Alcohol and Alcoholism Advance Access published July 30, 2008

Transcription

1 Alcohol and Alcoholism Advance Access published July 30, 2008 Alcohol & Alcoholism Vol., No. 0, pp. 1 8, 2008 doi: /alcalc/agn058 Automated Measurement of Carbohydrate-Deficient Transferrin Using the Bio-Rad %CDT by the HPLC Test on a Variant TM HPLC System: Evaluation and Comparison with Other Routine Procedures François Schellenberg 1,, Louise Mennetrey 2, Catherine Girre 3, Bertrand Nalpas 4 and Jean Christophe Pagès 1 1 Laboratoire de Biochimie et Biologie Moléculaire, Pôle de Biologie, Hôpital Trousseau, CHRU de Tours, France, 2 CCAA, CHRU, Tours, France, 3 Addictology Unit, Hôpital Fernand Widal, Paris, France and 4 INSERM U567, Hôpital Necker, Paris, France Corresponding author: Laboratoire de Biochimie et Biologie Moléculaire, Pôle de Biologie, Hôpital Trouseau, Tours, France. Tel: ; Fax: ; f.schellenberg@chu-tours.fr (Received 09 November 2007; first review notified 03 January 2008; in revised form 22 May 2008; accepted 24 June 2008) Abstract Aims: In this study, we evaluated the new %CDT by the HPLC method (Bio-Rad, Germany) on a Variant TM HPLC system (Bio-Rad), checked the correlation with well-known methods and calculated the diagnostic value of the test. Methods: Intra-run and day-to-day precision values were calculated for samples with extreme serum transferrin concentrations, high trisialotransferrin and interfering conditions (haemolysed, lactescent and icteric samples). The method was compared with two routine procedures, the %CDT TIA (Bio-Rad, Hercules, CA, USA) and the Capillarys TM CDT (Sebia, France). A total of 350 clinical sera samples were used for a case-control study. Results: Precision values were better in high CDT and medium CDT pools than in low CDT pools. The serum transferrin concentration had no effect on CDT measurement, except in samples with serum transferrin <1 g/l. Haemolysis was the only interfering situation. The method showed high correlation (r 2 > 0.95) with the two other methods (%CDT TIA and CZE %CDT). The global predictive value of the test was >0.90 at 1.9% cut-off. Conclusions: These results demonstrate that the %CDT by the HPLC test is suitable for CDT routine measurement; the results from the high-throughput Variant TM system are well correlated with other methods and are of high diagnostic value. INTRODUCTION Among the biological markers of alcohol abuse investigated over the past decades, CDT (carbohydrate-deficient transferrin) (Stibler, 1991) is considered the most accurate marker for the detection of alcohol abuse and possible relapse (Stibler et al., 1991; Helander 2003; Bortolotti et al., 2006). The main transferrin (Tf) glycoform comprises one polypeptide chain bearing two bi-antennary N-linked oligosaccharide chains, and each antenna is terminated with a sialic acid residue (i.e. four sialic acid residues). The overall number of sialic acid residues (Hatton et al., 1982) is theoretically between 0 (no oligosaccharide chain) and 8 (two tetra-antennae oligosaccharide chains terminated with four sialic acid residues on each). The Tf structure is modified following an alcohol-induced mechanism, resulting in the synthesis of Tf molecules lacking one (disialotransferrin) or both (asialotransferrin) glycans. As disialotransferrin (DiST) is the main modified glycoform (Stibler et al., 1979) and the increase of asialotransferrin (AST) can be observed only in samples with considerable amount of DiST, it was proposed as the target molecule for CDT standardization and calibration (Jeppsson et al., 2007). Over recent years, the use of CDT as a routine test for alcohol abuse has been hampered by laborious and time-consuming methods (Bortolotti et al., 2006). The methods based on microcolumn ion-exchange chromatography allow the global quantification of asialo-, monosialo-, disialo- and a variable amount of trisialotransferrin. The final result is expressed as the ratio between CDT and total Tf. More recently, new methods based on capillary zone electrophoresis (CZE) (Bortolotti et al., 2007; Schellenberg et al., 2007), high-performance liquid chromatography (HPLC) (Jeppsson et al., 1993, Helander and Bergström, 2006) and direct immunonephelometric measurement (Delanghe et al., 2007) have been developed that allow reliable and automated measurement of CDT. HPLC methodology allows the quantitative spectrometric determination of each Tf glycoform and has been proposed as a candidate reference method (Helander et al., 2003). The aim of this study was to evaluate the analytical features and diagnostic value of the commercial kit %CDT by HPLC on the routine high-throughput Variant TM HPLC system (Bio-Rad, Munich, Germany). MATERIALS AND METHODS