A targeted PRM approach was applied for the quantitation of intact N-glycopeptides of serum haptoglobin to characterize the alterations in site-specific glycopeptide forms between cirrhosis and HCC.
The experimental workflow is shown in Fig. 1. Haptoglobin was purified from a 20µL aliquot of serum from each patient by using an HPLC-based antibody-immobilized column 1,9. Serum haptoglobin has four glycosylation sites (Asn184, Asn207, Asn211, and Asn241), which can be observed after a two-step enzymatic digestion using trypsin and Glu-C 9. In this work we focused on the glycosylation sites Asn184, Asn207, and Asn241; where we were unable to identify any N-glycopeptides from the glycosylation site Asn211 in a reproducible number of samples to apply reliable evaluation 9,11. A pooled sample was analyzed and used to identify the N-glycopeptides present in the samples based on monoisotopic mass, charge, retention time, and MS2 spectra, Supplementary Table S1.
The specific information for each detected N-glycopeptide structure was targeted in the LC-MS-PRM approach on the Orbitrap QExactive (Thermo). The selections of the precursor ions for all identified glycopeptide structures was based on the signal intensity so that each case the most abundant ion was used in the PRM strategy. Three charged precursor ions were the common ionic glycopeptide species in the three evaluated glycosylation sites, except for the glycopeptides with tri- and tetra-antennary and mono-, di-, and trisialylated glycans attached to the glycosylation site Asn241; where these structures presented four charged ions, Supplementary Table S1. The energy level was tested in the HCD cell of the QExactive before the application of the PRM strategy. The most appropriate collision energy to produce stable and abundant Y fragment ions of the glycan portion was 25 eV. Additionally, the fragmentation produced allowed us to confirm the site-specific glycosylation by the observation of abundant Y1 ions with m/z values of 1940.9333, 1176.5484, and 1998.0920 for the sites Asn184, Asn207, and Asn241 respectively. The peptide fragments were observed in low abundance in the three glycosylation sites, Supplementary Fig. S1. Six of the most representative and abundant fragment Y ions were selected for each N-glycopeptide, and their peak area calculated using Xcalibur (Thermo) software, see Supplementary Table S1. After the area under the curve was computed the data was normalized based on the total area of all identified glycopeptides and their relative abundance calculated. Significant differences in abundance between cirrhosis and HCC samples were calculated using the Wilcoxon Test. Any p – values < 0.05 were considered to be significant changes; Supplementary Table S2 shows the data obtained for cirrhosis samples and Supplementary Table S3 shows the data obtained for HCC samples. The retention time of the evaluated structures was such that N-glycopeptides for the glycosylation site Asn184, eluted between 42 and 54 min, for the Asn207 site between 40 and 60 min, and for the Asn241 site between 61 and 88 min. For all sites a retention time pattern based on the glycan moiety was observed, Supplementary Table S1. For all the glycosylation sites the first glycopeptides to be eluted were the structures with small biantennary mono-, and di-sialylated glycans attached. The glycopeptides with sialylated-fucosylated glycans attached eluted according to the number of sialic acid molecules presented in the glycan moiety regardless of the antennae number, Supplementary Table S1. The core or branch fucosylation of important N-glycopeptides was performed by the evaluation of their fragmentation, an example can be found in Supplementary Fig. S2. A pooled sample was used to determine the sialic acid linkage of some important structures by the application of α2,3 neuraminidase enzyme digestion (a description can be found in Supplementary Fig. S3). To facilitate the glycan and glycopeptide descriptions, the nomenclature used is as follows. For glycans, a four-digit nomenclature represents the number of HexNAc, Hex, Fuc, and Neu5Ac molecules (N-acetylhexosamine, Hexose, Fucose, and N-acetylneuraminic acid respectively). In the case of glycopeptides, the letters MVS, NFL, and VVL were added to the glycan nomenclature and represent the peptide chain where they were observed (MVS = MVSHHN184LTTGATLINE, NLF = NLFLN207HSE, and VVL = VVLHPN241YSQVDIGLIK).
