LC-MS/MS Identification of the O-Glycosylation and Hydroxylation of Amino Acid Residues of Collagen α-1 (II) chain from Bovine Cartilage

Materials

A pepsinized immunization grade bovine collagen α-1 (II) chain (CO2A1) was acquired from Chondrex (Redmond, WA). Ammonium bicarbonate, sodium phosphate, and MS-grade formic acid were purchased from Sigma-Aldrich (St. Louis, MO). HPLC-grade solvents, including methanol and isopropanol, were procured from Fisher Scientific (Pittsburgh, PA). HPLC grade water was acquired from Mallinckrodt (Hazelwood, MO). HPLC grade acetonitrile was acquired from J.T.Baker (Phillipsburg, NJ). Mass spectrometry grade Trypsin Gold was obtained from Promega (Madison, WI) while endoproteinase GluC was purchased from New England Biolabs (Ipswich, MA).

Enzymatic Digestion

CO2A1 samples were digested with GluC and trypsin. Two 20-μg aliquots of CO2A1 were prepared in 50mM ammonium bicarbonate buffer (ABC) while a 20-μg aliquot was prepared in 50mM phosphate-buffered saline (PBS). In both cases, denaturation was performed at 65° C for 2 hours. The samples prepared in PBS and one of the samples prepared in ABC were subjected to GluC digestion by adding a 1-μl aliquot of the enzyme (enzyme/substrate ratio of 1:20 w/w) to each sample and incubating at 37.5°C for 16 hours. The second sample prepared in ABC was subjected to tryptic digestion by adding a 0.4-μl aliquot of trypsin (enzyme/substrate ratio of 1:50 w/w) at 37.5° C for 16 hours. To ensure complete enzymatic digestion, microwave digestion was performed at 45°C and 50W for 30 min.⁵¹ Both GluC and tryptic digestion were quenched through the addition of 0.5-μl aliquots of neat formic acid to the samples.

LC-MS/MS

Trypsin and GluC digested peptides were subjected to LC-MS/MS analysis using Dionex 3000 Ultimate nano-LC system (Dionex, Sunnyvale, CA) interfaced to LTQ Orbitrap Velos mass spectrometer (Thermo Scientific, San Jose, CA) equipped with a nano-ESI source. The samples were initially online-purified using a PepMap 100 C18 cartridge (3 μm, 100Å, Dionex). The purified peptides were then separated using a PepMap 100 C18 capillary column (75 μm id × 150 mm, 2 μm, 100Å, Dionex). The separation of peptides was achieved at 350 nl/min flow rate, using the following gradient: 0-10 min 3% solvent B (98% ACN with 0.1% formic acid), 10-35 min ramping solvent B 3-10%, 35-40 min ramping solvent B 10-15%, 40-43 min ramping solvent B 15-25%, 43-46 min ramping solvent B 25-80%, 46-50 min maintaining solvent B at 80%, 50-51 min decreasing solvent B 80-3%, and 51-60 min sustaining solvent B at 3%. Solvent A was a 2% ACN aqueous solution containing 0.1 % formic acid. The separation and scan time was set to 60 min.

The LTQ Orbitrap Velos mass spectrometer was operated with three scan events. The first scan event was a full MS scan of 380-2000 m/z at a mass resolution of 15,000. The second scan event was CID MS/MS of parent ions selected from the first scan event with an isolation width of 3.0 m/z, a normalized collision energy (CE) of 35%, and an activation Q value of 0.250. The third scan event was set to acquire HCD MS/MS of the parent ions selected from the first scan event. The isolation width of HCD experiment was set to 3.0 m/z while the normalized CE was set to 45% with an activation time of 0.1ms. The CID and HCD MS/MS were performed on the 8 most intense ions observed in the MS scan event. In a separate LC-MS/MS, ETD MS/MS was conducted in conjunction with CID and HCD. The first scan event was a full MS scan and 15 scan events were followed alternating between CID, HCD, and ETD. This enabled MS/MS of the 5 most intense ions observed in the first scan. For ETD, the isolation width was set to 4.0 m/z and the default charge state was set to 4. The reaction time was set to 150 ms with a supplemental activation. The LTQ Orbitrap Velos mass spectrometer was externally calibrated, permitting < 2ppm mass accuracy.

