Potential blood-based markers of celiac disease

Following written informed consent from parents/legal guardians, blood and duodenal biopsy specimens were collected from pediatric patients (Table 1) investigated for suspected CD or at follow-up on a gluten-free or gluten-containing diet, both for diagnostic purposes and for research purposes.

For diagnostic purposes, blood was collected in a blood tube containing polymer gel and clot activator (Becton, Dickinson and Company, Franklin Lakes, NJ) and centrifuged at 2400 × g for 5 min for serum isolation. The sera were stored at room temperature for a maximum of 10 hours after centrifugation, and then at 4°C until the presence of CD-specific antibodies (anti-TG2 and antibodies against DGP [anti-DGP]) and antibodies against GL (anti-GL) was investigated, which occurred within 5 days of centrifugation. In six cases, the antibody tests were performed on plasma collected for research purposes. Multiple biopsy specimens were collected using an endoscope in all but one case, where a pediatric Watson capsule was used to extract a single biopsy specimen. The tissue was formalin-fixated and paraffin-embedded, and histopathologically assessed.

For research purposes, blood was collected both in EDTA blood tubes (Becton, Dickinson and Company) for DNA and plasma isolation, and in Tempus Blood RNA tubes (Life Technologies, Carlsbad, CA) for RNA isolation. An aliquot of EDTA blood for DNA isolation was removed, and the remaining blood centrifuged at 1500 × g for 10 min for plasma isolation. For two cases, one in the group with no indications of CD and one in the group with active CD (Table 1), blood for RNA purification was collected in EDTA blood tubes instead of Tempus tubes. Additionally, a biopsy specimen immersed in pre-chilled RNAlater RNA Stabilization Reagent (Qiagen, Hilden, Germany) was collected from all cases in the study. Biopsies and stabilized blood for RNA purification were kept at 4°C for about 18 hours, and then at -20°C. RNA from EDTA blood was, however, purified without prior storage. Plasma was stored at -80°C. A maximum of two freeze-thaw cycles was accepted for all protein analyses.

The study was conducted under the approval of the Regional Ethical Review Board in Linköping.

DNA was isolated from EDTA blood using the EZ1 DNA Blood 350 μL Kit and BioRobot EZ1 (Qiagen) according to the manufacturer’s instructions.

RNA from stabilized blood was purified using the Tempus Spin RNA Isolation Reagent kit (Life Technologies), and RNA from EDTA blood was purified using the QIAamp RNA Blood Mini kit (Qiagen), in both cases according to the manufacturer’s instructions. The quality of the RNA from stabilized blood and EDTA blood was verified, and the RNA was reverse transcribed using a previously documented procedure [15]. The resulting cDNA and the remaining RNA were stored at -80°C.

Biopsies were assessed by a single experienced pathologist, blinded to all case data, in accordance with instructions for quality assurance and standardization assembled by the Swedish Society of Pathology. The status of the villi and crypts and the number of IELs were assessed for each biopsy. In cases where hematoxylin-eosin staining revealed an IEL number close to the ULN (25 IELs per 100 epithelial cells), an additional staining for CD3 was performed to better assess the number of IELs; when using CD3 staining, there should be >30 IELs per 100 epithelial cells to be indicative of CD. Hematoxylin-eosin staining was performed using the Tissue-Tek DRS 2000 Slide Stainer (Sakura, Alphen aan den Rijn, The Netherlands), and CD3 staining was performed using antibodies against CD3 (Dako, Glostrup, Denmark) and intelliPATH FLX (Biocare Medical, Concord, CA). The histological changes were reported according to the modified Marsh scale (0, 1, 2, 3A, 3B, or 3C) [16].

Detection of IgA anti-TG2, IgA anti-GL, and Immunoglobulin G (IgG) anti-DGP in serum or plasma was performed using EliA Celikey IgA (positive result ≥ 7 U/mL), EliA Gliadin IgA (positive result ≥ 7 U/mL), and EliA GliadinDP IgG (positive result ≥ 10 U/mL), respectively, on Phadia250 (Thermo Fisher Scientific, Waltham, MA) as described by the manufacturer. In cases with total IgA levels below 0.07 g/L, detection of IgG anti-TG2 replaced IgA anti-TG2 (EliA Celikey IgG, Thermo Fisher Scientific). In order to distinguish results below the detection limit of an assay from missing data, the former were replaced with the detection limit divided by two.

DNA from each case was HLA-typed for DRB1, DQA1 and DQB1 using a sequence-specific primer PCR method and capillary gel electrophoresis [17, 18]. The risk gradient for CD based on HLA type was calculated for each case using relative genotype risks extracted from a Scandinavian population [8].