Analytical procedures %CDT by the HPLC technique. The Bio-Rad %CDT by HPLC was run on a Variant TM HPLC system. This system comprises a two dual-piston pump system, a thermostated (20 C) 100-sample auto sampler and a column compartment (35 C) and a dual wavelength (460 nm and 690 nm) detector. The system is connected to the CDM software (Bio Rad). Before analysis, the samples are pre-treated by 30-min incubation with a ferrous and dextran sulphate solution, followed by 10-min (10,000 g) centrifugation, for both iron saturation and lipid precipitation. The supernatant is then ready for chromatographic analysis. The Tf fractions are separated using an ion-exchange column (600 tests) protected by a guard column (100 tests). The absorbance of the ferrous iron Tf complex is measured at 460 nm; the secondary 690 nm wavelength is used for background noise reduction. The system calculates the area of each Tf glycoform peak following baseline integration from disialo- to pentasialotransferrin. The hexasialotransferrin peak is not integrated; asialotransferrin, when present, is integrated in a separate baseline mode (Fig. 1A and C). The area of each identified peak is expressed as the ratio of its surface to the total under-peak area. The total analysis time is 10 min. %CDT TIA technique. This assay is based on separation of the Tf glycoforms on micro anion exchange chromatography C The Author Published by Oxford University Press on behalf of the Medical Council on Alcohol. All rights reserved

2 2 Schellenberg et al. Fig. 1. Transferrin glycoform pattern obtained using the %CDT by the HPLC test on a Variant TM system. The Y-axis in arbitrary absorbance units. (A) Chromatogram of a homozygote transferrin-c control with a low CDT concentration (1.12%). (B) Chromatogram of a patient with suspected liver cirrhosis, low serum transferrin (1.23 g/l) and DST as a shoulder on the TriST peak, leading to erroneous integration (5.40%). (C) Chromatogram of an alcohol-abusing patient with elevated DST (15.60%) and AST (4.77%) levels. (D) Chromatogram of a serum with high (2 g/l) haemoglobin content. PST: pentasialotransferrin, TeST: tetrasialotransferrin, TriST: trisialotransferrin, DST: disialotransferrin, AST: asialotransferrin. columns followed by nephelometric, or turbidimetric, measurement of the Tf in each eluate, allowing CDT quantification (Schwarz et al., 2003). The Tf glycoforms contained in the CDT eluate are mainly di-, mono- and asialotransferrin, and a minor part of trisialotransferrin (Aldén et al., 2005). In this study, nephelometric determinations were run using an Immage 800 (Beckman Coulter, Brea, CA, USA). The cut-off limit was 2.60% according to the manufacturer s recommendations (Schwan et al., 2004). Capillarys TM CDT technique. This capillary zone electrophoresis (CZE) technique has been developed by SE- BIA (Evry, France) on the routine multicapillary system Capillarys TM (Bortolotti et al., 2007). In this assay, the Tf glycoforms are separated by high voltage electrophoresis (8200 V) in an alkaline buffer (ph 8.8) according to ph and electroosmotic flow. The Tf glycoforms are detected by absorbance measurement at 200 nm wavelength. Samples Six pools of sera with low (1 and 2), medium, critical and high (1 and 2) CDT concentrations were prepared from anonymous routine samples, in which CDT had been previously measured by the routine method %CDT TIA (Bio-Rad). CDT values were <1.9% for the low pools, % for the normal pool, % for the critical pool and >5% and >9% for the high pools 1 and 2, respectively. A panel of 402 samples was analysed in this study: (a) 169 samples from patients with an AUDIT (Alcohol Use Disorders Identification Test) score >10, daily ethanol intake >50 g or

3 Evaluation of the New CDT by the HPLC Method 3 BAC (Blood Alcohol Concentration) >0, which constituted the alcohol abusers group; (b) 112 hospital patients who had no history or suspicion of alcohol abuse and no biological signs (excluding CDT levels) of alcohol abuse, and 69 patients with an AUDIT score <7 and a daily alcohol intake <40 g, constituting a control group (181 patients); (c) 52 samples, surplus of routine examinations, characterized by serum Tf concentration outside the alert values ( g/l) or trisialotransferrin >6% (according to CZE CDT determination). Thirty lipaemic samples were also included upon criteria of high triglyceride concentration, lactescent aspect and high lipemic index determined by a routine chemistry analyser Olympus 2700 (Olympus Corp., Japan). Thirty samples with various CDT concentrations were enriched with nonconjugated bilirubin to a final bilirubin concentration of µmol/l. Thirty samples with various CDT concentrations were enriched with haemoglobin