Data Processing and Quantitation

The identification of CO2A1 peptides/glycopeptides was achieved using MASCOT database and Scaffold 3 software. Proteome Discoverer version 1.2 software (Thermo Scientific, San Jose, CA) was used to generate a mascot generic format file (*.mgf) which was subsequently employed for database searching using MASCOT version 2.3.2 (Matrix Science Inc., Boston, MA) and Scaffold 3 (Proteome Software, Inc., Portland, OR). Parent ions were selected from a mass range of 350-5000Da with a minimum peak count of 1. The parameters from Mascot Daemon were set to search against the Swissprot database. Oxidation of methionine was set as a variable modification. Additionally, several PTMs of CO2A1 were set as variable modifications, including hydroxylation of proline and lysine, and glycosylation of hydroxylysine with 1 Hex and 2 Hex. An m/z tolerance of 5 ppm was set for the identification of peptide/glycopeptide with 2 missed cleavages. Also, tandem MS ions were searched within 0.8 Da mass tolerance. Additionally, manual searching of theoretical m/z was performed using mass accuracy < 5ppm. The m/z values were based on every theoretical peptide backbone sequence using the PeptideMass tool from the ExPASy website. Then, hydroxylation and glycosylation modifications were applied, and theoretical m/z values were generated. These m/z values were employed to create extracted ion chromatograms (EICs) using Xcalibur Qual Broswer 2.1 (Thermo Scientific, San Jose, CA). A mass accuracy of 5ppm or better was then employed to confirm ions prior to manual evaluation of tandem MS data. Manual inspection of tandem MS data was employed for these cases. We only considered and assigned a fragment ion if its S/N is equal or higher than 3. Moreover, ions with such S/N are only considered if they appear in more than one tandem mass spectra associated with a particular m/z value. From multiple MS/MS spectrum, the unique y/b ions or c/z ions of specific modification were subsequently confirmed. CID/HCD (/ETD) MS/MS was repeated using the mass list of the theoretical m/z values. Assignments of peptide modifications were attained using both MASCOT database searching and manual interpretations.

For quantitation, peak heights were acquired using Xcalibur Qual Browser. The software was employed to generate EICs. Mass range was set to full FTMS scan with smoothing of 7 points enabled and mass tolerance of 10 ppm allowed. Peak apex was determined as peak height with a complete peak shape. Invariably detected peptides were selected as internal standards against which all data were normalized to offset ESI spray variations. Three tryptic peptides with the following amino acid sequences GEAGAQGPMGPAGPAGAR, GFpGLpGPSGEPGK, and GFTGLQGLpGPpGPSGDQGASGPAGPSGPR were selected to normalize quantitative data. These peptides were detected at 18.66 min, 34.67min, and 46.44 min, respectively. Every modified peptide sequences possessing a glycosylation site was considered for semi-quantitation.

Identification of O-glycosylation and Hydroxylation of Amino Acid Residues

CO2A1 possesses 24 potential glycosylation sites with a Gly-Xaa-HyK motif.^{1, 2, 19} Here, LC-MS/MS analyses of trypsin and GluC digested bovine CO2A1 allowed the identification of 23 identified glycosylation sites with sequence coverage of ca. 90% as shown in Figure 1. Of the 24 potential glycosylation sites associated with CO2A1, only K1130 was not detected. We believe that peptide sequences generated by the protease containing K1130 were either too short or too large to be effectively observed by LC-MS/MS. The abovementioned sequence coverage does not include signal and propeptides from 1 to 201 and the C-telopeptide region from 1253 to 1487, since the sample used in this study is pepsinized. In Figure 1, detected sequences are underlined, and hydroxylated P/K are designated with bold letters. Letters P or K listed above the sequence indicate the detection of the residue as both modified and unmodified. 104 HyP residues were Yaa residue in Gly-Xaa-Yaa motif while 14 HyP residues were Xaa residue. Previously, two of Xaa positions of Gly-HyP-HyP from CO2A1 have been reported which are found at P1144 and P1186.⁵² We also determined 7 unusual HyP residues in Gly-HyP-Ala and a HyP residue in Gly-HyP-Val motifs, which were also reported by Yang et al,⁴⁹. However, we here observed a HyP in Gly-HyP-Gln at P220 which has not been previously reported. As suggested by Yang et al⁴⁹ and observed here, it appears that unusual HyPs at Xaa position would occur in different motifs associated with different types of collagens. The ETD or CID mass spectra of 9 peptides with unusual hydroxyproline residues are included as Supplementary Spectra. The glycosylation sites of CO2A1 identified here were compared to what has been reported by Butler and Francis et al,^21-23 Seyer et al,⁵³ and Taga et al⁴⁷ (see Supplementary Table 1).