Potential reference genes for the mRNA analysis were investigated using a Human Endogenous Control Plate (Life Technologies) containing assays for 32 potential reference genes, and cDNA from a total of nine blood RNA samples including three samples from cases with no mucosal injury (Marsh 0) and six with varying degrees of mucosal injury (Marsh 2-3C). Three potential reference genes (Additional file 1) were selected based on low sample-to-sample variation in mRNA levels, as investigated using the NormFinder algorithm [19] in version 5.4.2 of the Genex software package (MultiD Analyses, Göteborg, Sweden). The three selected reference genes were analyzed in the complete dataset, and the final selection of reference gene/s was established using the complete dataset by means of both low sample-to-sample variation and an absence of group differences in expression.

Genes for analysis of mRNA and protein levels were selected by reviewing published studies on blood mRNA/protein expression in CD, and by functional context (Additional file 1).

Multiplex detection of proteins in plasma was performed using Milliplex kits (Millipore, Billerica, MA) based on the Luminex xMAP technology, according to the manufacturer’s instructions (Additional file 1). The analyses were performed on the Bio-Plex 200 system (Bio-Rad, Hercules, CA). The CD163 soluble protein was detected using an enzyme-linked immunosorbent assay (ELISA) according to instructions from the manufacturer (Additional file 1), and the results were obtained and analyzed using a Sunrise microplate absorbance reader combined with version 7.0 of the Magellan software package (Tecan Group Ltd, Männedorf, Switzerland). Standard curves were included in all protein analyses, and optimized using Bio-Plex Manager 6.1 (Bio-Rad) for all Milliplex assays, and the Akima method for curve fitting in Magellan v.7.0 for the ELISA. In order to distinguish results below the detection limit of an assay from missing data, the former were replaced with the lowest detected value of the assay divided by two.

Levels of mRNA were investigated using custom made TaqMan Array Cards (Life Technologies) containing 47 gene expression assays including assays for the reference genes, or by using single assays and a previously documented procedure [15] (Additional file 1). Gene expression analysis on the TaqMan Array Cards was performed using the TaqMan Universal Master mix II without UNG and the recommended thermal profile (Life Technologies) on 300 ng cDNA in duplicates. Cards were prepared as recommended by the manufacturer, including analysis on the 7900HT Fast Real-time PCR system (Life Technologies).

For both TaqMan Array Cards and single assay results, quantification cycle (C_q) values were established using version 1.0.2 of the ExpressionSuite software package (Life Technologies). The auto-baseline algorithm in the software was used to compensate for background noise for each amplification curve, and the thresholds were adjusted to the log-linear range and set to the same level for all samples within one assay. Missing C_q values due to low copy numbers were replaced by the highest C_q value available for the gene in question, increased by one cycle. The resulting C_q values were normalized against selected reference gene/s (Genex).

Version 10 of the STATISTICA software package (StatSoft, Tulsa, OK) was used in all statistical analyses. Differential expression was investigated using Kruskal–Wallis one-way analysis of variance by ranks, except for the analysis of differential expression in cases stratified based on having one or two DQB1*02 alleles, where the Mann–Whitney U test was used. Post-hoc comparisons of mean ranks of all pairs of groups were performed (significance level p < 0.05, two-sided significance levels with a Bonferroni adjustment) [20]. Spearman rank correlation was used to investigate relations between mRNA/protein levels and histopathology and CD risk gradient based on HLA type. For all statistical analyses, except the post-hoc comparisons, a false-discovery rate was used and set to 5% [21].

The diagnostic performance of individual assays and logistic regression models of assay combinations was evaluated using receiver operating characteristic (ROC) curve analysis (MedCalc Statistical Software version 13.1.2, MedCalc Software, Ostend, Belgium).

Detectable levels were found for all protein markers (n = 22, Additional file 1), and all mRNA markers except for IL25 (n = 48, Additional file 1). CDKN1B was selected as the most stable reference gene, and used for normalization of all target mRNA levels.

In cases with active CD (Table 1, Active CD), significantly increased levels were observed for CXCL11 protein (p = 0.003, Figure 1, Additional file 2) and TNFSF13B mRNA (p = 0.001, Figure 1, Additional file 2), in comparison to cases without a CD diagnosis (Table 1; Not CD). Additionally, TNFSF13B mRNA levels were significantly elevated in cases with confirmed CD and normalized histology (Table 1; Normalized CD), in comparison to cases without a CD diagnosis (p = 0.001, Figure 1, Additional file 2). Levels of TNFRSF9 mRNA were significantly decreased in cases with active CD and in cases without a CD diagnosis, in comparison to CD cases with a normalized histology (p = 0.005 and p = 0.02, respectively, Figure 1, Additional file 2).

