Secondary Logo

Journal Logo

Original Articles: Gastroenterology

Discriminant Score for Celiac Disease Based on Immunohistochemical Analysis of Duodenal Biopsies

Tosco, Antonella; Maglio, Mariantonia; Paparo, Francesco; Greco, Luigi; Troncone, Riccardo; Auricchio, Renata

Author Information
Journal of Pediatric Gastroenterology and Nutrition: May 2015 - Volume 60 - Issue 5 - p 621-625
doi: 10.1097/MPG.0000000000000675
  • Free


What Is Known/What Is New

What Is Known

  • Celiac disease (CD) represents a spectrum that includes cases with minor histological abnormalities (potential CD).
  • The count of intraepithelial lymphocytes (IELs) in an otherwise normal small bowel biopsy proved not to be sufficient to define diagnosis of CD.
  • Among the immunohistochemical markers, one of the best predictors of CD diagnosis is the increase in γδ IELs.
  • What Is New
  • An improved immunohistochemical approach, based on the count of CD3+, γδ + IELs, and lamina propria CD25+ cells, may be of help in identifying subjects with gluten-dependent enteropathy among those with normal mucosal architecture.

Celiac disease (CD) is characterized by a small intestinal mucosal lesion secondary to a T-cell–mediated reaction to gliadin in genetically susceptible individuals (1). Until the publication of the new guidelines for the diagnosis of CD in January 2012 (2), the criterion standard for the diagnosis was biopsy with conventional histology showing a clear villous atrophy with crypt hyperplasia. It is now, however, widely accepted that, from a histological point of view, CD represents a spectrum that can include minor mucosal abnormalities (3); consequently, the presence of villous atrophy cannot represent the only diagnostic criterion. In fact, according to the new guidelines, clinical gluten-dependent symptoms, serology, genetics (mainly human leukocyte antigen [HLA]), and histology, all contribute to the diagnosis of CD (2).

In this context, a special challenge is posed by the so-called patients who potentially have CD because one increasingly frequently finds patients with positive anti-tissue transglutaminase (anti-TG2) antibodies and/or anti-endomysial antibodies (EMA) but with normal villous architecture (Marsh 0 or 1) who will or will not develop CD (4). It then became necessary to have finer tools to correctly identify cases with real gluten-dependent enteropathy among patients with minimal histological abnormalities (5). The count of intraepithelial lymphocytes (IELs) in an otherwise normal small bowel biopsy proved not to be sufficient. When a Marsh 1 lesion is observed, only in approximately 10% of cases, it represents the initial presentation of CD (5,6). Similar results have been obtained in pediatric patients (7). The count of IELs at villous tip was reported to be more specific for CD (8,9). Among the immunohistochemical markers, one of the best predictors of CD diagnosis is the increase in γδ intraepithelial lymphocytes (10) but the specificity of this finding is not high (11). The detection of intestinal deposits of immunoglobulin A (IgA) anti-TG2 by immunofluorescence was reported to be the best marker to identify, among patients with potential CD, those who will eventually develop a gluten-dependent enteropathy (12–14).

The aim of this work is to evaluate the contribution that the combination of a few immunohistochemical markers in duodenal biopsies may provide for the correct identification of gluten-related minor enteropathy (Marsh 0 and 1).



Our study involved 112 children (59 girls, median age 6.3 years, range 0.5–17.3 years) who underwent a small intestinal biopsy for gastrointestinal symptoms, at the Department of Translational Medical Science, Pediatrics Section, University Federico II, Naples. Written informed consent was obtained from the parents of the children enrolled. The study protocol was approved by the ethical committee of the University Federico II. Fifty-six of the 112 patients were consecutively enrolled among those who presented with villous atrophy and received a diagnosis of CD: 37/56 were classified as Marsh 3b grade according to Marsh classification modified by Oberhuber, and 19/56 as Marsh 3c (3). All of the celiac children showed serum anti-TG2 autoantibody levels higher than cutoff and positive EMA.

In the same period, we enrolled, as nonceliac controls, 56/112 patients matched for age and sex; they were taken among children with no serum anti-TG2 antibodies and negative EMA who performed a gastroduodenoscopy at our department and received a diagnosis different from CD. This group included 11 patients with signs of epithelial infiltration (Marsh 1) and 45 with completely normal mucosa (Marsh 0). They received diagnoses of inflammatory bowel disease (IBD) (n = 8), esophagitis (n = 19), food allergy (n = 9), and functional abdominal pain (n = 20). A third group of 61 children, who underwent a small intestinal biopsy for gastrointestinal symptoms, with architecturally normal intestinal mucosa (Marsh 0 or Marsh 1) served as a validation cohort (38 girls, median age 7.4 years, range 1.2–17.5 years). These patients were retrospectively enrolled and the pathologists blindly analyzed their biopsies without knowledge of their medical history or serological status.

