Top-down characterization of histone H4 proteoforms with ProteinGoggle 2.0

XIAO Kaijie, TIAN Zhixin

(School of Chemical Science & Engineering, Shanghai Key Laboratory of Chemical Assessment andSustainability, Tongji University, Shanghai 200092, China)

?

Top-down characterization of histone H4 proteoforms with ProteinGoggle 2.0

XIAO Kaijie, TIAN Zhixin*

(SchoolofChemicalScience&Engineering,ShanghaiKeyLaboratoryofChemicalAssessmentandSustainability,TongjiUniversity,Shanghai200092,China)

Top-down characterization of combinatorial and dense post-translational modifications (PTMs) on core histones has long been a big analytical challenge because of enormous putative proteoforms for identification and simultaneously enormous putative sites of each individual PTM for localization. ProteinGoggle 2.0, as implemented with the isotopic mass-to-charge ratio and envelope fingerprinting algorithm, has multiple unique strengths for top-down characterization of histone PTMs together with high-resolution tandem mass spectrometry. Here we report our database search and proteoform identification of HeLa core histone H4 using ProteinGoggle 2.0. The Theoretical database containing all putative proteoforms was created with shotgun annotation from the human H4 flat text downloaded from UniProt; information including the amino acid sequence, putative PTMs (methylation, di-methylation, tri-methylation, acetylation and phosphorylation) and amino acid variation (A77 to P) was adopted from the flat text file. A total of 426 proteoforms were confidently identified with a spectrum level false discovery rate 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. By and large, ProteinGoggle 2.0 can be adopted for database search and proteoform identification of proteins with multiple combinatorial PTMs as well as amino acid variation.

ProteinGoggle 2.0; top-down; histone H4; proteoform

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 throughn-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 isotopicm/zfingerprinting (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 them/zdeviation and relative abundance deviation of these experimentally isotopic peaks should be within user-specified threshold values. These two parameters are designated as isotopic peakm/zdeviation (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.

1 Experimental

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 6 672 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 throughPScore cutoff of combined PrSMs from the target-decoy searches both forward and random databases.Pscore 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 ofPscore,Pf,n, is shown in Equations 1 and 2.

(1)

(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; “2 000” is them/zscan 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 experimentalm/zdeviation of an isotopic peak relative to its theoretical value; and “M” is the relative molecular mass (in Da) of the proteoform.

2 Results and discussion

With database search of the H4 dataset using ProteinGoggle 2.0, 7 139 and 41 PrSMs were obtained from the forward and random searches, respectively. These PrSMs were combined and ranked byPscore from high to low; a cutoffPscore, 169.09 (negative log value), was then chosen to obtain forward PrSMs with a spectrum level FDR≤1. Above this cutoffPscore, there are 5 969 and 29 PrSMs from the forward and random searches, and the FDR=29×2×100%/(5 969+29)=0.97%. The 5 969 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), isolationm/z, experimentalm/z, theoreticalm/z, IPMD (ppm),z, theoretical monoisotopic mass (Da), sequence, PTMs, number of matching product ions (MPs), number of non-MPs,Pscore, 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 (10m/zisolation 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.

Fig. 1 Overlap of H4 proteoforms identified by ProSightPC 2.0 and ProteinGoggle 2.0

For ETD spectrum 1 320 with MS scan 1 310 (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 1 368 with MS scan 1 365 (Table 1), the identification difference between the two search engines is similar to that in spectrum 1 320. 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); whilethe 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.

Table 1 Side-by-side comparison of ProSightPC 2.0 with ProteinGoggle 2.0 in interpretation of the tandem mass spectra 1320, 1368, 1439, 1810, 2885 and 2919 in the H4 dataset

*SGRGKGGKGLGKGGAKRHRKVLRDNIQGITKPAIRRLARRGGVKRISGLIYEETRGVLKVFLENVIRDAVTYTEHAKRKTVTAMDVVY-ALKRQGRTLYGFGG; **K12AcK16AcK20Me, K8AcK16AcK20Me, K8AcK12AcK20Me, K5AcK16AcK20Me, K5AcK12AcK20Me, K5AcK8AcK20Me, R3MeK12AcK16Ac, R3MeK8AcK16Ac, R3MeK8AcK12Ac, R3MeK5AcK16Ac, R3MeK5AcK12Ac, R3MeK5AcK8Ac, S1AcK16AcK20Me, S1AcK12AcK20Me, S1AcK8AcK20Me, S1AcK5AcK20Me, S1AcR3MeK16Ac, S1AcR3MeK12Ac, or S1AcR3MeK5Ac.

