Background: Anti-tissue transglutaminase (tTG) assays that use human tTG as antigen have recently become available. We evaluated commercially available assays with human tTG antigen to estimate their diagnostic accuracies and to determine whether they agree sufficiently to be used interchangeably.
Methods: Ten commercially available second-generation anti-tTG assays were evaluated. The following populations were studied: celiac disease (CD) patients at the time of diagnosis without (n = 70) or with (n = 5) IgA deficiency; diseased controls (n = 70); and CD patients without (n = 28) or with (n = 2) IgA deficiency during follow-up. All individuals included in the study underwent intestinal biopsy. Technical performance (linearity, interference, precision, correlation, and agreement) and diagnostic accuracy (sensitivity and specificity) were compared. Anti-gliadin and anti-endomysium antibodies were also measured.
Results: IgA anti-tTG results correlated well overall, but numerical values differed. Diagnostic sensitivity ranged between 91% and 97% and specificity between 96% and 100%. These were higher than the sensitivity and specificity of the IgA endomysium assay and the IgA gliadin assay. Generally, IgG anti-tTG was less sensitive but more specific than IgG anti-gliadin for the diagnosis of CD in the small group of IgA-deficient patients.
Conclusions: Overall diagnostic performance of IgA tTG assays is acceptable and comparable among the different assays, but numerical values differ. Standardization is needed.
Celiac disease (CD)1 is a lifelong intolerance to gluten that occurs in genetically predisposed individuals. This immunologically mediated small-bowel enteropathy is widespread in Western countries and is often underdiagnosed(1). Recent studies have suggested that the prevalence of CD in Europe and North America may be as high as 1 in every 100–300 individuals(2)(3). Early diagnosis of the disease and treatment with a gluten-free diet may decrease the risk for complications, malignancies, and mortality(4).
Histologic demonstration of characteristic mucosal lesions on a small intestine biopsy specimen remains the gold standard for diagnosing CD(5). Serologic detection of anti-endomysium antibodies (EMAs) is widely used to support the diagnosis and to screen populations at risk(5)(6). Since the identification of tissue transglutaminase (tTG) as the major target antigen recognized by EMAs in 1997(7), ELISAs detecting the presence of IgA/IgG anti-tTG antibodies have been developed. The first-generation assays used guinea pig liver tTG as antigen. Although initial studies reported good performance of these assays(8)(9), several reports described both false-negative and false-positive results(10)(11)(12)(13)(14)(15)(16).
More recently, second-generation assays have become available and have been introduced in clinical laboratories. These assays use either recombinant human tTG (h-tTG) or purified h-tTG as antigen. The h-tTG-based ELISAs are superior to guinea pig liver-tTG-based assays for the detection of anti-tTG antibodies(10)(13)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36). Few studies, however, have compared the available h-tTG-based tests(10)(17)(25)(26)(27), and these studied only a limited number of the available tests [three methods in Ref. (27), four in Refs. (10)(25)(26), and six in Ref. (17)]. The reports also had limitations with respect to selection of patients and controls. For example, in some studies, positive and negative samples were selected based on the EMA results(10)(17)(26). This precluded calculation of diagnostic characteristics such as sensitivity and specificity of the anti-tTG assay or comparing its diagnostic accuracy with that of the EMA. Other studies had included only a limited number of diseased controls(10)(17), had no data on biopsy results for all diseased controls(10)(25)(26), or did not include IgA-deficient patients(10)(17)(26). Most studies were restricted to the study of either children or adults(10)(17)(25)(26)(27). Comprehensive information on the diagnostic characteristics of different commercially available second-generation anti-tTG assays is lacking, and it is not known whether the different assays yield comparable results.
The aim of this study was to evaluate 10 different commercially available second-generation (i.e., h-tTG-based) IgA/IgG tTG ELISA tests for the diagnosis and follow-up of CD in children and adults, using the results of internal biopsy as the criterion standard. Technical and diagnostic performance (sensitivity and specificity) were compared among the tests. The results of the tTG assays were also compared with results obtained with tests for IgA EMAs and IgA/IgG anti-gliadin antibodies (AGAs).