There is a 98% amino acid sequence overlap (homology) between bovine and human CO2A1. Both bovine and human CO2A1 have 24 Gly-Xaa-HyK glycosylation motifs. In Supplementary Figure 1, the different PTMs associated with bovine CO2A1 as determined in this study were compared to those of human CO2A1, which were recently reported by Van den Steen et al.⁴⁶ They have detected 19 Gly-Xaa-HyK glycosylation motifs in human CO2A1 while peptides containing K419, K452, K731, K929, and K1130 as part of Gly-Xaa-HyK glycosylation motifs were not observed (Supplementary Figure 1). Lysine residues at K308, K374, K527, K764, and K1055 in human CO2A1 were observed to be unmodified; however, these same residues were observed to be modified in bovine CO2A1. K287, K299, K470, and K542 in both human and bovine CO2A1 were observed as hydroxylated and modified with either Glc-Gal or Gal moieties. The PTMs of the remaining lysine residues with Gly-Xaa-HyK glycosylation motifs in human and bovine CO2A1 were not comparable. For example, K773 in bovine CO2A1, as determined here, was detected as unmodified, but K773 in human CO2A1 was observed as hydroxylated and modified with either Glc-Gal or Gal moieties. Although bovine and human CO2A1 proteins have the same Gly-Xaa-HyK motifs, their PTMs were not exactly the same.

In some cases, the identification of PTMs associated with a specific lysine residue was facilitated by differences in the retention times. For example, the EICs of TGPPGPAGFAGPPGADGQPGAKGEQGE peptide with different modifications of proline and lysine residues are shown in Figure 2. The lysine residue at 848 (K848) of this peptide sequence is detected as unmodified, hydroxylated or glycosylated with either Glc-Gal or Gal moieties. Two of the proline residues (P839 and P830) associated with this peptide sequence were only observed as hydroxylated. The abovementioned peptide sequence modified with Glc-Gal moiety has an m/z value of 926.073 (+3) and retention time of 29.59 min (Figure 2A). The same peptide with a modification of Gal moiety has an m/z value of 872.055 (+3) and retention time of 30.27 min (Figure 2B). At m/z 818.037 (+3), two isomers were observed (Figure 2C). The peptide eluted at 28.64 min was the peptide sequence TGPPGPAGFAGPPGADGQPGAKGEQGE with the hydroxylation of 2 consensus proline residues (P839 and P830) and an additional proline residue (P845). The other isomer observed at 30.95 min was the same peptide sequence with the hydroxylation of 2 consensus proline residues (P839 and P830) and the lysine residue (K848). The retention time of TGPPGPAGFAGPPGADGQPGAKGEQGE peptide with HyK was higher than that with HyP. This is in agreement with what has been previously reported.⁵⁴ The peptide sequence containing the unmodified K848 was the most retained structure (31.38 min) as shown in Figure 2D. Accordingly, it appears that the hydroxylation and glycosylation of this peptide sequence reduces hydrophobicity of the structures and subsequently decreases retention times.

Isomeric separation deamidated Asn residues at N612 and N800 with different retention times was also observed (Supplementary Table 2). Such isomers are observed as either Asp or isoAsp. Although ETD spectrum did not provide enough information to assign the elution orders of these isomers, the isoAsp containing peptide are believed to elute before Asp counterpart similar to what have been previously reported.^55-57 In other cases CID, HCD, and ETD tandem MS has facilitated the identification of many PTMs.