Anti-TG2 Antibodies and EMA

Serum anti-TG2 IgA levels were measured by enzyme-linked immunosorbent assay according to the manufacturer's instructions (Eurospital Kit Eu-tTg IgA, Trieste, Italy). Cutoff value was set up at 7.0 U/mL. Serum IgA EMA levels were determined by indirect immunofluorescence on 7-μm thick frozen sections of human umbilical cord as source of antigen. Samples were considered positive if a thin fluorescent network appeared around the smooth muscle fibers.

Duodenal Biopsy and Immunohistochemical Analysis

Four duodenal and 1 bulb biopsies were taken by gastroduodenoscopy from all of the patients. Three fragments were fixed in 10% formalin, included in paraffin, and then treated for histological and morphometrical analyses. Two experienced pathologists analyzed 4-μm thick paraffined hematoxylin-stained sections from the 3 fragments by light microscopy, and villous height crypt depth ratio (Vh/CrD) was determined. Marsh score was used. Biopsies were classified as Marsh 0 (normal), Marsh 1 (epithelial infiltration), Marsh 2 (crypt hyperplasia), and Marsh 3 (villous atrophy = Vh/CrD < 2) of increasing severity (a, b, and c) (3,15). This evaluation was made blind to any serology results. One of the duodenal specimens was embedded in an optimal cutting temperature (OCT) compound (Tissue-Tek, Sakura, Alphen aan den Rijn, the Netherlands), stored in liquid nitrogen, and used for immunohistochemical stainings for CD3+ and TCRγδ+ intraepithelial cells and CD25+ lamina propria cells as previously reported (16). Briefly, 4-μm cryostat sections were fixed in acetone for 10 minutes. After incubation with normal rabbit serum (1:100; Dako, Milan, Italy) for 20 minutes, sections were covered with anti-CD3 (1:200; Dako), anti-TCRγδ (1:80; Thermo Scientific Inc, Rockford, IL), or anti-CD25 (1:20; Exalpha Biological, Boston, MA) monoclonal antibodies for 1 hour, and then with rabbit anti-mouse immunoglobulins (1:25; Dako) for 30 minutes. As a negative control, mouse IgG2a/IgG1 (1:100; Dako) replaced the primary antibody. After washing with Tris pH 7.4, the sections were covered with monoclonal mouse peroxidaseantiperoxidase (1:100; Dako) or monoclonal mouse alkaline phosphatase antialkaline phosphatase (1:40; Sigma-Aldrich, Milan, Italy) for 30 minutes. 2-Amino-9-ethyl-carbazole (AEC) (Sigma-Aldrich) and new fuchsin were used as peroxidase and phosphatase alkaline substrate, respectively. Monoclonal antibodies were diluted in Tris pH 7.4. All of the incubations were conducted at room temperature in a humid chamber. Finally, sections were counterstained with Mayer hematoxylin and mounted with Aquamount (BDH, Poole, UK). The density of cells expressing CD3 and TCRγδ in the epithelial compartment was determined by counting the number of stained cells per millimeter epithelium; cutoff values for CD3+ and TCRγδ+ cells are 34 mm/epithelium and 3.4 mm/epithelium, respectively. The number of cells expressing CD25 in the lamina propria was evaluated within a total area of 1 mm2 of lamina propria using a microscope with a calibrated lens aligned parallel to the muscularis mucosae; the cutoff value for CD25+ cells is 4/ mm2 lamina propria. To determine the cutoff values, 100 children with untreated CD and 50 noncoeliac control children were studied. Percentiles were obtained using the SSPS software (IBM SPSS Statistics, Armonk, NY). Cutoff values represent the 90th percentile of the non-CD patients (16).