Materials and Methods
Patients who had undergone intestinal biopsy and serologic tests for CD during the period from 1997 to 2003 were retrospectively identified from the serology laboratory database. The main reasons for performing these tests were diarrhea, flatulence, weight loss, anemia, iron deficiency, anorexia, failure to thrive, and small stature.
The first group consisted of 75 consecutive patients with biopsy-confirmed CD and included 48 adults (13 males; age range, 16–77 years; median, 40 years) and 27 children (8 males; age range, 0–15 years; median, 5 years). All samples were diagnostic samples, which means that they were obtained at the time the first biopsy was performed and the diagnosis was established for each patient. All patients were consuming a typical diet at the time of blood sampling. Five of the 75 patients were IgA deficient. Histologic grading of the biopsy specimens was Marsh I for six patients, Marsh II for 18 patients, Marsh IIIa for 7 patients, Marsh IIIb for 10 patients, and Marsh IIIc for 34 patients.
The second group consisted of 30 patients on a gluten-free diet in whom the diagnosis of CD was well established. It contained 24 adults (9 males; age range, 16–66 years; median, 47 years) and 6 children (4 males; age range, 3–7 years; median, 5 years). Two of these patients were IgA deficient. In all of these patients, a small-bowel biopsy and a simultaneous blood sampling were performed. The time of investigation was between 3 months and 1 year after the start a gluten-free diet for 16 patients and was more than 1 year after the start a gluten-free diet for the other patients. One patient underwent two additional serum samplings without a simultaneous biopsy.
Diagnosis of CD was established according to the original European Society of Paediatric Gastroenterology and Nutrition (ESPGAN) criteria(5) for the pediatric population and according to the revised ESPGAN criteria(5) for the adult population.
As a control group, we used an appropriate spectrum of patients to whom the anti-tTG test would be applied in clinical practice. This diseased control group included 50 consecutive persons: 31 adults (12 males; age range, 16–80 years; median, 43 years) and 19 children (15 males; age range, 0–15 years; median, 7 years) who were investigated during the period from August to September 2003. In all of these persons, CD has been considered in the differential diagnosis. In addition, 20 patients with Crohn disease (10 males; age range, 23–56 years; median, 42 years) were included. All 70 diseased controls were examined by small-bowel biopsy and were found not to have histologic changes consistent with CD. The total IgA concentration was within reference values in this control population. Final diagnosis in the diseased control group was Crohn disease (20 patients), gastroesophageal reflux disease (13 patients), infectious disease (6 patients), irritable bowel syndrome (6 patients), functional dyspepsia (4 patients), gastritis (3 patients), psychosomatic complaints (3 patients), small-bowel bacterial overgrowth (3 patients), neurologic disease (3 patients), lactose intolerance (2 patients), ferriprive anemia attributable to loss of blood (2 patients), unexplained symptoms (2 patients), gastric ulcer (1 patient), fecal impaction (1 patient), and weight loss attributable to diabetes mellitus (1 patient).
In tertiary care centers in Belgium, the threshold to perform an upper endoscopy for patients with suspicion of upper gastrointestinal tract disorders or malabsorption is low. In cases of atypical symptoms and negative endoscopic (macroscopic) findings, duodenal biopsies are obtained almost routinely. This can explain why duodenal biopsies were often available for patients who ultimately had only rather “mild” diagnoses. On the contrary, if, for example, a gastric or esophageal tumor was found during endoscopy, duodenal biopsies were not taken. This explains why patients with severe diseases were lacking in the control group.
The study was approved by our hospital ethics committee.
IgA/IgG ema indirect immunofluorescence assay
EMAs were measured by indirect immunofluorescence using monkey esophagus tissue as substrate (Immco). A reticular pattern of immunofluorescence observed in the muscularis mucosae with a serum dilution ≥1:20 was reported as being positive. The intensity of this fluorescence was reported in a semiquantitative way (negative, 1+, 2+, or 3+). All slides were viewed by two independent observers who were blinded to the results of the biopsy.
IgA/IgG aga elisa
IgA/IgG AGAs were measured by commercial ELISA (The Binding Site). Results were expressed in arbitrary units with a reference interval of 0–20 arbitrary units.
