Histones, as chromatin proteins, play important roles not only in structural organization of DNA, but also in regulation of almost all DNA activities through post-translational modifications (PTMs) [1, 2]. Aberrant histone PTMs lead to diseases and cancers [3, 4]. These histone PTMs often cross-talk and function through combinatorial histone code [5], which necessitates intact molecular characterization. Modern ESI high-resolution tandem mass spectrometry together with various peripheral hyphenated techniques (especially high-performance liquid chromatography) has evolved into a state-of-the-art instrumental analytical tool for both qualitative and quantitative characterization of histone PTMs [6-20].
Tian et al. [9] have so far done the most comprehensive top-down analysis of core histones. The four families (H4, H2B, H2A, and H3) of HeLa core histones were first fractionated offline by reversed-phase liquid chromatography; various proteoforms with combinatorial PTMs (such as acetylation, methylation, and phosphorylation) within each family fraction were then separated by pH-gradient weak cation exchange chromatography (WCX) which was coupled online through n-ESI to an Orbitrap mass spectrometer. Five data-dependent datasets (one for H4, H2B, H2A, H3_1 and H3_2 each) using alternative collision-induced dissociation (CID) and electron-transfer dissociation (ETD) were acquired and deposited online (http://www.peptideatlas.org/). In total, 708 proteoforms (including 105 from the H4 family) identified with protein database search engine ProSightPC 2.0 were reported; ever since, the authors have developed a top-down intact protein database search engine ProteinGoggle 2.0 using distinctly alternative search algorithm, isotopic mass-to-charge ratio and envelope fingerprinting (iMEF) [21]. The iMEF algorithm interprets mass spectra and identifies both precursor and product ions by fingerprinting theoretical isotopic envelopes onto the corresponding experimental ones, which are measured directly by MS and thus pre-processing of de-isotoping as uniformly practiced in all the other search engines is avoided. The iMEF algorithm is a combination of isotopic m/z fingerprinting (iMF) and isotopic envelope fingerprinting (iEF). The former is used to fish out putative precursor or product ion candidates from the theoretical database with fingerprinting of the theoretically highest isotopic peak; while the latter is used to identify matching precursor or product ions with fingerprinting of all the isotopic peaks above a theoretical abundance threshold. This threshold is designated as isotopic peak abundance cutoff (IPACO); for a matching precursor or product ion, theoretical isotopic peaks with relative abundance above IPACO should be experimentally observed; also the m/z deviation and relative abundance deviation of these experimentally isotopic peaks should be within user-specified threshold values. These two parameters are designated as isotopic peak m/z deviation (IPMD) and isotopic peak abundance deviation (IPAD). Passing through the search parameters of IPACO, IPMD and IPAD, every matching precursor or product ion has an ideal experimental isotopic envelope. This ensures confident protein identification as well as site localization of PTMs if any. The iMEF algorithm also has its intrinsic unique strength of resolution of extremely dense data in protein tandem mass spectra and distinct exclusion of non-ideal data [22]. The newest ProteinGoggle 2.0 with full capacity has been applied for both qualitative identification and quantitative analysis of differentially expressed proteins in hepatocellular carcinoma [23].
Here we reported our database search and proteoforms identification of the aforementioned histone H4 using ProteinGoggle 2.0. A total of 426 proteoforms were confidently identified with a spectrum level false discovery rate (FDR) of less than 1%, which represents the most comprehensive H4 proteoforms reported so far. Side-by-side comparison of these proteoforms with those identified by ProSightPC 2.0 was also made.
The histone H4 dataset was downloaded from Peptide Atlas (http://www.peptideatlas.org/) with the dataset identifier as PASS00070. The dataset has been published online together with its database search results from ProSightPC 2.0. Originally, the dataset was acquired with WCX-tandem mass spectrometry (alternative CID and ETD) analysis of HeLa histone H4 family which was obtained from RPLC fractionation of HeLa core histones mixture. ProSightPC 2.0 search was conducted in the “absolute mass” mode, and the adopted mass tolerance of precursor and product ion were ±1 Da and 10 ppm (10×10-6), respectively. Methylation (mono-, di-, and tri-), acetylation, and phosphorylation were treated dynamically and all annotated in the customized database. With a spectrum-level FDR≤1% using the reverse database in the decoy search, 105 H4 proteoforms were identified with “Number of Best Hits=1”, i. e. all proteoforms were uniquely given their respective matching product ions and possible candidates annotated in the customized database.
In database search of this dataset with ProteinGoggle 2.0, a flat text file of human histone H4 (P62805) containing amino acid sequence together with candidate PTMs was first downloaded from UniProt. The customized forward and random databases were created with mono-methylation (R4, K21), di-methylation (R4, K21), tri-methylation (K21), acetylation (S2, K6, K9, K13, K17, K32, K92), and phosphorylation (S2, S48, Y52, T81) as dynamic PTMs and max PTMs per proteoform were limited to 6. Initial methionine was either kept or removed; and mutation of A77 to P was also considered. With all these options taken into account, a total of 6672 candidate proteoforms were annotated. Initial protein spectrum matches (PrSMs) search in both the forward and the random databases was carried out with the following tolerance parameters: IPACO, IPMD, IPAD for the precursor ions and product ions were 40%/15 ppm/100% and 20%/15 ppm/50%, respectively; percentage of matching product ions (PMPs)≥5; PTM score≥1 or Proteoform score≥1.
