Epiproteomics: quantitative analysis of histone marks and codes by mass spectrometry

Abstract

Histones are a group of proteins with a high number of post-translational modifications, including methylation, acetylation, phosphorylation, and monoubiquitination, which play critical roles in every chromatin-templated activity. The quantitative analysis of these modifications using mass spectrometry (MS) has seen significant improvements over the last decade. It is now possible to perform large-scale surveys of dozens of histone marks and hundreds of their combinations on global chromatin. Here, we review the development of three MS strategies for analyzing histone modifications that have come to be known as Bottom Up, Middle Down, and Top Down. We also discuss challenges and innovative solutions for characterizing and quantifying complicated isobaric species arising from multiple modifications on the same histone molecule.

Introduction

The fundamental repeating unit of eukaryotic chromatin, the nucleosome, is formed by wrapping 147 bp of DNA around a histone octamer, consisting of two copies of histones H2A, H2B, H3, and H4. Linker DNA between nucleosomes is further bound with histone H1 for hierarchical folding of chromatin. These histones are highly modified with posttranslational modifications (PTMs) that dynamically modulate local and global chromatin structure [1]. The large number of recurrent mutations identified in histone modifying enzymes has spurred great interest in cancer epigenetics [2]. A recent survey of literature has identified total of 519 distinct modifications from 237 sites in core and linker histones [3], creating a perception of vast combinatorial complexity. However, we focus here on methods to readout those PTMs that form the discrete marks and their combinations (often referred as ‘codes’ in literature) present at high levels, sharply curtailing the combinatorial explosion of histone forms theoretically possible.

Biochemical methods relying on modification-specific antibodies are the gold standards for analyzing histone modifications in biological systems. However, these approaches suffer from antibody cross-reactivity arising from the high similarities in both modification chemical structures themselves (e.g., mono-, di-, and trimethylation) and their surrounding amino acid sequences (e.g., H4 K5/8/12/16 acetylation, S₁GRGKGGKGLGKGGAKR₁₇) [4]. Antibody selectivity can be negatively affected by co-occurring histone modifications on the same molecule [5]. In addition to interrogating individual sites like the antibody-based approaches, liquid chromatography with MS (LC–MS) can directly measure multiple modifications on the same molecule (i.e., ‘codes’), which has proven central in chromatin biology [6^••,7^••,8^••]. Here, we focus in discussing recent advances in analyzing histone marks and codes by quantitative MS and LC–MS. Other aspects of this topic are well covered elsewhere [9].

Development of mass spectrometric analysis of histone modifications

Over the past 15 years, three analytical modes for histone modifications have come to be known as Bottom Up, Middle Down, and Top Down (Figure 1). The first two terms refer to the use of proteases to create small- or medium-sized peptides, whereas Top Down designates use of no proteases prior to tandem MS (MS/MS). Early MS analyses of histones used MALDI-TOF [10,11] before the field shifted heavily toward LC–MS/MS. Because histones are highly enriched with arginine and lysine residues (especially on the N-termini where most PTMs are known to reside), trypsin digestion results in peptides <5 residues, most of which are too hydrophilic to be retained on the reverse phase LC columns in widespread use. The method of chemical derivatization followed by trypsin digestion developed in the Hunt laboratory [12] has now been widely adopted for Bottom Up analysis of histones [6^••,13,14,15^••]. In this method, free amine group on the N-termini and ε-amine group of unmodified and monomethylated lysine residues react with propionic anhydride to form propionyl amides [12]. This method produces Arg-C like peptides (cleaved only C-terminal to arginine residues, see Figure 2a) with high efficiency and reproducibility. The presence of small levels of acetic anhydride and low levels of endogenous arginine methylation are two analytical interferences that should be noted for this approach. The first causes the artifactual introduction of acetyl groups and the second can lead to fluctuation of the total amount of the peptides generated by cleavage C-terminal to modified arginine residues in this Bottom Up procedure. Such effects are generally operative at low levels (<3%) because the stoichiometry of arginine methylation is known to be relatively low in yeast, mammalian cells, and other model eukaryotes. For the most widely studied histone marks, the Bottom Up/propionylation procedure allows readout of ~20–50 of them in a single LC–MS run [15^••,16^•,17]. Most quantitative studies use a label-free roll-up of the data, where individual modified species are normalized against total signal from all Arg-C peptides sharing the same underlying sequence (Figure 2a) and standard Z-scores are used to create heat maps to survey for changes across many conditions [17].

