Integrated GlycoProteome Analyzer (I-GPA) for Automated Identification and Quantitation of Site-Specific N-Glycosylation

Abstract

Human glycoproteins exhibit enormous heterogeneity at each N-glycosite, but few studies have attempted to globally characterize the site-specific structural features. We have developed Integrated GlycoProteome Analyzer (I-GPA) including mapping system for complex N-glycoproteomes, which combines methods for tandem mass spectrometry with a database search and algorithmic suite. Using an N-glycopeptide database that we constructed, we created novel scoring algorithms with decoy glycopeptides, where 95 N-glycopeptides from standard α1-acid glycoprotein were identified with 0% false positives, giving the same results as manual validation. Additionally automated label-free quantitation method was first developed that utilizes the combined intensity of top three isotope peaks at three highest MS spectral points. The efficiency of I-GPA was demonstrated by automatically identifying 619 site-specific N-glycopeptides with FDR ≤ 1%, and simultaneously quantifying 598 N-glycopeptides, from human plasma samples that are known to contain highly glycosylated proteins. Thus, I-GPA platform could make a major breakthrough in high-throughput mapping of complex N-glycoproteomes, which can be applied to biomarker discovery and ongoing global human proteome project.

Figure 1. Structural and functional components of the Integrated N-glycoproteome analyzer (I-GPA).

(a) Schematic workflow for site-specific glycoform analysis of intact N-glycopeptides by I-GPA. N-glycopeptides enriched from tryptic digests of glycoproteome samples using HILIC were analyzed by high-resolution MS with a combination of HCD and CID fragmentation, followed by analysis using the automated search engine I-GPA. (b) Schematic algorithm of site-specific glycoform analysis of N-glycopeptides by I-GPA. I-GPA consists of a GPA-DB (e.g., 254,826 N-glycopeptides in human plasma), an identification component (id-GPA), a quantitation component (q-GPA), and comparative GPA (c-GPA). The N-glycopeptide database for I-GPA was automatically constructed using the program GPA-DB-Builder. The id-GPA search algorithm was used for N-glycopeptide identification from the MS and MS/MS (HCD/CID) data using M-, S-, and Y-scores as criteria for N-glycopeptide identification. The identified N-glycopeptides were then quantitated by q-GPA by summation of the top three-isotope peaks from the MS1 spectral points. c-GPA can compare the abundance of an N-glycopeptide between glycoproteome samples by the top three-isotopes quantitation (3TIQ).

Figure 2. Computational algorithm of id-GPA for identification of standard α1-acid glycoprotein (AGP).

id-GPA was designed for identification of N-glycopeptides in high-throughput analysis using three scoring systems (M-, S-, and Y-scores). (a) M-score for N-glycopeptide selection based on 15 glycan-specific oxonium ions from the total HCD-MS/MS spectra; N = 1,674, number of selected N-glycopeptide spectra for subsequent analysis with M-score > 1.3 from a total of 5,818 HCD spectra, at a false discovery rate (FDR) of 2.5% (as determined by manual validation). (b) S-score for the selection of N-glycopeptide candidates by matching the isotope distribution of N-glycopeptides in the database; n = 924, number of selected precursor ions of N-glycopeptide candidates with an S-score > 98.0 at an FDR of 19.7% (determined by manual validation). (c) Y-score for identification of N-glycopeptides by matching fragment ions in the CID and HCD spectra with the original 924 N-glycopeptide candidates. Ultimately, N = 456 N-glycopeptide spectra were identified with Y-score > 69.5 at an FDR of 0.0% (as determined by manual validation).

Figure 3. q-GPA algorithm for label-free quantitation of standard α1-acid glycoprotein (AGP).

(a) The identified N-glycopeptides of AGP were quantitated based on the combined intensity of top three isotope peaks. For example, if we suppose to have two N-glycopeptides (#1 = CANLVPVPITNATLDQITGK_6503(+4) and #2 = CANLVPVPITNATLDQITGK_6522(+4)) with monoisotope mass of 1247.288 and 1247.533 Da, respectively, they might be distinguished in Fig. 3a (top) by MS due to their mass difference over 150 ppm. However, N-glycopeptide #2 in XIC (m/z = 1247.533) of Fig. 3a (bottom) can be interferenced by second isotope (m/z = 1247.538) of N-glycopeptide #1 due to their mass difference of 3–4 ppm. For correct quantification of N-glycopeptide #2, we introduced TIQ (Top Three Isotopes Quantification) method as shown in Fig. 3b,c, where 3TIQ uses the combined intensity of top three isotope peaks at three highest MS spectral points (Fig. 3b,c). Since we are evaluating the spectral pattern of selected MS spectra with S-score more than 98.0, it is possible to remove signal interference effectively from co-eluted peaks from similar m/z ions as shown in Fig. 3b, even though it has only mass difference of 3–4 ppm between the monoisotope ion of N-glycopeptide #2 and second isotope ion of N-glycopeptide #1. (d) Number of MS spectral points (1, 2, 3, 5, and 7) needed for TIQ, compared with the XIC manually extracted for quantitation. N is the number of selected N-glycopeptides used for quantitation of the AGP standard. (e) Gray bars indicate linear regression with XIC quantitation. Red line indicates the percentage of identified N-glycopeptides that were quantitated. The highest number of N-glycopeptides was quantitated by 3TIQ, which yielded the best linear regression (R² = 0.959) with XIC.

Figure 4. c-GPA algorithm for quantitation of three different HILIC-enriched batches of standard α1-acid glycoprotein (AGP).

(a) Schematic algorithm of c-GPA for quantitation of N-glycopeptides from multiple samples. (b) Correlation coefficient of linear regressions (R²) between areas of manual 3TIQ and XIC were above 0.93 for several N-glycopeptides from an AGP standard sample enriched by HILIC. (c) The Pearson correlation coefficient (r) of a scatter graph of the coefficients of variation (CVs) was the value r = 0.8199, indicating the similarity between the results obtained by 3TIQ and XIC. The correlation coefficient of the two methods had a p-value below 0.0001.

Figure 5. Analysis of N-glycopeptides in normal and hepatocellular carcinoma (HCC) plasma samples by I-GPA.

(a) Venn diagrams of the number of unique N-glycopeptides identified in depleted or non-depleted samples of human plasma. (b) Results of label-free quantitative analysis of N-glycopeptides, using c-GPA on a depleted human plasma sample. (c) Comparison of label-free quantitation using 3TIQ on depleted human plasma after N-glycopeptides with coefficients of variation >30% were filtered out. (d,e) Volcano plot showing log (fold change) versus log (P-value) of differentially expressed glycoproteins (d) or N-glycopeptides (e) differentially expressed N-glycopeptides from AGP, AACT, HPX, CLU and AFP in red, purple, cyan, blue and green circle, respectively.