Figure 2 describes the data treatment using the glycopeptide structure NLFLN207HSE + HexNAc5,Hex6,Neu5Ac3 (NLF_5603) as an example. Figure 2a and Fig. 2b depict the extracted ion chromatograms (EICs) for the glycopeptide NLF_5603 derived from all cirrhosis and HCC patients, respectively. According to the statistical test, the isomeric structure number 2 of this glycopeptide form had significant changes in abundance between cirrhosis and HCC samples, Supplementary Table S4; this result is further discussed in this section. The corresponding precursor ion for the structure NLF_5603 has a triple charge m/z value of 1181.7977, it eluted in a retention time of about 72 min, and it produced monocharged Y fragment ions principally of the glycan portion. Figure 2c shows the capability of our strategy to resolve three isomeric structures for the glycopeptide NLF_5603. Representative TICs of other important isomeric structures observed in the samples are described in Supplementary Fig. S4. Figure 2d shows the sensitivity enhancement of our PRM approach versus the full scan (MS1) signal observed in the pooled sample injected with the same concentration. This increase on sensitivity allowed us to address an accurate quantitation of low abundant sialylated fucosylated glycopeptide structures that have special importance in HCC detection.
Haptoglobin Microheterogeneity
Based on our early research efforts as described by Huang et al. and Zhu et al., the isomeric glycan profile of serum haptoglobin and its glycosylation site heterogeneity were determined 1,9. These works demonstrated the importance of the isomeric structures, either glycan or glycopeptides in the differentiation of patients with NASH cirrhosis and NASH related-HCC. Huang et al. reported seven sialylated fucosylated isomeric glycans with the structures HexNAc4Hex5FucNeuAc, HexNAc5Hex6FucNeuAc3, and HexNAc5Hex6Fuc2NeuAc3. All of these structures showed significant changes in abundance between the two disease states 1. In a related work, Jin et al. found 12 isomeric glycans with significant differences between healthy and HCC patients 13. The development of an LC strategy that allows the separation of isomeric N-glycopeptides was addressed based on the use of a long C18 column (50cm) and high temperature. This strategy was adopted from the work of Ji et al., where they reported an important separation of isomeric O- and N-glycopeptides from tryptic digested α1-acid glycoprotein (AGP), and evaluated the effects of temperature on the isomeric separation as well 26.
We were able to accurately identify and quantify a total of 73 isomeric structures that corresponded to 42 N-glycopeptide forms from the tryptic/Glu-C digested haptoglobin, as shown in Table 1. The distribution of the N-glycopeptides in the glycosylation sites were as follows: 13 isomeric structures that corresponded to 11 N-glycopeptide forms in the glycosylation site Asn184; 44 isomeric structures that corresponded to 19 N-glycopeptide forms in the glycosylation site Asn207; and 16 isomeric structures that corresponded to 12 N-glycopeptide forms in the glycosylation site Asn241. The glycosylation site Asn207 was found to have the most abundant diversity of N-glycopeptides as well as the most abundant isomeric glycopeptide structures.
Table 1
Observed haptoglobin glycopeptide structures by glycosylation site including isomeric forms. The number in parentheses “()” represents the number of isomeric forms in the glycopeptide structure. Glycan nomenclature; HexNAc, Hex, Fuc, NeuAc (N-acetylhexosamine, Hexose, Fucose, N-acetylneuraminic acid).
Glycopeptide backbone | MVSHHN184LTTGATLINE | NLFLN207HSE | VVLHPN241YSQVDIGLIK |
N-glycosylation Microheterogeneity | --- | [3-4-0-1](2) | --- |
--- | [4-4-0-1](3) | [4-4-0-1] |
[4-5-0-0] (2) | [4-5-0-0](2) | [4-5-0-0] |
[4–5–1–1] | [4–5–1–1](2) | --- |
[4–5–1–2] | [4–5–1–2](2) | [4–5–1–2](2) |
[4-5-0-1] | [4-5-0-1] | [4-5-0-1] |
[4-5-0-2] | [4-5-0-2](2) | [4-5-0-2] |
[4-6-0-1] | --- | --- |
[5–6–1–1] | [5–6–1–1](2) | --- |
[5–6–1–2] | [5–6–1–2](2) | [5–6–1–2] |
[5–6–1–3] | [5–6–1–3](2) | --- |
[5-6-0-1] | [5-6-0-1](2) | [5-6-0-1] |
[5-6-0-2](2) | [5-6-0-2](2) | [5-6-0-2] |
--- | [5-6-0-3](3) | [5-6-0-3](3) |
--- | [6–7–1–1](4) | --- |