IgA/IgG tTG elisa tests
IgA/IgG ELISAs were obtained from 10 different manufacturers: The Binding Site (Birmingham, UK), Biofons (Turku, Finland), D-tek (Mons, Belgium), Euroimmun (Lübeck, Germany), Eurospital (Trieste, Italy), Genesis (Cambridgeshire, UK), Immco (Buffalo, NY), Inova (San Diego, CA), Orgentec (Mainz, Germany), and Pharmacia (IgA only; Freiburg, Germany). All ELISAs were performed according to the manufacturers’ instructions. The technical details of the different tTG assays are summarized in Table 1⇓ . All assays were carried out on an automated ELISA instrument, the PhD. The manufacturers’ recommended cutoff values and the cutoff values as determined by ROC plots were used to calculate the diagnostic performance of each test.
Biopsy specimens were obtained from the second duodenal portion during gastroduodenoscopy. Histologic evaluating was performed according to the modified Marsh classification as described by Rostami et al.(37).
ROC plot analysis, Pearson correlations, and Altman–Bland analysis were performed with Analyze-It, Ver. 1.62 (Smart Software). The areas under the ROC curves were compared by use of the methodology of DeLong et al.(38). This method takes into account possible correlations between the areas under the ROC curves, created by the fact that the same persons underwent all tests.
technical performance of the IgA tTG ELISAs
We determined assay linearity by diluting a sample containing a high concentration of anti-tTG antibodies with increasing amounts (from 0% to 100%, in increments of 10%) of a sample that did not contain anti-tTG antibodies. Generally, the IgA tTG ELISAs from the different manufacturers showed good linearity (Table 1⇑ ), the best result being achieved by Euroimmun. The linearity for the Inova ELISA was not good at high values (>80 units).
We assessed the imprecision by testing 20 aliquots of two serum samples (one sample with a high tTG value and one with a low tTG value) in one run, as proposed by NCCLS guideline EP5-A(39). Within-run CVs varied between 4.6% and 21% (Table 1⇑ ). The best results were obtained with the Genesis and Eurospital tests and the worst results with the D-Tek test. The between-run imprecision could not be assessed because we performed only a limited number of runs.
Possible interfering substances that were investigated were hemoglobin, unconjugated bilirubin, and triglycerides. We evaluated interference by adding various concentrations of hemoglobin, bilirubin, and triglycerides to a pool of serum samples negative for IgA tTG antibodies as described previously(40). Final concentrations were 0–48.7 g/L for hemoglobin, 0–300 mg/L for bilirubin, and 0.25–6.07 g/L for triglycerides. Increasing amounts of the potential interfering substances had no clear effect on the IgA tTG results except for the Immco assay, in which hemoglobin and triglycerides produced a small positive interference (Fig. 1⇓ ).
The range of anti-tTG antibody titers and median anti-tTG titer in diagnostic CD samples varied substantially among the measurement methods (Table 1⇑ ).
Quantitative data distribution showed that diagnostic samples of CD patients mostly had tTG values near by or exceeding the upper detection limit: the proportion of results exceeding the upper limit were 73%, 81%, 70%, 31%, 0%, 71%, 59%, 0%, 81%, and 44% for the Biofons, Genesis, Orgentec, D-tek, Eurospital, Immco, The Binding Site, Inova, Euroimmun, and Pharmacia methods, respectively. Samples from patients without CD generally had tTG values below the cutoff values proposed by the respective manufacturers. We observed no significant difference in tTG concentration distribution between children and adults.
Using the Pearson correlation, we assessed linear correlation among the different assays. The correlation coefficients (r) for all possible comparisons are given in Table 2⇓ . The lowest correlation coefficients were observed when the different methods were compared with D-tek or with Inova. The correlation coefficient ranged between 0.43 and 0.75 for comparison with D-tek and between 0.58 and 0.89 for comparison with Inova. For all other comparisons (thus excluding D-tek and Inova), we observed a very good correlation (mean r = 0.92; median = 0.93; range, 0.80–0.98).
A high correlation does not mean that two methods have a high degree of agreement. We therefore performed Deming regressions and Altman–Bland analyses. The results are shown in Table 2⇑ . These analyses showed, overall, low agreement among assays. Additionally, Altman–Bland analysis, which was done by plotting the difference between observations against the mean, revealed that the difference between assays increased for increasing means (heteroscedasticity). Of all results, the best agreement was observed between the results obtained with the Eurospital and Pharmacia methods.
diagnostic performance of the tTG ELISAs
IgA tTG in patients with untreated CD without IgA deficiency.
