A B-cell gene signature correlates with the extent of gluten-induced intestinal injury in celiac disease

ORIGINAL RESEARCH A B-Cell Gene Signature Correlates With the Extent of Gluten-Induced Intestinal Injury in Celiac Disease Mitchell E. Garber, 1,2 Alok Saldanha, 3 Joel S. Parker, 4,5 Wendell D. Jones, 6 Katri Kaukinen, 7,8,9 Kaija Laurila, 7 Marja-Leena Lähdeaho, 9,10 Purvesh Khatri, 11,12 Chaitan Khosla, 2,13,14 Daniel C. Adelman, 1,15 and Markku Mäki 7,9,16 1 Alvine Pharmaceuticals, Inc, San Carlos, California; 2 Department of Chemistry, 11 Institute for Immunity, Transplantation and Infection, 12 Division of Biomedical Informatics, Department of Medicine, 13 Department of Chemical Engineering, 14 Stanford ChEM-H, Stanford University, Stanford, California; 3 InterSystems Corporation, Cambridge, Massachusetts; 4 Lineberger Comprehensive Cancer Center, 5 Department of Genetics, University of North Carolina, Chapel Hill, North Carolina; 6 EA Genomics, Division of Q 2 Solutions, Morrisville, North Carolina; 16 Tampere Center for Child Health Research, 7 University of Tampere Faculty of Medicine and Life Sciences, Tampere, Finland; 10 Department of Pediatrics, 8 Department of Internal Medicine, 9 Tampere University Hospital, Tampere, Finland; 15 Division of Allergy/Immunology, Department of Medicine, University of California San Francisco, San Francisco, California SUMMARY A gluten challenge in celiac patients provides a unique opportunity to study the immunology associated with the transition from relative health to autoimmunity. This study showed that a B-cell population in peripheral blood correlated inversely with gluten-dependent small intestinal lesions, implicating a protective mechanism. BACKGROUND & AIMS: Celiac disease (CeD) provides an opportunity to study autoimmunity and the transition in immune cells as dietary gluten induces small intestinal lesions. METHODS: Seventy-three celiac disease patients on a longterm, gluten-free diet ingested a known amount of gluten daily for 6 weeks. A peripheral blood sample and intestinal biopsy specimens were taken before and 6 weeks after initiating the gluten challenge. Biopsy results were reported on a continuous numeric scale that measured the villusheight–to–crypt-depth ratio to quantify gluten-induced intestinal injury. Pooled B and T cells were isolated from whole blood, and RNA was analyzed by DNA microarray looking for changes in peripheral Band T-cell gene expression that correlated with changes in villus height to crypt depth, as patients maintained a relatively healthy intestinal mucosa or deteriorated in the face of a gluten challenge. RESULTS: Gluten-dependent intestinal damage from baseline to 6 weeks varied widely across all patients, ranging from no change to extensive damage. Genes differentially expressed in B cells correlated strongly with the extent of intestinal damage. A relative increase in B-cell gene expression correlated with a lack of sensitivity to gluten whereas their relative decrease correlated with gluten-induced mucosal injury. A core B-cell gene module, representing a subset of B-cell genes analyzed, accounted for the correlation with intestinal injury. CONCLUSIONS: Genes comprising the core B-cell module showed a net increase in expression from baseline to 6 weeks in patients with little to no intestinal damage, suggesting that these individuals may have mounted a B-cell immune response to maintain mucosal homeostasis and circumvent inflammation. DNA microarray data were deposited at the GEO repository (accession number: GSE87629; available: https:// www.ncbi.nlm.nih.gov/geo/). (Cell Mol Gastroenterol Hepatol 2017;4:1–17; http://dx.doi.org/10.1016/j.jcmgh.2017.01.011) Keywords: Oral Tolerance; Mucosal Immunity; Autoimmunity; Regulatory B Cell. Human beings ingest a wide variety of food proteins in the diet that are considered foreign with respect to the immune system. Immune cells in the small intestine survey the contents in the intestinal environment looking for pathogens. Oral tolerance is a mechanism that balances the need to promote tolerance to orally administered, foreign, yet harmless, food proteins with the need to provide a host defense against harmful pathogens in the intestine. 