Main

The development of genomics1, proteomics2,3 and metabonomics4 is transforming complex disorder biomarker discovery. Adding to this development, we present here the adaptation of our DNA sequencer/fragment analyzer–based N-glycan profiling tool5 into a platform for 'clinical glycomics'. We report that the total serum protein N-glycome yields an excellent biomarker for the detection of liver cirrhosis. To further clinical implementation, we developed protocols similar in complexity to typical DNA-fragment analysis diagnostic tests (all steps in 96-well plates).

Some considerations led us to test our technology on chronic liver disorders. First, the vast majority of the glycoproteins in serum are produced either by hepatocytes or by immunoglobulin-secreting plasma cells. Second, the asialoglycoprotein receptor and the mannose/N-acetylglucosamine (GlcNAc) receptor in the liver have important roles in the clearance of aberrantly glycosylated proteins6,7. Thus, changes in the serum N-glycome profile should mainly reflect changes in liver and/or B-lymphocyte physiology.

Currently, the diagnostic work-up of first-presentation patients with a chronic liver disorder requires a liver biopsy to assess fibrosis stage and necroinflammatory activity, and to detect cirrhosis8. In a large subgroup of these patients (especially those with chronic viral hepatitis), fibrosis progresses to cirrhosis with variable rates, which then can lead to severe complications9 and is a major risk factor for the development of hepatocellular carcinoma (HCC)10. Because liver biopsy is an uncomfortable and sometimes risky procedure11, it is unsuitable to incorporate it into the routine follow-up examinations of chronic liver disease patients. There is therefore a demand for serum markers that can routinely assess progression of liver fibrosis and reliably detect the stage of liver cirrhosis. Binary logistic regression models such as Fibrotest, which is based on a range of clinical chemistry analytes, have recently been intensively studied for these purposes12,13. However, these markers have low sensitivity at the >95% specificity levels required to obviate the need for biopsy in cirrhosis patients, or to reliably detect the onset of cirrhosis during follow-up. Additional serum markers with high specificity and good sensitivity are therefore required. In this study, we used Fibrotest to complement our new GlycoCirrhoTest, and show that the combination of both markers can achieve the desired >95% specificity with high sensitivity.

Results

Cirrhosis detection

A desialylated serum protein N-glycan profile (Fig. 1) was obtained for the 248 subjects under study (Supplementary Note online and Supplementary Table 1 online). We quantified 14 peaks that were detectable in all samples, and normalized their abundance to the total peak height intensity.

Figure 1: Examples of total serum protein N-glycan profiles.
figure 1

Top, malto-oligosaccharide reference. Middle, typical electropherogram of desialylated N-glycans derived from proteins in control serum sample. Nine peaks are clearly visible in the full detection range, with five more in the ×10 blow-up of the latter part of the electropherogram. Bottom, representative electropherogram obtained from cirrhosis case. Structures of N-glycans of relevance to this study are shown below the panels; peaks that are important for fibrosis/cirrhosis markers are boxed. Monosaccharide unit symbols (also valid for Fig. 5): ○, β-linked GlcNAc; ●, β-linked galactose; □, α-linked mannose; ▪, β-linked mannose; ▵, α-1,6-linked fucose.

Full size image

Five of the fourteen peaks in the serum N-glycome showed the desired trend of increasing or decreasing abundance in the following order: healthy blood donors </> noncirrhotic chronic liver disorder </> compensated cirrhosis </> decompensated cirrhosis (Supplementary Fig. 1 online).

Further analysis (Supplementary Note online) allowed us to summarize the relevant information in three variables (Fig. 2), each requiring the measurement of just two peaks in the profile (making final clinical use of the profiles easier). One variable in particular (log(peak 7/peak 8), renamed GlycoCirrhoTest) stood out: receiver operating characteristic (ROC) analysis14 indicated a classification efficiency of 85–90% for GlycoCirrhoTest, which is very similar to the Fibrotest marker (Fig. 3a). We used the cut-off values obtained from these ROC curves to classify the patients (Fig. 3b). The combination of our glycomics-based GlycoCirrhoTest and the Fibrotest binary logistic regression model yielded 100% specificity for the detection of compensated cirrhosis, with a sensitivity of 75% (18/24). The overall classification efficiency was 93% (76/82). Decompensated cirrhosis patients were also classified using the above calculated cut-off values (Fig. 3b), revealing 100% sensitivity.