Spectrum-level FDR control is achieved through P Score cutoff of combined PrSMs from the target-decoy searches both forward and random databases. P score is used to evaluate the probability of a random proteoform match from a MS/MS spectrum and its scoring is built on Poison distribution. The exact computation of P score, Pf, n, is shown in Equations 1 and 2.
In Equations 1 and 2, “x” is the random match probability of an isotopic peak in the MS/MS spectrum; “f”is the total number of isotopic peaks in the spectrum; “n” is the number of peaks hitting a theoretical peak for random; “NTheo” is the number of theoretical product ions of the matched proteoform; “2000” is the m/z scan range of the MS spectrum, which could be read in from the raw experimental data; “z” is the charge state of the corresponding precursor ion for the MS/MS spectrum; “IPMD” is the short name of isotopic peak mass-to-charge ratio deviation, and is the experimental m/z deviation of an isotopic peak relative to its theoretical value; and “M” is the relative molecular mass (in Da) of the proteoform.
With database search of the H4 dataset using ProteinGoggle 2.0, 7139 and 41 PrSMs were obtained from the forward and random searches, respectively. These PrSMs were combined and ranked by P score from high to low; a cutoff P score, 169.09 (negative log value), was then chosen to obtain forward PrSMs with a spectrum level FDR≤1. Above this cutoff P score, there are 5969 and 29 PrSMs from the forward and random searches, and the FDR=29×2×100%/(5969+29)=0.97%. The 5969 forward PrSMs were grouped with amino acid sequence and PTMs to remove duplicates and obtain the final 426 proteoforms. The detailed information (including spectrum index, retention time (min), isolation m/z, experimental m/z, theoretical m/z, IPMD (ppm), z, theoretical monoisotopic mass (Da), sequence, PTMs, number of matching product ions (MPs), number of non-MPs, P score, PTM score, and proteoform score) for each of the 426 proteoforms was provided in Supplemental Table S1.
Statistically, these 426 proteoforms, with 84 unique molecular formulae, were identified from 244 precursor ions across 168 MS/MS spectra in an elution window of 52.69 minutes. Each MS/MS spectrum (10 m/z isolation window, actually) may contain multiple precursor ions, and each of these precursor ions may contain multiple co-eluting isomeric proteoforms. As for the distribution of these proteoforms in terms of amino acid sequence, 254 proteoforms have the normal sequence (without initial methionine, no amino acid variation); 106 proteoforms have amino acid variation of A77→P (without initial methionine); 4 proteoforms have initial methionine (with or without A77→P), and the last proteoform has R79→C. It should be noted that mutation of this A77 to P was originally reported and reviewed in UniProt, and the identification here is supported by both precursor ion fingerprinting tolerance and fragmentation. About 70% of the 426 proteoforms have 3 or more PTMs each (Fig. S1); acetylated proteoforms elute in the order of decreasing acetylation number per proteoform (Supplemental Fig. S2). The more acetylation a proteoform has, the less proton charges it has, and thus be eluted earlier in this WCX separation. A total of 10 proteoforms (S1AcK12AcY88P, R3dMeK5AcY88P, R3dMeS47P, S1AcS47P, S1AcK20MeY88P, S1PK-20dMeS47PY51PY88P, S1AcK5AcY88P, S1AcK-20dMeS47P, K12AcK20dMeS47P, S1AcK20dMeY-88P) containing combinatorial PTMs beyond the N-terminal tail have been identified in this study. These proteoforms may not be identified by the alternative bottom-up approach where the enzymatic tail (SGRGKGGKGLGKGGAKRHRKVLR) was analyzed. As an example, the iEF map and graphical fragmentation map for the proteoform with S1AcK12AcY88P are shown in Fig. S3a and S3b, respectively.
For comparison of ProteinGoggle 2.0 and ProSightPC 2.0 in the database search of the H4 dataset, the results from the two search engines have a good proteoforms overlap (Fig. 1). However, the former has 327 unique proteoforms not identified by the latter; while it is 6 vice versa. Detailed side-by-side comparison of the two search engines in identification of the 6 proteoformsis provided in Table 1 and Fig. 2-4.
For ETD spectrum 1320 with MS scan 1310 (Table 1), ProSightPC 2.0 and ProteinGoggle 2.0 identified the proteoforms of S1AcK5AcK8AcK-12AcK16Ac and S1AcK5AcK8AcK12AcK16AcK-20dMe with 6 and 12 matching product ions, respectively. The precursor ion iEF maps of two proteoforms are shown in Fig. 2a and 2b. The theoretically highest isotopic peaks in the precursor ion isotopic envelopes are two peaks away from each other, i. e. the nominal mass difference between the two proteoforms is 2 Da. The graphical fragmentation map from ProteinGoggle 2.0 is shown in Fig. 2c, where all PTMs are uniquely localized.
