3 recognized H3K4cr.⁴⁶ They found that SIRT3 had the highest affinity for this PTM and demonstrated that it is capable of catalyzing decrotonylation of H3K4 in cells.

When using these AfBPs, removal of the modification by eraser enzymes in the lysate is a potential complication. To address this concern, Olsen and coworkers developed a series of peptide probes containing non-cleavable thioamide or hydrazide acyl lysine analogues.⁴⁷ Since deacylases cannot remove these marks, it is plausible that they may bind and then become trapped, leading to higher affinity of erasers for the probe. Similarly, the Wang lab designed probes with Kme1/2/3-alpha-ketoglutarate mimics to capture demethylases. ⁴⁸ Using this probe, the authors pulled down several known histone demethylases and homologous proteins. Interestingly, they also pulled down HDAC1/2 and components of several other repressive complexes including the Polycomb repressive complex (PRC2), suggesting that readers of Kme2 could have a transcriptionally repressive effect via recruiting other PTM-altering proteins.⁴⁸ These types of probes represent one of the latest iterations of peptide AfBPs and may help to identify previously unknown histone PTM erasers.

Recombinant Nucleosome Probes

Extensive mutagenesis screens have identified the ‘acidic patch’ of the nucleosome as a docking point for a large number of binding proteins.⁴⁹ Additionally, many chromatin effectors bind both DNA and histones.⁵⁰ Such multivalent binding behavior cannot be recapitulated using a peptide probe alone. Indeed, studies in which the authors used a nucleosome probe and the corresponding peptide probe found that, while the probes identified many of the same reader proteins, the nucleosome probes yielded a better signal-to-noise ratio and identified additional hit proteins.^{51, 52}

Nucleosome probes are most often composed of one modified histone (i.e., H3, H4, H2A, or H2B; both copies modified), and the other histones are wild type. The modified histone can be prepared via several strategies, but most commonly by native chemical ligation (NCL).⁵³ The modified tail containing the PTM of interest is synthesized via solid-phase peptide synthesis (SPPS) and ligated to the truncated histone. An alternative strategy is genetic code expansion, which Chatterjee and coworkers used to make a benzophenone-labeled H3K23ac histone.⁵⁴ Interestingly, most of the recombinant nucleosome probes published to date do not incorporate a photocrosslinker moiety, relying on chromatin interactors’ higher affinity for a nucleosome compared to a histone peptide to capture these interactions.^{51, 52, 55, 56} Nucleosome probes can have the affinity tag on the histone tail or incorporate biotinylated DNA.^{38, 51, 52, 56}

One of the first studies to use a nucleosome probe investigated whether DNA methylation and histone methylation have a combinatorial effect on interactor binding.⁵⁶ The authors used H3K4me3, H3K9me3, or H3K27me3 mononucleosomes (MNs), and they assembled these nucleosomes with biotinylated ‘601’ or ‘603’ nucleosome-positioning DNA to test for sequence specificity. They also methylated the DNA enzymatically. From their SILAC data, they identified several known methyl-CpG binding proteins and saw some degree of sequence specificity. They also found proteins, such as MeCP2, that had a much greater fold-change in interaction with methylated DNA when the DNA was in a nucleosome context. They identified several proteins (e.g., ORC complex) that recognize H3K9me3, H3K27me3, and methylated DNA but bind with the highest affinity to a nucleosome with both histone and DNA methylation. They also identified some proteins (e.g., KDM2A) that recognize histone methylation but cannot do so when DNA methylation is present. Their approach reveals something of the complexity and nuance underlying the crosstalk between chromatin modifications. Interestingly, members of DNA or histone binding complexes tended to cluster together in their data set, allowing uncharacterized proteins that clustered similarly to be putatively identified as members of complexes.⁵⁶