Statistical Analysis

Data with a Gaussian distribution were compared using the Student t test, whereas Pearson χ2 test was used for not-normal data. Stepwise canonical discriminant analysis was adopted to select variables that discriminate between the patients with CD and controls. The discriminant Wilk λ estimates the capacity of each variable to discriminate between the 2 groups; it ranges between 0 and 1, in which 1 = complete overlap and 0 = complete separation. The variables are multicorrelated: the stepwise multivariate procedure selects the first variable that minimizes Wilk λ, and then includes the following ones progressively, according to the contribution to lower the Wilk λ. The variance ratio ‘F’ provides an estimate of the contribution to discrimination of each variable. Level of significance was set at P < 0.01. The discriminant equation was obtained by a progressive analysis of the discriminant capacity of the single variables and then adding, one by one, the other variables. The contribution of each combination of variables was estimated by the percentage of cases/controls correctly predicted and classified by the equation into their original group. The final equation was obtained including at step 0 all of the variables obtained by the immunohistochemical analysis of each biopsy: intraepithelial CD3+ cells, γδ+ cells, γδ+/CD3+ ratio, and lamina propria CD25+ cells. The combination of all of the 4 parameters provided the best discrimination between groups. This equation is composed of the actual value of the variables observed (eg, count of CD3) multiplied by an unstandardized regression coefficient proposed by the best multivariate combination, to produce a discriminant score (Dscore), which is the measure of the distance between cases and controls (or other groups).

We performed a cross-validation of the discriminant equation obtained: each case is classified by the functions derived from all of the cases other than that case excluding progressively each case from the analysis and then applying the classification procedure. Data analysis was performed using SPSS software (version 21.0).


Patients With CD and Controls: Immunohistochemical Analysis

All of the immunohistochemical parameters were statistically different between patients with CD and controls, as shown in Table 1. As expected, the immunohistochemical parameters were correlated with each other (Table 2): the strongest correlation was found between the presence of high levels of γδ+ lymphocytes and an increased γδ+/CD3 ratio (Pearson coefficient = 0.895). The count of CD3+ cells was the immunohistochemical parameter better related with values of serum anti-TG2 autoantibodies (Pearson correlation coefficient = 0.538).

Immunohistochemical analysis in patients with CD and controls
Pearson correlation coefficient between immunohistochemical parameters in biopsies and anti-TG2 levels

Discriminant Score

Among the 112 cases of the study population, we attempted to identify those variables that help to discriminate between the patients with CD and controls. The count of CD3+ cells was able to identify correctly the patients of the original groups in 92.0% (103/112) of cases; this value progressively decreased for γδ+, CD25+, and γδ+/CD3+ considered individually, obtaining a correct classification in 90.2% (101/112), 88.4% (99/112), and 84.8% (95/112) of the patients, respectively. When we considered the combination of CD3+ and CD25+, 92% (103/112) of original grouped cases were correctly classified: (3.6% [2/56] of controls classified as patients with CD and 12.5% [7/56] of CD cases classified as controls). The combined analysis of CD3+ and γδ+ lymphocytes appeared to be more accurate: 94.6% (106/112) of patients were correctly classified. When all of the 4 parameters were considered, a discriminant equation was obtained as follows: Dscore = (CD3 × 0.06) − (γδ × 0.119) + (CD 25 × 0.012) + (γδ/CD3 × 0.131) − 4.709

By this equation, we obtained a correct predicted classification in the groups of CD or controls in 97.3% (109/112) of cases: indeed, only 1/56 patient with CD was misclassified as a control and only 2/56 controls were misclassified as patients with CD; the original diagnosis of these 2 misclassified patients was gastroesophageal reflux and IBD, respectively. The cross-validation confirmed high efficiency of the discriminant equation, obtaining a correct classification of 96.4% of cases, when the iterative exclusion procedure was applied. Figure 1 shows the probability to be celiac according to Dscore values.

Probability of being celiac predicted by the Dscore obtained by immunohistochemistry. Dscore = discriminant score.

Classification of Patients With Normal Mucosa and Unknown Diagnosis

We checked the cohort of 61 patients with normal intestinal mucosa and unknown diagnosis, in order to validate the model. The aim was to identify patients with potential CD (those with positive CD antibodies) among them. Only after the application of Dscore by the pathologists we revealed that 34 of these received a diagnosis of potential CD based on positive serology and compatible HLA, and 27 patients without positive CD serology received different diagnosis such as IBD (n = 5), gastroesophagitis (n = 7), food allergy (n = 3), and functional abdominal pain (n = 12). The results of the comparison of the single immunohistochemical parameters between patients with potential CD, patients with untreated CD, and controls are shown in Figure 2. Table 3 reported the sensitivity, specificity, positive predictive value, and negative predictive value of the different markers in our population.