Fig. 2 (a, b) iEF maps for the two precursor ions identified for ETD spectrum 1320 by ProSightPC 2.0 and ProteinGoggle 2.0; (c) graphical fragmentation map for H4 proteoforms with PTMs of S1AcK5AcK8AcK12AcK16AcK20dMe identified from MS/MS spectra 1320 by ProteinGoggle 2.0; (d, e) iEF maps for the two precursor ions identified for CID spectrum 1368 by ProSightPC 2.0 and ProteinGoggle 2.0; (f) iEF map for the same precursor ions identified for CID spectrum 1810 by ProSightPC 2.0 and ProteinGoggle 2.0

Fig. 3 (a, b) iEF maps for the two precursor ions identified for ETD spectrum 1439 by ProSightPC 2.0 and ProteinGoggle 2.0; (c, d) graphical fragmentation map for H4 proteoforms with PTMs of S1AcK8AcK12AcK16AcK20dMe and S1AcK5AcK12AcK16AcK20dMe identified from MS/MS spectra 1439 by ProteinGoggle 2.0

Fig. 4 (a) Graphical fragmentation map for H4 proteoforms with PTMs of R3MeK12AcK16Ac identified from MS/MS spectra 1810 by ProteinGoggle 2.0; (b) iEF map for the product ion b101+ identified for CID spectrum 1810 by ProteinGoggle 2.0; (c, d) iEF maps for the two precursor ions identified for ETD spectrum 2885 by ProSightPC 2.0 and ProteinGoggle 2.0; (e) iEF map for the same precursor ion identified for CID Spectrum 2919 by ProSightPC 2.0 and ProteinGoggle 2.0; (f) graphical fragmentation map for H4 proteoforms with PTMs of S1AcR3MeK8Ac identified from MS/MS spectrum 2919 by ProteinGoggle 2.0

For ETD spectrum 1 439 with MS scan 1 431 (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 1 810 with MS scan 1 805 (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 1 810 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 2 885 with MS Scan 2 883 (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 2 919 with MS Scan 2 916 (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.

3 Conclusions

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.#edu.cn/.

Supplementary Information Supplemental information including three figures and detailed tabular information for the identified proteoforms (20 pages in total) are provided at http://www.Chrom-china.com/UserFiles/File/1609012SupportingInfo(2).pdf.

[1] Venkatesh S, Workman J L. Nat Rev Mol Cell Bio, 2015, 16(3): 178

[2] Tessarz P, Kouzarides T. Nat Rev Mol Cell Bio, 2014, 15(11): 703

[3] Falkenberg K J, Johnstone R W. Nat Rev Drug Discov, 2014, 13(9): 673

[4] Hojfeldt J W, Agger K, Helin K. Nat Rev Drug Discov, 2013, 12(12): 917

[5] Jenuwein T, Allis C D. Science, 2001, 293(5532): 1074

[6] Zheng Y P, Fornelli L, Compton P D, et al. Mol Cell Proteomics, 2016, 15(3): 776

[7] Kwak H G, Dohmae N. Biosci Trends, 2016, doi: 10.5582/bst.2016.01090

[8] Moradian A, Kalli A, Sweredoski M J, et al. Proteomics, 2014, 14(4/5): 489

[9] Tian Z X, Tolic N, Zhao R, et al. Genome Biol, 2012, 13(10): R86

[10] Bonet-Costa C, Vilaseca M, Diema C, et al. J Proteomics, 2012, 75(13): 4124

[11] Han J, Borchers C H. Proteomics, 2010, 10(20): 3621

[12] Pesavento J J, Bullock C R, LeDuc R D, et al. J Biol Chem, 2008, 283(22): 14927

[13] Pesavento J J, Kim Y B, Taylor G K, et al. J Am Chem Soc, 2004, 126(11): 3386