To compare an array of nucleosomes to a histone peptide probe, Fischle and coworkers generated a 12-mer nucleosome array of either H3K4me3 or H3K9me3 nucleosomes and the corresponding histone tail peptides alone.⁵¹ While both types of probes yielded a similar number of interactors from the SILAC experiment, the overlap between the hits for the two types of probes was surprisingly low, and the peptide probes yielded slightly more hits overall than the chromatin probes. Together their data suggests that the chromatin context, rather than recruiting more factors due to the presence of additional binding sites (e.g., DNA, acidic patch), provides “the more stringent and more discriminating binding surface.” In comparing their data to previous work, they found that there was more overlap between their peptide data and the Kouzarides MN data set than between their array data and the Kouzarides MN data⁵⁶; however, other experimental differences between the two studies made it difficult to conclude much about probe performance from these comparisons. As expected, their data shows that different proteins are recruited to H3K4me3 and H3K9me3, and their list of hits included many known interactors of these marks. The authors postulated that arrays of nucleosomes containing a variety of PTMs, potentially in combination with methylated DNA, could allow profiling of multidentate complexes that recognize complex patterns of histone PTMs. However, they did not test this hypothesis, in no small part due to the synthetic challenges such complex arrays present.⁵¹

The approaches described above compared the interactomes of a modified histone to an unmodified histone. Ren and coworkers instead compared H3K4me1 and H3K4me3 MNs to find H3K4me1-specific interactors.⁵² They found members of the BAF complex and other chromatin remodeling complexes as well as two subunits of cohesin, which is directly involved in transcriptional activation and enhancement. Choosing to focus on the BAF complex over their other hits, they further provided evidence that H3K4me1 may play a unique role in recruiting BAF to enhancers.⁵²

Non-traditional Affinity Probes

Some probes eschew the traditional peptide or nucleosome scaffold. In 2007, Cravatt and coworkers designed a probe based on the histone deacetylase (HDAC) inhibitor suberoylanilide hydroxamic acid (SAHA), to which they attached a benzophenone crosslinker and alkyne handle.⁵⁷ Their probe was able to pull down HDAC1, 2, and 6. Additionally, they found that their probe pulled down other members of HDAC complexes, including CoREST and MBD3.⁵⁷ Sun and coworkers developed an acetyllysine probe to profile lysine deacetylases. Their probe consisted of acetyllysine, protofluorophore, diazirine photocrosslinker, and alkyne. When the probe is deacetylated, the free amine reacts with the protofluorophore to produce fluorescence.⁵⁸ The authors used this probe to pull SIRT1 from 293T cell lysate as visualized by an anti-SIRT1 immunoblot; however, the pulldown efficiency was poor due to low affinity of the probe for SIRT1.⁵⁸ The advantage of this probe is that it allows for in-gel analysis of the enzymatic function of pulled-down proteins. However, the number of complex pieces makes synthesis challenging, and the probe appears to have low affinity for its cellular targets. This downside can be ameliorated somewhat by conducting pulldowns from nuclear fractions as opposed to whole cell lysate.

In Situ Approaches

While AfBPs continue to be useful tools for studying histone PTM-reader interactions, in situ approaches capture interactions within the intact nuclear environment. Rather than employing a synthetic histone probe, endogenous histone-interactor complexes within solubilized chromatin can be immunoprecipitated by an affinity handle^59-61 or a PTM-specific antibody^62-65 for subsequent MS analysis (ChIP-MS). While these approaches typically target a specific histone PTM and then look for interactors, the inverse strategy in which an interactor of interest is crosslinked to chromatin and immunoprecipitated to identify which histone PTMs are enriched has also been reported.⁶⁶

In the first strategy, an affinity handle (e.g., FLAG or TAP) is genetically fused to the histone of interest. An early example of this strategy from Figeys and coworkers profiled interactors of H2A and the variant Htz1p in yeast.⁶⁰ Kapoor and coworkers used a similar approach to identify H3 and H4 interactors in mammalian cells with the added feature of employing amber suppression to incorporate a diazirine photocrosslinker into these histones to improve capture of low-affinity interactions.⁵⁹ Impressively, they determined that their modified H3 constitutes approximately 1/3 of all chromatin-incorporated H3.⁵⁹ However, genetic code expansion strategies are limited by the selection of nonstandard amino acids for which a corresponding aminoacyl-tRNA exists. Additionally, this approach was unable to interrogate a specific PTM interactome since the modification state of the diazirine-containing histones was not controlled. Indeed, it is currently challenging to install multiple different modified amino acids (e.g., acetyllysine and diazirine-modified lysine) using genetic code expansion.⁶⁷

