A method for systematic mapping of protein lysine methylation identifies new functions for HP1β in DNA damage repair

SUMMARY

Lysine methylation occurs on both histone and non-histone proteins. However, our knowledge on the prevalence and function of non-histone protein methylation is poor. We describe here an approach that combines peptide array, bioinformatic and mass spectrometric analyses to systematically identify lysine methylation sites in proteins and methyllysine-mediated protein-protein interactions. We demonstrate the utility of this approach by identifying a methyllysine-driven interactome of the heterochromatin protein (HP) 1β and uncovering, simultaneously, numerous methyllysine sites on non-histone proteins. The HP1β interactome is enriched with proteins involved in DNA damage repair and RNA splicing. We showed that lysine methylation played a pivotal role in the function of the DNA-dependent protein kinase catalytic subunit (DNA-PKcs) and its interaction with HP1β during DNA damage response. Moreover, by combining heavy methyl SILAC with Multiple Reaction Monitoring (MRM) mass spectrometry (MS), we showed that lysine methylation underwent widespread and large changes in response to DNA damage. Our work indicates that lysine methylation is a highly dynamic post-translational modification occurring frequently on non-histone proteins and that the approach presented herein may be extended to many methyllysine-binding modules to systematically uncover lysine methylation events in the cell.

INTRODUCTION

Lysine acetylation and lysine and arginine methylation are post-translational modifications (PTMs) that occur on histone tails and some non-histone proteins. Histone acetylation is often associated with a relaxed chromatin state and enhanced gene transcription whereas deacetylation corresponds to gene repression or silencing (Yang, 2004; Yang and Seto, 2003). Histone methylation, in contrast, may lead to either activation or repression of gene transcription, depending on the site and state of the modification (Barski et al., 2007). Lysine methyltransferases (KMTs) catalyze methylation of a Lys residue in an S-adenosyl-L-methionine (AdoMet)-dependent manner (Paik et al., 2007; Smith and Denu, 2008). The ε-amino group of Lys may be mono-, di- or tri-methylated, with each methylation-state often associated with a distinct biological outcome (Lake and Bedford, 2007). Arg residues may be mono- or di-methylated (either asymmetrically or symmetrically) on its guanidino nitrogens by protein arginine methyltransferases (PRMTs) (Bedford and Richard, 2005).

Different combinations of histone modifications encode information, referred to as the “histone code”, that is interpreted by downstream factors to translate the modification patterns into distinct biological outcomes (Taverna et al., 2007a). These effector molecules often harbor modular domains that recognize the PTMs on histones and are referred to as chromatin-binding modules (CBMs) (Taverna et al., 2007b). The Royal superfamily, including the chromo domain (CD), the tudor domain(s) and the MBT (malignant brain tumor) domain, forms the largest group of CBMs that recognize methyllysine (Kme) marks. For instance, chromo domains have been shown to recognize di- and tri-methyls in the H3K9 and H3K27 contexts (Jacobs and Khorasanizadeh, 2002; Jacobs et al., 2001). Tudor domains, which are usually found in tandem in a protein, may recognize mono-, di- or tri-methyllysine (Kim et al., 2006; Liu et al., 2010).

While tremendous strides have been made in recent years in the understanding of how histone methylation regulates a plethora of biological functions such as chromatin structure, gene expression, cell growth and differentiation (Martin and Zhang, 2005), our knowledge on the methylation of non-histone proteins and its physiological function has lagged behind. It remains to be determined, at the proteome level, whether Lys and/or Arg methylation occurs as widely as other PTMs such as Ser/Thr/Tyr phosphorylation and what methylation entails for the modified proteins. In this regard, a number of recent studies provided examples demonstrating the importance of methylation in regulating the function of a non-histone protein. For example, methylation has been shown to increase p53 stability and alter its subcellular localization (Lake and Bedford, 2007). Specifically, K372 in p53 is mono-methylated by SET7/9 (Chuikov et al., 2004), K370 is mono-methylated by Smyd2 (Huang et al., 2006a), and K382 is mono-methylated by SET8 (Shi et al., 2007). The MBT repeats of L3MBTL1 reads out the K382me1 signal while the double tudor domains of 53BP1 recognizes the K382me2 mark. Besides p53, methylation sites have also been reported for pRb (Munro et al., 2010) and TAF10 (Couture et al., 2006). Methylation of the retinoblastoma tumor suppressor protein pRb on Lys810 impedes binding of cyclin-dependent kinases and its subsequent phosphorylation (Carr et al., 2011).

On the genome level, data on protein methylation are sparse compared to protein phosphorylation. A total of 169 Kme sites from 72 proteins and 263 Rme sites from 92 proteins, including both histone and non-histone proteins, have been identified in humans based on published data collected at PhosphoSitePlus (Hornbeck et al., 2012). These numbers dwarf in comparison to the tens of thousands of pSer, pThr and pTyr sites identified to date. Given that the human genome encodes approximately 50 histone methyltransferases (Shi, 2007), it is reasonable to assume that the size of the protein methylome would be much greater than what is suggested by the number of sites identified to date. If this is indeed the case, the majority of methylation must occur on non-histone proteins. Using methyl-specific antibody for affinity-purification followed by mass spectrometry (MS) analysis, Mann and colleagues identified 59 Arg methylation sites on a variety of non-histone proteins (Ong et al., 2004). To the best of our knowledge, no high throughput method has been reported to identify lysine methylation in non-histone proteins, owing partly to the lack of an antibody that specifically and efficiently bind to the methyllysine.

To tackle this important problem, we have developed an integrated approach that combined peptide array screening and bioinformatics with mass spectrometry to identify, de novo, Lys methylated proteins and determine the corresponding methylation sites. Our approach is based on the specific recognition of methyllysine sites by modular domains (called Kme-binding domains or KMBD herein) such as the chromodomain (CD). The specificity profile of the KMBD, determined by peptide arrays, is used to predict potential sites of lysine methylation in the pool of proteins co-purified with the domain. These candidate sites are then validated by MRM, a MS method that can be used to identify proteins and PTM events based on the detection of multiple product ions from one or more precursor ions (Yocum and Chinnaiyan, 2009).

We report here the application of our approach to the heterochromatin protein 1β or HP1β (Nakayama et al., 2001; Peters et al., 2001). Through its chromodomain (CD), HP1β binds the H3K9me3 mark with high specificity and affinity (Kwon and Workman, 2008; Lomberk et al., 2006) and thereby plays an essential role in transcriptional repression by helping package the silenced genes into repressive, heterochromatin domains (Grewal and Elgin, 2002). Using the CD as bait, we identified the first methyllysine-driven interactome of HP1β. Our approach led to the identification of 109 HP1β associated proteins, among which 29 are methylated on 40 Lys residues. The HP1β interacting proteins are segregated into distinct functional groups that include DNA damage response and RNA splicing, suggesting HP1β play an important role in these processes. The physiological relevance of the identified HP1β interactome and the methylation sites is demonstrated by biochemical and cellular data showing that HP1β bound to DNA-PKcs in a methyllysine-dependent manner and that this interaction played a critical role in cellular DNA damage response. By coupling heavy methyl SILAC (Ong et al., 2002) with MRM, we showed that proteins belonging to different functional groups in the HP1β interactome exhibited distinct dynamics of methylation in response to etoposide-induced DNA damage.