Figure 1.

Three general strategies to analyze histone modifications using liquid chromatography and mass spectrometry (LC–MS): Top Down, Middle Down, and Bottom Up. Histones, usually enriched by acid extraction of isolated nuclei, or fractionated into sub-sets of variants by RP-HPLC can be used as starting material. Example MS data of histones from human cell lines are shown for all three approaches (at right). Top right: partial spectrum of the 18+ charge state of whole histone H3.1. The multiple species observed arise from methylation and acetylation (each peak differs by a nominal mass of 14 Da). Center right: partial spectrum of the 8+ charge state of 1-50 N-terminal tail from histone H3.1. Bottom right: an extracted ion chromatogram (XIC) of a LC–MS run is shown for 2+ ion of H3 K18un-K23un (peptide is referred by the highly-modified lysine residues within the sequence for clarity; un, unmodified) with its spectrum along the chromatogram. Chemical derivatization is widely adopted in the Bottom Up approach to block unmodified and monomethylated lysine residues to generate Arg-C like peptide. ^prX, propionylation in the N-terminal amino group of an amino acid; K^pr, propionylation in the ε-amino group of lysine.

Figure 2.

Analytical challenges of analyzing histone modifications by Bottom Up and LC–MS. (a) 42 unique modified species are shown as sticks with modifications over specific residues in histone H3.1 (ac, acetylation; me, methylation). It represents a Bottom Up experiment outcome when unmodified and monomethylated lysines are blocked by chemical derivatization (e.g., propionic anhydride or deuterated acetic anhydride) before trypsin digestion. Using this workflow, eight highly-modified lysine residues in histone H3.1 (K4, K9, K18, K23, K27, K36, and K79) fall into five groups of peptides which share the same underlying sequence. The combination of modifications when two modified lysines are present in the same peptide generates isobaric species (red bracket). In addition, small mass difference between acetylation and trimethylation produce near isobaric species (purple bracket). (b) Fragment ions (such as y5 or y6) between K18 and K23 generated by tandem mass spectrometry (MS/MS or MS2) can be used to distinguish isobaric species of H3 K18un-K23ac from H3 K18ac-K23un. In the left panel, an extracted ion chromatogram of the most abundant monoisotopic peak from these doubly charged isobaric peptide species is used to determine the combined abundances, which can be further divided by fragment ion relative ratios obtained in multiplexed collision-induced dissociation (CID) tandem mass spectrum (shown in the inset) to quantify the abundance of individual isobaric species. In the right panel, SRM chromatograms of specific transitions (e.g., precursor → y5 and precursor → y6) can be used to quantify each isobaric species. Noting that data obtained by these two methods for a same sample (human multiple myeloma cell line, KMS11-TKO) is consistent to show that H3 K18un-K23ac is about 7-fold more abundant than the co-eluted H3 K18ac-K23un (retention time labeled along SRM peaks). (c) Trimethylation and acetylation are indistinguishable in low resolution instrument, such as triple quadrupole used for SRM. An example of SRM data from a human cell line (MCF-7) demonstrates that transitions designated to quantify H3 K9me3-K14un can detect two peaks. Using the retention time profile determined by synthetic peptides, these near isobaric specices (me3 vs. ac) can be distinguished.

Our laboratory has published extensively on the Top Down approach for all histones a decade ago, establishing a ‘basis set’ of histone proteoforms [18] that arise from multiple modifications and are present at approximately 5% levels and above [18,19,20,21^•,22]. For histone H4, 42 proteoforms were described from 50% down to 0.05% relative abundance in bulk chromatin using weak cation exchange-hydrophilic interaction liquid chromatography (WCX-HILIC) to separate primarily by extent of acetylation before Top Down MS [23]. The labor intensive nature of this ‘off-line’ approach using direct infusion nanospray motivated the Garcia and Pasa-Tolic laboratories to automate ‘on-line’ LC solutions using ‘saltless’ pH gradient of WCX-HILIC [24^•] or two dimensional LC [25^•], respectively.