by addition of haemolysed and centrifuged erythrocytes to a final haemoglobin concentration of 1 4 g/l. STUDY DESIGN Precision Intra-run precision was calculated from the repeated measurement of low[1], medium and high[1] pools. These samples were tested 10 times each day for three consecutive days. Intra-day precision was also calculated for each pool. The total precision was determined as described in the CLSI protocol EP5-A2 using low[2], critical and high[2] pools. Each pool was analysed in duplicate 5 days a week by two runs separated by at least 2 h for four consecutive weeks. Interfering situations A. Tf concentration. A panel of samples covering the range of Tf serum concentrations was selected among samples of the (b) and (c) panels to determine if there was a relationship between total Tf concentration and CDT values. The samples were selected according to the following criteria: serum Tf concentrations g/l in 0.1 g/l increments; CDT was assumed as having no reason to be increased, based on a lack of suspected alcohol abuse, according to usual liver parameters (ALT, GGT) and normal CDT values when measured by the routine laboratory method. The relationship between Tf concentration and CDT values was assessed by measuring the slope of the curve CDT = f(tf). B. Serum aspect. Thirty lipaemic samples were analysed in duplicate before and after centrifugation (60,000 g, 30 min). Prior to CDT determination, all samples were submitted to the pre-treatment described by the manufacturer. Thirty samples with various CDT concentrations were enriched with non-conjugated bilirubin to a final bilirubin concentration of µmol/l. CDT levels were measured before and after bilirubin addition using the %CDT by the HPLC technique. Thirty samples with various CDT concentrations were enriched with haemoglobin and the interference with haemolysis was measured by the %CDT by the HPLC technique before and after haemoglobin addition. C. High trisialotransferrin. Thirty samples with a trisialotransferrin (TriST) level >6% of the total Tf, determined by the Capillarys CDT technique, were selected in the (b) and (c) panels to validate the accuracy of the present method in high TriST samples. Correlation study A total of 302 samples from panels (a), (b) and (c) covering the full range of CDT values were analysed in parallel with the %CDT by HPLC, the %CDT TIA and the Capillarys CDT techniques. Highly discrepant results were reanalysed. The methods were compared using the Pearson correlation coefficient and the Passing and Bablock method for regression. The linearity was assessed by the Cusum test for linearity, which tests the random (or not) distribution of the points on each side of the regression line. Case-control study A case-control study was performed to evaluate the diagnostic efficacy of the method and to validate the cut-off proposed by the manufacturer. Samples of panels (a) and (b) were analysed in this purpose. They were 350 participants, aged years, divided into 181 control subjects (115 males, 66 females) and 169 patients with alcohol abuse disorders (AUD) (136 males, 33 females). The participation of these patients and control individuals in the clinical study was approved by the ethical committee of Hôpital Fernand Widal (Paris, France). The %CDT by the HPLC procedure was compared to the %CDT TIA and Capillarys CDT procedures. Statistical calculations were performed using MedCalc 9.2 (MedCalc Software, Mariakerke, Belgium) and Analyse-it (Analyse-it Software, Leeds, UK) software. Pearson s correlation coefficient for non-gaussian distribution was used in the correlation studies. RESULTS Precision Intra-run precision (Table 1) was inversely correlated to the CDT concentrations of the sample. In the low[1] pool, individual measurements varied from simple to double between runs, resulting in daily coefficient of variation (CV) ranging from 9.01% to 19.01%. The precision of CDT determinations in samples with higher CDT concentrations was better, with CVs ranging from 7.84% to 11.74% in the medium pool and 1.43% to 2.90% in the high[1] pool. The high CDT concentration in this pool made possible the integration of the AST peak. The results presented in Table 1 are the sum AST + DiST. Excluding AST did not significantly increase the precision of the technique (data not shown). The EP5-A protocol was applied for day-to-day precision evaluation. In this protocol, within-run, between-run and between-day are not considered as independent factors and consequently a poor within-run precision, as observed in low[2] pool, did not allow the calculation of between-run and betweenday precision (Table 2). We made the same calculations using the EP Evaluator TM software and obtained similar results. Consequently, we calculated separately the within-run precision using the mean of each duplicate as individual value; the CVs ranged from 16.7% (low[2] pool) to 3.60% (high[2] pool excluding AST). Similarly, we used the mean of results