RESULTS

A General Strategy to Identify in vivo Lysine Methylation Sites

Due to lack of an effective enrichment method, systematic identification of lysine methylation sites has proven difficult. This is in contrast to the numerous phosphorylation sites and lysine acetylation sites identified by mass spectrometry to date (Bodenmiller and Aebersold, 2010; Shaw et al., 2011). In this work, we developed a proteomic method that combined MS analysis with peptide arrays to identify functional lysine methylation sites in histones and non-histone proteins (Fig. 1). We reasoned that a significant fraction of methyllysine sites would be recognized by methyllysine (Kme)-binding domains (KMBD) (Liu et al., 2010). Therefore, one could use the appropriate KMBD to enrich for lysine methylated proteins prior to MS analysis for protein identification (Fig. 1). Because proteins identified by the KMBD-affinity purification coupled with MS analysis (AP-MS) contain both methylated and non-methylated proteins, we developed a method to distinguish the former from the latter. In particular, we first determined the specificity of the KMBD by both methyllysine peptide arrays and/or permutation arrays of known ligands of the KMBD. In a permutation array (Jia et al., 2005), every residue in the methyllysine (Kme)-containing peptide substrate is replaced, respectively, by the 20 naturally occurring amino acids. The resulting array was then probed for binding to the corresponding KMBD. Quantification of the binding signals in the permutation array generates a scoring matrix that represents the preference of an amino acid at a given position of the peptide substrate. This specificity information was then imported to the SMALI program (Li et al., 2008) to predict candidate methyllysine sites from the list of proteins identified by AP-MS. Methyllysine peptides that produced a SMALI score above the cut-off (to be determined empirically) were subsequently verified by MRM-MS (Lange et al., 2008) (Fig. 1). To demonstrate the utility of this method, we used it to identify methyllysine-dependent interactions for HP1β.

Figure 1. Schematic of a strategy that combines mass spectrometry with peptide arrays and bioinformatics to identify lysine methylation (Kme) sites in histone and non-histone proteins.

Nuclear proteins are affinity-purified by an immobilized methyllysine binding domain (KMBD) and the associated proteins identified by MS/MS. In complementary experiments, peptide arrays, including permutation arrays based on a known ligand sequence, were used to determine the specificity of the KMBD. The specificity information was imported to the online program SMALI for in silico prediction of methyllysine sites in candidate proteins identified from LC-MS/MS analysis. Methyllysine sites/peptides scored above the cut-off value were confirmed by Multiple Reaction Monitoring (MRM)-mass spectrometry.

Identification of a Methyllysine-mediated HP1β Interactome

HP1β contains a chromodomain (CD) that recognizes the H3K9 and H3K23 marks in a methylation-dependent manner (Liu et al., 2010). To identify methyllysine-dependent binding partners for HP1β, we expressed the HP1β-CD in E.coli, purified it to homogeneity, covalently linked it to the MS300/carboxyl magnetic beads (JSR Co., Japan), and employed the beads to isolate proteins from the HEK293T cells. Because HP1β is localized in the nucleus (Luciani et al., 2005), we employed the nuclear fraction in the the HP1β-CD pull-down experiment to isolate physiologically relevant complexes. To minimize false positives, we eluted the proteins from the beads using an H3K9me3 peptide, TKQTAR[Kme3]STGGKA, to ensure that only specifically bound proteins were collected (Fig. S1). The corresponding unmethylated peptide, which was included as a control, was incapable of eluting methylated proteins (Fig. S1). The eluate was then digested with trypsin and analyzed by LC-MS/MS, which led to the identification of 112 proteins from three independent runs of MS/MS analysis. This number was reduced to 109 when the three proteins (i.e., HMGA2, RBM25, and SAS10) that showed up in the H3K9 peptide eluate were subtracted (Supplementary File 1-Table S1). These proteins were grouped based on functions annotated in the UniProt database (Consortium, 2012), and connected according to the STRING database (Szklarczyk et al., 2011). Histones H1 and H3, which are known interactors of HP1β (Daujat et al., 2005; Nielsen et al., 2002), emerged in the list of specific binding proteins, validating the effectiveness of the AP-MS/MS approach to identify functionally relevant proteins. Intriguingly, the majority of proteins in the identified interactome are non-histone proteins, suggesting that the functions of HP1β go well beyond its known roles in binding methyl H3K9 in chromatin biology (Billur et al., 2010). Based on the connectivity and function, the HP1β-binding proteins were segregated into four groups, namely DNA damage repair, RNA splicing, ribosomal proteins and miscellaneous proteins (Fig. 2). The identification of multiple components of the DNA damage repair and RNA splicing machineries suggest that our AP-MS approach led to the isolation of functional complexes that are nucleated by the HP1β-CD. Although commonly recognized as cytoplasmic proteins, ribosomal proteins can localize in the nucleus (Moreland et al., 1985). The identification of a cluster of ribosomal proteins is in agreement with Egorova et al (Egorova et al., 2010) who counted ribosomal proteins as one of the largest groups of non-histone proteins that are lysine methylated (vide infra).

Figure 2. An HP1β interactome identified from affinity purification of HP1β-CD binding proteins followed by LC-MS/MS analysis.

Proteins in the interactome are grouped according to functions as annotated in the UniProt database (Consortium, 2012). Color codings are as follows: green, RNA splicing; red, DNA damage repair; blue, ribosomal proteins; grey, miscellaneous; yellow, RNA splicing and DNA damage repair. Purple circles denote proteins that contain Kme sites as determined by subsequent LC-MRM-MS analysis (see also Table 1). Interacting proteins are connected via lines based on the STRING database (Szklarczyk, 2011) using the high confidence cut-off. Interactions within each functional cluster are identified in black lines whereas those between clusters in grey lines. Proteins that co-immunoprecipitated with HP1β (Fig. 4) are identified in cyan. See Supplementary File-1 (Table S1) for additional information on the HP1β interactome. The interactome was generated using NAViGaTOR (Brown et al., 2009).

Predicting Direct Binders of HP1β Based on the Specificity of its Chromodomain

Similar to other AP-MS experiments, the proteins pulled down by HP1β-CD may or may not bind to HP1β directly. To obtain a higher resolution interactome, we set out to identify direct binders of HP1β. A direct binder of the HP1β-CD should contain a methyllysine residue embedded in a sequence that conforms to the specificity of the domain. Although the HP1β-CD was found to recognize the methylated H3K9 mark (Nielsen et al., 2002), its specificity has not been fully defined. We began by determining the minimal motif recognized by the HP1β-CD based on the H3K9me3 peptide. To this end, we synthesized a series of peptides on a cellulose membrane that represented different truncated analogs of the H3K9me3 (Fig. 3A). The binding profile of the resulting peptides to HP1β-CD suggests the minimal binding motif lies between residues −3 and +1, relative to the Kme3 residue (Fig. 3A). To further delineate the specificity of the CD, we generated a peptide spot array in which each residue, except Kme3, in the K^-5QTAR-Kme3-STG⁺³ peptide, was replaced by a naturally occurring amino acid. This permutation peptide array was then probed by the HP1β-CD. As shown in Fig. 3B, apparent selectivity was observed for positions −4 through +1, with position −2 favoring only the Ala and Gly residues. It is notable that the wild-type sequence was strongly favored, suggesting that the H3K9me3 peptide is optimized for binding to the HP1β-CD. To investigate if the methylation state of a peptide would affect its specificity, we synthesized permutation arrays in which the Kme3 residue was replaced by either Kme1 or Kme2. As shown in Fig. S2A, essentially identical binding patterns were obtained from permutation arrays anchored with either Kme-1, −2, or −3. Collectively, these data indicate that the methylation state of lysine does not affect specificity conferred by the flanking residues.

Figure 3. Predicting Lys methylation sites based on the specificity of the HP1β-CD.

(A) Binding of HP1β-CD to a peptide array containing a series of truncated versions of the H3K9me3 peptide, RTKQTAR[Kme3]STGGKA. (B) Binding profile of HP1β-CD to a permutation peptide array based on the peptide, RKQTAR[Kme3]STGG. (C) Correlation between the experimental binding signal as reported by Liu et al (Liu et al., 2010) and the SMALI score for a group of histone methyllysine peptides. (D) Determination of SMALI cut-off for predicting Lys methylation sites. The SMALI score of the candidate methyllysine peptides are plotted against the Ala/Gly content at the −2 position (relative to the methyllysine residue). A SMALI cut-off was set at 7 at which point 80% of peptides contained an Ala or Gly at the position −2. (E) Distribution of lysine-containing peptides in the HP1β interactome against SMALI scores. Peptides with a score > 7.0 were predicted to contain a methyllysine. The remaining peptides (>98%) were filtered out.