Figure 2: Trends in derived diagnostic variables.
figure 2

Serum samples were classified in eight clinically relevant groups. Three diagnostic variables were derived from profiles in Figure 1: log(peak 1/peak 8), log(peak 2/peak 8) and log (peak 7/peak 8). The latter was renamed GlycoCirrhoTest. Ordinate scale is logarithmic. Error bars represent 95% confidence intervals for the mean.

Full size image
Figure 3: Classification efficiency.
figure 3

(a) ROC curve analysis to evaluate efficiency of GlycoCirrhoTest and Fibrotest in differentiating between group with fibrosis stages F0–F3 and group with compensated cirrhosis. AUC, area under the curve. (b) Two-dimensional scatterplots classifying cirrhosis sample group and precirrhotic chronic liver disease group, based on cut-off values obtained from ROC analysis. Top, compensated cirrhosis cases, with or without HCC, were detected with 75% sensitivity and 100% specificity. Bottom, decompensated cirrhosis cases, with or without HCC, were detected with 100% sensitivity. (c) Test of GlycoCirrhoTest cut-off for general population sample (Red Cross blood donors) and patients with nonliver autoimmune disease.

Full size image

In summary, our novel marker combination for cirrhosis detection did not yield any false positives among patients who were liver biopsy candidates, and the sensitivity of detection was 75% for compensated cirrhosis and 100% for decompensated cases.

Less advanced fibrosis stages

The other two serum N-glycome–derived variables, log(peak 1/peak 8) and especially log(peak 2/peak 8), rose gradually with increasing fibrosis stage15, from F0/F1 onward, whereas GlycoCirrhoTest stayed relatively stable from F0/F1 to F3, increasing only in cirrhosis (Fig. 4). The correlation between fibrosis stage and values for log(peak 2/peak 8) approaches linearity (ρ = 0.76 by Spearman rank correlation), making it a very promising marker for the follow-up of fibrosis. To establish this more firmly requires a longitudinal study with paired liver biopsies, which we are currently organizing. As expected, for the Fibrotest model we observed a sigmoidal increase with increasing fibrosis stage (binary logistic regression models regress on a 0–1 interval, with ensuing asymptotic behavior at both extremes). The transition in the sigmoidal curve has been reported at about F2 (ref. 12). We found that the transition occurred at about F3. The reason for this difference is unclear.

Figure 4: Values of some glycome markers gradually increase with fibrosis stage.
figure 4

Data for three glycome-derived markers were plotted against METAVIR fibrosis stage. Stage F4+ indicates decompensated cirrhosis. Horizontal lines in right two panels represent ROC-determined cut-off values for cirrhosis detection. GlycoCirrhoTest is relatively stable from F0/F1 to F3, increasing only from F4. Sigmoidal behavior of Fibrotest model was expected. Log(peak 2/peak 8) gradually increased with fibrosis stage.

Full size image

General population sample and other diseases

To gain a broader understanding of the characteristics of GlycoCirrhoTest, we classified a control group of Red Cross blood donors (general population sample, negative for HIV and hepatitis B and C) by the cut-off value optimized for cirrhosis detection. Only 2 of 60 (3%) donors scored positive (Fig. 3b). Regular alcohol consumption does not disqualify from blood donation and may go unnoticed, but it is the main cause of cirrhosis in regions with low incidence of hepatitis C, such as Flanders, Belgium, where the study was performed. Combined with the rather high median age of this group (61 years), 2–3% compensated cirrhosis may be expected16.