--- | [6–7–1–2](3) | --- |
--- | [6–7–1–3](5) | --- |
--- | [6-7-0-1] | [6-7-0-1] |
--- | [6-7-0-2](3) | [6-7-0-2] |
--- | --- | [6-7-0-3](2) |
--- | [6-7-0-4] | --- |
We found 7 common glycan structures between the three glycosylation sites Asn184, Asn207, and Asn241, 10 between the sites Asn184 and Asn207, 11 between sites Asn207 and Asn241, and 7 between sites Asn184 and Asn241, Table 1. The glycan structure HexNAc4, Hex6, Neu5Ac was only observed in glycosylation site Asn184; the structure HexNAc3, Hex4, Neu5Ac was only observed in the site Asn207; and the structure HexNAc6, Hex7, Neu5Ac3 was observed only in the site Asn241 which matched with the findings of Zhu et al. 9 In addition, some of the common structures showed a different number of isomeric peaks among the glycosylation sites. For example, the glycan composition HexNAc4, Hex5, Neu5Ac2 (4-5-0-2) was present in the three sites, but interestingly, two isomeric glycans were observed only in the site Asn207, Fig. 3. Similar results were observed for the other common glycan structures among the three glycosylation sites, Table 1. The results demonstrated the ability of our analytical strategy to unravel the glycan microheterogeneity of the glycosylation site NLFLN207HSE, and clearly described the distribution of all glycan structures among the three evaluated haptoglobin glycosylation sites. According to previous research, 35 of the 42 identified N-glycopeptides in our research were common with the reported work of Zhu et al., who evaluated similar analytical samples with an LC-EThcD-MS2 analytical strategy and found a large number of N-glycopeptides; however, their work did not report isomeric structures. 9
Principal component analysis (PCA) was performed to evaluate the ability of the obtained data to differentiate between cirrhosis and HCC. The analysis was performed with the MarkerView® (AB Sciex) software using the normalized data. The initial evaluation of the complete data showed partial separation between both cohorts, the cirrhosis and HCC samples, Supplementary Fig. S5. A further PCA analysis of the data by glycosylation site, Fig. 4, showed interesting results. For the sites Asn184 and Asn241, the data was not able to separate the disease cohorts as shown in Fig. 4a and Fig. 4c. In comparison, the PCA analysis for the glycosylation site Asn207 showed an important separation between cirrhosis and HCC samples. As discussed previously, the site Asn207 showed an important number of isomeric N-glycopeptides suggesting that the differentiation of the two cohorts is possible due to the microheterogeneous development of the glycosylation site Asn207, Fig. 4b. Further PCA comparisons between cirrhosis and early HCC (TNM stage 1) data of the site Asn207 showed important separation of the two cohorts, Fig. 4d. This results suggested that our analytical strategy is effective in differentiating between cirrhosis and early HCC. This PCA analysis used a reduced number, 7 HCC (TNM stage 1) against all cirrhosis samples, but these promising results should next be confirmed with a larger number of samples.
Heat maps were used to evaluate the changes in abundance of the individual N-glycopeptides quantified in cirrhosis and HCC samples. Site-specific heat maps were obtained for each glycosylation site (Asn184, Asn207, and Asn241), Supplementary Fig. 6a - c. The N-glycopeptides identified with a bold arrow indicate structures with significant changes in abundance between both diseases, cirrhosis and HCC (p < 0.05), with a blue line separating cirrhosis from HCC samples. The rectangles identify representative groups of glycopeptides such as sialylated and sialylated-fucosylated structures, the green color represents low abundance (down regulation), and the red color represents high abundance (up regulation). For the site Asn184 the heat map allows us to identify two principal changes in the glycopeptide abundances between the two sample groups. The bi- and triantennary sialylated glycopeptides (MVS + 4-5-0-2, 5-6-0-1, 5-6-0-2, and 5-6-0-3) were more abundant in HCC than in cirrhosis samples. Otherwise, the bi- and triantennary sialylated glycopeptides (MVS + 3401, 4401, 6701, and 6702) were more abundant in cirrhosis than in HCC samples.