The permutation array-binding signals were quantified to build a Position-Specific Scoring Matrix (PSSM) in a manner similar to what was described previously for the SH2 domains (Li et al., 2008). The scoring matrix was then imported into the online program SMALI (Li et al., 2008) to predict HP1β-CD binding peptides/proteins. To evaluate the accuracy of prediction, we used SMALI to predict interactions that have been experimentally investigated in a previous study employing a histone methyl-peptide array (Liu et al., 2010). As shown in Fig. 3C, the three binding peptides, namely H3K23me3, H1K25me3 and H3K9me3, identified by the peptide array screen, were clustered in the upper-right corner of the graph. Moreover, each of these peptides had a SMALI score greater than 7.0 whereas the non-binders formed a separate cluster with scores smaller than 7.0. This indicates that SMALI based on the specificity of the chromodomain could be used to accurately predict binding partners for the HP1β-CD.

To determine whether the template sequence of the permutation array would affect the PSSM and consequently, the accuracy of SMALI prediction, we employed the H3K23me3-derived PSSM to repeat the prediction and found that the result correlated well with that produced using the H3K9me3-derived PSSM (Fig. S2B). This suggests that the PSSM-based prediction is not significantly affected by the template sequence used in the permutation array analysis. To test the generality of the permutation peptide array strategy, we applied it to another methyllysine-binding module, the JMJD2A-double Tudor domains which have been shown to bind to the H3K23me3, H3K4me3 and H4K20me3 marks (Huang et al., 2006b; Lee et al., 2008; Liu et al., 2010). As shown in Fig. S2C, distinct binding patterns were obtained for the JMJD2A Tudor domains on the H3K23me3, H3K4me3 and H4K20me3 peptide arrays. Of note, the specificity profiles revealed by the H3K4me3 and H4K20me3 permutation arrays was in excellent agreement with the crystal structures of the corresponding peptide-Tudor domain complexes (Lee et al., 2008). It is expected that the specificity information would be useful in predicting methyllysine sites recognized by the JMJD2A Tudor domains in a manner similar to what is described above for the HP1β-CD.

The 109 proteins in the HP1β interactome contain a total of 5743 lysine residues (Supplementary file 1-Table S3). Because any of these Lys residues may be methylated, it is necessary to distinguish the sites with a high probability to be methylated (and thereby conferring HP1β-CD binding) from those with a low probability. This filtering process would significantly reduce the workload and at the same time, maximize the chance of success in subsequent MRM-MS experiments designed for the identification of methyllysine sites. We, therefore, ranked the peptides corresponding to the 5743 Lys sites with SMALI using a 11-residue window centered at the Lys site (Li et al., 2008). Because the presence of either Ala or Gly at the −2 position is crucial for HP1β-CD binding (Fig. 3B), we plotted the A/G-2 content against the corresponding SMALI score for all the Lys-containing peptides. As seen in Fig. 3D, the correlation between A/G-2 content and SMALI followed a sigmoidal pattern. Based on this data, a cut-off SMALI score of 7.0 was applied to all peptides to identify candidate Lys sites with a high probability (>80%) of methylation. This filtration process collapsed the potential Lys methylation sites from 5743 to 95 (Fig. 3E).

Identification of Methylated Proteins and Lys Methylation sites in the HP1β Interactome

We next set out to determine, using MRM, how many of the 95 predicted Lys residues were indeed methylated. A new batch of HP1β-CD beads was used to pull down proteins from the nuclear fraction of HEK293T cells. The sample was divided into four fractions, each of which was digested with a different enzyme, namely ArgC, GluC, elastase, and chymotrypsin, prior to MRM-MS analysis. The Skyline program (MacLean et al., 2010) was used to generate theoretic peptide fragments that contain at least one Lys for each protease and a list of transitions to monitor the corresponding precursor ions (Supplementary File 2). Specific b/y product ions were chosen to distinguish a Lys methylated fragment from the unmethylated form. Using MRM-IDA (Information Dependent Acquisition) method, MRM peaks were verified in the corresponding MS/MS spectra. Representative MRM and MS/MS spectra were shown in Fig. S3 and Fig. S4, respectively. A positive identification was made when three or more MRM transitions corresponding to either b- or y-type product ions of a precursor containing the methyllysine residue under investigation were detected. For example, the peptide IIEKKHVSLNKAK(1150)R corresponding to the K1150 of DNA-PKcs was detected in a sample digested with ArgC. As shown in Fig. 4A, 9 product ions from the K1150me3 precursor ion were observed in the MRM-IDA experiment. Because there are multiple lysines in this peptide and any lysine methylation will give the same molecular mass, it was important to include daughter ions specific for the K1150 methylation in the MRM analysis. For instance (Fig. 4A), the b14 ion contained the intended methylation site (K1150me3) whereas the b13 ion did not. The detection of the b14 ion was therefore crucial for the identification of K1150 trimethylation. The extracted ion chromatograms (EIC) of all MRM transitions (listed in Fig. 4B) observed for the K1150me3 peptide is shown in Fig. 4C.

Figure 4. Identification of Lys methylation sites by MRM and validation of the HP1β interactome.

(A) Determining Kme sites in the HP1β interactome by MRM-MS. A diagram showing the MRM transitions detected for a peptide from the K1150me3 site in DNA-PKcs. (B) A list of precursor and product ions as depicted in (A) showing the corresponding m/z and charge. The b13 and b14 ions unequivocally identified the trimethylated K1150 residue. (C) Extracted Ion Chromatographs (XIC) showing the MRM transitions detected for the K1150me3 peptide. (D) A pie chart showing that 67 (83.75%) of the 80 non-methylated proteins in the HP1β interactome interact with the 29 Lys methylated proteins according to the STRING database.

(E) Validation of representative interactions in the HP1-β interactome. HP1β was immunoprecipitated (IP) from HEK293T cells and blotted (IB) for the indicated proteins, respectively. An anti-HP1β and an anti-lamin A/C blot were included as controls. (F) Co-immunoprecipitation of endogenous DNA-PKcs with HP1β from U2OS cells. WCL, whole cell lysate.

Using the same approach, we examined the methylation status of the 95 candidate peptides by MRM. In the control experiment 9 peptides with an SMALI score below the cut-off value (7.0) were randomly chosen for MRM (Supplementary File 1-Table S3). The MRM analysis led to the identification of 40 methyllysine sites (with di- or/and tri-methylation) in 29 proteins, or a 40% confirmation rate of the SMALI prediction, which is highly significant (p = 4.02E-76, Hypergeometric test). In contrast, none of the control peptides was found methylated. While the MRM experiments were optimized for 4000 QTRAP instrument, we also repeated the assays on the 5500 QTRAP machine and obtained a 92.5% validation rate (Table 1). Together, these data suggest that specificity-based prediction is highly effective in filtering out non-methylated Lys residues and identifying authentic Lys methylation sites. Intriguingly, the identified Lys methylation sites spread in all functional clusters of the HP1β interactome, suggesting that Lys methylation may regulate different aspects of HP1β function (Fig. 2). A full list of methylated proteins and the corresponding Lys methylation sites are provided in Table 1 with the corresponding MRM transitions listed in Supplementary File 2.

Table 1.