We studied a control group of 24 autoimmune disease patients because undergalactosylation of IgG is well documented in several of these diseases, especially rheumatoid arthritis17. IgG glycan modifications could be reflected in the glycan pattern of total serum glycoproteins, as IgG is the major non-liver-produced glycoprotein in serum (11 g/l). We micropurified and N-glycan-profiled immunoglobulins from the sera of these patients and of six healthy blood donors (Supplementary Note and Supplementary Table 2 online). As expected, a relatively strong undergalactosylation was detected in the autoimmune group, as reflected by the greater abundance of nongalactosylated core-α-1,6-fucosylated glycan (Fig. 1; on average 55% higher than in controls; P = 0.01). We also found a large increase in the level of substitution of the immunoglobulin N-glycans with a bisecting GlcNAc residue, which on average doubled the abundance of the glycan (Fig. 1; P < 0.0001). In most cases, however, these increases did not disturb the N-glycan pattern of total serum glycoproteins sufficiently to score positive on GlycoCirrhoTest (4/24 positive). Nevertheless, careful interpretation of GlycoCirrhoTest results in arthritis patients might be warranted.

To assess whether increased levels of serum carbohydrate-deficient transferrin (CDT)18 influence GlycoCirrhoTest, we obtained serum samples from chronic alcoholism patients with and without elevated CDT. The means for GlycoCirrhoTest in the CDT-positive and CDT-negative groups were not significantly different (Fig. 2; P > 0.1 by Student t-test).

Structural analysis of the differentially abundant N-glycans

Exoglycosidase sequencing of the serum N-glycome of three of the patients is shown in Figure 5 and interpreted in Supplementary Note online. The structural analysis shows that N-glycans that are either undergalactosylated or modified with a bisecting GlcNAc residue increase in liver cirrhosis, whereas fully galactosylated bi-and triantennary N-glycans decrease.

Figure 5: Partial structural analysis of differentially regulated N-glycans.
figure 5

Shown are results of exoglycosidase array sequencing on N-glycans derived from glycoproteins in three serum samples, which were chosen to reflect the quantitative range of the observed alterations. Leftmost sequencing column is representative for the control's profiles. Middle, relatively mild alteration that already trespasses the cut-off value for the cirrhosis-diagnostic variable described in the text. Right, one of the most affected samples. Peaks depicted in black do not carry a bisecting GlcNAc residue; peaks depicted in grey are all modified with a bisecting GlcNAc residue (also indicated by arrowheads). Reference panels under sequencing columns (middle and right) were assembled from different electropherograms, each resulting from a specific exoglycosidase digestion on reference glycans (see Supplementary Note online). See Figure 1 legend for definition of monosaccharide unit symbols.

Full size image

Further technology development: toward routine application

Using a standard heated-lid PCR thermocycler, we developed a procedure requiring only fluid addition or removal and dilution to produce ready-to-analyze, labeled N-glycans from serum in less than 8 h, with little hands-on time (see Methods). This procedure exploited the margin offered by the very high concentration of serum glycoproteins, as well as the small quantities of labeled glycans needed for our glycan analytical method (15 fmol are easily detectable). The protocol was tested on 20 serum samples that had been previously analyzed with our standard Immobilon-P plate–based sample preparation method. The values for GlycoCirrhoTest obtained using both methods were very strongly linearly correlated (Pearson's r = 0.993; Supplementary Fig. 2 online), indicating that the new method is a very good substitute. We have assembled the reagents in a kit form that allows the parallel preparation of 48 or 96 samples.

Capillary gel electrophoresis–based DNA analyzers are widely used in molecular pathology laboratories for diagnostic assays (such as tumor characterization and infectious disease diagnosis) involving DNA sequencing or high-resolution DNA fragment analysis. These analyzers, which are more automated and easier to use, are replacing the older gel-based instruments and are more suitable for clinical diagnostic laboratories. We therefore optimized the analysis of the desialylated total serum N-glycome on the ABI 310 single-capillary DNA analyzer. Only small modifications to standard DNA analysis procedures were required to obtain robust profiling in 18 min, with even higher resolution than that obtained on the gel-based ABI 377 (Supplementary Fig. 3 online). In combination with the optimized sample preparation, we achieved a 24-h turnaround time for 48 samples, with no more handling than needed for typical DNA fragment analysis. The use of multicapillary sequencers could easily increase throughput.