4 4 Schellenberg et al. Table 1. Precision of CDT measurement with the %CDT by the HPLC method intrarun precision Low [1] Medium High [1] day 1 Mean (extreme values) 1.13 ( ) 1.56 ( ) ( ) CV 17.13% 7.84% 1.43% day 2 Mean (extreme values) 1.20 ( ) 1.66 ( ) ( ) CV 19.01% 11.74% 2.80% day 3 Mean (extreme values) 1.12 ( ) 1.65 ( ) ( ) CV 9.01% 8.00% 2.90% Intra-run precision scores for low, medium and high CDT (DiST + AST when detected) values measured by the %CDT by HPLC kit. Intra-run precision was calculated by consecutively analysing 10 aliquots of each pool on three consecutive days. Table 2. Precision of the %CDT by the HPLC test evaluated by the EP5-A2 protocol Within run Between day Total Pool n Mean SD CV(%) SD CV(%) SD CV(%) Low [2] Critical High [2] Disialo High [2] Di- + Asialo Precision scores of the %CDT by the HPLC test according to EP5-5A protocol. Three pools (Low[2], Critical and High[2]) were analysed in duplicate in two separate runs on 5 days per week for 4 weeks. Between-run precision could not be evaluated by the protocol. The missing between-day precision value is due to the high within-run CV for this pool. from each day as individual value to calculate the day-to-day precision. The CVs ranged from 7.18% (low[2] pool) to 2.96% (high[2] pool excluding AST). As observed previously in the intra-run precision study, the coefficients of variation were inversely correlated to CDT concentration in the samples, and the total imprecision according to the EP-5A protocol spread out from 16.7% (low[2] pool) to 5.1% (high[2] pool). In the high[2] pool, AST was not quantifiable in all samples. The exclusion of AST in this pool, as previously recommended (Jeppsson et al., 2007), improved the precision of the technique, resulting in a total imprecision of 4.3%. Interfering situations Peak detection. Technically, the limit of detection (LOD) has been set in the Variant s software at 500 area counts (AC). The limit of quantification (LOQ) results from both the limit of detection and the total Tf concentration. A normal Tf concentration (2.5 g/l) gives a total area of 140, ,000 (AC); consequently, the LOQ is closed to 0.4%. This value is significantly higher than that from the candidate reference method (Helander et al., 2003). This can be attributed to the lower sample volume of the %CDT by the HPLC test, 17 µl versus 28.6 µl of neat serum injected in the column and to a possible superior optical quality of the detection system used for the reference technique (Agilent 1100 LC). Tf concentration. The relationship between serum Tf concentration and CDT measurement was evaluated in 40 samples (Tf range g/l). CDT levels could not be quantified in eight of these samples; in seven samples (Tf range g/l), the DiST peak area was under the integration threshold, and the uneven baseline of one sample (Tf 1.65 g/l) did not allow DiST quantification. The seven samples with no quantifiable CDT had a low Tf concentration and an AC for the sum of all glycoforms below 100,000. Therefore, this value of the total area peak can be considered as a limit for CDT quantification. In the 32 other samples, all CDT values were <1.90%, no significant correlation was observed between Tf and CDT values (r 2 = 0.098), and the slope of the regression line was not significantly different from zero. Lipaemic, icteric and haemolysed samples. Among the 30 lipaemic samples analysed by the %CDT by the HPLC test before and after de-lipidation by ultracentrifugation, the results of untreated and de-lipidated samples were highly correlated (r 2 = 0.971). CDT values showed no difference as attested by the absence of a significant slope (0.944) of the regression equation and by a non-significant paired t-test (P = 0.445). %CDT in the analysed samples was from 0.7% to 11.9%; only two results could be considered as erroneous (difference >10% between both determinations) and were obtained from two samples with high concentrations of triglycerides and chylomicrons, with severely lactescent aspect. Addition of unconjugated bilirubin to non-lipaemic, nonhaemolysed samples had no interfering effect on the %CDT by the HPLC test. The results of supplemented and original samples were highly correlated (r 2 = 0.985), the regression equation showed no significant slope (0.967) and the paired t-test did not reveal any difference (P = 0.439). By contrast, addition of haemoglobin to normal samples resulted in a progressive increase in the baseline of the chromatogram (Fig. 1D), and did not allow a correct integration of the DiST peak for haemoglobin concentrations >2 g/l. In addition, mean values were slightly increased after haemoglobin addition (3.68% versus 3.30%, P = 0.049). High trisialotransferrin. Thirty samples with elevated TriST ( %) were analysed with the %CDT by the HPLC test. In 16 samples, DiST appeared as a shoulder on the TriST peak (Fig. 1B). A low Tf concentration (<1.5 g/l) in 13 of these samples was an additional factor that made quantification of the minor peak difficult, and 9 of these could not be properly