Methyllysine sites identified in the current study

UniProtKB ID	Protein Name	Lys Position	Protease	40000 QTRAP		5500 QTRAP
				Kme2	Kme3	Kme2	Kme3
P46013	MKI67	2652	Arg-C	√		√
P78527	DNA-PKcs	1150	Arg-C	√	√	√	√
A4UGR9	XIRP2	2799	Chymotrypsin	√	√	√	√
P38919	eIF4A3	374	Chymotrypsin	√		√
P38919	eIF4A3	70	Arg-C	√		√
Q9BYG3	MKI67IP	179	Chymotrypsin		√		√
Q14978	Nopp140	80	Glu-C		√		√
Q14980	NUMA1	1803	Chymotrypsin	√	√	N/A	√
A4UGR9	XIRP2	1944	Chymotrypsin		√		√
A4UGR9	XIRP2	1877	Glu-C		√		√
Q7L014	DDX46	113	Arg-C		√		√
O43143	DHX15	18	Arg-C	√	√	√	√
P52272	HNRPM	651	Chymotrypsin		√		√
P46013	MKI67	620	Arg-C		√		√
P46013	MKI67	1616	Glu-C		√		√
P46013	MKI67	263	Chymotrypsin		√		N/A
P13010	Ku80	7	Chymotrypsin		√		√
P42166	LAP2A	94	Arg-C		√		√
Q9Y383	LUC7L2	166	Chymotrypsin		√		√
P61326	MAGOH	14	Chymotrypsin		√		N/A
Q9NX24	NHP2	56	Chymotrypsin		√		√
P78527	DNA-PKcs	3248	Chymotrypsin	√		√
P78527	DNA-PKcs	2746	Chymotrypsin	√		√
O75533	SAP155	290	Glu-C	√	√	N/A	√
Q3T8J9	GON4L	1126	Chymotrypsin	√		√
P62906	CSA-19	106	Chymotrypsin	√		√
P62424	RPL7A	101	Chymotrypsin		√		√
Q8IY81	FTSJ3	763	Arg-C		√		√
Q8IY81	FTSJ3	749	Chymotrypsin	√	√	√	N/A
Q13247	SRp55	242	Elastase		√		√
Q8IYB3	SRm160	425	Elastase	√		√
Q9UQ35	SRm300	239	Arg-C	√		√
Q9Y2W1	Trap150	756	Arg-C		√		N/A
P61313	RPL15	54	Chymotrypsin	√	√	√	N/A
P68431	H3.1	10	Arg-C	√	√	√	√
P68431	H3.1	24	Arg-C	√	√	√	√
P10412	H1.4	26	Arg-C	√	√	√	√
P67809	CBF-A	92	Chymotrypsin		√		√
Q08211	RHA	120	Glu-C	√		√
Q7L014	DDX46	389	Chymotrypsin		√		√

Footnote: N/A, insufficient MRM transitions (<3) detected to allow identification with confidence

Validation of Interactions in the HP1β Interactome

We next employed peptide-binding, bioinformatic analysis and co-immunoprecipitation (Co-IP) to validate the interactions of the HP1 interactome. Because chromodomains bind to their targets in a methylation-dependent manner (Jacobs and Khorasanizadeh, 2002), we first examined whether the identified Kme sites represented high-affinity binding sites for the HP1β-CD. Peptides representing all 40 Lys methylation sites were therefore synthesized and measured for binding to purified HP1β-CD or the defective mutant W42A (Nielsen et al., 2002) by fluorescence polarization. As shown in Supplementary File 1-Table S3, while 35 methyl peptides bound to the HP1β-CD with affinities smaller than 100 µM, none bound to the W42A mutant with Kd <150 µM. This suggests that the majority of the Kme sites are capable of mediating direct binding between HP1β and the corresponding proteins. By inference, the first layer of binding proteins in the interactome would come from the 29 methylated proteins and the second layer of interactions comprises those between the methylated proteins and the remaining proteins of the interactome (Fig. 2). Should this be the case, one would expect that the two layers of proteins to interact. By searching the STRING database (Szklarczyk et al., 2011) at a cut-off of medium confidence, we found that 67 of 80, or 83.5% the non-methylated proteins were indeed connected to the 29 methylation proteins (Fig. 4D). To validate the 1^st layer of interactions, we examined the binding between HP1β and five methylated proteins by co-IP. As shown in Fig. 4E & F, eIF4A3, LAP2A, Ku80 (regulatory subunit of DNA-PK), and DNA-PKcs (catalytic subunit of DNA-PK) all bound as tightly as H3K9me3 (a positive control) to HP1β.

HP1β Binds to and Regulates DNA-PKcs Function during DNA Damage Response

More than a dozen proteins involved in DNA damage response were pulled down by HP1β-CD, suggesting a central role for HP1β in DDR. Of note, the entire complex of DNA-PK (Sibanda et al., 2010), including the catalytic domain DNA-PKcs and the regulatory proteins Ku70/Ku80, was found to precipitate with HP1β (Fig. 4E & F). In addition, three lysine residues, K1150, K2746 and K3248, in the DNA-PKcs were methylated (Table 1), suggesting that these sites may mediate DNA-PKcs binding to HP1β-CD. To confirm this possibility, we synthesized spot arrays of peptides representing these Lys sites in both methylated and unmethyl forms and tested their binding to purified HP1β-CD. Peptides representing the two known targets of HP1β-CD, namely H3K9me3 and H3K23me3 (Liu et al., 2010), were included as positive controls. As shown in Fig. 5A, the di- and tri-methylated, but not the unmethylated forms of the DNA-PKcs-derived peptides bound to HP1β-CD, and the binding signal strength for the three peptides correlated well with their SMALI scores (Table 1). The Kd values between the K1150me2/3 peptide and HP1β-CD are even smaller than that between H3K9me2 and HP1β (Fig. 5B; Supplementary File 1-Table S3).

Figure 5. HP1β binds to DNA-PKcs and regulates its function during DDR.

(A) HP1β-CD binds to methylated peptides taken from DNA-PKcs, but not the non-methylated counterparts. An array of peptides corresponding to DNA-PKcs K3248, K2746, K1150 and histone H3 K9 and K23 were synthesized on a cellulose membrane in non-methylated (Me0), di-(Me2) or trimethylated (Me3) forms and probed for binding to GST-HP1β-CD. Shown is data from an anti-GST Western blot. (B) Binding curves of the HP1β-CD to the K1150 peptide in different methylation states as measured by fluorescence polarization (FP) in solution. The H3K9me2 peptide was shown as a positive control. (C) Wild-type (wt) HP1β, but not a mutant containing the W42A mutation, pulled down DNA-PKcs from the cell lysate. HP1β or the mutant was expressed in E.coli with an (His)₆ tag, purified and used to pull down DNA-PKcs from U2OS cells. (D) DNA-damage promoted the DNA-PKcs-HP1β interaction. (His)₆-tagged HP1β was used to pull down DNA-PKcs from cells treated (+) or not (−) with etoposide. (E) A Western blot showing the expression level of DNA-PKcs and K-to-R mutants in the V3 cell lines. (F) Sensitivity of the V3 (DNA-PKcs null) cells or cells stably expressing the wilt-type (WT) or a K-to-R mutant of DNA-PKcs to ionizing radiation. A colony formation assay (Franken et al., 2006) was used to compare the radiation sensitivities of the V3-vector, V3-WT, V3-K1150R, V3-K2746R, V3-K3248 and V3-K(1150/2746/3248)R cells.

To ascertain that the DNA-PKcs-HP1β interaction is dependent on the CD domain, we generated a mutant of HP1β, HP1β-W42A, in which the ligand-binding residue, Trp42, in the chromodomain was mutated to an Ala (Nielsen et al., 2002). Both the wild-type (wt) and mutant proteins were expressed in E.coli and purified to homogeneity. The proteins were then used to pull down DNA-PKcs from the U2OS cell lysate. The wt, but not the W42A mutant, was capable of binding to DNA-PKcs (Fig. 5C), indicating that the HP1β-DNA-PKcs interaction is dependent on the chromodomain. Because both HP1β and DNA-PKcs are involved in DNA double strand break (DSB) repair, we were interested in finding out whether the interaction was affected by DNA damage. To induce DSB, we treated U2OS cells with etoposide, a topoisomerase inhibitor and an anti-cancer agent (Vock et al., 1998). Interestingly, more DNA-PKcs was pulled down by HP1β (Fig. 5D) from cells treated with etoposide than from the untreated cells, despite no change in the total amount of DNA-PKcs was observed.