[14] Garcia B A, Shabanowitz J, Hunt D F. Curr Opin Chem Biol, 2007, 11(1): 66

[15] Britton L M, Gonzales-Cope M, Zee B M, et al. Expert Rev Proteomics, 2011, 8(5): 631

[16] Sidoli S, Cheng L, Jensen O N. J Proteomics, 2012, 75(12): 3419

[17] Bonaldi T, Noberini R. Expert Rev Proteomics, 2016, 13(3): 245

[18] Onder O, Sidoli S, Carroll M, et al. Expert Rev Proteomics, 2015, 12(5): 499

[19] Liu Z W, Zhu M R, Zhai L H, et al. Chinese Journal of Chromatography, 2016, 34(9): 825

[20] Wang G J, Zhang K, He X W, et al. Chinese Journal of Chromatography, 2013, 31(6): 514

[21] Li L, Tian Z X. Rapid Commun Mass Spectrom, 2013, 27(11): 1267

[22] Xiao K J, Yu F, Fang H Q, et al. Sci Rep, 2015, 5: 14755

[23] Fang H Q, Xiao K J, Li Y H, et al. Anal Chem, 2016, 88(14): 7198

基于ProteinGoggle 2.0的組蛋白H4蛋白質(zhì)變體的自上而下表征

肖開捷, 田志新*

(同濟(jì)大學(xué)化學(xué)科學(xué)與工程學(xué)院, 上海市化學(xué)品分析、風(fēng)險(xiǎn)評估與控制重點(diǎn)實(shí)驗(yàn)室, 上海 200092)

由于大量可能蛋白質(zhì)變體以及每一個(gè)翻譯后修飾大量可能位點(diǎn)的存在,核心組蛋白上密集的組合式翻譯后修飾的自上而下表征一直是一個(gè)巨大的分析挑戰(zhàn)。結(jié)合高分辨串級質(zhì)譜,基于同位素質(zhì)荷比和輪廓指紋比對的整體蛋白質(zhì)數(shù)據(jù)庫搜索引擎ProteinGoggle 2.0在組蛋白翻譯后修飾的自上而下鑒定方面擁有諸多獨(dú)特的優(yōu)勢。該文報(bào)道ProteinGoggle 2.0對HeLa核心組蛋白H4的數(shù)據(jù)庫搜索及蛋白質(zhì)變體的鑒定結(jié)果?；趶腢niProt網(wǎng)站下載的人類核心組蛋白H4的純文本文件和“鳥槍法”注釋,ProteinGoggle 2.0首先創(chuàng)建包含所有可能蛋白質(zhì)變體的理論數(shù)據(jù)庫;從純文本文件中提取的信息主要是氨基酸序列、可能的翻譯后修飾(單甲基化、二甲基化、三甲基化、乙?；土姿峄?及氨基酸變異(A77→P)。在控制質(zhì)譜水平假陽性率低于1%的前提下,共鑒定到426個(gè)蛋白質(zhì)變體,這是目前為止H4蛋白質(zhì)變體的最全報(bào)道。這些ProteinGoggle 2.0鑒定到的H4蛋白質(zhì)變體也與之前報(bào)道的ProSightPC 2.0的鑒定結(jié)果進(jìn)行了肩并肩比較?？偠灾?ProteinGoggle 2.0可以對具有復(fù)雜組合修飾及氨基酸變異的蛋白質(zhì)組進(jìn)行數(shù)據(jù)庫搜索和蛋白質(zhì)變體鑒定。

ProteinGoggle 2.0;自上而下;組蛋白H4;蛋白質(zhì)變體

10.3724/SP.J.1123.2016.09012

Foundation item: National Natural Science Foundation of China (No. 21575104); China State Key Basic Research Program (No. 2013CB911203); Shanghai Science and Technology Commission (No. 14DZ2261100).

O658 Document code: A Article IC:1000-8713(2016)12-1254-09

Special issue for commemorating Professor ZOU Hanfa (Ⅰ)·Article

* Received date: 2016-09-04

* Corresponding author. Tel: +86-21-65986992, E-mail: zhixintian@#edu.cn.