The use of PTM-specific antibodies has been used to interrogate histone PTM interactomes. In 2013, Soldi and Bonaldi used H3K4me3- and H3K9me3-specific antibodies to immunoprecipitated (IP) histones from Hela cells after formaldehyde crosslinking for subsequent MS analysis to identify both known and novel readers.⁶⁵ In 2015, Young and coworkers used this approach with an expanded suite of histone PTM antibodies and found known, implicated, and novel interactors of the six histone PTMs targeted.⁶² In the same year, Poot and coworkers added a protein-protein crosslinking step (by disuccinimidyl glutarate, DSG) to the typical ChIP-MS workflow and utilized antibodies for four histone PTMs to identify interacting proteins.⁶⁴ While Young and coworkers failed to identify some low abundance TFs, Poot and coworkers were able to do so in their study, suggesting improved sensitivity of their specific workflow. A recent study incorporates proximity labeling into the ChIP-MS workflow.⁶³ In this method, chromatin from cell lysates is incubated with a PTM-specific antibody fused to APEX2 to enable biotinylation of proteins in the vicinity of chromatin containing the PTM. The main advantage of proximity biotinylation is that it may allow for better detection of low abundance interactors even under fully native conditions (i.e., without crosslinking). However, ChIP is only as specific as the antibody used, and there are not sufficiently high-fidelity antibodies available yet for every histone PTM. Nevertheless, ChIP-MS is a highly accessible method that analyzes proteins bound to endogenous histone modifications within the cellular chromatin context.

Recently, the Muir and coworkers used split inteins to synthesize a nucleosome probe in nucleo. ⁶¹ They transfected mammalian cells with a histone protein fused to a split intein, which was incorporated into cellular chromatin. Nuclei were isolated, and a delivery construct bearing a synthetic histone tail with the PTM of interest, diazirine, biotin, and the other half of the split intein was delivered to the nuclei. The intein undergoes trans-splicing, yielding the complete histone protein with the PTM at the desired position within the chromatin of the intact nuclei. The authors benchmarked their method by showing that they can pulldown heterochromatin protein 1 alpha (HP1α), a known reader of H3K9me3. They then used the approach to identify additional known H3K9me3 readers and putative new readers, TAF15 and TMA7.⁶¹ It would be informative to analyze the genomic localization of the histone-intein fusion since the location of the histone-intein fusion may impact the identification of binding proteins.

Finally, Heck and coworkers developed an MS approach that requires neither probe nor ChIP.⁶⁸ Their XL-MS method entails simply generating protein-protein crosslinks by treating isolated nuclei with disuccinimidyl sulfoxide (DSSO) before MS analysis. Histone PTM interactors are identified by looking for modified histone peptides crosslinked to another peptide from an interacting protein. While the authors identified some H3K23ac and ubiquitinated histone readers using XL-MS, the method will require sensitivity improvements to target low abundance PTMs and interactors.

Methods for profiling binding preferences of specific chromatin PTM interactors

Peptide Arrays

While AfBPs approach histone PTMs from a ‘modification first’ perspective, another strategy is to characterize interactions from a ‘reader first’ perspective. Traditional approaches for measuring affinity of proteins for a particular peptide containing a PTM include relatively low-throughput methods such as isothermal titration calorimetry (ITC) or fluorescence polarization (FP) assays.⁶⁹ Using arrays of immobilized peptides bearing a variety of histone sequences and PTMs enables high-throughput screening. In these assays, peptides are immobilized and then incubated with the protein of interest (POI), which is usually fused to an affinity tag for downstream detection (Figure 3a).^{69, 70} If the POI binds a particular peptide, it can be detected using an optical readout (e.g., fluorescent or HRP-conjugated antibodies).

Figure 3.

a) Schematic of a sample workflow for profiling the PTM binding preference of a protein using a peptide array. Tagged protein is incubated with the immobilized peptide array and then washed several times. While readout intensity tends to scale with the binding affinity, higher-sensitivity experiments such as isothermal titration calorimetry or fluorescence polarization are needed for exact measurement of binding affinities of individual protein-peptide interactions. b) Schematic of a sample workflow for using a DNA-barcoded library to profile the PTM binding preference of a protein