1 Although not well understood, oral tolerance to a food protein is an active immune response whose function is to suppress inflammatory immune responses to the same food protein when presented to the immune system for a second time. In celiac disease (CeD), a lack of oral tolerance develops to a family of cereal proteins collectively referred to as gluten, 2 resulting in a pathogenic and inflammatory immune response. The small intestinal mucosa consists of an epithelium and its underlying structures, which are immediately adjacent to the intestinal lumen and in contact with digested food. To increase surface area for nutrient absorption, the mucosa projects finger-like extensions called villi into the lumen of the gut. At the base of the villi are proliferative crypts. In patients with CeD, gluten ingestion results in blunting of the villi and hypertrophy or elongation of the crypts. The ratio of the height of the villi (Vh) to the depth of the crypt (Cd), expressed as Vh:Cd, has been used to quantify the extent of intestinal damage in CeD. 3–8 In severe cases, villi shrink completely with extensive crypt elongation, resulting in a flat mucosa and a Vh:Cd measurement approaching zero. HLA-DQ is an important genetic factor that predisposes individuals to CeD and type 1 diabetes. 9,10 A total of 5%–10% of individuals with type 1 diabetes develop CeD, 11 which is significantly higher than the chance of developing CeD in the overall Caucasian population, estimated to be 1%. 12 Most CeD patients (90%) express HLA-DQ2.5, whereas the remainder express HLA-DQ2.2 or HLA-DQ8. 13 Gluten peptides that are deamidated by the self-protein transglutaminase 2 (TG2) bind strongly to HLA-DQ and the resulting complex is presented to HLA-DQ–restricted CD4þT cells, 14 resulting in a T-cell response to deamidated gluten. 15 In addition to the T-cell response, glutendependent, disease-specific B cells appear early during disease pathogenesis. They precede gut damage, often are predictive of impending disease, 16–20 and produce antibodies specific for deamidated gluten and TG2. 21 It is unclear whether gluten-dependent auto-antibodies against TG2 contribute to the disease and there is little evidence that T cells with specificity for self-antigens drive the disease. The B cell recognizes a specific protein antigen, such as gluten or TG2, through direct interactions with its B-cell receptor containing a membrane-bound immunoglobulin. 22 The immunoglobulin determines antigen specificity and is unique for each B-cell clone. B-cell receptor signaling contributes strongly to B-cell proliferation and differentiation. B cells can be antigen-presenting cells 23 and have the ability to express HLA-DQ2 or HLA-DQ8. Because gluten peptides are excellent substrates for TG2 and the two proteins form a transient covalent complex, the possibility exists that a deamidated glutenor TG2-specific B cell binds the complex through the B-cell receptor, internalizes, and then presents gluten on HLA-DQ2 or HLA-DQ8 at the B cell surface to a gluten-specific HLA-DQ2–or HLA-DQ8–restricted CD4þT cell. 15,24 In this scenario, the resulting B cell is a functional antigen-presenting cell whose cell phenotype, as determined by B-cell–receptor signaling and direct Band T-cell co-stimulatory interactions, ultimately may be under the control of gluten. An inflammatory, gluten-induced immune response in the gut may propagate systemically in peripheral blood. The plausible trafficking of B and T cells and anti-TG2 to sites beyond the gut may help to explain several systemic clinical manifestations of CeD, 25 including dermatitis herpetiformis, a gluten-dependent blistering skin condition. CeD also may manifest as a bone disease, 26 in the central nervous system as ataxia and brain atrophy, 27 or as an isolated subclinical 28 or severe liver disease. 