The identifications of the majority of the modified peptide sequences summarized in Supplementary Table 2 were achieved employing different fragmentation mechanisms, including CID, HCD and ETD. For example, Figure 3 depicts the CID tandem MS of the peptide sequence TGPPGPAGFAGPPGADGQPGAKGEQGE with different modifications. Figure 3A illustrates CID tandem MS of Glc-Gal-HyK modified peptide with m/z 926.073 (+3). This ion exhibited a very complex fragmentation pattern since Glc and Glc-Gal are easily lost during the CID process. For this ion, fragment ions originating from the loss of Glc and Glc-Gal were observed in the CID tandem MS. The fragment ion representative of a neutral loss of Glc was observed at m/z 872 (+3) while a fragment ion representative of neutral loss of Glc-Gal was also observed at m/z 818 (+3). Both fragment ions were observed at high intensities. As depicted in Figure 3B, the fragment ions originating from the Gal modified peptide sequence (m/z 872.055, +3) were mainly peptide fragments with intact Gal moiety. A very limited number of y or b ions with a loss of Gal were observed in the spectrum. Figure 3C exhibits CID tandem MS of the isomer observed at 28.64 min with a hydroxylation on a different proline and not on K848. On the other hand, Figure 3D is a CID tandem MS of the other isomer detected at 30.95 min, in which K848 is hydroxylated. The CID of unmodified K848 peptide is shown in Figure 3E.

There are many isomeric peptides/glycopeptides associated with CO2A1 because there are many possible combinations for the same masses with multiple hydroxyproline and hydroxylysine residues. As shown in Figure 2C, the retention times of two isomeric peptides were notably different, although they only differ in the hydroxylation of a single amino acid residue. They were unequivocally identified by tandem MS as depicted in Figures 3C and D. Each of these isomers has a unique series of y ions (y6, y7, and y8) and b ions (b19, b20, and b21) associated with the PGAK sequence. For example, the y6 fragment ion at m/z 647 (+1) corresponds to the N-terminus HyP fragment (Figure 3C) while the y6 fragment ion at m/z 663 (+1) corresponds to the N-terminus HyK fragment (Figure 3D). The ion with m/z 647 (+1) is a fragment ion unique for the HyP modified peptide, while the ion with m/z 663 (+1) is a fragment ion unique to the HyK modified peptide. The ETD or CID mass spectra of 48 peptides containing unmodified, glycosylated, or/and hydroxylated lysine residues are available as

Microheterogeneity of Glycosylation Sites

The assignment of glycosylation sites by CID tandem MS when multiple potential sites are present in a peptide sequence may not be feasible because of the limited number of b and y fragment ions detected. Also, fragments involving a loss of Glc and Gal prompt highly complex patterns, thus making the identification of the glycosylation sites quite challenging. Therefore, ETD experiments were employed to confirm the glycosylation sites of CO2A1 and their micro- and macroheterogeneities.

Figure 4A and 4B depict an ETD tandem mass spectrum of two glycoforms associated with a peptide sequence possessing two possible glycosylation sites (K299 and K308). The fragment ions originating from the first glycoform with a Gal moiety on K299 and Glc-Gal moiety on K308 are labeled in Figure 4A while the fragment ions originating from the second glycoform with Glc-Gal moiety on K299 and Gal moiety on K308 are labeled in Figure 4B. These two glycoforms co-eluted as suggested by the co-existence of both characteristic ions in the same spectrum. In Figure 4A, the m/z difference between c8 and c9 fragment ions and z27 and z28 fragment ions is 306, which corresponds to the presence of Gal moiety at K299 site. The m/z difference between c17 and c18 fragment ions and z18 and z19 fragment ions is 468, which corresponds to the presence of Glc-Gal moiety at K308 site. On the other hand, the characteristic fragment ions suggesting the other glycoform, including c8 and c9 fragment ions and z27 and z28 fragment ions (a difference of 468) as well as c17 and c18 fragment ions and z18 and z19 fragment ions (a difference of 306), are shown in Figure 4B. In addition to those diagnostic ions of glycosylated lysine residues, annotations of a series of different fragment ions between c9 and c17 and z19 and z27 confirmed the co-existence of two glycoforms on K299 and K308.