Median and interquartile range of immunohistochemical parameters in patients with potential CD compared with celiacs and controls. CD = celiac disease.
Specificity, sensitivity, PPV, and NPV of different immunohistochemical parameters for CD diagnosis (untreated and potential)

When we applied the discriminant equation to their immunohistochemical data 96% (26/27) of those with normal anti-TG2 values were correctly classified as nonceliac. Looking at all of the patients with an elevated Dscore value 92.9% (13/14) turned out to be patients with potential CD. Sixty-two percent of patients with potential CD, however, were not recognized by the equation as they presented a normal Dscore (Fig. 3). The latter were more often patients with Marsh 0 score; in fact, when Marsh classification was considered, 11/20 (55.0%) of the patients with potential CD with a Marsh 1 lesion were correctly classified as celiacs versus only 2/14 (14.3%) of those reported to be Marsh 0.

Dscore blindly applied to patients with normal intestinal mucosa and unknown diagnosis: 27 patients with final diagnosis different from CD and 34 patients with final diagnosis of potential CD. CD = celiac disease; Dscore = discriminant score.

In subjects with normal mucosa, a positive score almost invariably identifies a subject with potential CD, but a normal score does not exclude this condition.


According to new ESPGHAN guidelines for the diagnosis of CD, small bowel biopsy may be omitted only in symptomatic patients who have high IgA anti-TG2 levels (10 times above upper limit of normal), verified by EMA positivity, and are HLADQ2 and/or HLA-DQ8 heterodimer positive. Histological evaluation remains necessary for all of the suspected cases that do not have these conditions. The same guidelines suggested additional supportive evidence including increased γδ+ cell count in the case of mild lesions; in fact, the immunohistochemical analysis of the biopsies remains still useful in cases with minor alterations and should be implemented. In this work, we analyzed the value of different immunohistochemical parameters when considered singularly or in combination. The number of lamina propria CD25+ is considered a marker of mucosal inflammation, but it is not specific for CD. The increased density of intraepithelial CD3+ cells in a mucosa otherwise normal renders mandatory a clinical workup for CD (17,18), even if only in 9% of the cases of a large population it proved to be associated with a definitive diagnosis of CD (7). In fact, Guz-Mark et al (19) have shown that slightly elevated duodenal IEL counts are common in the pediatric population even in children with no other identified cause that may have contributed to duodenal lymphocytosis (eg, food allergies, IBD, recognized infection with Giardia, autoimmune diseases). Finally, as far as IEL γδ+ cells are concerned also in our cohort, their increased density remains the best marker of a gluten-dependent enteropathy.

As expected, the analysis of the single markers is significantly different between Marsh 3 patients with untreated CD and controls. Moreover, patients with potential CD presented median values significantly different from patients with untreated CD and controls, with the exception of lamina propria CD25+ cells whose density was not different between patients with potential CD and controls. We attempted a step forward by combining the various parameters into a multivariate model. This analysis proved to be effective in discriminating nonceliac patients and patients with overt CD. Applying this analysis to the population of unknown patients with normal villous architecture, 93% of those with positive score were correctly classified as potential celiac. Twenty-one of the 34, however, who had a positive serology, in >1 occasion, were not predicted as CD by this score. More often, they presented a Marsh 0 score. It would be interesting to evaluate the predictive value of this analysis toward the development of a frank damage of the intestinal mucosa. In any case, only 1 non-CD patient was diagnosed as celiac based on the positive score, confirming the high specificity of the immunohistochemical analysis and its determinant contribution to the diagnosis of CD. One of the drawbacks of this approach is represented by the need (for the count of intraepithelial γδ+ cells) of frozen bioptic material embedded in OCT. The problems related to the need of frozen biopsies apply also to other more recently introduced techniques, also valuable in identifying patients with CD, as the detection of intestinal deposits of IgA anti-TG2. The report of the possibility of counting γδ+ cells in paraffin-embedded material (20) is particularly promising in this context. Fernandez-Banares et al (21) showed that the analysis of IEL flow cytometric pattern is a fast, accurate method for identifying CD in the initial diagnostic biopsy of patients with lymphocytic enteritis. The authors suggested that γδ + IEL quantification is significantly better than anti-TG2 IgA intestinal deposits for identifying CD in patients with lymphocytic enteritis (21). In conclusion, we think that an improved immunohistochemical approach, in particular based on CD3 and γδ staining, may be of help in identifying subjects with potential CD among those with normal mucosal architecture, although a negative result does not exclude this condition. All of those involved in the care of patients with CD know how patients with potential CD represent a challenge from both the diagnostic and the therapeutic point of view. They should be referred to tertiary care centers with experience in the management of CD and facilities to perform such analysis.