29 Evidence indicatesthatimmunecellsmigratetoandfromthegutin peripheral blood. For example, disease-specific T cells expressing the gut-homing b7 integrin migrate transiently to the periphery upon gluten challenge in CeD patients; these T cells are inflammatory in nature. 30,31 The peripheral blood therefore may be a good source to obtain biomarkers of the disease. It is not clear how gluten interacts with mechanisms of peripheral immune tolerance and whether B or T cells are responsible for disrupting the tolerogenic environment of the small intestine. The objective of this work was to determine if peripheral blood B and T cells modify gene expression in response to a 6-week gluten challenge in patients with treated CeD, and to correlate any changes in peripheral Bor T-cell gene expression with the extent of gluten-induced histological damage to the small intestine. Materials and Methods All authors had access to the study data and reviewed and approved the final manuscript. Abbreviations used in this paper: Cd, crypt depth; CeD, celiac disease; cRNA, complementary RNA; TG2, transglutaminase 2; TR, T-cell receptor; Vh, villus height; Vh:Cd, ratio of villus height to crypt depth. Most current article © 2017 The Authors. Published by Elsevier Inc. on behalf of the AGA Institute. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 2352-345X http://dx.doi.org/10.1016/j.jcmgh.2017.01.011 2 Garber et al Cellular and Molecular Gastroenterology and Hepatology Vol. 4, No. 1 Human Clinical Study Intestinal biopsy specimens and whole blood samples were collected from a single clinical site at the University of Tampere, Tampere, Finland, with patient consent (ethics approval ETL R09084M, ETL R10135M, and Eudra CTs 2009-012221-10 and 2010-023127-23). For each time point, small-bowel biopsy specimens (4–7 specimens) were sampled from the descending duodenum, sectioned, and scored by the same evaluator using standardized morphometric tools. 5,32 Of the 73 patients included in this analysis, 37% were male and 63% were female. The age range was 23–74 years with a median age of 59 years. For 6 weeks, patients ingested 6 g/day (20 patients), 3 g/day (26 patients), or 1.5 g/day (27 patients) wheat gluten with a meal. At the baseline time point, patients (85%) were negative for antibodies against transglutaminase 2. All patients completed the full 6-week study. Band T-Cell Isolation CD19þB cells and CD3þT cells were purified together from freshly prepared, venous whole blood (14 mL) using whole-blood CD19 and CD3 magnetic microbeads and the whole-blood column kit as specified by the manufacturer (Miltenyi Biotec, Bergisch Gladbach, Germany). EDTA was used as a blood anticoagulant. The final cell pellet containing purified B and T cells was resuspended in TRIzol (Fisher Scientific) reagent (5 mL) to lyse the cells and protect the RNA from degradation. The denatured Band T-cell lysate was stored at -70C for up to 3 months before purification of total RNA. For convenience, whole-blood samples may be stored on ice for up to 1 hour before adding magnetic beads. Preparation time from the point of adding magnetic beads to freezing the Band T-cell lysate was less than 2 hours. Total RNA Preparation Pooled Band T-cell lysates (5 mL) denatured in TRIzol were extracted with