Sensitivity and specificity were calculated for each IgA tTG method (Table 3⇓ ). The areas under the curves were 0.991, 0.993, 0.989, 0.986, 0.993, 0.980, 0.980, 0.997, 0.991, and 0.996 for the Biofons, Genesis, Orgentec, D-tek, Eurospital, Immco, The Binding Site, Inova, Euroimmun, and Pharmacia methods, respectively. The areas under the curves were not statistically significantly different among assays. On the basis of the ROC curves, optimum cutpoints (with the highest sum of sensitivity and specificity) differed among assays (Table 3⇓ ), with a range of <4 to >56 kilounits. Diagnostic sensitivities and specificities were calculated for each assay at these optimum cutoff values (Table 3⇓ ). Sensitivities and specificities were 91–100%. With reagents from two of the manufacturers (Inova and Genesis), only 2 samples (from the same patients) of 70 diagnostic samples were negative for the IgA tTG test; for these two samples, the IgA tTG results of the other manufacturers was also negative, as were the IgA EMA results. The IgA AGA results were positive in the two samples, and the IgG AGA results were positive in one of the two samples. The Marsh histologic classification was Marsh I in one patient and Marsh II in the other.
The diagnostic performance characteristics of the anti-tTG assays were compared with those of IgA EMA and IgA/IgG AGA assays. The IgA EMA test had a specificity of 100%, but the sensitivity was only 90%. Retesting of samples negative for IgA EMAs at a cutoff dilution of 1:2.5 did not increase this sensitivity. The IgA AGA test had a sensitivity of 87% and a specificity of 91%. The IgG AGA test had a sensitivity of 94% but a low specificity of 76%.
Comparison of the IgA EMA and the IgA tTG results showed that low EMA titers corresponded to low anti-tTG titers and high EMA titers to high anti-tTG titers. Repeated analysis of a sample with a high tTG titer and a negative EMA result revealed the presence of anti-smooth muscle antibodies, which masked the presence of EMAs in indirect immunofluorescence analysis.
In general, patients with more severe lesions on histologic examination (grading) had higher anti-tTG values than patients with less severe lesions (Table 4⇓ ). This was observed for all anti-tTG assays tested.
IgA tTG in CD patients without IgA deficiency during gluten-free diet.
IgA tTG antibodies were determined in four consecutive samples from one CD patient: at diagnosis, 3 months after the start of a gluten-free diet, during gluten challenge, and 1 year after restarting of the gluten-free diet. The results (Fig. 2⇓ ) show a decrease in the IgA tTG antibody titer after starting of a gluten-free diet and an increase after the reintroduction of gluten in the food. Similar results were obtained with all methods.
In addition, we tested 28 follow-up samples from patients who underwent a biopsy at the time of serum sampling. In 24 patients, the biopsy revealed persistent mucosal lesions, whereas in 4 patients the biopsy showed a normal small bowel histology. In the patient group in which the biopsy revealed persistent lesions, 67–79% and 67–88%, respectively, of these samples were positive for IgA tTG when we used the cutoff as proposed by the manufacturer or the cutoff based on the ROC curve. The EMA test, the IgA AGA test, and the IgG AGA test had concordances of 58%, 37%, and 83%, respectively.
In the four samples obtained from patients in whom the simultaneous biopsy showed a normal small bowel histology, the number of positive IgA tTG test results was 0 or 1, depending on the assay. All four of these samples were positive for IgG AGAs.
IgA tTG in IgA-deficient patients with untreated CD.
We measured IgG EMAs, IgG AGAs, and IgG tTG, using nine different assays, in five untreated IgA-deficient CD patients at the time of diagnosis. All five samples were positive for IgG AGAs (Table 5⇓ ), and two were positive for IgG EMAs. These two samples were also positive in all nine IgG tTG assays. Three samples were negative for IgG EMAs. One of these three samples was positive in eight of the nine IgG tTG assays, one sample was positive in four of the nine assays, and the third sample was positive in only two (Orgentec and D-tek) of the nine IgG tTG assays. Analysis of 69 diseased controls revealed that the assays from Orgentec and D-tek had the lowest specificities. For the other manufacturers, the specificity of the IgG tTG test was higher than the specificity of the IgG AGA test.