Discussion

We have developed clinical glycomics techniques suitable for direct implementation in well-equipped DNA diagnostic laboratories. Using this approach with total serum, we differentiated patients with cirrhosis from patients with noncirrhotic chronic liver disease, with a very high classification efficiency, by measuring just two easily detectable peaks (resulting in the GlycoCirrhoTest biomarker). Combining this new marker with the Fibrotest marker allowed detection of 75% of the compensated cirrhosis cases in the study, with 100% specificity. As always with new biomarkers, further and larger studies are necessary to fully validate this result. However, the extremely high specificity for cirrhosis indicates that liver biopsy could become truly unnecessary in patients that score positive for the GlycoCirrhoTest-Fibrotest marker combination. Moreover, the partial structural analysis of the differentially regulated glycans permits linking the observed profile changes to known aspects of cirrhosis.

Most importantly, we detected an increase in the modification of serum N-glycans with a bisecting GlcNAc residue. In normal liver, the enzyme responsible for this modification, N-acetylglucosaminyltransferase III (GnT-III), is expressed only in nonparenchymal liver cells, but in regenerating liver (modeled by two-thirds hepatectomy in rats), GnT-III is also expressed in hepatocytes19. GnT-III activity is also increased in the sera and liver tissue of cirrhosis and HCC patients20. Consequently, it is conceivable that GnT-III expression in hepatocytes, and thus GnT-III modification of serum glycoproteins, is a hallmark of the presence of the regenerative nodules that histologically define liver cirrhosis. This proposal is underpinned by our observation that GlycoCirrhoTest, which is mainly influenced by this increase in substitution with bisecting GlcNAc, is not increased in the earlier fibrosis stages, which lack these regenerative nodules. The progressively stronger undergalactosylation of the serum glycoproteins during the stages of liver fibrosis could result from a deficiency in glycoprotein clearance. A receptor for glycoproteins with GlcNAc- and mannose-terminated glycans is present on the liver sinusoidal endothelium (liver mannose receptor6), and this receptor is required in a mouse model for the clearance of GlcNAc-displaying proteins in serum7. Analogous to the observations with the asialoglycoprotein receptor21,22, we propose that the number of GlcNAc/mannose receptors is lower in cirrhotic liver, leading to the observed accumulation of glycoproteins with GlcNAc-terminated glycans. Experimental confirmation of this rationalization of our statistically derived diagnostic variable, in terms of known aspects of liver pathology, is warranted.

The log(peak 2/peak 8) glycome marker, which is dependent on both features described above (increased bisecting GlcNAc and undergalactosylation; note the structure corresponding to peak 2 in Fig. 1), correlates strongly with the fibrosis stage. Its value as a noninvasive marker of fibrosis progression will be further studied.

We believe that the highly specific and sensitive combination of liver cirrhosis markers introduced here could be of considerable importance in monitoring the progression of chronic liver disease. Its routine application would yield a warning signal for cirrhosis-related, life-threatening complications, and would help in focusing HCC screening efforts on cirrhosis patients, in whom HCC is most likely to develop. Moreover, these markers could help evaluate the efficacy of therapies for advanced fibrosis and cirrhosis, because they are much more convenient than repeated liver biopsies.

It will equally be of interest to apply the technology to other pathologies and other body fluids, or proteome fractions obtained thereof.

Methods

Serum samples and clinical diagnosis.

The clinical study was approved by the local ethical committee of Ghent University Hospital. Informed consent was obtained from all serum donors. A detailed characterization of the patients and the clinical diagnostic procedures that were used can be found in Supplementary Note online.