chloroform (0.2 vol) and the crude RNA was precipitated from the aqueous phase with the addition of isopropanol (0.7 vol). The RNA pellet was purified further using the RNeasy Plus Kit (Qiagen, Hilden Germany) according to the manufacturer’s specifications. Purified total RNA was eluted from the spin column with the addition of 2 aliquots (35 mL) RNase-free water. An Agilent BioAnalyzer (Santa Clara, CA) was used to determine total RNA yield (median, 2.2 mg), purity (median A260/A280, 2.1), and size (median RNA integrity number, 9.5) for all 146 samples. Purified total RNA was stored at less than -60C. Microarrays RNA samples were converted into labeled target complementary RNA (cRNA) using the Illumina TotalPrep-96 RNA Amplification Kit (Ambion). Briefly, 100 ng of total RNA was converted to double-stranded complementary DNA using an oligo-d(T) primer-adaptor. This complementary DNA was purified using magnetic beads and used as a template for in vitro transcription using T7 RNA polymerase and biotin–uridine triphosphate. The resulting biotinylated cRNA was purified using magnetic beads and quantified by spectrophotometry. For hybridization, 750 ng of purified biotinylated cRNA was added to Hybridization Cocktail Buffer (Illumina), applied to arrays, and incubated at 58C for 16 hours. After hybridization, arrays were washed and stained using standard Illumina procedures before scanning on an Illumina iScan instrument using the DirectHyb Tiff setting. Scanned images were processed by the Gene Expression module of GenomeStudio (v 1.6; Illumina) using default parameters without normalization. A total of 146 RNA samples were analyzed by microarray representing 73 patients with high-quality microarray data at baseline and 6-week time points (146 arrays). The 146 arrays included 39 from HT-12 version 3 and 107 from HT-12 version 4 (Illumina). The version 3 arrays were discontinued by the manufacturer, requiring the change to version 4. Only probes having identical sequence content between versions were retained. Potential artifacts caused by the 2 array versions and batch were removed by assuming the batch medians for a given probe were equal (covariates were not associated with batch). This was facilitated by the fact that both baseline and 6-week time points for 60 of 73 patients (82%) were analyzed on the same batch. After median shifting each probe relative to batch medians, probe quantile normalization was performed across all samples. Background then was subtracted with a monotonic transform of probe values below background to the (0,1) interval so that all normalized probes had positive values before log transformation. After normalization, batch adjustment, and background subtraction, the difference in expression between baseline and the 6-week time points was calculated as log 2 (6-wk/baseline) for each subject resulting in a matrix of 20,624 features and 73 samples. Gene Lists According to Bindea et al, 33 genes highly enriched in T cells were segregated into immune subpopulations: T cell, T CD8, T gd, T helper, T helper 1, T helper 2, T helper 17, T central memory, T effector memory, T follicular helper, and T regulatory. These categories were taken verbatim from the publication as multiple lists representing T-cell subpopulations and used separately to analyze the CeD data set. The T-regulatory category was characterized by one gene (FOXP3), which was not present on the Illumina array and therefore dropped from the analysis. According to Newman et al, 34 genes highly enriched in T cells also were segregated into immune subpopulations: CD8, CD4 naive, CD4 memory resting, CD4 memory activated, T follicular helper, T regulatory, and T gd. In this case, genes representing T-cell subpopulations were consolidated into one gene list. Genes