There are two main methods for synthesizing peptide arrays. In the first, peptides are synthesized using standard SPPS with an N-terminal biotin moiety and immobilized on a streptavidin-coated surface. ^{69, 71-73} The other major approach for array synthesis (called ‘SPOT’) involves synthesizing the peptides directly on a cellulose membrane.^{70, 74} While this results in more peptide per surface area, the peptide cannot be purified or characterized before use in an assay. In addition to these two approaches, peptide arrays synthesized by SPPS can be left on the resin for screening via an ‘on bead Western’ workflow.⁷⁵ Conveniently, several peptide arrays containing a variety of PTMs in the context of the H3 and H4 tails are commercially available.⁷⁰ Such histone peptide arrays have been used extensively over the last 13 years to profile the PTM preferences of chromatin interactors, including chromodomains^{74, 76}, Tudor domains^{73, 74, 77}, BRDs^{69, 74, 78, 79}, and PHD domains^{74, 75}, to name only a few examples of applications of these tools. Below we describe several examples of how histone peptide arrays have been used and improved upon.

Using an array of acylated histone peptides, Knapp and coworkers found that some BDs strongly prefer a diacetylated peptide.⁷⁸ Together their array and additional structural data suggests that BRDs can read a variety of acylation patterns.⁷⁸A subsequent study screened 49 BRD-containing human proteins against a peptide array containing six different lysine acylations in several different histone sequence contexts.¹³ Cochran and coworkers found that most BRD-containing proteins can recognize and bind lysine propionylation (Kpro) with similar affinity to Kac, and they did not observe any proteins that could significantly distinguish between the two. Interestingly, they found a set of BRD-containing proteins that would bind butyrylation (Kbu) but not crotonylation (Kcr) as well as two proteins without a strong preference between the two marks (TAF1, TAFL1).¹³ Similar peptide arrays have been used to study the specificity and cross-reactivity of antibodies against histone PTMs,⁸⁰ the acyl-binding preferences of the BET family,⁵⁵ and the PTM preferences of a selection of chromodomain and Tudor domain-containing proteins.⁸¹

Using BRDs for testing, Strahl and coworkers achieved some technical improvements to the standard peptide array workflow to improve detection of low-affinity interactions.⁶⁹ They found that blocking with milk and using a formaldehyde crosslinking step with high salt washes together lowered background and increased the threshold for detection by as much as five-fold. For example, they found binding of BRD4 to di- and triacetylated H4 peptides was detected with their optimized protocol but not using a more conventional procedure. A recent innovation coupled a peptide array to a Matrix-assisted Laser Desorption/ionization (MALDI) MS readout.⁸² Peptides were immobilized on a gold surface using alkanethiolate chemistry such that each spot contained one histone peptide and a reporter peptide that can be deacetylated by KDAC8. A fusion protein of KDAC8 and a POI was exposed to the peptide array. The low affinity of the reporter peptide for KDAC8 means that deacetylation of the reporter is only observed when KDAC8 is brought in proximity to the reporter by the POI. The degree of deacetylation is then read out directly from the array surface by MALDI-MS. The technique was benchmarked by testing against several chromodomain-containing proteins whose substrate specificities were known and PHD1 (an H3K4ac reader).

DNA-Barcoded Nucleosome Libraries

While peptide libraries have remained a popular tool due to the ease of array construction, advances in the synthesis of modified histones have enabled the production of relatively large nucleosome libraries. Although the nucleosome libraries to date are still generally smaller than most peptide arrays, they boast the major advantage of providing a chromatin context for histone PTMs. Furthermore, the impact of PTMs on different histones in the same nucleosome (e.g., H3K4me3 and H4K20ac) on binding can be assessed.

Since the nucleosome already requires DNA, adding a unique barcode sequence to each library member is straightforward and allows for 1) pooling of library members for one-pot assays and 2) assay readout using next-generation DNA sequencing (NGS; Figure 3b). The Muir lab pioneered this DNA-barcoded nucleosome library approach.⁸³ While the synthesis of the modified histones was still quite labor intensive, they modified the MN assembly process to streamline library preparation. In the first iteration, they prepared a library with 54 unique members (i.e., different histone PTMs, variants, and/or mutants). To conduct the assay, an affinity-tagged POI was incubated with the library, the POI was immunoprecipitated, and then NGS was performed to identify MNs that coimmunoprecipitated with the POI. The authors tested their assay by investigating sequence and PTM specificity of various BRDs and found trends agreeing with previous data.⁷⁸ When they incubated the library in nuclear extracts and monitored de novo acetylation of H3K14 of the library, they detected stimulation by several H4 acetyl marks and by H3K4me3.⁸³ The second iteration of the Muir lab’s modified MN library contained 115 unique MNs and has been used for several applications to date.^{84, 85,86} In this version, a masked restriction site was included in the DNA such that octamer sliding by chromatin remodelers could be detected via restriction enzyme efficiency as quantified by NGS. This feature allowed for a detailed assessment of the substrate preferences of members of the ISWI and SWI/SNF families of chromatin remodelers. Kadoch, Muir, and coworkers found that cBAF avoids remodeling H3K4me3 nucleosomes, particularly in combination with H4 acetyl marks in the same MN.⁸⁴ Importantly, this ‘crosstalk’ of PTMs on different histones in the same MN would not be detectable using a peptide array.