Serum protein N-glycome sample processing.

The N-glycans present on the proteins in 5 μl of the 248 serum samples were released after protein binding to an Immobilon P–lined 96-well plate, derivatized with 8-aminopyrene-1,3,6-trisulfonic acid, desialylated and analyzed on an ABI 377A DNA sequencer5 (Applied Biosystems). The optimized protocol for glycan release and labeling using a PCR thermocycler is as follows: 1 μl of 20 mM NH4Ac buffer (pH 7) containing 10% SDS was added to 5 μl serum in PCR tubes. The tubes were heated at 95 °C for 5 min in a standard PCR thermocycler with heated lid. After cooling, 1 μl of 10% NP-40 solution was added to neutralize the denaturing effect of SDS on peptide N-glycosidase F (PNGase F, Glyko). After the addition of 1 IUBMB mU of PNGase F, the tubes were closed and incubated in the thermocycler at 37 °C for 3 h. Subsequently, 8 μl of 50 mM NaAc (pH 5) and 2 μU Arthrobacter ureafaciens sialidase (Glyko) were added, and the tubes incubated in the thermocycler at 37 °C for 3 h. One microliter of the resulting solution was transferred to a new PCR tube and evaporated to dryness with the thermocycler open at 65 °C and the tubes open. The evaporation was complete within 5 min, after which 1.5 μl of the labeling solution5 was added to the bottom of the tubes. The tightly closed tubes were then heated at 90 °C for 1 h (the elevated temperature ensures fast reaction kinetics). Water (150 μl) was then added to each tube to stop the reaction and dilute the label to about 100 pmol/μl. The resulting solution was used for analysis on the ABI 377, as described above, or on an ABI 310 equipped with standard 47-cm ABI DNA analysis capillaries, according to the following specifications. As a separation matrix, we used a 1:3 dilution of the proprietary POP6 polymer in genetic analyzer buffer (all materials from Applied Biosystems). The injection mixtures were prepared by 1:25 dilution of the APTS-derived serum glycan solutions (see previous paragraph) in deionized formamide. Injection was for 5 s at 15 kV, followed by separation for 18 min at 15 kV and 30 °C. As an internal standard, the rhodamine-labeled Genescan 2,500 reference ladder (Applied Biosystems) was used in the dilution specified by the manufacturer. No alterations of the sequencer hardware or software (see below) were needed to perform these analyses, except for the coupling of an external cooling bath to the ABI 377 as described before5.

Data processing.

Data analysis was performed using the Genescan 3.1 software (Applied Biosystems). We quantified the heights of the 14 peaks that were detectable in all samples to obtain a numerical description of the profiles, and analyzed these data with SPSS 11.0 software. The assumption of normality of the variables over the studied populations was assessed using the Kolmogorov-Smirnov test at the 0.05 significance level. One-way analysis of variance was followed by Tukey Honestly Significant Difference tests. We used ROC curve analysis in SPSS 11.0 to assess the classification efficiency of the potential diagnostic variables. The curves in Figure 4 were obtained using the nonparametric Lowess regression.

Partial structural analysis of the N-glycan pool by exoglycosidase array sequencing.

Batches (1 μl) of APTS-labeled N-glycans, obtained according to the procedure described above, were subjected to digestion with different mixtures of exoglycosidases in 20 mM NaAc (pH 5.0). The enzymes used were Arthrobacter ureafaciens sialidase (2 U/ml; Glyko), Diplococcus pneumoniae β-1,4-galactosidase (1 U/ml; Boehringer Mannheim), jack bean β-N-acetylhexosaminidase (30 U/ml; Glyko) and bovine epididymis α-fucosidase (0.5 U/ml; Glyko). Unit definitions were as specified by the enzyme suppliers. After complete digestion, the samples were evaporated to dryness, reconstituted in 1 μl water and analyzed on an ABI 377 as described above.

Note: Supplementary information is available on the Nature Medicine website.