highly enriched in B cells were taken verbatim from the publication as a single list (Bindea et al 33 ) or the B-cell subpopulations were consolidated into a single list (Newman et al 34 ). Bindea et al 33 also provided gene lists representing a diverse set of immune cell expression phenotypes, including mast, natural killer, neutrophils, macrophage, eosinophils, dendritic, and cytotoxic cells, as well as SW480 cancer cells and normal mucosa, which were included as July 2017 B Cell Gene Signature in Celiac Disease 3 controls. The University of North Carolina IgG gene list was used previously as a predictor in breast cancer. 35 All gene lists used to analyze the CeD data set are shown in Supplementary Table 1. In whole blood, the estimate is that T cells vastly outnumber B cells. In addition, the purified CD19þB-cell population used in this study should have little to no fully differentiated CD19plasma B cells, further reducing the number of B cells. Given that B and T cells were purified together, Band T-cell genes were selected from Newman et al 34 based on the following characteristic: highly enriched in at least one Bor T-cell subpopulation relative to other leukocytes. In addition, for B-cell gene selection, the gene had to show little to no expression in T cells or expression that greatly exceeded that in T cells because the contribution of B cells to the overall pool of total RNA was relatively small. Genes obtained from the Affymetrix (Santa Clara, CA) array platform (Bindea et al 33 and Newman et al 34 studies) that were not expressed in the CeD data set or not present on the Illumina array platform were excluded from analysis. There were compatibility issues with T-cell receptor (TR) gene segments between array platforms. The T-cell gene list obtained from Newman et al 34 was modified to retain TR transcripts by deleting Affymetrix TR probes and substituting all TR probes on the Illumina array if these probes were expressed in the CeD data set. It was common for genes obtained from Bindea et al 33 and Newman et al 34 to correspond to multiple Illumina DNA microarray probes. All relevant probes were included in the analysis. Correlation Between Gene Lists and DVh:Cd Approximately one third of genes mapped to more than one microarray probe. In these cases, we took two different approaches. First, the gene was consolidated to a single probe by taking only the probe with the greatest SD of expression across the 73 patients. The rationale behind this approach is that probes that perform poorly show a nearly constant, low level of signal across the data set, and thus can be screened out by their low variation. This approach was referred to as genes. Second, the gene was analyzed using all corresponding probes, as opposed to the one with highest variation. This approach was referred to as probes. In both cases, expression levels for all genes or probes in a given list were averaged on log signal values, and the average expression profile across 73 patients was correlated to DVh:Cd using Spearman rank correlation. Nearly identical results were obtained using the median rather than the mean expression profiles for both genes and probes (data not shown) to collapse a given gene list. For determination of the core B-cell gene module, expression levels for single probes also were correlated to DVh:Cd using Spearman rank correlation. Statistical Analyses Spearman correlations were calculated in R using the standard cor.test routine. The significance of the correlation was assessed using the Student tdistribution by setting exact ¼FALSE. The GSA module in R was used for file parsing. The Student ttest used for correlations with antiTG2 also was performed in R. Chi-squared analysis was performed using Microsoft Excel (Redmond, WA). Celiac Disease Serum Antibodies Serum antibodies directed against TG2-IgA were measured by enzyme-linked immunosorbent assay (Quanta Lite h-TG-IGA; Inova Diagnostics, San Diego, CA). 