Recently, the Fischle lab developed a method to overcome some difficulties associated with synthesizing a modified MN library. They made MNs with wild-type H2B and H4, truncated H3 (residues 33-135), and an H2A-IntC fusion designed to allow splicing at residue 18.⁸⁷ These MNs enable orthogonal installation of chemically-modified H3 and H2A tails in two enzymatic steps: protein trans-splicing to install the H2A tail and Sortase-A mediated installation of the H3 tail. They immobilized their library via biotinylated DNA with each member in one well of a multiwell plate, enabling parallel synthesis of library members. They used their technique to generate a library of 280 individually-modified MNs and measured the binding of HP1, a known Kme2 reader. They found that HP1 binding to H3K9me2 was reduced by H3S10phos and enhanced by H3S28phos.⁸⁷ While they did not observe crosstalks between H2A and H3 PTMs, this may be simply because they were focusing on one specific POI. This intein/sortase strategy could be combined with traditional histone semi-synthesis and DNA barcoding to further expand the PTM space covered. Looking beyond MNs, synthesis of libraries of nucleosome arrays containing differently modified nucleosomes^{88, 89} would bring us even closer to capturing the full complexity of chromatin landscapes in cells.

Post-synthetic modifications of Cytosine

Cytosine bases in DNA can be post-synthetically modified, typically at the 5 position (Figure 1c).^{90, 91} These modifications almost always occur in the context of a 5’-CG-3’ dinucleotide (also referred to as a CpG motif).^90-93 CpG dinucleotides are found throughout the genome but are especially enriched in regions near promoters and transcription start sites, in regions called ‘CpG islands.’^{91, 94} Cytosine modifications can occur within gene bodies and CpG islands.^{92, 93} The most well-studied of these modifications is 5-methyl cytosine (5mC; Figure 1c).^{91, 94} In humans, cytosine methylation is catalyzed by the DNA methyl transferase (DNMT) family of enzymes.^{90, 91} DNA methylation, especially in CpG islands, is traditionally viewed as a transcriptionally repressive mark.^{91, 94} However, high levels of DNA methylation at CpG sites in gene bodies has been correlated with higher levels of transcription, suggesting a more nuanced role for 5mC in transcription.⁹¹ Importantly, aberrant DNA methylation has been linked to cancer, neurological diseases, metabolic diseases, and autoimmune diseases.^{90, 94}

When cytosine is modified in the context of a CpG motif, both strands of DNA typically bear the modification (i.e., symmetrically modified).^{90, 91, 94, 95} However, recent work has demonstrated that hemimodified CpG motifs can exist as a stable mark.⁹⁵ Other work has revealed the presence of oxidized forms of 5-methylcytosine including 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC), and 5-carboxycytosine (5caC).^{93, 96-98} The proteins that catalyze these sequential oxidations are the ten eleven translocase (Tet) family of proteins.^{99, 100} Since 5fC and 5caC can be recognized by thymine DNA glycosylase (TDG), an enzyme that performs the first catalytic step of base-excision repair, the Tet family of proteins was identified as a step in active DNA demethylation.^{99, 100} However, the fact that 5hmC is not recognized by TDG led to the hypothesis that it may be a separate epigenetic mark with its own unique transcriptional role.^{93, 98} Indeed, 5hmC is present in most human cell lines and tissues but is most prevalent in pluripotent stem cells and neurons and is implicated in regulating pluripotency and differentiation. ^{92, 98, 101-103}