5 Evaluation of the New CDT by the HPLC Method 5 %CDT HPLC (HPLC - CZE) / Average % SD Mean SD AVERAGE of HPLC and CZE HPLC=0.989*CZE %CDT CZE A %CDT HPLC (HPLC - TIA) / Average % SD Mean SD AVERAGE of HPLC and TIA 8 4 HPLC=1.190*TIA B %CDT TIA Fig. 2. Scattergrams of the results with regression lines and 95% confidence intervals for CDT measured by %CDT by the HPLC test and either CZE (A) or TIA (B). The dotted lines correspond to y = x. The insets show the Bland Altman plots of the between-method percentage of the difference. quantified. Six samples were obtained from AUD patients and CDT results were increased in these patients by Capillarys CDT and %CDT by HPLC techniques. Correlation study A total of 302 samples were included in the study. Two were rejected because their chromatogram revealed a BC phenotype for Tf. The CDT concentrations of 17 samples were particularly low and could not be quantified by the %CDT by the HPLC method. The remaining 283 samples covered a wide range of Tf ( g/l) and %CDT ( %) concentrations. We used the paired t-test to compare the overall bias of the results obtained with the different methods. The t values were and 7.822, when the %CDT by the HPLC test was compared with %CDT TIA and Capillarys CDT methods, respectively. These differences were highly significant (P < ), indicating a gap between the results of the compared methods. Fig. 2 (A and B) shows comparisons of the %CDT by the HPLC test with the %CDT TIA and Capillarys CDT methods. The Pearson s correlation coefficients were [95% confidence interval (CI) ] for the %CDT TIA comparison and (95% CI ) for the Capillarys CDT comparison, indicating high concordance between the new %CDT by the HPLC test and the two other methods. Using the Passing and Bablock regression analysis, the slopes of the linear regression lines were calculated as (95% CI ) for the %CDT TIA comparison and (95% CI ) for the Capillarys CDT comparison. This indicated that the difference between %CDT by the HPLC test and %CDT TIA increases with CDT concentration, whereas the difference between %CDT by the HPLC test and the Capillarys CDT method remains approximately constant. The intercepts calculated by the same regression equation were (95% CI to 0.960) for %CDT TIA and (95% CI to 0.503) for Capillarys CDT, both being significantly different from zero. The linearity of the relationship between paired CDT measurements was checked using the Cusum test, which assesses the fit of the data to a linear model. As expected from the Bland and Altman diagrams in Fig. 2 (A and B), the test revealed a non-linear regression between %CDT by the HPLC test and %CDT TIA, and a linear regression between %CDT by the HPLC test and Capillarys CDT. Table 3. Mean CDT values according to sex and alcohol abuse %CDT n mean 95% CI Control subjects Men Women AUD patients Men Women Distribution of %CDT by HPLC values in the study subgroups. A significant difference (t-test P < 0.001) between men and women. All differences between control and AUD groups were significant for CDT (P < 0.001). Case control study Among the 350 subjects studied, 13 had a CDT concentration below the LOQ of the %CDT by the HPLC technique. Two control subjects had a Tf BC phenotype, and their results were not considered in the study. One sample had an uneven baseline that made DiST peak integration not possible. One additional sample from an AUD patient was rejected owing to an abnormally high TriST, resulting in one single peak for TriST and DiST (15.1%). For the remaining 333 subjects included, 165 control subjects (107 males, 58 females) and 168 AUD patients (136 males, 32 females), CDT was between 0.61% and 47.64%. There were no differences in age between the four subgroups of subjects, whereas there was a difference between mean CDT values for male and female subjects in both the control and the AUD patients groups (Table 3). CDT values ranged from 0.61% to 2.66% in the control group and from 0.65% to 47.64% in the AUD group. CDT results in this latter group showed a tailed distribution towards low values, which could be attributed to non-responding patients for CDT, or to an erroneous report of the time since the last alcohol consumption. When calculating the diagnostic values using the cut-off proposed by the manufacturer (1.70%), the sensitivity was 0.74 and the specificity was Another optimized cut-off at 2.0% was proposed by the statistical software from the results of the case-control study reported on the ROC curve (Fig. 3). We next calculated the global predictive value (GPV) of the test, which represents the fraction of subjects correctly classified by