The isomeric separation achieved by our analytical strategy increased the number of glycopeptide structures in the glycosylation site Asn207 and thus the structural information in the site (microheterogeneity). The heat map for this site describes three different groups of glycopeptides with interesting changes in abundance between cirrhosis and HCC samples. The sialylated glycopeptides had different abundance patterns in this site; the triantennary mono-, di-, and trisialylated structures (NLF + 5-6-0-1, 5-6-0-2, and 5-6-0-3) were more abundant in cirrhosis than HCC samples. However, the mono-, di-, and tetra antennary mono- and disialylated glycopeptides (NLF + 3-4-0-1, 4-4-0-1, 6-7-0-1, and 6-7-0-2) were more abundant in HCC than cirrhosis samples. In the case of sialylated fucosylated glycopeptides, the heat map showed similar changes in abundance for almost all structures; these glycopeptides were low abundance in cirrhosis samples, and increased considerably for HCC samples, as expected from prior works 1,2,11. The heat map for the glycosylation site Asn241 only showed a change in abundance in the glycopeptide structures VVL + 4-5-0-1 and 4-5-0-2. These glycopeptides were more abundant in cirrhosis than HCC samples.
The glycome changes associated with the progression of cirrhosis toward HCC was evaluated using pie graphs. For comparison, the data were separated into three main glycan groups attached to the glycopeptide sites; sialylated, sialylated fucosylated, and other structures, Supplementary Fig. S7. The evaluation of the complete data (sites Asn184, Asn207, and Asn241) was applied using the sum of abundances of the glycopeptides with the same glycan moiety; the changes in abundance of the identified glycan structures between the two cohorts are described in Supplementary Fig. S7a. Relative abundance of 91.9% and 7.4% were observed for the sialylated and sialylated fucosylated glycans respectively in cirrhosis samples. Otherwise, the relative abundance of the same glycan groups in HCC samples changed considerably to 83.4% and 16.2% respectively. The changes of the mentioned groups of glycopeptides were also evaluated by glycosylation site. As expected, according to the PCA plots and heat map analyses, the sites Asn184 and Asn241 did not show significant changes in abundance for sialylated and sialylated fucosylated glycopeptides, Supplementary Fig. S7b and Fig. S7d.
For the site Asn207, we observed relative abundance values of 87.2% and 12.1% for sialylated and sialylated fucosylated glycopeptides respectively in cirrhosis samples. In comparison, the relative abundance values of sialylated and sialylated fucosylated glycopeptides changed for HCC samples as 77.5% and 21.7%, Supplementary Fig. S7c. According to the results observed in the pie graphs, two main changes were identified in the glycome of serum haptoglobin from both analyzed cohorts. First is the high abundance of sialylated glycans presented in cirrhosis samples, which decreased by around 9.6% in HCC. The second important change was an increase of 8.8% in abundance of sialylated fucosylated glycans in cirrhosis samples compared to HCC samples.
The results observed in the PCA plots, heat maps, and pie graphs showed consistent changes in the haptoglobin glycopeptide heterogeneity between samples from patients with cirrhosis and HCC. The development of the microheterogeneity of the glycosylation site NLFLN207HSE of serum haptoglobin provided the position of the protein glycosylation changes to differentiate cirrhosis and HCC. Additionally, most of the changes were observed in isomeric glycopeptides, which has substantially increased the number of possible biomarker structures with respect to previous research in the field.
Biomarker Candidates as Differentiated between Cirrhosis and HCC
The statistical Wilcoxon Test was applied to the normalized data of each identified N-glycopeptide in samples from patients with NASH cirrhosis and NASH related-HCC, Supplementary Table S4. Twelve N-glycopeptides showed statistically significant differences in abundance between both analyzed cohorts (p – value < 0.05). To simplify the discussion, we named the following N-glycopeptides “biomarker candidates”: two N-glycopeptides in the site Asn184, MVS + (4-5-1-2 and 5-6-1-3); nine N-glycopeptides for the site Asn207, NLF + (5-6-1-2 Iso 2, 5-6-1-3 Iso 1, 5-6-1-3 Iso 2, 5-6-0-1 Iso 2, 5-6-0-2 Iso 2, 5-6-0-3 Iso 2, 6-7-1-1 Iso 4, 6-7-1-2 Iso 1, and 6-7-1-2 Iso 3); and for the site Asn241, the N-glycopeptide VVL + 4-4-0-1. Eight of the biomarker candidates were sialylated fucosylated glycopeptides; among these were observed di-, tri-, and tetra-antennary structures. These indicated that the increase in abundance of sialylated fucosylated glycans in serum haptoglobin during HCC stages occurs regardless of the antennae number. Four of the biomarker candidates were sialylated glycopeptides with significant decreases in abundance from cirrhosis to HCC samples. Moreover, nine of the candidates were derived from isomeric glycopeptides, and all of them from the glycosylation site NLFLN207HSE. Figure 5 shows the Box plots of the biomarker candidates. Surprisingly, the glycopeptides NLF_6711 Isomer 4 and NLF_6712 Isomer 1 showed an unusual decrease in abundance for HCC samples. These structures presented four and three iso-forms respectively, where their decrease in abundance was caused by the ratio redistribution of the isomeric forms of these glycopeptides. The evaluation of the relative response without isomeric consideration of the structures NLF_6711 and NLF_6712 (Supplementary Tables S2 and S3) demonstrated that both glycopeptides showed the expected increase in abundance in HCC samples.