Figure 4C and 4D depict the ETD tandem mass spectrum of the glycoforms observed for another peptide sequence possessing two glycosylation sites (K452 and K464). The fragment ions labeled in Figure 4C originate from the occupation of K452 with Glc-Gal moiety and the hydroxylation of K464. These modifications were determined based on assignments of c14 and c15 fragment ions and z18 and z19 fragment ions (a difference of 468) while those of c26 and c27 fragment ions and z6 and z7 fragment ions (a difference of 144). In Figure 4D, the different m/z values of those fragment ions from the same ETD spectra support the occupancy of each K452 and K464 with a Gal moiety. Additionally, fragment ions suggesting the hydroxylation ofK452 and occupancy K464 with Glc-Gal were not observed.

Macroheterogeneity of Glycosylation Sites

The macroheterogeneity of a glycoprotein represents the variability of glycan occupancies at multiple glycosylation sites of the same protein, resulting in different copies of the protein. Enzymatic or chemical proteolysis, in bottom-up approach, facilitates MS or tandem MS analysis of modified amino acid sequences. However, it is quite challenging to study the macroheterogeneity of glycosylation sites since there is a lack of information on different copies of protein modification. Here in the case of CO2A1, the tryptic digestion generated several peptides containing two glycosylation sites, such as K299 and K308, K452 and K464, K464 and K470 as well as K857 and K884, thus enabling the assessment of the macroheterogeneity of these sites.

As shown in Figure 4A and 4B, K299 and K308 co-exist in a tryptic peptide with an m/z value of 969.9250 (+4) with different glycoforms. The tandem MS data suggest that there are two glycoforms associated with this m/z value. The first has K299 modified with Gal moiety and K308 modified with Glc-Gal moiety while the second has K299 modified with Glc-Gal moiety and K308 modified with Gal moiety. The chromatographic conditions employed here were not sufficient to resolve such isomers. Based on the intensities of fragment ions observed in the spectrum, the isoform with K299 modified with Gal moiety and K308 modified with Glc-Gal moiety appears to be more abundant than the other isoform. A tryptic peptide containing K299 modified with Glc-Gal moiety and K308 as hydroxylated was also detected at m/z value of 929.4114 (+4). ETD spectrum of this m/z value suggested the presence of only one form (data not shown). Another tryptic peptide with m/z value of 1010.4374 (+4) corresponds to both K299 and K308 modified with Glc-Gal moieties. Other peptides with the different number of HyP residues and with deamidation were also observed with the same macroheterogeneities (data not shown). Accordingly, the macroheterogeneity associated with those two sites are 2 Glc-Gal moieties on K299 and K308, Gal moiety on K299 with Glc-Gal moiety on K308, Glc-Gal moiety on K299 with Gal moiety on K308, and Glc-Gal moiety on K299 with hydroxylation on K308.

Isoforms associated with the glycosylation of K452 and K464 were observed at m/z 869.1708 (+4) as illustrated in Figure 4C and 4D. The tandem MS depicted in Figure 4C suggests K452 modification with Glc-Gal moiety and hydroxylation of K464. On the other hand, the tandem MS shown in Figure 4D suggests that both K452 and K464 are modified with a Gal moiety. The other possible isoform consisting of hydroxylated K452 and K464 modified with Glc-Gal moiety was not observed. By comparing intensities of fragment ions observed in ETD spectrum, both isoforms appear to be present at comparable levels. Another isoforms associated with K452 and K464 was detected at m/z value of 909.6833 (+4). These isoforms were determined to possess K452 modified with Gal moiety and K464 modified with Glc-Gal moiety as well as K452 modified with Glc-Gal moiety and K464 modified with Gal moiety. The levels of these isoforms appear to be comparable. Multiple glycosylation with 2 Glc-Gal moieties on K452 and K464 was also detected at m/z value of 946.1980 (+4) and 950.1942 (+4). These tryptic peptides contain unmodified and hydroxylated K470, respectively. Consequently, macroheterogeneities associated with K452-K464 appear to be 2 Glc-Gal moieties on K452 and K464, Gal moiety on K452 with Glc-Gal moiety on K464, Glc-Gal moiety on K452 with Gal moiety on K464, Glc-Gal moiety on K452 with hydroxylation on K464, and 2 Gal moieties on K 452 and K464.