32 The positive threshold was 20 intensity units. Results Gluten-Dependent Intestinal Damage The data set consisted of 73 CeD patients following a strict gluten-free diet for at least one year. Each patient ingested 6, 3, or 1.5 g wheat gluten daily for 6 weeks. A whole blood sample, which was used to purify B and T cells, and intestinal biopsy specimens were taken before (baseline) and 6 weeks after initiating the gluten challenge. The median Vh:Cd at baseline was 2.7 (see Table 1 for patient data). Net change in intestinal biopsy from baseline to 6 weeks, defined as DVh:Cd, showed wide variation across all patients from no change or slight improvement to extensive mucosal damage (Figure 1). The largest DVh:Cd (-2.9) was observed in 3 patients who transitioned from a relatively healthy mucosa (Vh:Cd, 3.1) to a nearly flat mucosa (Vh:Cd, 0.2) in 6 weeks. Daily gluten dose for 2 of these patients was 6 g (roughly 2 slices of wheat bread). Although the 6 g gluten dose in these 2 patients resulted in extensive mucosal damage, in other patients it resulted in no damage (Figure 1, blue bars). Similar observations were made for the other 2 gluten doses, 3 g (yellow) and 1.5 g (grey). As a result, gluten dose was distributed across the full spectrum of intestinal damage. Regression analysis of DVh:Cd vs gluten dose showed that gluten dose explained roughly 18% of the variation in mucosal damage (adjusted R 2 , 0.18; P¼.0001). How an individual responded to the gluten challenge partly reflected the magnitude of the gluten dose. We were interested in wide variations in gluten-induced DVh:Cd across the data set, which we observed. The question we investigated in this study was whether mucosal damage correlated with an immune response, irrespective of the nominal amount of gluten administered daily. Genes Differentially Expressed in B and T Cells The experimental design was to isolate only CD19þB cells and CD3þT cells from whole blood, which enabled a simplified analysis of gene expression data focusing solely on genes highly enriched in B and T cells. With this design, our goal was to reduce the noise associated with analyzing global gene expression. For this purpose, we used two published gene lists derived from DNA microarray analyses of fractionated human leukocytes (Bindea et al 33 and Newman et al 34 ). The overlap in the B-cell gene lists between Bindea et al 33 (23 genes) and Newman et al 34 4 Garber et al Cellular and Molecular Gastroenterology and Hepatology Vol. 4, No. 1 Table 1.Patient Data Patient ID Vh:Cd (time, 0 days) Vh:Cd (time, 42 days) DVh:Cd (time 0 - time 42) Gluten challenge, g/day Age range, ySex Anti–TG2-IgA (time, 0 days) Anti–TG2-IgA (time, 42 days) Anti–TG2-IgA fold change Anti–TG2-IgA positivity Z 3.1 0.2 2.9 6 41–50 F 7 216 31 Pos X 3.1 0.2 2.9 6 51–60 F 6 8 - Neg AV 3.1 0.2 2.9 3 61–70 F 17 183 11 Pos O 3.5 0.7 2.8 6 51–60 F 8 146 18 Pos P 2.7 0.2 2.5 6 41–50 F 11 227 21 Pos AS 3.2 0.7 2.5 3 51–60 M 11 8 - Neg M 2.9 0.5 2.4 6 61–70 M 12 19 - Neg Y 2.8 0.4 2.4 6 51–60 F 10 10 - Neg G 2.6 0.3 2.3 6 31–40 M 6 240 40 Pos W 2.6 0.3 2.3 6 41–50 F 6 53 9 Pos AF 2.7 0.4 2.3 3 41–50 M 23 40 - Baseline pos R 3 0.8 2.2 6 61–70 F 7 10 - Neg AW 2.8 0.6 2.2 3 41–50 F 15 85 6 Pos K 2.4 0.3 2.1 6 41–50 F 15 282 19 Pos CM 2.8 0.7 2.1 1.5 41–50 F 10 72 7 Pos AQ 2.4 0.4 2 3 61–70 F 10 22 2 Pos AY 2.6 0.6 2 3 61–70 M 12 36 3 Pos AC 2.3 0.3 2 3 61–70 M 7 11 - Neg V 2.6 0.7 1.9 6 61–70 F 10 133 13 Pos CX 2.6 0.7 1.9 1.5 61–70 F 12 34 3 Pos AP 3 1.1 1.9 3 51–60 F 8 88 11 Pos CS 2.5 0.6 1.9 1.5 61–70 F 9 8 - Neg AH 2.7 0.9 1.8 3 41–50 F 10 114 11 Pos H 2 0.4 1.6 6 51–60 F 