For CID spectrum 1368 with MS scan 1365 (Table 1), the identification difference between the two search engines is similar to that in spectrum 1320. ProSightPC 2.0 identified H4 proteoform with K8AcK12AcK16AcK20MeY51p; whereas ProteinGoggle 2.0 identified two ambiguous proteoforms, S1AcR3MeK5AcK8AcK12AcK16Ac and R3dMeK5AcK8AcK12AcK16AcK20tMe, due to limited matching product ions. The precursor ion found by ProSightPC 2.0 is a relatively lower experimental isotopic envelope (Fig. 2d); while the precursor ion found by ProteinGoggle 2.0 is a relatively higher one (Fig. 2e). The two precursor ions are 4 Da (4 isotopic peaks) away from each other.
For ETD spectrum 1439 with MS scan 1431 (Table 1), the two search engines identified proteoforms with nominal mass difference of 2 Da, i. e. the theoretically highest isotopic peaks in the precursor ion isotopic envelopes are two peaks away from each other. ProSightPC 2.0 identified proteoform S1AcK8AcK12AcK20tMe with A77 mutated to P; whereas ProteinGoggle 2.0 identified two co-eluting proteoforms (S1AcK8AcK12AcK16AcK20dMe and S1AcK5AcK-12AcK16AcK20dMe) of the normal sequence with high confidence localization of all PTMs (Fig. 3c and 3d). The latter proteoforms have much better precursor ion fingerprinting between the experimental and theoretical isotopic envelopes (Fig. 3b) as well as many more matching product ions than the former proteoform (Fig. 3a).
For CID spectrum 1810 with MS scan 1805 (Table 1), the same experimental isotopic envelope was matched for the same precursor ion by the two search engines to give two proteoforms with PTMs of S1AcR3MeK16Ac and R3MeK12AcK16Ac, respectively. The iEF map of the precursor ion is shown in Fig. 2f. For the H4 proteoform R3MeK12AcK16Ac identified by ProteinGoggle 2.0, matching product ion b101+(iEF map shown in Fig. 4b) containing only R3Me proves that S1 does not have acetylation; and all the three PTMs are uniquely localized with sufficient sit-determining product ions as illustrated in the graphical fragmentation map (Fig. 4a). When spectrum 1810 is searched by ProteinGoggle 2.0 to fit the CID spectrum to the proteoform S1AcR3MeK16Ac as identified by ProSightPC 2.0, no matching product ion is found between S1 and K12 to unambiguously localize the acetylation on either of the two sites, which excludes co-elution possibility of this proteoform and also proves that identification by ProSightPC 2.0 is not right.
For ETD spectrum 2885 with MS Scan 2883 (Table 1), ProSightPC 2.0 identified proteoform of S1AcY88p (also A77→P) with the correspondingprecursor ion iEF map shown in Fig. 4c; while ProteinGoggle 2.0 identified three candidate proteoforms (S1AcK20dMeY88P, S1AcK20dMeY51P, and S1AcK20dMeS47P) sharing the same precursor ion with the corresponding iEF map shown in Fig. 4d. The nominal mass difference between the two different precursor ions is 2 Da; the precursor ion identified by ProteinGoggle 2.0 clearly has a much better fingerprinting between the theoretical and the experimental isotopic envelopes.
For CID spectrum 2919 with MS Scan 2916 (Table 1), the same experimental isotopic envelope was found for the same precursor ion by the two search engines, and the corresponding iEF map is shown in Fig. 4e. From the CID spectrum, ProSightPC 2.0 identified H4 proteoform with S1AcR3MeK8Ac exclusively; however, the graphical fragmentation map for this proteoform from ProteinGoggle 2.0 (Fig. 4f) clearly shows that location of methylation and acetylation (except the one on S1) could not be uniquely localized with the existing experimental data, and there should be as many as 20 putative proteoforms.
Different deconvolution algorithms often report shifted monoisotopic masses, which compromise the confidence of identification. Protein database search with iMEF and ProteinGoggle 2.0 not only removes uncertainties in deisotoping, but also possesses unique intrinsic capabilities of efficient resolution of overlapping iEs and unambiguous separation of confident product ions with ideal experimental iEs from ambiguous product ions with non-ideal experimental iEs. Confidence of PTM location assignment is leveraged by enforcement of both PTM score and Proteoform score. With the inherent strengths of iMEF, ProteinGoggle 2.0 displays superior performance in the database search of challenging histone H4; 426 proteoforms with unique localization of each PTM were confidently identified. ProteinGoggle 2.0 could be adopted for qualitative identification of any intact protein or proteome without size limitation. ProteinGoggle 2.0 is currently freely available at http://proteingoggle.tongji.edu.cn/.
2016,
Vol. 34 Issue (12): 1254-1263
PDF
2016,
Vol. 34 Issue (12): 1254-1263
PDF