A tryptic peptide possessing both K464 and K470 was observed. ETD tandem MS data suggested the presence of different forms, including 2 Glc-Gal moieties on both K464 and K470 (m/z 1057.4783, +4), Glc-Gal moiety on K464 and a Gal moiety on K470 (m/z 1016.9632, +4), and Glc-Gal moiety on K464 with hydroxylated K470 (m/z 976.4493, +4). No other combinations were observed. Another tryptic peptide with K857 and K884 was observed as having 2 Glc-Gal moieties on each K857 and K884 (m/z 1048.9772, +4), Gal moiety on K857 and Glc-Gal moiety on K884 (m/z 1008.4650, +4), and hydroxylation on K857 with Glc-Gal moiety on K884 (m/z 967.9489, +4) with 2HyPs and m/z 963.9534 (+4) with 1HyP). Accordingly, ETD tandem MS data have permitted the assignment of partial macroheterogeneity of several glycosylation sites.

Semi-quantitation of O-linked Glycopeptides from Collagen α-1 (II) Chain

Several glycoforms and isoforms were observed for each glycosylation site. Therefore, the peak heights of all observed peptides representative of a glycosylation site were added to depict the abundance of such site. We are aware of the fact that there are two major challenges associated with label-free quantitation: spray variability and the discrepancy in the ionization efficiencies of peptides/glycopeptides. Therefore, we have applied normalization and relative comparison of abundances to improve the reliability in quantitation.^58-60 Raw intensities were normalized using invariably present peptides detected at different retention times to offset the spray variability. The quantitative data were compared as percentages of different modifications at the same glycosylation site, thus compensating for the different ionization efficiencies of peptides. Hence, this semi-quantitative approach can be employed to compare different modifications of each glycosylation site associated with different samples.

Table 1 and Figure 5 summarize the abundances of all the CO2A1 peptide sequences possessing Gly-Xaa-HyK motif with different posttranslational modifications. The sequences, m/z values, retention times, and mass accuracies of all the modified peptides used for quantitation are summarized in Supplementary Table 2. All modified peptide sequences were supported by MS/MS and/or at least 5ppm mass accuracy. Using the abovementioned semi-quantitative approach, the data indicate that the glycosylation of hydroxylysine residues is not only dependent on the extent of hydroxylation of the lysine residues but also dependent on other factors such as steric hindrance. As summarized in Table 1 and shown in Figure 5, K1118, K929, K452, K419, K374, K308, K299 and K287 were only observed as hydroxylated or glycosylated. Unmodified forms of K731 (<1%) and K803 (2%) were only observed at trace levels. K884 (97.7%), K731 (92%), K419 (82%), K374 (81%), K857 (81%), K299 (79%), K764 (70%), K542 (67%), and K608 (63%) were mainly detected as glycosylated with either Glc-Gal or Gal moieties. On the other hand, K1118 (3%), K1055 (14%), K956 (6%), K452 (11%) K308 (14%) and K287 (18%) exhibited low total glycosylation. K1118 (97%), K452 (89%), K929 (88%), K308 (86%), K287 (82%), K470 (74%), K848 (62%) and K620 (54%) are mainly observed as hydroxylated (Table 1 and Figure 5). K773 (100%) was only observed as unmodified while K527 (54%) and K464 (39%) were mainly observed as unmodified.

Semi-quantitation of micro- and macroheterogeneity of bovine CO2A1 glycosylation correlated to proline or lysine hydroxylations could help in better understanding the biological roles of these modifications in fibrillogenesis, cross-linking, or matrix mineralization. This comprehensive characterization could enable better understanding the biological roles of CO2A1 posttranslational modifications in several diseases, including rheumatoid arthritis or collagen-induced arthritis.