Identifying readers of cytosine modifications

Approaches similar to the ones described for identifying and characterizing histone interactors have been developed for readers of cytosine modifications.^{104, 105} Early attempts to find interactors of 5mC used radiolabeled oligonucleotides with or without 5mC. These probes were incubated with cell extracts, and electrophoretic mobility shift assays (EMSAs) were performed to search for DNA-protein complexes.^{106, 107} These efforts identified Methyl-CpG binding protein 1 (MeCP1)¹⁰⁶ and Methylated DNA Binding Protein 1 (MDBP1, now referred to as methyl binding domain protein 1, MBD1).¹⁰⁷ Other 5mC binding proteins were discovered serendipitously.¹⁰⁸ Still others were discovered through sequence homology approaches (e.g., MBD2, MBD3, and MBD4).⁹⁶ These initial approaches were limited in their scope and only identified a few proteins at a time. Another inherent limitation was that hypotheses at that time typically assumed that cytosine modification primarily functioned by recruiting proteins with methyl binding domains, not necessarily by excluding proteins from binding or by other mechanisms.^{91, 94} In the last two decades, however, the advent of NGS and MS technologies have allowed for the design of high-throughput experiments.. Moreover, the discovery of oxidized 5mC in cells has spurred the field to search for interactors of these modifications as well.¹⁰⁹

In 2009, Mann and coworkers used biotinylated oligonucleotide probes with and without 5mC to affinity purify proteins from nuclear extracts and characterized these proteins by SILAC (Figure 4a).¹¹⁰ Stunnenberg and coworkers similarly used this approach to identify known 5mC interactors^{96, 106, 108} as well as a novel interactor, RBP-J, a component of the Notch signaling pathway.³⁶ However, further analysis revealed that the probe contained an RBP-J consensus binding sequence with a T to C mutation. Within that context, 5mC was acting as a thymine mimic.³⁶ While the biological significance of 5mC recognition by RBP-J remained unclear, this experiment did highlight a potential pitfall for high-throughput screens for modified cytosine binders.

Figure 4.

a) Schematic of a workflow for identifying reader proteins for a cytosine modification using an oligonucleotide affinity probe. b) Sample workflow for characterizing consensus sequence and modification preferences of a protein of interest using a modified oligonucleotide library.

Taipale and coworkers used a high-throughput approach to test how the binding affinities of TFs to their consensus binding sequences are affected by cytosine methylation.¹¹¹ They prepared a library of random 40-bp oligonucleotides and purified ~500 TFs (Figure 4b). They then used a Systematic Evolution of Ligands by Exponential Enrichment (SELEX) workflow to measure binding affinities of TFs for their preferred binding sequences. They then repeated their method with the addition of a cytosine methylation step in each round of SELEX to measure the effect on TF binding. Some TFs bound to their consensus sequence with higher affinity in the presence of methylation, whereas others bound with lower affinity. Still others seemed to bind without regard to methylation state or preferred different consensus sequences when their primary sequence was methylated.¹¹¹ TFs that bound better to methylated sequences tended to be involved in embryonic and tissue development, whereas those that bound better to unmethylated sequences tended to be involved in replication and differentiation. The authors validated their in vitro work by comparing to previously collected ChIP-Seq and bisulfite sequencing data.^{92, 93, 111} They found their results were mostly consistent, though they pointed out that in vivo TF binding sites tended to be lower in 5mC content than the surrounding regions of chromatin.¹¹¹ They hypothesized that this phenomenon occurs when TFs recruit DNA-demethylating enzymes, leading to activation of the target gene.

With the discovery of 5hmC and its oxidized derivatives, SILAC-based studies were designed to identify their readers. One approach used a biotinylated oligonucleotide sequence containing 5mC, 5hmC, 5fC, or 5caC to pulldown proteins that bind modified cytosine in a CpG island context.¹¹² The limitation of this experiment was that it could not provide information about sequence specificity, but it did enable pulldown of a variety of previously unidentified 5mC and 5hmC readers. The authors also identified readers for 5fC and 5caC, many of which were proteins associated with DNA repair or active DNA demethylation. They commented that the high number of active base excision repair enzymes that interacted with 5fC and 5caC potentially indicates a high turnover of these modifications in cells, since consistent levels can be observed.¹⁰⁰