AFP is a currently used biomarker in the detection of HCC, where the levels of this glycoprotein increase considerably in late stages of HCC. Unfortunately, AFP does not have sufficient sensitivity and specificity to differentiate between liver cirrhosis and early-HCC. To demonstrate the effectiveness of the biomarker candidates developed by our analytical strategy, receiving operating characteristic (ROC) curves were used to evaluate the sensitivity and specificity of their response, Table 2. The area under the curve (AUC) values obtained for the single biomarker candidates and AFP alone were compared, where only the glycopeptide structures NLF_5612 Isomer 2 (0.84) and NLF_5613 Isomer 1 (0.84) showed comparable AUC values with AFP (0.85). Sialylated biomarker candidates NLF_5601 Isomer 2, NLF_5602 Isomer 2, and NLF_5603 Isomer 2 were expected to show better performance based on their high abundance. However, their high standard deviation values led to a low sensitivity and specificity which is reflected with low AUC values in the ROC analyses, Table 2 and Supplementary Table S4. ROC curves for the twelve N-glycopeptides with significant changes in abundance are shown in Supplementary Fig. S8.
Table 2
Area under the curve (AUC) comparing one-biomarker models and two-biomarker models (alpha-fetoprotein “AFP” + biomarker candidate) for the evaluation of cirrhosis and HCC samples “Top”. AUC for single biomarker candidates observed in the evaluation of cirrhosis and early-HCC samples “Down”. MVSHHN184LTTGATLINE = MVS, NLFLN207HSE = NLF, VVLHPN241YSQVDIGLIK = VVL, Iso = isomer, and AFP = alpha-fetoprotein. Glycan nomenclature: HexNAc, Hex, Fuc, NeuAc (N-acetylhexosamine, Hexose, Fucose, N-acetylneuraminic acid).
Single biomarker model | Two-biomarker model |
Marker | AUC | p-value | AFP + Marker | AUC | CI (low) | CI (high) | p-value |
AFP | 0.85 | --- | --- | --- | --- | --- | --- |
MVS_4512 | 0.73 | 0.331 | AFP + MVS_4512 | 0.92 | 0.80 | 1.00 | 0.382 |
MVS_5613 | 0.76 | 0.420 | AFP + MVS_5613 | 0.92 | 0.81 | 1.00 | 0.402 |
NLF_5612 Iso 2 | 0.84 | 0.897 | AFP + NLF_5612 Iso 2 | 0.88 | 0.74 | 0.97 | 0.690 |
NLF_5613 Iso 1 | 0.84 | 0.930 | AFP + NLF_5613 Iso 1 | 0.95 | 0.86 | 1.00 | 0.202 |
NLF_5613 Iso 2 | 0.80 | 0.664 | AFP + NLF_5613 Iso 2 | 0.92 | 0.79 | 1.00 | 0.418 |
NLF_5601 Iso 2 | 0.74 | 0.346 | AFP + NLF_5601 Iso 2 | 0.87 | 0.73 | 0.97 | 0.819 |
NLF_5602 Iso 2 | 0.77 | 0.493 | AFP + NLF_5602 Iso 2 | 0.86 | 0.72 | 0.96 | 0.890 |
NLF_5603 Iso 2 | 0.72 | 0.282 | AFP + NLF_5603 Iso 2 | 0.84 | 0.69 | 0.96 | 0.929 |
NLF_6711 Iso 4 | 0.73 | 0.312 | AFP + NLF_6711 Iso 4 | 0.88 | 0.74 | 0.98 | 0.697 |
NLF_6712 Iso 1 | 0.72 | 0.296 | AFP + NLF_6712 Iso 1 | 0.91 | 0.77 | 0.98 | 0.517 |
NLF_6712 Iso 3 | 0.73 | 0.289 | AFP + NLF_6712 Iso 3 | 0.94 | 0.84 | 1.00 | 0.256 |
VVL_4401 | 0.79 | 0.601 | AFP + VVL_4401 | 0.89 | 0.75 | 0.98 | 0.657 |
Single biomarker model, evaluation of cirrhosis and early-HCC samples |
Marker | AUC | CI (low) | CI (high) |
AFP | 0.66 | 0.37 | 0.96 |
MVS_5613 | 0.86 | 0.66 | 1.00 |
NLF_5612 Iso 2 | 0.84 | 0.60 | 1.00 |
NLF_5613 Iso 1 | 0.88 | 0.69 | 1.00 |
NLF_6712 Iso 3 | 0.80 | 0.55 | 1.00 |
VVL_4401 | 0.84 | 0.61 | 1.00 |
AUC values were obtained from bootstrap method (n = 1000), 95% confidence interval (CI). |
