Potential Celiac Children: 9-Year Follow-up on a Gluten-containing Diet

Renata Auricchio MD, PhD; Antonella Tosco MD; Emanuela Piccolo MD; Martina Galatola PhD; Valentina Izzo PhD; Mariantonia Maglio PhD; Francesco Paparo PhD; Riccardo Troncone MD, PhD; Luigi Greco MD, PhD


Am J Gastroenterol. 2014;109(6):913-921. 

In This Article


Study Protocol

We prospectively enrolled patients with at least two tests positive for anti-TG2 antibodies, confirmed by EMA IgA antibodies, and with normal or slightly infiltrated small bowel mucosa (Marsh stages 0–1). All had HLA DQ2 or DQ8 haplotypes. Total serum IgA was within the normal range. Children with symptoms suggestive of CD immediately started a gluten-free diet. All the remaining asymptomatic patients were left on a gluten-containing diet (Figure 1) and were the prospective study cohort. Every 6 months, antibodies and clinical conditions were checked, and a small bowel biopsy was taken every 2 years, if the occurrence of symptoms did not require it earlier.

Figure 1.

Study cohort progression during 9 years of observation. GFD, gluten-free diet.

One hundred and six patients of the initial cohort have been reported previously at a 3-year follow-up.[12]


Anti-TG2 Antibodies and EMA. To measure serum anti-Tg2 antibodies, an enzyme-linked immunosorbent assay kit was used, based on a human recombinant antigen (Eu-tTg IgA, Eurospital, Trieste, Italy). The cutoff point for positivity was ≥7 IU.

Serum IgA EMA was measured by indirect immunofluorescence on 7-μm-thick frozen sections of human umbilical cord as the source of antigen. Samples were considered positive if a thin fluorescent network appeared around the smooth muscle fibers. Positive results were quantified further by titration.

Genotyping. HLA typing was performed using six single-nucleotide polymorphisms (SNPs) to identify DQ2.2, DQ2.5, DQ7, and DQ8 risk variants based on strong linkage disequilibrium at the HLA loci. SNPs rs2187668, rs2395182, rs4713586, rs7775228, rs4639334, and rs7454108 were genotyped with TaqMan assays (Applied Biosystems, Foster City, CA) on a 7800HT Fast Real-Time PCR system (Applied Biosystems). HLA-DQ risk types were predicted using the method described by Monsuur et al.[14] and validated in several populations of European origin.

All markers had a genotyping success rate >99% and did not deviate (P>0.05) from the Hardy–Weinberg equilibrium, with the exception of SNP rs4713586, which failed to produce clear allele clusters in the TaqMan assay and was therefore excluded from further analyses. The genotype at the rs4713586 locus allows for discrimination between DQ2.2 and the rare DQ4 type and is therefore needed to predict DQ2 risk carriage in DQ7/DQ2.2 individuals.

Thirteen SNPs for non-HLA genes (rs2816316, rs917997, rs6441961, rs17810546, rs9811792, rs1464510, rs6822844, rs2327832, rs1738074, rs3184504, rs842647) were genotyped using the TaqMan technology (Applied Biosystems) as described elsewhere.[15,16]

All reactions were performed on a 7900HT Fast Real-Time PCR System (Applied Biosystems) in a volume of 20 μl, containing SNP master mix, TaqMan reagents, and about 50 ng of genomic DNA template. Each plate also contained genotyping controls.

Duodenal Biopsy and Immunohistochemical Analysis. In each patient, esophagogastroduodenoscopy with five biopsies of the bulb and distal duodenum was carried out. According to our protocol, four of five fragments were fixed in 10% formalin, embedded in paraffin, and then stained with hematoxylin. The histological and morphometrical analysis by light microscopy was performed by two experienced pathologists. A villous height:crypt depth ratio ≥2 was considered normal.[11] The evaluation of these four fragments was made blinded to any serology results.

The fifth duodenal biopsy was put in an optimal cutting temperature compound (Killik, Bio-Optica, Milan, Italy), stored at −80 °C, and used for immunohistochemical staining for CD3+, TCR γδ+, and CD25+ cells, as previously reported.[17] The number of stained cells per millimeter of epithelium determined the density of cells expressing CD3 and TCRγδ in the intraepithelial compartment. Cutoff values for CD3+ and TCR γδ+ cells were 34 mm and 3.4 mm per epithelium, respectively. On the other hand, the number of cells expressing CD25 in the lamina propria was evaluated within a total area of 1 mm2. The usual cutoff value for CD25+ cells is 4 mm2 lamina propria. To determine the cutoff values to be used, 100 children with untreated CD and 50 non-CD control children were studied. Percentiles were obtained using the SPSS software (IBM, Chicago, IL). Cutoff values represented the 90th percentile of non-CD patients.

Intestinal Deposits of Anti-TG2 IgA Antibodies. Duodenal biopsies from all patients were also investigated for the presence of extracellular deposits of anti-TG2 IgA antibodies as previously described.[18] The evaluation of the deposits, performed considering the pattern and the intensity of the staining, was graded semiquantitatively as follows: negative (1), very weak (2), weak with patchy distribution (3), strong with patchy distribution (4), and strong with a homogenous distribution (5).

Ethics Statement. The Study protocol was approved by the Ethics Committee of the University of Naples Federico II.

Statistics. From our previous work, we estimated that 30% of potential CD patients would develop full-blown disease over 3 years of follow-up.[12] To evaluate factors that might affect the outcome, with a minimum expected difference in the outcome of 10%, with a 99% power, and a 0.05 first-degree error, we would require 122 cases. As we estimated 40% censored data up to 5 years (incomplete follow-up), we added 49 cases to the desired sample size.

We compared percentages by the χ 2-test. First-degree error was set to 0.05. Continuous variables were screened for normality before parametric analysis. Antibody concentrations were converted by decimal logarithm when required.

Survival analysis was performed to observe the occurrence of an event (relapse) over long periods of time within a starting cohort of individuals. The simple computation, with no survival analysis, of events×100/total population produces an accurate estimate of the occurrence of events only if the starting cohort is entirely followed up until the end of the observation period. As this is not the case in population studies, the survival analysis was performed to account for the real-life scenario of different follow-up times in different individuals. Actually, survival analysis is efficient to make the best use of the shortest and longest times of follow-up, without producing the disastrous bias of excluding cases who did not reach the final time point.

Hence, survival was based on the full use of all available, unselected information: at each time point (5 years, 6 years, and so on), we computed the total number of individuals contributing to the analysis in that single time period, and we added as 'censored' those who stopped in that time interval. Then, the number of events in that time interval was computed and compared with the cohort at that time point. It is obvious that the precision of our survival estimate decreased with increasing time and decreasing size of the cohort followed up. Therefore, the confidence intervals also increased with the follow-up time. However, this system of survival estimation is quite robust and gives the best possible estimate when the cohort at the last time point is not reduced drastically, independently of how many entered the follow-up at first.

Kaplan–Meier tables were used for the survival analysis. The Wilcoxon (Gehan) statistic was used to compare risk factors for survival. Stepwise canonical discriminant analysis was adopted to select variables that discriminated between the cases who progressed to flat mucosa and those who remained potential celiacs. Wilks' lambda was used to estimate the capacity of each variable to discriminate between the two groups. It ranges between 0 and 1, where 1=complete overlap and 0=complete separation. The stepwise multivariate procedure selects the first variable that minimizes Wilks' lambda, and then includes the subsequent variables progressively, according to their contribution to lowering Wilks' lambda. The variance ratio 'F' provides an estimate of each variable's contribution to the discrimination between groups.