IgA tTG in IgA-deficient CD patients during gluten-free diet.
Follow-up samples from two IgA-deficient patients were analyzed, one from a patient in whom the simultaneously performed biopsy revealed a normal mucosa and one from a patient in whom the simultaneously performed biopsy revealed persistent mucosal lesions. The sample from the patient in whom the biopsy demonstrated persistent mucosal lesions was positive in all IgG tTG tests, the IgG EMA test, and the IgG AGA test. The sample from the patient in whom biopsy showed a normal mucosa was negative in the IgG AGA test and the IgG EMA test but positive in all IgG tTG tests. This suggested that the IgG AGA assay and the IgG EMA assay better reflected disease activity than the IgG tTG assay in IgA-deficient patients.
In this study, we evaluated 10 commercially available second-generation IgA/IgG tTG ELISAs and evaluated whether the results agreed. All methods use recombinant h-tTG as antigen, except for the Inova, which uses purified human antigen.
The strength of this study lies in the fact that the same patients and the same controls were tested in 10 different anti-tTG assays and in the anti-EMA and anti-AGA tests and that serum from each individuals originated from the same blood sampling. In addition, patient selection was based on final clinical diagnosis in which biopsy was included. Thus, selection of the samples was not based on results of another laboratory test, e.g., anti-EMA test. Controls were diseased controls for whom the possibility of CD was considered and for whom a biopsy was performed to exclude CD. All assays were performed on an automated ELISA instrument, excluding operator-dependent variation.
Generally, the IgA tTG ELISAs demonstrated good technical performance. Linearity was good, and interference by hemoglobin, bilirubin, and lipemia was absent or minimal. We found a high correlation among the methods from the different manufacturers, but agreement was low, which means that the result of one ELISA method cannot be replaced by the result of another. Thus, anti-tTG titers should not be used interchangeably. The international scientific community and the commercial private sector should undertake efforts to harmonize the assays.
Overall, the diagnostic performance of the 10 tTG assays evaluated was excellent, with only minor differences in sensitivity and specificity. The area under the ROC curve was high and not statistically different among the 10 assays evaluated, signifying that the diagnostic performance of the various second-generation tTG assays was comparable. For some assays, an adaptation in the cutoff value may be desirable. Overall, the diagnostic performance the IgA tTG assays was superior to the performance of the IgA EMA and IgA/IgG AGA tests. Therefore, a positive tTG test result does not need confirmation by EMA testing.
It has previously been suggested that EMA could recognize other endomysial antigens in addition to tTG(41)(42). In our study, however, the assays did not identify samples that were negative for IgA tTG and positive for IgA EMAs. By contrast, the assays detected five samples negative for IgA EMAs and positive for IgA tTG. This indicated that the second-generation tTG assays are more sensitive than the EMA assay. We observed a correlation between the IgA tTG titer and the EMA titer. Samples that were missed by the EMA test had low anti-tTG titers, except for one sample in which the EMAs were masked by anti-smooth muscle antibodies. We also found a correlation between the tTG value and the severity of the lesions on histologic grading.
In two CD patients without IgA deficiency, IgA tTG antibodies (with all methods) and IgA EMA results were negative. Follow-up data were available for one of these two patients and confirmed a negative anti-EMA result in two subsequent determinations (10 and 16 months after diagnosis). In these two samples, IgA AGA were positive. This indicated that, although more sensitive than EMA, tTG does not have 100% sensitivity. Such finding has been reported previously in the literature(29).
In our opinion, there is no need to test systematically for IgA AGAs in addition to IgA tTG. For example, we calculated that in a population of patients with diabetes mellitus type 1 with a CD prevalence of 5%, the risk for the presence of CD in an individual with a negative IgA tTG result and a positive IgA AGA results was 1.7% (Bayes theorem). This is too low to justify a biopsy in all of these patients. When a patient is clinically suspected as having CD, an intestinal biopsy should always be performed independently of the serologic results. If the IgA tTG test is negative and the biopsy is suggestive for CD, an additional determination of IgA AGAs should be considered. Thus, systematic IgA/IgG AGA pair testing is not a cost-effective approach and provides only minor, if any, help. One indication to perform an IgA AGA test, however, is in (diagnosing and) monitoring of histologically confirmed CD patients without IgA deficiency who were negative for IgA tTG at diagnosis.