Mutation of the Methyllysine Sites in DNA-PKcs or Depletion of HP1β Compromised the Function of DNA-PKcs

To understand the significance of lysine methylation in DNA-PKcs function, we generated a series of mutants in which Lys1150, Lys2746 and Lys3248 were replaced by Arg individually or all together. These mutants were stably expressed in a DNA-PKcs-deficient (V3) cell line (Fig. 5E) and the corresponding cell lines were examined for sensitivity to DNA damage induced by ionizing radiation (IR) upon exposure to the ¹³⁷Cs c-ray (Chen et al., 2007). As shown in Fig. 5F, the DNA-PKcs null V3 cells were sensitive to radiation as the survival fraction of cells decreased proportionally with the increasing dosage of IR. In contrast, cells expressing wild type DNA-PKcs showed strong resistance to radiation. The three single mutant cell lines, V3-K1150R, V3-K2746R and V3-K3248R, showed less resistance than the wild type cell line to radiation induced cell death, indicating that these lysine sites are important in DNA damage repair. The triple mutant underwent rapid degradation in the cell and the corresponding line, V3-Tri-R, showed the same sensitivity as the DNA-PKcs-null V3 line towards radiation from ¹³⁷Cs c-ray. These data suggest that the three Lys residues play a critical role in the proper function of DNA-PKcs in DNA damage response.

To investigate the biological significance of the HP1β-DNA-PKcs interaction in DDR, HP1β was depleted from U2OS cells. A specific siRNA, but not the corresponding scrambled siRNA, knocked down HP1β expression without affecting the expression of DNA-PKcs (Fig. 6A). DNA-PKcs is one of the first responsive proteins following a DSB. Upon DNA damage, DNA-PKcs is recruited to the damage foci together with the regulatory subunits Ku70/Ku80. DNA-PKcs or ataxia-telangiectasia-mutated (ATM) then phosphorylates S139 in H2Ax, creating γH2Ax (Martin et al., 2012). We therefore used γH2Ax as a hallmark of DSB (Dinant et al., 2008) and examined whether depletion of HP1β in U2OS cells had an effect on γH2Ax prior to or following etoposide treatment. Compared to cells treated with the vehicle (DMSO), cells incubated with etoposide were characterized with a markedly enhanced γH2Ax level (Fig. 6B). Intriguingly, this pattern was not affected by UNC0638, a specific inhibitor of G9a/GLC (Hauser and Jung, 2011) to block H3K9 methylation (Fig. 6C). Nevertheless, depletion of HP1β rendered the cells irresponsive to etoposide treatment, as both the treated and untreated cells displayed approximately the same level of γH2Ax (Fig. 6B). This suggests that HP1β is indispensable for DDR.

Figure 6. HP1β plays a role in the localization of DNA-PKcs to DNA double strand breaks (DSB).

(A) siRNA-mediated knockdown of HP1β from U2OS cells. A specific siRNA, but not the scrambled control oligo, was able to reduce HP1β expression. The expression of DNA-PKcs was not affected. (B) HP1β depletion affected DSB repair. The γH2Ax level was compared by Western blots for HP1β-depleted or control U2OS cells with or without etoposide treatment. (C) The H3K9me2 level in U2OS cells were decreased upon incubation with the G9a/GLC inhibitor UNC0638, but not the solvent DMSO. (D) HP1β recruited the activated DNA-PKcs (marked with pT2609) to DSB foci (marked by γH2Ax). Confocal immunofluorescence microscopic images of cells transfected with an HP1β-specific siRNA or the scrambled control and treated with etoposide (+ETO) or DMSO (-ETO) to monitor the colocalization of pT2609 with γH2Ax. Lower panel, cells were treated with the G9a/GLC inhibitor UNC0638 together with ETO. (E) Quantification of data in (D). The confocal images were quantified by the Pearson correlation coefficient using the JACoP plugin in the ImageJ software. *, p<0.005, n=5.

T2609 in DNA-PKcs is phosphorylated in response to DNA damage (Chan et al., 2002), which is required for the full activation of DNA-PKcs and the subsequent DSB repair (Chen et al., 2007). To investigate the role of HP1β on DNA-PKcs function during DSB repair, we used specific antibodies to immunostain pT2609-DNA-PKcs and γH2Ax foci in U2OS cells. While essentially no γH2Ax foci or pT2609 immunofluorescence signal was detectable in the control U2OS cells (-ETO), etoposide treatment (+ETO) of cells transfected with a control siRNA promoted the formation of pT2609 and γH2Ax foci that mark DNA-PKcs activation and DSB, respectively. The immunofluorescence signals for these two markers colocalized, confirming the presence of DNA damage (Fig. 6D). When HP1β was depleted by siRNA, etoposide was still able to generate γH2Ax and pT2609 foci, suggesting DNA-PKcs and H2Ax phosphorylation occur in the absence of HP1β. However, the pT2609 immunostaining signals no longer colocalized with the γH2Ax foci (Fig. 6D & E), indicating that HP1β plays a role in the localization of active DNA-PKcs to sites of DSB. Interestingly, application of UNC0638 did not significantly alter the collocalization pattern of the γH2Ax and pT2609 foci (Fig. 6D & E), suggesting that it is unlikely that H3K9 methylation plays an important role in the recruitment of activated DNA-PK to the damage foci. We also examined the colocalization of HP1β, γH2Ax and DNA-PKcs. As shown in Fig. S5A, they colocalized with one another upon etoposide treatment. The role of DNA-PKcs lysine methylation on its localization to DNA damage foci was investigated by confocal microscopy using the DNA-PKcs K/R mutant cell lines (Fig. S5B). These mutants only partially colocalized with the γH2Ax foci, suggesting these Lys residues play a role in the proper localization of DNA-PKcs during DSB repair.

DNA Damage Induced Wide-spread Changes in Lysine Methylation

Approximately 50 methyltransferases and a similar number of demethylases exist in the human genome, suggesting that lysine methylation may be as dynamic as phosphorylation (Shi, 2007). To explore this possibility, we measured the changes in lysine methylation for proteins of the HP1β interactome under cellular stress. Because HP1β plays an important role in the DNA damage response, we examined lysine methylation dynamics in response to DNA damage. To quantify changes in lysine methylation, we adapted the heavy methyl SILAC (Stable Isotope Labeled Amino Acid Culture) method developed by Mann and colleagues (Ong et al., 2002). Specifically, one batch of HEK293T cells was cultured in a medium containing Met-methyl-d3 (heavy medium), and a second batch was fed with normal methionine (Met-methyl-d0, light medium). Methionine is metabolized to S-adenosine methionine in cells and the latter is used as a methyl donor (Cantoni, 1952; Wood and Shilatifard, 2004). To induce DNA damage, cells grown in the heavy medium were treated with etoposide, whereas cells cultured in the light medium were treated with the solvent DMSO. This method enabled us to compare the change in methylation for a given lysine site induced by DNA damage.