To learn more about the context specificity of reader proteins involved in recognition of modified cytosine, Reik and coworkers sought to identify readers of the different oxidation states of 5mC.¹¹³ Their probes were designed by selecting two enhancer regions known to have high levels of 5hmc (Pax6 and Fgf15). The Fgf15 enhancer region contains a CpG island (the ‘typical’ environment for 5hmc) whereas the Pax6 does not.¹¹³ They performed PCR on both regions to incorporate biotin and each modification. While this approach is straightforward and relatively easy to perform, it should be noted that the non-specific incorporation of modified cytosine bases does pose the risk of generating a high number of modified cytosines in a non-CpG context, which could potentially complicate data interpretation. Following incubation with nuclear lysate from murine embryonic stem cells (ESCs), the probes were pulled down and interactors were analyzed using MS. The authors first verified that their approach was able to enrich for the known 5mc/5hmc binder UHRF1¹¹⁴ and known binders for unmodified cytosine and 5fC.¹¹² They identified a variety of 5mc and 5fc binders in the CpG island context of Fgf15, such as the NuRD complex, that are associated with transcriptional repression, suggesting that 5fc plays a repressive role in much the same way as 5mC.¹¹³ However, this was not the case with the Pax6 probe, suggesting that the CpG island context is important for recognition.

A recently reported technology multiplexed the measurement of interactions between modified DNA and proteins.¹¹⁵ Zhu and coworkers developed a DNA-barcoded protein library and a library of barcoded oligonucleotides that are either unmodified or bear 5mC, 5hmC, 5fC or 5caC. They prepared separate libraries for unmodified, hemimodified, and symmetrically modified oligonucleotides to profile the effects of each modification. While the library preparation was onerous, it allowed all the DNA-protein interactions to be quantified in a single high-throughput DNA sequencing reaction. In brief, the technique involves incubating the DNA-barcoded protein library with the oligonucleotide library, crosslinking, and then adding a DNA ligase.¹¹⁵ Since the DNA barcode from the protein is in close proximity to the barcoded oligonucleotide, the ligase attaches the protein barcode to the oligonucleotide also bearing its own barcode, enabling identification of protein-DNA interactions using NGS. The authors benchmarked their approach by testing it against a library of 31 human TFs and an unmodified DNA library, successfully identifying the previously characterized consensus sequences for each of those TFs with good reproducibility. Next, the authors expanded their protein library to 1543 human TFs and the full library of modified oligonucleotides. One key takeaway from this work is that cytosine modifications can modulate TF binding activity by promoting or inhibiting TF-DNA interactions, depending on the type of modification and the specific TF. For example, sequences containing 5caC were pulled down by HMBOX1 more frequently than the unmodified control sequences, suggesting 5caC enhances HMBOX1-DNA interactions.¹¹⁵Another key conclusion is that cytosine modifications can change the consensus sequence a TF prefers. For example, the protein ETS1 binds the unmodified consensus sequence 5’-CCGGAAGT, whereas cytosine formylation promotes binding of ETS1 to the sequence 5’-CCGA(fC)GTA. Families of proteins usually exhibited the same binding trends and preferences. Interestingly, the authors also identified several proteins with selectivity towards hemimodification, such as NR2E1 and YBX1.¹¹⁵ These findings highlight the growing recognition of the importance of cytosine hemimodification.

While many studies have focused on recognition of modified cytosine by TFs, studies of other families of reader proteins, such as DNA repair enzymes and chromatin remodelers, are also yielding important insights. For example, the MBD family of proteins, especially MeCP2, is important in recognizing asymmetrically modified cytosine in non-CpG contexts. Loss or mutation of MeCP2 leads to development of Rett syndrome.¹¹⁶ EMSAs have demonstrated that MeCP2 has reasonably high affinity for methyl-CpA and hydroxymethyl-CpA motifs. Improvements in bisulfite sequencing and ChIP-seq have demonstrated that methyl- and hydroxymethyl-CpA motifs are more abundant in murine neurons and brain samples^{117, 118} and that MeCP2 binds to these modifications in vivo.¹¹⁹ While many putative functions and modes of action have been proposed for MeCP2^{116, 120}, one recent study demonstrates that MeCP2 is capable of binding to CpA-rich regions of DNA and of preventing nucleosome formation.¹¹⁹ These results highlight the importance of unbiased studies capable of identifying non-TF proteins that interact with cytosine modifications.