6 6 Schellenberg et al. Sensitivity Specificity Fig. 3. ROC plot (bold line) representing the ability of the %CDT by the HPLC method to distinguish 165 control subjects from 168 AUD patients. The fine lines delimitate the 95% confidence interval of the ROC curve. The arrow indicates the optimum cut-off level calculated by the software (2.0%). Global Predictive Value % Prevalence 1.70% 1.80% 1.90% 2.00% Fig. 4. Global predictive value (GPV ) of the %CDT by HPLC assay in 165 control subjects and 168 AUD patients at 9 degrees of prevalence (10 90%) of AUD for four cut-off values ( %). The global predictive value is the probability for a subject included in the study to be correctly classified by the test as alcohol abuser or not. the test in the studied sample. We used four different cut-off values (1.7%, 1.8%, 1.9% and 2.0%) and nine levels of prevalence (10% to 90%), to check the possible consequences of low or high cut-off levels on the diagnostic value of the test. As indicated in Fig. 4, there is a linear negative relationship between prevalence and GPV, with different slopes according to the cut-off level. DISCUSSION The present study was aimed at evaluating the new %CDT by the HPLC method focusing on the analytical aspects and to compare its diagnostic value to two established methods. Our results suggest that the precision of the test is dependent on the signal, the peak area and thus on CDT concentration. The lack of precision of the %CDT by the HPLC test in low pools does not hamper the clinical interpretation of the results, for no individual result was above the cut-off (1.70%). On the other hand, in the medium pool, CDT was above the cut-off in 7 of the 30 determinations, leading to possible misinterpretation of the result. Optical characteristics of the system probably contribute to this imprecision, since a greater precision was observed (Helander and Bergström, 2006) using the same kit on a high performance HPLC system (Agilent 1100 LC). Our results confirm the better precision of CDT measurement when exclusively quantifying the DiST, the recently proposed (Jeppsson et al., 2007) target analyte. The absence of significant slope of the regression line between Tf and CDT clearly indicates that the total Tf concentration does not by itself introduce a bias in CDT measurement. Nevertheless, this technique does not allow CDT quantification in some samples with Tf <1 g/l while in the 333 samples with Tf >1.5 g/l, only 4 had DiST peak area <500 AC. So, when Tf is in the normal range, the number of samples with no technically relevant result is acceptable ( 1.5%). In 14 patients with high TriST and well-separated TriST and DiST peaks, the %CDT by the HPLC technique gave increased CDT values in 6 AUD patients and below the cut-off in the 8 remaining samples, while the %CDT TIA method gave high CDT in all samples. This discrepancy can probably be attributed to the partial co-elution of TriST in the CDT fraction in the %CDT TIA method (Aldén et al., 2005) that leads to false-positive CDT results in patients with high TriST. In this study, we confirm the findings of a recent study (Arndt et al., 2008) for atypical Tf glycoform distribution. Among the 16 samples with high TriST and poor TriST DiST separation (di-tribridge), 9 of them had low serum Tf (AC< 120,000) and retrospective investigation of their clinical history revealed a liver disease and a suspicion of cirrhosis for 7 out 9 patients. Therefore, high TriST is not by itself a source of inappropriate results, except in patients with low Tf and di-tribridging phenomenon. The high correlation between the %CDT by the HPLC test and the two other techniques is in agreement with previous results (Helander et al., 2001; Schwarz et al., 2003). We observed a linear correlation between the two methods based on Tf glycoforms pattern determination (%CDT by HPLC and Capillarys CDT), and a non-linear correlation of these techniques with the %CDT TIA method. This lack of linearity could be attributed either to a non-linear calibration of the immunoassay or to the incomplete separation between TriST and DiST in the %CDT TIA test. Figure 2 shows a lower dispersion of the results between the HPLC and CZE methods, as opposed to the HPLC and TIA methods. These differences can be attributed to, at least, three factors influencing the TIA method: (i) a poor precision level at low values (CDT < 1.8%), (ii) the partial co-elution of TriST in the CDT fraction and (iii) the cumulative imprecision of the immunoassays used for the separate quantification of Tf and CDT. By contrast, in the HPLC and CZE methods, CDT is a ratio between areas obtained in a unique glycoform separation. Independent of the correlation coefficients, the three compared methods give significantly different CDT results, as