In the follow-up of CD patients without IgA deficiency, the IgA tTG test showed a higher correlation with histologic findings than did the IgA EMA test and the IgA AGA test. However, seroconversion of a positive IgA tTG test cannot be considered as a marker for normalization of the mucosa. At present, the only way to demonstrate mucosal recovery is histologic examination. In all patients in whom the follow-up biopsy was normalized, the IgG AGA result remained positive. This confirms the suggestion of Volta et al.(43), who reported that persistence of IgG AGAs should be regarded as an immunologic memory.
Patients with selective IgA deficiency are at increased risk for CD. It is therefore important to identify these individuals (by measuring total IgA) and to measure IgG-class antibodies instead of IgA-class antibodies. We tested only a limited number of samples obtained from IgA-deficient CD patients. Nevertheless, we were able to demonstrate that in spite of the better specificity compared with the IgG AGA test, the sensitivity of the IgG tTG test was limited for diagnosing CD. All methods gave a positive test result for samples in which IgG EMAs were present. In one patient negative for IgG EMAs, eight of the nine methods gave positive IgG tTG results. This probably reflected the higher sensitivity of IgG tTG compared with IgG EMAs. In two patients negative for IgG EMAs, IgG tTG was negative in >50% of the methods. This suggests that some IgA-deficient patients do not have IgG tTG antibodies.
For monitoring CD in IgA-deficient patients, the IgG tTG test was less reliable than the IgG AGA test. The IgG tTG tests remained positive (for all assays tested) in a patient in whom the histologic lesions were resolved, whereas IgG AGAs were normalized. This confirmed the finding of Korponay-Szabo et al.(30), who reported that the decrease in IgG tTG antibodies in IgA-deficient CD patients was very slow. We conclude that the current assays for detection of IgG anti-tTG antibodies perform less well than the IgG AGA assays for diagnosis and follow-up of CD in IgA-deficient patients.
In conclusion, the second-generation human IgA tTG tests are highly reliable serologic assays for the diagnosis and follow-up of CD patients without IgA deficiency. The assay is quantitative and is not affected by subjectivity, as is the case with EMAs. Absolute values cannot be used interchangeably. Assays for IgG tTG are less accurate for the diagnosis and follow-up of CD in IgA-deficient patients.
We thank Biofons, Genesis, Orgentec, D-tek, Eurospital, Immco, The Binding Site, Inova, Euroimmun, and Pharmacia for the generous donation of their IgA/IgG tTG assays. We are indebted to Lieve Godefridis for expert technical assistance. Severine Vermeire is a postdoctoral fellow of the Fund for Scientific Research Flanders (FWO-Vlaanderen) Belgium. X.B. is a Senior Clinical Investigator of the Fund for Scientific Research Flanders (FWO-Vlaanderen) Belgium.
1 Within-run CV was assessed by measurement of 20 aliquots of two samples in one run. One sample contained 60 kilounits/L and the other 116 kilounits/L (Euroimmun).
2 AP, alkaline phosphatase; pNPP, p-nitrophenyl phosphate; HRP, horseradish peroxidase; TMB, 3,3′,5,5′tetramethylbenzidine.
1 Results exceeding the upper limit were excluded from the statistical analysis.
1 The units for the ranges and median values are units for the Biofons and Inova methods and kilounits/L for all other assays.
1 Results of the IgG tTG assays, the IgG EMA assay, and the IgG AGA assay for samples from five IgA-deficient CD patients at diagnosis.
2 ND, not determined.
↵1 Nonstandard abbreviations: CD, celiac disease; EMA, anti-endomysium antibody; tTG, tissue transglutaminase; h-tTG, human tTG; AGA, anti-gliadin antibody; and ESPGAN, European Society of Paediatric Gastroenterology and Nutrition.
- © 2004 The American Association for Clinical Chemistry