Previously established CD markers (anti-TG2, anti-DGP, and anti-GL) were differentially expressed in active CD cases in comparison to cases with normalized CD (p < 0.005) as well as cases without a CD diagnosis (p < 0.00005).

Stratification based on having one or two DQB1*02 alleles revealed no significant differences in levels of potential CD markers or in levels of previously established CD markers (anti-TG2, anti-DGP, and anti-GL), either when examining active CD cases only, or when looking at all CD cases together. Differences in expression in normalized CD cases alone could not be investigated, because all but two cases had one DQB1*02 allele.

Significant correlations were observed between Marsh grade (all cases) and levels of CXCL11 protein (Figure 2; Spearman rank correlation coefficient [r_s] = 0.50), and IL21 and IL15 mRNA (r_s = -0.46 for both, data not shown).

Clinical antibody levels (anti-TG2, anti-DGP, and anti-GL) displayed significant correlations with Marsh grade (r_s = 0.73, 0.75, and 0.75, respectively; Figure 3).

Levels of potential and previously established CD markers were not significantly correlated with HLA genotype risk for CD.

All cases on a gluten-containing diet with anti-TG2 levels 10 times ULN or more (n = 13) showed a Marsh grade of 3A-3C and received a CD diagnosis. Correspondingly high levels of anti-DGP (n = 7) and anti-GL (n = 4) were less common, but all occurred in cases with an active CD diagnosis and a Marsh grade 3A-3C.

Of all cases with active CD on a gluten-containing diet (n = 18) and all cases without CD diagnosis (n = 14), anti-TG2 failed to identify one case with active CD (Marsh 3C, anti-TG2 3.4 U/mL) and gave a false positive result in two cases without CD diagnosis (Marsh 0–1, anti-TG2 11–23 U/mL).

Anti-DGP yielded the same number of misclassifications as anti-TG2, but affected partially different cases, whereas anti-GL yielded substantially more misclassifications with four false negative results, and one false positive result (data not shown).

Two cases with remaining enteropathy on a GFD (Marsh 2 - 3A, anti-TG2 2.0 - 5.4 U/mL) had normalized levels of all antibodies.

The case with active CD on a gluten-containing diet that was misclassified based on levels of anti-TG2 (see Clinical antibody tests) showed levels of the two CD up-regulated markers CXCL11 protein and TNFSF13B mRNA (Figure 1) that were above the 80% central range (CR) of the group without CD and within the 80% CR for the group with active CD (Additional file 2). For one of the two cases without CD that were misclassified based on levels of anti-TG2 (see Clinical antibody tests), levels of CXCL11 protein were below the 80% CR of the group with active CD and thus corresponded to the “Not CD” group (Additional file 2).

One of the two CD cases with remaining enteropathy on a GFD that were misclassified based on anti-TG2 (see Clinical antibody tests) showed a level of the TNFRSF9 mRNA marker for normalized CD (Figure 1) that was below the 80% CR of the group with normalized CD and within the 80% CR for the group with active CD (Additional file 2).

The remaining results fell within the 80% CR for more than one group.

Cases included in the group under investigation (Table 1, Under investigation) were on a gluten-containing diet and under continuous monitoring for suspected CD. Considering markers CXCL11 protein and TNFSF13B mRNA, the median of the group fell within the 80% CR for both the group without a CD diagnosis and the group with active CD (Additional file 2).

ROC curve analysis of discrimination between cases with active CD and without CD (Figure 4A and B) revealed a larger area under the curve (AUC) for anti-TG2 (AUC = 0.97) in comparison to CXCL11 protein (AUC = 0.81), TNFSF13B mRNA (AUC = 0.85), and a logistic regression model based on CXCL11 protein and TNFSF13B mRNA (AUC = 0.91). A logistic regression model based on anti-TG2, CXCL11 protein, and TNFSF13B mRNA resulted in the highest AUC (0.98). However, compared to anti-TG2 alone, this improvement was not significant (p = 0.54). ROC curve analysis of discrimination between cases with active CD and normalized CD (Figure 4C) revealed an AUC of 0.90 for anti-TG2 and 0.78 for TNFRSF9 mRNA. Compared to anti-TG2 alone, a logistic regression model based on anti-TG2 and TNFRSF9 mRNA resulted in an increased AUC of 0.93 (p = 0.54).

The human genes discussed in this article are presented in Additional file 1.