A compromise between Bottom Up and Top Down is the Middle Down approach. These strategies and terms are trending in proteomics generally, but were piloted first in ‘epiproteomics’ using histones where endoproteinase Glu-C or Asp-N clips the first 1-50 or 1-23 N-terminal residues off H3 or H4, respectively (Figure 1, middle). The rationale is that these peptides preserve the co-occurrence of most modifications in the same molecule because the majority of them occur on the N-terminal tails. One early Middle Down report focused on histone H3.2, and characterized >150 codes by direct infusion of WCX-HILIC fractions [26]. Shortly after, 72 unique forms of the H4 tail were profiled using on-line RP-HPLC separation of Asp-N digested histone H4 [27]. More recently, the feasibility of nanocapillary LC–MS/MS for faster Middle Down has been demonstrated for unfractionated histones digested by Glu-C or Asp-N using simple reverse phase HPLC [28] or innovative hybrid LC [29^••]. A direct comparison has concluded that Bottom Up and Middle Down can achieve comparable accuracies of quantifying individual histone modifications [30].

Challenges and strategies of quantifying histone modifications using MS

Although a large number of histone proteoforms exist due to the combination of multiple modifications, it is not true that every random combination of modifications exists. Otherwise a cell would produce an astronomical number of proteoforms, far exceeding the ~30 million nucleosomes in a human cell. Nonetheless, many combinatorial forms of methylation and acetylation have been observed in peptides carrying two residues modified with high occupancy (e.g., K9-K14, K18-K23, and K27-K36 in H3 as shown Figure 2a) or four modified residues (e.g., K5, K8, K12, and K16 of H4 4-17 in Figure 3a and b). The complexity originating from large amount of isobaric or near isobaric species (Figure 2a) found in these combinatorially modified species presents a central challenge for MS-based epiproteomics.

Figure 3.

The increasing analytical challenges encountered in a larger histone peptide or a whole histone protein (number of modified residues usually exceeds two) cannot be solved by a routine tandem mass spectrometry (MS2) strategy. (a) Four lysines (residue 5, 8, 12, and 16) in the N-terminal of histone H4 are known to be acetylated. In a monoacetylated peptide (residue 4–17) generated in Bottom Up/derivatization experiment, there are four possible positional isomers. K5ac and K16ac can be determined by MS2 fragment ions from regions 1 and 3, respectively (blue flag). However, fragment ions from region 2 can only determine the combined amount of K5ac + K8ac and K12ac + K16ac. Therefore, multiple linear equations are required to quantify these positional isomers. (b) Long-standing problem of positional isomers of diacetylated H4 4-17. The number of equations formulated from ratios of MS2 measurement is fewer than the parameters needed to be solved for ratios of these isomers. Recent efforts used MS3 (refragmentation of an isolated MS2 fragment ion) [38^••] and unique fragment ion patterns associated with each isomer to deconvolute them [39^••]. (c) When two methyl groups are detected in MS2 fragment ions with multiple lysines, it could be a dimethylation on a single lysine or two monomethylation on two residues. Such complication was called ‘convolving’ effect of multistate modification [7]. Two hypothetical scenarios of equal molar ratio mixture carrying two methyl groups are depicted here to illustrate this problem. The relative levels of measured methyl groups present in the fragment ions (N-termini to the cleavage site) are shown in the right panel. Because ‘meX’ (i.e., me1,2,3) is used in literature for mono-, di-, and trimethylation on a single residue, we use ‘Xme’ to indicate ‘X’ number of methyl groups is inferred from MS data. As shown here, identical fragment ion pattern is observed for both scenarios, which makes modification assignment ambiguous. However, both scenarios can pass current search algorithms relying on the presence of ‘site-determining’ MS2 fragment ions. Therefore, non-existed modification forms could be ‘created’ due to the false interpretation of MS2 evidences that identify both mixtures when only one of them actually existed.