To improve the performance of AFP to differentiate between cirrhosis and HCC stages, researchers have explored two-biomarker models. A Bootstrap test was applied to the response of each candidate structure + AFP to calculate the combined AUC values of the two-biomarkers. The results showed an increase in sensitivity and specificity for all combinations “biomarker candidate + AFP,” so that all combinations presented AUC values higher than AFP alone (0.85) and higher than any of the single candidates (0.84), Table 2. Six combinations showed the most significant increase in the AUC values with respect to AFP alone: AFP + NLF_5613 Isomer 1 (0.95), AFP + NLF_6712 Isomer 3 (0.94), AFP + NLF_5613 Isomer 2 (0.92), AFP + MVS_4512 (0.92), MVS_5613 (0.92), and NLF_6712 Isomer 1 (0.91). The correlation of all biomarker candidates was evaluated using the pairwise Pearson’s correlation (Supplementary Table S5), but even combinations of two candidates with coefficients of 0.92 and 0.89 (NLF_5602 Isomer 2 + NLF_5603 Isomer 2 and NLF_5612 Isomer 2 + NLF_5613 Isomer 1 respectively) showed considerably lower AUC values than all “biomarker candidate + AFP” curves. Regarding the sialylated biomarker candidates, it is noteworthy that these structures cannot be combined with AFP; while sialylated glycopeptides are high abundance in cirrhosis samples AFP is not, and vice versa. Therefore, these correlation differences along with the high standard deviations among samples make the sialylated glycopeptides non-viable biomarker candidates.
The sample heterogeneity of the HCC cohort allowed us to evaluated the data considering a subdivision based on early-HCC TNM stage 1, and seven samples for each cohort were used to calculate all statistical tests, Table 2 and Supplementary Table S5. Although this is a small sample number, such outstanding results must be validated. Five N-glycopeptides showed statistically significant differences in abundance between both cohorts, and all of these were observed in the evaluation of the whole data: one N-glycopeptide in the site Asn184, MVS + 5-6-1-3; three N-glycopeptides in the site Asn207, NLF + (5-6-1-2 Iso 2, 5-6-1-3 Iso 1, and 6-7-1-2 Iso 3); and one N-glycopeptide, VVL + 4-4-0-1, for the site Asn241. The area under the curve (AUC) values obtained for the single biomarker candidates and AFP were compared, Table 2. As was expected, the AUC value for AFP decreased considerably to 0.66. Without the high values reported in TNM stages 3 and 4, the selectivity and specificity of this glycoprotein decreased its effectiveness to detect HCC in early stages, Table 2. Conversely, the glycopeptides with significant changes in abundance between the two cohorts “biomarker candidates” showed substantially larger AUC values than AFP alone: the glycopeptide MVS_56123 (0.86), the glycopeptide NLF_5612 Isomer 2 (0.84), the glycopeptide NLF_5613 Isomer 1 (0.88), the glycopeptide NLF_6712 Isomer 3 (0.80), and the glycopeptide VVL_4401 (0.84), Table 2. Due to the low AUC value of AFP, the two-biomarker model “glycopeptide candidate + AFP” did not show significant increase in the AUC values of the combinations tested; rather, the AUC values observed in the ROC curves basically represented the signal of the glycopeptide candidates.