We first examined the change in the global methylation status of K1150 in DNA-PKcs associated with etoposide treatment. To this end, equal amounts of lysate from the treated (heavy) and control (light) cells were mixed for immunoprecipitation using a DNA-PKcs specific antibody. We then monitored the MRM signals of K1150me3 after enzymatic digest of the DNA-PKcs immunoprecipitate. Both the heavy and light K1150me3 peptides showed the same MS/MS spectrum, confirming they had the same amino acid sequence (Fig. S4A). The difference in m/z between the normal and deuterated methyl allowed us to distinguish the light from the heavy K1150me3 peptide by MRM. The total ion chromatograms for the two peptides eluted at identical time provided a reliable measurement of their relative quantities (Fig. 7A). The peak area for the deuterated K1150me3 peptide is 22 fold of that for the light peptide. Because the total protein level of DNA-PKcs did not change significantly with the etoposide treatment (Fig. 5D), this suggests that DNA damage led to more than 20-fold increase in trimethylation for K1150 in DNA-PKcs.

Figure 7. Lysine methylation is a dynamic PTM.

(A) The dynamic change in K1150 trimethylation in DNA-PKcs induced by DNA damage. Shown are extracted ion chromatograms (XICs) of the MRM transitions corresponding to the deuterated K1150me3 peptide (red trace, from cells cultured in the CD3-Met-containing medium and treated with etoposide) and the control methyl peptide (purple trace, from cells cultured in the CH3-Met medium without etoposide treatment). (B) Dynamic changes in methylation during DDR observed for 16 Lys residues in different of proteins of the HP1β interactome. Shown are ratios of methylation between etoposide-treated and the control samples. A negative number denotes a reduction whereas a positive number an increase in the methylation level. Proteins/sites are color-coded according to function. Red, DNA damage response; green, RNA splicing; grey, miscellaneous.

To determine changes in lysine methylation associated with DNA damage on a larger scale, we examined the dynamics of other methyllysine sites in the HP1β interactome using the SILAC-MRM approach. To this end, the nuclear lysate from the etoposide-treated, Met-methyl-d3-fed cells was mixed in an equal amount with that from the control cells. This lysate mixture was then incubated with the HP1β-CD beads. Proteins bound to HP1β-CD were subjected to on-bead tryptic digest and analyzed subsequently by MRM-MS. This led to the quantification of changes for 19 methyllysine sites in 14 proteins. As shown in Fig. 7B, while different Lys sites exhibited distinct dynamics of methylation, proteins belonging to the same functional cluster showed a similar trend. In particular, we found that proteins involved in DNA damage repair displayed the largest range of dynamics in Lys methylation. For example, the level of K3248me2 in DNA-PKcs was up-regulated 100 fold while that of K7me3 in Ku80 down-regulated by nearly 100 fold with etoposide treatment. Curiously, the levels of MKI67IP-K179me2, SAP155-K290me2 and HNRPM-K651me2 increased while those of HNRPM-K651me3 and XIRP2-K1944me3 decreased dramatically with the same treatment, suggesting that DDR is a complex process involving not only enzymes in DNA damage repair but also those regulating alternative splicing, cell proliferation and nuclear functions.

DISCUSSION

Systematic Identification of Lysine Methylation and Methyllysine-driven Protein-protein Interactions

The human genome encodes approximately 50 Lys- and Arg-methyltransferases, a similar number of demethylases (Di Lorenzo and Bedford, 2011; Kooistra and Helin, 2012), and hundreds of modular domains dedicated to the recognition of methyllysine or methylargine sites (Albert and Helin, 2010). The toolkits of protein methylation are therefore comparable in size to those for tyrosine phosphorylation which comprises 90 tyrosine kinases, 31 protein tyrosine phosphatases, and more than a hundred phosphotyrosine-binding domains (Alonso et al., 2004; Janas and Van Aelst, 2011; Liu et al., 2006; Uhlik et al., 2005). The parallels between these two types of post-translational modifications hint at the possibility that the protein methylome may be comparable in size to the phosphotyrosyl proteome (Jones et al., 2006). By interference, the methylation-driven protein interactome may match in size and complexity to that driven by tyrosine phosphorylation. The lack of a high throughput method by which to identify methyllysine sites has so far hindered efforts to elucidate the protein methylome and the methyl-driven interactome in a systematic manner.

In this regard, the method described herein provides an approach that may be used to speed up the pace of discovery in this burgeoning area of epigenetic research. Most current AP-MS methods for the identification of PTM events rely on the enrichment of the modified peptide/proteins using a specific antibody or an affinity matrix. Our approach, in contrast, takes advantage of the specific recognition of methylated residues by a modular domain. Several unique features stand out for the latter. First, the methylome is fractionated prior to MS analysis, thereby increasing the chance of success in identification. Of the hundreds of methyllysine and methylarginine binding domains that may be employed by our approach, each has unique specificity and preferred binding partners in the cell. These modular domains provide a valuable tool to separate the methylome according to their binding specificities prior to MS analysis. Second, our approach may be used to map the methylation site and the methyl interactome simultaneously. In contrast to antibody or metal affinity-based enrichment of phosphopeptide or phosphoproteins, methyllysine-binding domains are used to enrich for methylated proteins and the associated complexes. This allows for the identification of the interacting proteins and methylation sites simultaneously. Third, our approach may be used to generate a high-resolution methyl interactome. Because the bait domain binds to a protein substrate in a methylation-dependent manner, our method identifies the methylation sites and thereby, the direct binders for the bait domain. In the case of HP1β-CD interactome, we showed that the first layer of interactors is composed of most or all of the 29 methylated proteins to which the remaining proteins in the second layer are connected to. Although it is likely that some methylated proteins may be associated with HP1β indirectly through another methylated protein, as is the case for Ku80 that may be coupled to HP1β via DNA-PKcs, methylation would be a pre-requisite for proteins in the first layer of the HP1β-CD interactome. In contrast, the antibody-based AP-MS approach cannot provide information on direct physical interactions. Fourth, the unique strength of the MRM-MS approach to identify Lys methylation sites in a targeted manner is demonstrated by the large-numbe of methylation sites identified in the HP1β-CD interactome. In contrast, the shotgun strategy, which is widely used for large-scale identification of protein complexes and post-translational modifications (MacCoss et al., 2002; Walther and Mann, 2010), was not able to identify any methyllysine site in a parallel experiment. It is likely that the digested peptides did not contain optimized HP1β-CD binding sequences or that they were not in sufficient quantities to allow the detection by the shotgun approach which usually detects the most abundance species in a sample. Because of an MRM/SRM assay is built by in silico prediction of peptide precursors and transitions from spectral libraries originated from shotgun proteomic experiments (MacLean et al., 2010; Picotti and Aebersold, 2012), it allows for levels of specificity and sensitivity hard to match by the more standard shotgun experiments (de Graaf et al., 2011). The reliable prediction of Lys methylation sites based on the specificity of the chromodomain is vital for the success of the MRM experiments because it drastically reduced the number of precursor peptides to be surveyed by MRM. Finally, we showed that the integration of heavy methyl SILAC and MRM offers an effective strategy by which to gauge the dynamic changes in the methylome, or the methyl dynactome, associated with a cellular process.

The number of methylated proteins (29) and lysine methylation sites (40) identified in our case study on HP1β not only provided the proof-of-principle for our method, but also indicates that the protein methylome may be much greater in size than current data, which are largely focused on histone proteins, would suggest. Although a systematic comparison between the methylome and phosphoproteome is necessary to gauge the size of the former in an accurate manner, it can be envisioned that, when extended to the hundreds of Kme- and Rme-binding domains, our approach would be a useful tool to enable such a comparative study. It should be noted, however, that all methylation sites are not recognized by modular binding domains. Therefore, our approach should be used in conjunction with complementary proteomic methods, such as shotgun proteomics and biochemical methods, in order to map the methylome at the genomic level.

The HP1β Interactome

Our data showed that the HP1β interactome identified is of high quality. This assertion is substantiated by a series of validation experiments, including peptide binding, bioinformatic analysis, and co-IP assays. That the interactions identified in this analysis have >80% overlap with the STING database of known and predicted protein-protein interactions, and that all five methylated proteins tested for binding to endogenous HP1β turned out positive have provided compelling evidence for the high quality of our data. Besides confirming known interactions, our analysis revealed 40 methylation sites in 29 proteins, the majority of which is novel. By mapping the methyllsine sites and measuring methyl-peptide binding affinities, we showed that most of the 29 methylated proteins would be direct binders of the HP1β-CD, or the first layer in the interactome. The remaining 80 proteins would comprise the second layer, which is linked to HP1β-CD via the methylated proteins.