The area under the curve, or simply peak area, in LC–MS has become a major quantitation strategy due to its high reproducibility, large linear dynamic range and multiplexed nature (i.e., hundreds of targets can be monitored in a single LC–MS/MS run). ‘Discovery’ and ‘targeted’ modes are two general data acquisition strategies for tandem MS. In ‘discovery’ mode, multiple precursor ions identified in the first MS survey scan (MS1) are selected for fragmentation in the subsequent MS events (MS2). This workflow, also called data dependent acquisition (DDA), is the current standard for peptide identification because its data collecting scheme enables powerful database searching, even in ‘error-tolerant’, discovery mode [31]. However, DDA has limited quantification capabilities due to stochastic precursor ion selection. More specifically, MS2 events are not collected frequently enough over the entire peptide elution window to generate the best quantitative metrics. As a result, DDA mainly provides MS1-based quantitation as shown for an example in Figure 2b. Therefore, Fragment Ion Relative Ratios (FIRRs) from discriminative fragment ions are required to further divide the total area determined in MS1 to quantify multiple isobaric species (see inset of left panel in Figure 2b) [21^•]. However, due to the complexity of highly modified histone peptides, the implementation of such an approach is challenging, relegating it to expert laboratories to perform large-scale studies involving hundreds of modified peptides. In an elegant methodology developed by the Garcia laboratory, an entire set of modified histone peptides was generated for a precursor ion and a superposition problem was formulated using mixed integer linear optimization to determine the relative fractions of the isobaric species present in the multiplexed MS2 spectra [32^••]. Recently, a software tool, EpiProfile, was further developed to facilitate this quantitation workflow [33^••].

Selected or multiple reaction monitoring (SRM/MRM) is a ‘targeted’ mode of LC–MS running mostly on triple quadrupole mass spectrometers, in which transitions (i.e., a precursor transitioning into a specific fragment ion) are used to generate a chromatogram for a predetermined target. The target list has grown rapidly from a single mark (H3K56ac) in the first report [14] to 21 shortly after [34] and 93 in the most recent effort [16^• ]. The development of software such as Skyline and Pinpoint has simplified the workflow for targeted proteomics using the venerable SRM/MRM approaches [35]. Examples of visualized SRM peaks in Skyline are shown in the right panel of Figure 2b. Skyline allows users to manually select peaks, which is useful for determining the correct peak in the presence of interfering signal, such as the trimethylation vs. acetylation example illustrated in Figure 2c. More recently, multiplexed parallel reaction monitoring (PRM) running on the Q-Exactive mass spectrometers achieves a SRM-like procedure to quantify histone modifications [15^••,36^••]. The simplification of data analysis afforded by SRM/PRM has allowed the large-scale screening of altered histone modifications in the burgeoning field of cancer epigenetics [15^••,37].

When MS2 is not enough

When ≥3 modified residues are present in a peptide/protein, MS2 fragment ions are not adequate to rigorously localize and quantify all modifications. As shown in Figure 3a, internal acetylation (K8ac and K12ac) on a monoacetylated H4 4-17 peptide cannot always be cleanly distinguished by MS2 fragment ions. Instead, multiple linear equations determined by ratios of fragment ions from all three regions between modified residues are required to calculate their abundances [27]. Two strategies have been reported recently to solve the long-standing problem of four ‘positional isomers’ of a diacetylated H4 (Figure 3b, [27]). The first one used an MS3 approach [38^••] and the second took advantage of the characteristic fragment ion pattern associated with each positional isomer [39^••]. This isobaric problem grows more complicated when methylation is involved. For example, when two methyl groups are inferred in the fragment ion (Δm = 28 Da), a single dimethylation on a single lysine residue or two monomethylation on two separate residues are ambiguously indicated. This so called ‘convolving effect’ of methylation has been shown to complicate the analysis of codes harboring ≥3 marks of the same type, making MS3 or other approach necessary [7^••]. A hypothetical example of the convolving effect is shown in Figure 3c to demonstrate the possibility of erroneous assignment of a code using current search algorithms for matching MS2 fragment ions.