7 Evaluation of the New CDT by the HPLC Method 7 shown by the calculated regression equations. Obviously, a common international standard is needed to obtain a linear correlation between the techniques and to allow comparison of results (Oberrauch et al., 2008). In the control group, we observed a moderate, but significant, difference in the CDT values among genders. This result appears in contradiction with a recent work (Bergström and Helander, 2008) that concluded to an absence of sex difference for CDT values. This might result from differences in the control group definition and from a possible increased alcohol intake in males as compared to women, within the limit of World Health Organization (WHO) recommendations assessed through a questionnaire. Concerning the diagnostic value of CDT measured by HPLC methods, two studies (Simonsson et al., 1996; Werle et al., 1997) reported higher specificity (>0.95), and one study (Werle et al., 1997) reported better sensitivity (0.80). It is, however, hard to determine the respective contributions of the technical features and of the population study selection criteria to explain these differences. To compare the analytical performance of a test, it is crucial to base the population selection on the sole WHO definition of alcohol abuse. The cut-off setting is also a major point as it controls the diagnostic values. The cut-off can be defined from the distribution of values in the control group or as the value that gives the best discrimination between cases and controls in the study population. In this case, the cut-off can be determined using an ROC curve analysis. In our study (Fig. 3), cut-off determination from the ROC curve suggested a value of 2.0%. Selection of this cut-off resulted in a sensitivity of 0.66 with a specificity of Because they are calculated from an established diagnosis, sensitivity and specificity values are nevertheless not highly informative for the physicians. Conversely, the global predictive value (GPV) measures the probability for a subject to be properly classified by the test, independent of his clinical status. With the exception of a test with equal sensitivity and specificity values, GPV fluctuates with the prevalence of the disease (AUD) in the population sample. As indicated in Fig. 4, the higher (2.0% and 1.9%) cut-off values yielded higher GPVs for prevalence values below 50%. Considering a population sample representative of the AUD prevalence in the general population (10%), GPVs were 0.93 for a cut-off at 2.0% or 0.92 for a cut-off at 1.9%. Although the higher cut-off value gives a slightly higher GPV, the heavy social and medical consequences of erroneous interpretations of false-positive CDT value led us to propose 1.9% as cut-off. With the same technique, Helander proposed a cut-off at 1.70% (Helander et al., 2006). Similarly, cut-off values ranging from 1% to 1.75% were selected using other HPLC techniques (Simonsson et al., 1996; Werle et al., 1997; Arndt and Keller, 2004; Helander et al., 2005). These discrepancies might be due to several methodological factors, including the basis for cut-off definition (CDT value distribution in the control group or discrimination between control and AUD groups) and the criteria for inclusion in the control group. The various cut-off values proposed in these studies, as well as in this study, should be considered as interim values in the absence of a value obtained from a study based on the determination of CDT using the candidate reference method (Jeppsson et al., 2007) applied to a population strictly selected upon the WHO definition of alcohol abuse. CONCLUSIONS The data from this study indicate a good validity of results obtained with the %CDT by the HPLC method on the Variant TM system. The precision level is acceptable for routine use and the results are not affected by lipaemic or icteric samples. High trisialotransferrin concentrations are not by themselves a cause of false-positive values. The serum Tf concentration does not affect CDT measurement for concentrations >1 g/l. Furthermore, the results correlate well with those obtained with other routine methods based on other analytical processes. This