The interactome identified in our study greatly expands the functional repertoire for HP1β. While additional follow-up studies are necessary to decipher the roles of the novel interactions in HP1β function, our initial interactome analysis has offered some clues. Three functional clusters are clearly identifiable in the interactome. These include a group of 14 proteins with annotated functions in DNA damage repair, a cluster of 39 proteins involved in RNA splicing, and a group of 8 ribosomal proteins. Proteins within each cluster are densely connected with one another, suggesting that functional complexes, instead of specific components of a complex, have been isolated in the chromodomain pulldown. The architecture of the HP1β interactome strongly suggests that, in addition to chromatin remodelling, HP1β plays important roles in DNA damage response, RNA splicing and protein synthesis.

A Novel Function for HP1β in DSB Repair

Recent studies indicate that HP1β is involved in DNA damage repair, although the underlying mechanism is not fully understood (Ayoub et al., 2008; Luijsterburg et al., 2009). In particular, the role of the HP1β-CD in this process remains rather controversial. Although HP1β has been shown to bind specifically to H3K9me3 through its chromodomain, this interaction appears to have little effect in DNA damage repair (Billur et al., 2010). Therefore it is likely that HP1β acts as an adaptor to recruit a non-histone, lysine methylated protein to DNA damage foci through its chromodomain (CD). The identification of the HP1β-DNA-PKcs interaction provided strong evidence in support of this notion. We showed that HP1β promoted the recruitment of the activated form of DNA-PKcs (marked by pT2609) to DNA damage foci (marked by γH2Ax). DNA-PKcs is a component of the DNA-dependent kinase complex. Together with Ku70 and Ku80, DNA-PKcs is responsible for the early recognition of DNA double strand breaks. Although the DNA-PKcs has redundant function with ATM in generating γH2Ax foci, it appears that ATM is only involved in less than 25% of DSB repair (Goodarzi et al., 2008). Because the total amount of γH2Ax was markedly reduced by depletion of HP1β, HP1β may play an important role in DSB repair through DNA-PK.

Several phosphorylation sites, including T2609, S2612, T2638 and T2647, have been identified in DNA-PKcs (Douglas et al., 2002). Although the function of DNA-PKcs phosphorylation remains unknown, these phosphorylation sites are important for DNA-PKcs dependent DSB repair (Uematsu et al., 2007). We have shown here that the function of DNA-PKcs is also regulated by methylation on the K1150, K2746 and K3248 residues, and these methyllysine sites mediate binding of the kinase to HP1β. Compared to the wild-type DNA-PKcs, cells containing a K-to-R mutant or the triple mutant displayed defects in DSB repair. It is likely that HP1β plays a role of an adaptor to recruit DNA-PKcs to the sites of DNA lesion in a methylation dependent manner in order to phosphorylate its substrates. Our work suggests a new mechanism by which DNA-PKcs functions through HP1β and provides an example in which methylation and phosphorylation may cross-talk to regulate an essential biological function (Tang et al., 2008). Future studies should focus on the identification of methyltransferases (KTMs) and demethylases (KDMs) that are responsible for controlling the methylation status of DNA-PKcs, and by the same token, of other methylated proteins involved in DNA damage response (Fig. 7). In this regard, it is interesting to note that Set8, which modulates the function of p53 by methylating Lys382, is dynamically regulated during DNA damage (Shi and Whetstine, 2007). Other KMTs involved in DNA damage response, such as DOT1 (Nguyen et al., 2011) and G9a/GLP (Huang et al., 2010), may also serve the role of the KTM for non-histone protein methylation during DDR.

Methylome Dynamics

The identification of the first histone demethylase (Shi et al., 2004) changed the long-held belief that lysine methylation is a stable and irreversible PTM (Byvoet et al., 1972). Although few studies have been devoted to the characterization of methylation dynamics in histone and non-histone proteins, dynamic histone methylations are thought to contribute to normal cellular function (Ng et al., 2009). Pesavento et al. (Pesavento et al., 2008) showed that H4K20 undergoes dynamic changes in methylation during a cell cycle. Our analysis of the methylation status of the HP1β interactome in DNA damage response provides strong evidence that lysine methylation on non-histone proteins is a highly dynamic event. We observed a large range of dynamics in lysine methylation occurring on a wide spectrum of proteins. For instance, the Ku80-K7me3 was up-regulated and the DNA-PKcs-K3248me2 down-regulated by two orders of magnitude under DNA damage induced by etoposide. Beli et al (Beli et al., 2012) recently examined the dynamic changes in the phosphoproteome and acetylome induced by DNA damage using a SILAC-based strategy and showed that phosphorylation generally displayed larger changes than acetylation. Intriguingly, the amplitude of the changes in lysine methylation observed in our study is significantly greater than observed for either acetylation or phosphorylation (Olsen et al., 2006). Moreover, DNA damage led to changes in methylation for not only the subunits of DNA-PK, but also proteins not directly involved in DSB repair. This is likely caused by the up- or down-regulation of certain methyltransferases or demethylases associated with DNA damage response. We predict that other cellular processes, such as proliferation, differentiation and apoptosis, would similarly involve global changes in the protein methylome. A future direction of research is therefore to decipher these dynamic changes and understand how methylome dynamics alone or in combination with other dynamic PTMs such as phosphorylation regulate cellular functions and development.

In summary, we have developed a high throughput method for systematic identification of methylation sites in proteins and the protein-protein interactions mediated by Lys methylation. Our study with the HP1β suggests that the protein methylome is likely to be larger than current data would suggest and that methylation is a highly dynamic PTM occurring on many non-histone proteins. Moreover, lysine methylation may be involved in a wide array of cellular functions.

EXPERIMENTAL PROCEDURES

Peptide and peptide Array Synthesis, Probing and SMALI

The H3K9me3 peptide and the unmethylated version were synthesized on the ABI430A peptide synthesizer following standard Fmoc chemistry. The peptides were subsequently purified on HPLC and confirmed for identify using mass spectrometry. Synthesis and probing of peptide array followed the same procedures as reported previously (Jia et al., 2005). Quantification of array binding signals and the calculation of PSSM for SMALI were conducted in a similar manner as reported elsewhere (Li et al., 2008).

Cell Lines and Antibodies

HEK293T, U2OS and CHO V3 cells were maintained at 37°C in DMEM medium supplemented with 10% fetal bovine serum (FBS), 2 mg/ml glutamine and 100 units of penicillin-streptomycin. The DNA-PKcs mutant stable cell lines were cultured in the same medium, except with 400 mg/ml G418 for selection and 200 mg/ml G418 for maintenance. To induce DNA damage, U2OS cells grown to 70–80% confluence were incubated with 20 nM etoposide for 2 hrs at 37 °C.

Mouse anti-Ku80 were purchased from Santa Cruz. Mouse anti-H3K9me2, rabbit anti-HP1β, rabbit anti-LAP2A and rabbit anti-eI4A3 were from Abcam. Mouse anti-PRKDC(pT2069) was from Abnova.

LC-MS/MS Mass Spectrometry

The HP1β chromodomain was expressed with an (His)₆ tag in E.coli, and purified on Ni-NTA beads (Qiagen Co.). The (His)₆ tag was subsequently cleaved using the TEV protease, and the HP1β-CD protein was further purified on a G75 column by FPLC. Purified HP1β-CD was immobilized onto the MagnosphereTM beads (JSR Co., Japan) through functionalized carboxylate groups. The HP1β-CD-beads was then incubated with the nuclear lysate of HEK293T cells. After extensive washes in TBST and TBS to remove non-specific binding proteins, the bound proteins were eluted by incubating with the H3K9me3 peptide (10 µM) or the unmethylated control peptide (10 µM). The sample obtained above was digested with trypsin and subject to mass spectrometry analysis as described below.