Caveats of Bottom Up quantitation

As noted about Bottom Up above, sample processing can affect the readout of low abundance histone marks. Derivatization reagents other than propionic anhydride, such as deuterated anhydride [40] should be used when measuring endogenous lysine propionylation [41]. A hybrid method using phenyl isocyanate to derivatize the newly generated N-termini after trypsin digestion has been developed to avoid the selective losses of very hydrophilic peptides, such as H3K4me2/3 [42]. More significantly, chemical derivatization using acid anhydride can partially alkylate the hydroxyl group of serine and threonine [43] and methanol used in the procedure can lead to the methyl esterification of glutamic acid and aspartic acid residues [14]. However, a direct comparison of 12 commercially available anhydrides with different hydrophobicity suggests propionic anhydride is still the best choice for the overall performance [44] yet can introduce a few percent of acetylation from contaminating acetic anhydride. To reduce such deleterious side reactions, N-hydroxysuccinimide (NHS) propionate has been demonstrated for its specificity [43]. An evaluation of side reactions using propionic anhydride and NHS-propionate at various conditions has helped to identify proper procedures for accurate quantitation [45].

A comprehensive evaluation of the detection efficiencies performed recently demonstrated a wide spectrum of differences (>1700-fold) using MS1 quantitation by extracted ion chromatograms [46^•]. Because these detection efficiencies are method and platform specific, they are not applicable to generally correct all quantitation methods. Luckily, the correction factors to reduce detection bias does not change the conclusions (i.e., up- or down-regulated) for most histone marks except some low abundant ones in a differential analysis [46^•]. In addition, spiked-in internal standards, such as stable-isotopic-labeled synthetic peptides [47] and histones purified from SILAC labeled cells [15^••], can be used to improve quantitation accuracy across the study. Top Down Epiproteomics using either LC–MS or direct infusion has fewer sample handling steps, introduces significantly fewer sources of bias and directly reads out (not inferring) proteoform-level dynamics and PTM crosstalk. Various free and commercial versions of a software, ProSight, have been developed to support a complete view of histone proteoforms by Top Down MS over the years [22,23,26].

Conclusions and outlook

With the steady advancements in three pillars of LC–MS analysis (instrumentation, chromatographic separation, and informatics), Middle/Top Down will play increasingly important roles for histone modification analysis because they can provide more comprehensive information that putatively links to function more directly (vide supra and Figure 4). The advantages go beyond the aforementioned readout of co-occurring, multiple modifications. Two very recent Middle/Top Down studies have identified unexpected results quite readily: the clipping of histone tails [48] and the hypermethylated trivalent marks in a histone methyltransferase overexpression system [7^••]. However, more challenges arise from dissecting the increasing number of isobaric species in Middle/Top Down, which demands innovations in MS3 or other approach, such as internal fragmentation [49] to distinguish them. Finally, the use of native Top Down MS [50] on whole nucleosomes may eventually be able to readout intermolecular codes imprinted multiple histone tails non-covalently bound together.

Figure 4.

Estimated levels of major H3/H4 modifications in the pool of bulk histones depicted on a log scale of absolute abundance. When properly calibrated, mass spectrometry can provide absolute abundance information, such as the numbers of nucleosomes per cell carrying a specific epigenetic mark. The ubiquitous presence of the major histone marks in their individual state (e.g., H3K27me3 present in >1 million of nucleosomes/cell) means their abundances are unlikely to change when transcriptions from a small subset of genes are altered (e.g., in response to the loss of UTX, a H3K27 demethylase). On the other hand, those combinatorial modifications (or ‘codes’) only present on 100–1000 nucleosomes, as an example shown in red, are more likely to correlate with the transcription activity from a small subset of genes. It is worth noting that comparing the frequency of combinatorial modification, consisting of two co-occurring marks, against the co-frequency of these two independent marks is a simple yet elegant method to identify positive and negative crosstalk between pairs of marks [8]. For example, the frequency of H3 K27me3-K36m3 is near 100-fold less than the co-frequency of H3K27me3 and H3K36me3, suggesting a strong negative crosstalk between these two marks.