1. Qiao SW, Iversen R, Ráki M, et al. The adaptive immune response in celiac disease. Semin Immunopathol 2012; 34:523–540.
2. Husby S, Koletzko S, Korponay-Szabò IR, et al. European Society for Pediatric Gastroenterology, Hepatology, and Nutrition guidelines for the diagnosis of coeliac disease. J Pediatr Gastroenterol Nutr 2012; 54:136–160.
3. Marsh MN. Gluten, major histocompatibility complex, and the small intestine. A molecular and immunologic approach to the spectrum of gluten sensitivity. Gastroenterology 1992; 102:330–354.
4. Tanpowpong P, Broder-Fingert S, Katz AJ, et al. Characteristics of children with positive coeliac serology and normal villous morphology: potential coeliac disease. APMIS 2013; 121:266–271.
5. Vande Voort JL, Murray JA, Lahr BD, et al. Lymphocytic duodenosis and the spectrum of celiac disease. Am J Gastroenterol 2009; 104:142–148.
6. Kakar J, Nehra V, Murray HA, et al. Significance of intraepithelial lymphocytosis in small bowel biopsy samples with normal mucosal architecture. Am J Gastroenterol 2003; 98:2027–2033.
7. Shmidt E, Smyrk TC, Faubion A, et al. Duodenal intraepithelial lymphocytosis with normal villous architecture in pediatric patients: Mayo clinic experience, 2000–2009. J Pediatr Gastroenterol Nutr 2013; 56:51–55.
8. Biagi F, Luinetti O, Campanella J, et al. Intraepithelial lymphocytes in the villous tip: do they indicate potential coeliac disease? J Clin Pathol 2004; 57:835–839.
9. Järvinen TT, Collin P, Rasmussen M, et al. Villous tip intraepithelial lymphocytes as markers of early-stage coeliac disease. Scand J Gastroenterol 2004; 39:428–433.
10. Maki M, Holm K, Collin P, et al. Increase in gamma/delta T cell receptor bearing lymphocytes in normal small bowel mucosa in latent coeliac disease. Gut 1991; 32:1412–1414.
11. Remes-Troche JM, Adames K, Castillo-Rodal AI, et al. Intraepithelial gammadelta+ lymphocytes: a comparative study between celiac disease, small intestinal bacterial overgrowth, and irritable bowel syndrome. J Clin Gastroenterol 2007; 41:671–676.
12. Korponay-Szabò IR, Halttunen T, Szalai Z, et al. In vivo targeting of intestinal and extraintestinal transglutaminase 2 by celiac autoantibodies. Gut 2004; 53:641–648.
13. Salmi TT, Collin P, Järvinen O, et al. Immunoglobulin A autoantibodies against transglutaminase 2 in the small intestinal mucosa predict forthcoming coeliac disease. Aliment Pharmacol Ther 2006; 24:541–552.
14. Tosco A, Salvati VM, Auricchio R, et al. Natural history of potential celiac disease in children. Clin Gastroenterol Hepatol 2011; 9:320–325.
15. Kurppa K, Collin P, Viljamaa M, et al. Diagnosing mild enteropathy celiac disease: a randomized, controlled clinical study. Gastroenterology 2009; 136:816–823.
16. Paparo F, Petrone E, Tosco A, et al. Clinical, HLA and small bowel immunohistochemical features of children with positive serum antiendomysium antibodies and architecturally normal small intestinal mucosa. Am J Gastroenterol 2005; 100:2294–2298.
17. Rosinach M, Esteve M, González C, et al. Lymphocytic duodenosis: aetiology and long-term response to specific treatment. Dig Liver Dis 2012; 44:643–648.
18. Aziz I, Evans KE, Hopper AD, et al. A prospective study into the aetiology of lymphocytic duodenosis. Aliment Pharmacol Ther 2010; 32:1392–1397.
19. Guz-Mark A, Zevit N, Morgenstern S, et al. Duodenal intraepithelial lymphocytosis is common in children without coeliac disease, and is not meaningfully influenced by Helicobacter pylori infection. Aliment Pharmacol Ther 2014; 39:1314–1320.
20. Lonardi S, Villanacci V, Lorenzi L, et al. Anti-TCR gamma antibody in celiac disease: the value of count on formalin-fixed paraffin-embedded biopsies. Virchows Arch 2013; 463:409–413.
21. Fernandez-Banares F, Carrasco A, Garcıa-Puig R, et al. Intestinal intraepithelial lymphocyte cytometric pattern is more accurate than subepithelial deposits of anti-issue transglutaminase IgA for the diagnosis of celiac disease in lymphocytic enteritis. PLoS One 2014; 9:e101249.

γδ intraepithelial lymphocytes; celiac diagnosis; diagnosis; histology

© 2015 by European Society for Pediatric Gastroenterology, Hepatology, and Nutrition and North American Society for Pediatric Gastroenterology,