The LC-MS/MS experiment used to identify the HP1β interactome was carried out on a Q-TOF Ultima global mass spectrometer (Waters/Micromass) coupled to a Waters NanoAcquity UPLC system. The instrument was run in DDA mode. The sample was separated on a 1.7µm BEH130 C18 column (75 µm × 250 mm, Waters) with a 60 min gradient of 5–60% acetonitrile (containing 0.1% formic acid ). The MassLynx 4.1 software (Waters) was used for data acquisition and processing. Proteins were identified by Mascot (Perkins et al., 1999) search against the IPI_human database (Kersey et al., 2004). LC-MS/MS analyses were carried out on three biological replicates to obtain the final list of binding proteins.

To detect the methyllysine sites by the shotgun method, nuclear proteins were pull down by HP1β-CD. Bound proteins were eluted by 1M NH4OH (200µM x3) and vacuum dried. The sample obtained was divided into four fractions and each fraction was digested with a different enzyme, ArgC, GluC, elastase, or chymotrypsin (to match the conditions used in the MRM experiments as described below). The digested fractions were then pooled and loaded onto a C18 column (strata C18-E, 55µM, 70A). The peptides were eluted by 50% acetonitrile and dried again. Re-dissolved peptides sample was incubated with HP1β-CD beads for 2h in PBS at 4 °C, followed by 1×TBST wash. NH4OH (200µM x3) was used to elute the peptides off the beads. Vaccum-dried sample was reconstituted in water and subject to MS/MS analysis on the QTof Global mass spectrometer run in DDA mode.

Multiple or Selected Reaction Monitoring (MRM or SRM)

To identify methylated peptides, proteins bound to the HP1β-CD beads were eluted by the H3K9me3 peptide, digested with ArgC, GluC, elastase, or chymotrypsin, and the mixture was analyzed by positive ESI LC-MS/MS on a triple quadruple mass spectrometer (4000 Q-TRAP or 5500 Q-TRAP, Applied Biosystems Inc.) utilizing Q3 as a linear ion trap. A nanoAcquity UPLC system (Waters) equipped with a C18 analytical column (1.7 µm, BEH130, 75 µm×250 mm) was used to separate the peptides at the flow rate of 300 nl/min and operating pressure of 8000 psi. Peptides were eluted using a 62 min gradient from 95% solvent A (H₂O, 0.1% formic acid) and 5% B (acetonitrile, 0.1% formic acid) to 50% B in 41 min, 6 min at 90 %B, and back to 5% for 10 min. Eluted peptides were directly electrosprayed (Nanosource, ESI voltage +2000V) into the mass spectrometer. The instrument was set to monitor 50 to 100 transitions in each sample with a dwelling time of 100 msec/transition.

The in silico protease digest patterns (i.e. to generate precursor ions) and the corresponding MRM transitions were compiled using the Skyline™ software made freely available to us by the McCoss Lab, Department of Genome Sciences University of Washington School of Medicine (MacLean et al., 2010). Transitions that are larger than the precursor ion were selected based on the Skyline predictions and the specific b/y ions that allow unambiguous identification of the methylated lysine site were included. A methylated peptide identified by MRM (with three or more transitions observed) was subsequently verified by MS/MS sequencing. Information dependent acquisition (IDA) was used to collect MS/MS spectra when a MRM signal exceeded 500cps.

Methyl SILAC

SILAC labeling medium (DMEM, GIBCO-21013) lacking glutamine, cysteine and methionine was reconstituted according to the manufacturer's instructions and supplemented with amino acid stocks prepared in PBS. [CD₃] methionine (Met-methyl-d3, Sigma Isotec) was added to the heavy medium while the natural abundance methionine (Sigma-Aldrich) was added to the light medium. Both media contained 10% dialyzed FBS, glutamine, cysteine and antibiotics. HEK293T cells were cultured in the heavy medium for at least ten cell doublings to allow adaptation and full incorporation of the stable isotope-containing methionine. The heavy-isotope labeled cells were treated with 20 nM etoposide for 2 hours while cells in the light medium treated with the solvent DMSO. Lysates of cells from the heavy and light media mixed in 1:1 ratio (based on the total protein content) were subject to pull-down by HP1β-CD beads. Bound proteins were digested by trypsin on beads and subject to MRM analysis on a 4000QTRAP instrument designed to specifically monitor the 40 identified methyllysine sites. The total intensity of MRM peaks was used to quantify the heavy or light isotope-labeled methylation sites.

Pull-down, Immunoprecipitation and Western Blotting

HEK293T or U2OS cells were lysed in 400 µl of cold lysis buffer (50mM Tris, pH 7.2, 150 mM NaCl, 2mM MgCl₂, 0.1 mM EDTA, 0.1 mM EGTA, 0.5 mM DTT, 10 mM NaF, 1% Nonidet P-40 and protease inhibitor cocktail (1:1,000) and fractionated into cytosolic and nuclear fractions using a cell fractionation kit (Pierce). The nuclear lysate was incubated either with 100 µg purified (His)₆-tagged HP1β, the CD or a mutant or with 5µl of rabbit anti-HP1β for 1h at 4 °C. The (His)₆-tagged protein complex was isolated using Ni-NTA beads (Qiagen) and the immunocomplex precipitated from solution using protein G-Sepharose 4B beads. The complexes were separated by SDS–polyacrylamide gel electrophoresis and transferred to PVDF membrane. Proteins were detected by immunoblotting with appropriate antibodies and visualized by the ECL method (Pierce).

Knockdown of HP1β by siRNA

U2OS cells were transfected twice with 30 nM siRNA duplexes using DharmaFECT1 (ThermoScientific) at the 0 and 24 hr, respectively. Control cells were transfected with the same amount of a scrambled siRNA (Sigma). Cells were grown for 72h prior to subsequent experiments. Total cell extracts were also prepared at this time to check the extent of protein depletion by Western blotting. The sequences of the sense strand of the siRNA duplex (Sigma) used was: forward: 5'-GACUCCAGUGGAGAGCUCAUGTT-3'; reverse: 5'-CAUGAGCUCUCCACUGGAGUCTT-3'.

Confocal Microscopy

U2OS cells were grown on glass coverslips in DMEM. Cells were fixed for 10 minutes in 2% formaldehyde and washed with ice-cold methanol. The methanol was rinsed off with two brief PBS washes, and then the blocking solution (3% BSA in PBS) was applied for 30 minutes at room temperature. After three washes with staining buffer (PBS + 0.1% NP-40), a primary antibody was added. The antibodies and dilutions used were: pT2609-DNA-PKcs (Abcam #ab18356), γH2Ax (Abcam # ab11174) and HP1β (Santa Cruz #SC10212) at 1:100. After 1-hour incubation at room temperature, the primary antibody was removed and the cells were washed three times in the staining buffer. The appropriate secondary antibody (Alexa Fluor-466 anti-rabbit or Alexa Fluor-647 anti-mouse or Cy3 anti-goat IgG) was added at a 1:500 dilution for 30 minutes at room temperature. The cells were washed three times with the staining solution, DAPI added at 1:10,000 dilution for one minute, the cells washed again and finally the glass cover slips inverted and placed on a microscope slide with the Invitrogen SlowFade mounting medium. Images were acquired on a Zeiss LSM 510 META confocal microscope. Unprocessed images taken from five cells for each sample/condition were used for colocalization quantification analysis. The colocalization level was quantified based on the Pearson correlation coefficient using the JACoP plugin in ImageJ software (Bolte and Cordelieres, 2006).