assay provides accurate clinical information, conjugated to the high throughput allowed by the Variant TM HPLC system. Acknowledgements The authors are grateful to Mrs Annie Détruit, Nadine Puche and Geneviève Wattelet for their technical contribution. They thank Bio-Rad for supporting this study by lending the Variant TM system and providing the reagents. They thank Sebia for providing kits for CZE determinations of CDT. The case-control study was partly supported by INSERM grant ATC Alcool no. ALC0205. REFERENCES Aldén A, Ohlson S, Påhlsson P et al. (2005) HPLC analysis of carbohydrate deficient transferrin isoforms isolated by the Axis-Shield %CDT method. Clin Chim Acta 356: Arndt T, Keller T. (2004) Forensic analysis of carbohydrate-deficient transferrin (CDT): implementation of a screening and confirmatory analysis concept is hampered by the lack of CDT isoform standards. Forensic Sci Int 146:9 16. Arndt T, Van Der Meijden BB, Wielder JPM. (2008) Atypical serum transferrin isoform distribution in liver cirrhosis studied by HPLC, capillary electrophoresis and transferrin genotyping. Clin Chim Acta 394:42 6. Bergström JP, Helander A. (2008) Influence of alcohol use, ethnicity, age, gender, BMI and smoking on the transferrin glycoform pattern: implication for use of carbohydrate-deficient transferrin (CDT) as alcohol biomarker. Clin Chim Acta 388: Bortolotti F, De Paoli G, Pascali JP et al. (2007) Fully automated analysis of carbohydrate-deficient transferrin (CDT) by using a multicapillary electrophoresis system. Clin Chim Acta 380:4 7. Bortolotti F, De Paoli G, Tagliaro F. (2006) Carbohydrate-deficient transferrin (CDT) as a marker of alcohol abuse: a critical review. J Chromatogr B 841: Delanghe JR, Helander A, Wielders JPM et al. (2007) Development and multicenter evaluation of the N Latex CDT direct immunonephelometric assay for serum carbohydrate-deficient transferrin. Clin Chem 53: Hatton MWC, Berry LR, Regoeczi E. (1982) The structural heterogeneity of the carbohydrate moiety of desialylated human transferrin. Can J Biochem 60: Helander A. (2003) Biological markers of alcohol abuse. JNeural Transm 66 (Suppl): Helander A, Bergström JP. (2006) Determination of carbohydratedeficient transferrin in human serum using the Bio-Rad %CDT by HPLC test. Clin Chim Acta 371: Helander A, Fors M, Zakrisson B. (2001) Study of Axis-Shield new %CDT immunoassay for quantification of carbohydrate-deficient transferrin (CDT) in serum. Alcohol Alcohol 36: Helander A, Husa A, Jeppsson JO. (2003) Improved HPLC method for carbohydrate-deficient transferrin in serum. Clin Chem 49: Helander A, Wielders JPM, Te Stroet R et al. (2005) Comparison of HPLC and capillary electrophoresis for confirmatory testing of the alcohol misuse marker carbohydrate-deficient transferrin. Clin Chem 51: Jeppsson JO, Arndt T, Schellenberg F et al. (2007) Toward standardization of carbohydrate-deficient transferrin (CDT) measurements: I. Analyte definition and proposal of a candidate reference method. Clin Chem Lab Med 45:

8 8 Schellenberg et al. Jeppsson JO, Kristensson H, Fimiani C. (1993) Carbohydrate-deficient transferrin quantified by HPLC to determine heavy consumption of alcohol. Clin Chem 39: Oberrauch W, Bergman AC, Helander A. (2008) HPLC and mass spectrometric characterization of a candidate reference material for the alcohol biomarker carbohydrate-deficient transferrin. Clin Chim Acta [Epub ahead of print]. Schellenberg F, Girre C, Nalpas B et al. (2007) Analytical evaluation of a new capillary electrophoresis method for carbohydrate-deficient transferrin. Clin Chim Acta 382: Schwan R, Loiseaux MN, Schellenberg F et al. (2004). Multicenter validation study of the %CDT TIA kit in alcohol abuse and alcohol dependence. Alcohol Clin Exp Res 28: Schwarz MJ, Domke I, Helander A et al. (2003) Multicentre evaluation of a new assay for determination of carbohydrate-deficient transferrin. Alcohol Alcohol 38: Simonsson P, Lindberg S, Alling C. (1996) Carbohydrate-deficient transferrin measured by high-performance liquid chromatography and CDTect TM immunoassay. Alcohol Alcohol 31: Stibler H. (1991) Carbohydrate-deficient transferrin in serum: a new marker of potentially harmful alcohol consumption reviewed. Clin Chem 37: Stibler H, Borg S, Allgulander C. (1979) Clinical significance of abnormal heterogeneity of transferrin in relation to alcohol consumption. Acta Med Scand 206: Stibler H, Borg S, Joustra M. (1991) A modified method for the assay of carbohydrate-deficient transferrin (CDT) in serum. Alcohol Alcohol 1 (Suppl 1): Werle E, Seitz GE, Kohl B et al. (1997) High-performance liquid chromatography improves diagnostic efficiency of carbohydratedeficient transferrin. Alcohol Alcohol 32:71 7.