Conclusions/Perspectives

Ultimately, greater insight into the role of epigenetic modifications in human health and disease will require experiments to identify key protein-DNA and protein-protein interactions as well as follow-up experiments to provide structural or biochemical data characterizing these interactions. The discovery of novel modifications (e.g., ADP ribosylation on DNA¹²¹ or new acylations on histones^{25, 26}) will continue to expand our understanding of the epigenetic code. The technologies described herein for identifying and profiling epigenetic readers will undoubtedly continue to prove invaluable in decoding these novel marks.

While there is evidence of ‘crosstalk’ between chromatin modifications and of readers that recognize complex combinations of marks, we are still far from a complete understanding of these complex signals. While most of the AfBP assays to date used SILAC for quantitation, adoption of techniques for multiplexed quantitation (such as isobaric tandem mass tagging¹²²) may allow for the design of experiments to assay more complex crosstalks. Methods for processing proteomics data are improving as well, though analyzing data from large-scale assays is still a challenge. The most recent release of MaxQuant, one software package commonly used for data processing for photoaffinity experiments, includes improved support for identifying crosslinked peptides directly in the data set, allowing for additional characterization of interactions directly from the MS data.¹²³ Improvements to software notwithstanding, innovative ways of processing, visualizing, and interpreting data is still needed in this field, especially as massively high-throughput experiments become the norm.

While large datasets from high-throughput experiments excel at revealing general trends and identifying potential therapeutic targets, it has become clear that there is no overarching general trend that explains the role of any particular chromatin modification in all circumstances. To elucidate the effect of each of these marks, further work studying the individual interactions between families of epigenetic reader proteins and modifications and the effects of those interactions on gene expression and disease is needed. Structural data can reveal the intermolecular interactions underpinning recognition of a particular mark by a protein, which in some cases can ultimately lead to the development of treatments or diagnostic tools for related diseases. For example, in the case of MeCP2 in Rett syndrome, structural studies have revealed that Arg111 and Arg133 are most closely involved in recognition of methyl- and hydroxymethyl-CpA motifs.^{119, 120} However, these residues are not commonly mutated in Rett syndrome, and the mutations common to Rett syndrome do not alter MeCP2 affinity to modified CpA motifs.¹²⁰ Since knockout of MeCP2 leads to increased nucleosome density in CA repeat regions, ultimately affecting the transcription of genes within CA-repeat regions,¹¹⁹ the data seem to support a model of Rett’s disease in which mutations lead to misfolding or aggregation of MeCP2, decreasing the effective concentration in cells.¹²⁰ In a similar fashion, the integration of large data sets with detailed biochemical and structural studies will be necessary to completely decipher the epigenetic code, to elucidate dysregulation of chromatin effectors, and to fully tap into the potential wealth of therapeutic targets that the chromatin holds.

Keywords

Nucleosome:

An octameric protein complex comprised of 2 copies each of histones H2A, H2B, H3, and H4, with DNA wrapped around the octameric protein core.

Chromatin:

The general term for eukaryotic DNA as packaged in cells, chromatin consists of many nucleosomes strung together and can take the form of tightly packed ‘heterochromatin’ or the more accessible, transcriptionally active ‘euchromatin’.

Epigenetics:

The study of genetic information that is contained or transmitted in a form other than that of the DNA sequence itself, including chemical modifications (epigenetic marks) on DNA and histones.

Epigenetic Reader:

A protein that recognizes epigenetic marks and effects some change in the structure of chromatin or the expression level of a gene.

Post-Translational Modification:

A chemical label, such as an acetyl or phosphoryl group, that is enzymatically added to a protein after translation.

Affinity-based Probes:

Probes that are used to pull down proteins that bind a particular motif, typically containing both a mimetic of the motif and a handle for pull down or other readout.

Affinity profiling:

The underlying premise of affinity profiling is using a library or array of potential binding partners to identify trends in the binding preferences of a particular protein or family of proteins.

Photo-crosslinking:

Covalently linking two proteins or peptides together using a chemical motif that is inert under normal conditions but becomes reactive when irradiated (typically with ultraviolet light).

Mass Spectrometry Proteomics:

The use of mass spectrometry to identify all proteins present (the proteome) in a complex sample, typically relying on internal standard methods such as SILAC or isobaric tagging to provide quantitative comparison of proteins identified between samples.