109 365 - Baseline pos AA 2.5 0.9 1.6 3 51–60 F 25 59 - Baseline pos AG 2.5 0.9 1.6 3 51–60 M 11 23 2 Pos BK 2.6 1 1.6 3 41–50 M 14 24 2 Pos S 2.8 1.2 1.6 6 61–70 F 11 14 - Neg AN 2.9 1.3 1.6 3 41–50 F 16 5 - Neg AU 1.7 0.2 1.5 3 61–70 M 22 21 - Baseline pos DK 3 1.5 1.5 1.5 61–70 F 9 8 - Neg AL 2.7 1.3 1.4 3 51–60 M 11 52 5 Pos AI 2.5 1.2 1.3 3 51–60 F 8 95 12 Pos AK 2.9 1.6 1.3 3 61–70 F 15 7 - Neg AE 2.4 1.1 1.3 3 71–80 M 8 67 8 Pos T 2.9 1.7 1.2 6 61–70 M 7 15 - Neg AD 2.3 1.1 1.2 3 51–60 F 50 41 - Baseline pos DF 2.6 1.5 1.1 1.5 51–60 M 6 4 - Neg L 1.4 0.3 1.1 6 61–70 M 23 209 - Baseline pos AM 3.3 2.2 1.1 3 51–60 F 6 6 - Neg CE 1.3 0.4 0.9 1.5 61–70 M 10 21 2 Pos DC 2.9 2 0.9 1.5 51–60 F 7 5 - Neg CA 3.6 2.8 0.8 1.5 31–40 F 8 7 - Neg DZ 3.2 2.4 0.8 1.5 61–70 F 6 5 - Neg CG 2 1.3 0.7 1.5 61–70 F 6 8 - Neg DQ 2.9 2.2 0.7 1.5 61–70 F 4 3 - Neg DY 3 2.3 0.7 1.5 61–70 F 7 5 - Neg DB 2.9 2.2 0.7 1.5 41–50 M 9 6 - Neg C 1.1 0.5 0.6 6 21–30 F 74 239 - Baseline pos July 2017 B Cell Gene Signature in Celiac Disease 5 (38 genes) was 15 genes. The overlap in the T-cell gene lists between the sum of all 11 T-cell subpopulations from Bindea et al 33 (185 genes) and the single T-cell list from Newman et al 34 (99 genes) was 33 genes. The two studies have more agreement than one would expect by chance but less than ideal agreement in defining differential Band Tcell gene expression. Lymphocytes are diverse, the methods used to purify these diverse cells vary, and differential gene expression is a relative measure that compares expression with other leukocytes. Given these complexities, it is beneficial to use both studies to define the most relevant genes. We started with comprehensive Band T-cell gene lists, and ultimately set out to refine these lists to obtain a gene signature that correlated more specifically with gluten-induced intestinal injury in CeD. The Band T-cell gene lists (see the Materials and Methods section and Supplementary Table 1) obtained from each publication were used separately to analyze the CeD data set. DVh:Cd Correlates With B-Cell Gene Expression By using DNA microarrays, we measured genome-wide gene expression in a purified pool of B and T cells taken from each patient at baseline and at the 6-week time points. We calculated the net change (baseline to 6 weeks) in Band T-cell gene expression for each of the published Band T-cell gene lists and correlated this change with the net change in Vh:Cd. Results showed that the net change in expression of B-cell genes from Bindea et al 33 (corr. -0.54; P¼8.7E-07) and Newman et al 34 (corr. -0.52; P¼1.9E-06) correlated (corr.) strongly with DVh:Cd (Figure 2Aand B). It was a negative correlation, which means that B-cell genes were expressed relatively strongly in patients with little to no DVh:Cd and relatively poorly in patients with large DVh:Cd. Two gene lists from Bindea et al 33 representing T-cell subpopulations showed weak positive correlation to DVh:Cd, including Th1 (corr. 0.34; P¼0.2.58E-03) and Th2 (corr. 0.31; P¼7.21E-03). In this case, opposite to B-cell genes, Table 1.Continued Patient ID Vh:Cd (time, 0 days) Vh:Cd (time, 42 days) DVh:Cd (time 0 - time 42) Gluten challenge, g/day Age range, ySex Anti–TG2-IgA (time, 0 days) Anti–TG2-IgA (time, 42 days) Anti–TG2-IgA fold change Anti–TG2-IgA positivity CN 3.1 2.5 0.6 1.5 71–80 F 14 11 - Neg AT 2.3 1.7 0.6 3 41–50 M 25 70 - Baseline pos DG 2.9 2.5 0.4 1.5 61–70 F 3 3 - Neg U 2.5 2.1 0.4 6 51–60 F 15 32 2 Pos DH 3.3 3 0.3 1.5 61–70 F 6 4 - Neg DW 2.6 2.3 0.3 1.5 61–70 M 8 6 - Neg A 2.9 2.6 0.3 6 41–50 F 47 43 - Baseline pos CB 3 2.7 0.3 1.5 31–40 F 17 32 2 Pos CJ 3 2.8 0.2 1.5 61–70 F 8 7 - Neg DR 2.9 2.7 0.2 1.5 61–70 F 8 6 - Neg AO 2.9 2.8 0.1 3 51–60 M 14 16 - Neg DJ 2.7 2.6 0.1 1.5 61–70 F 16 11 - Neg Q 2.8 2.7 0.1 6 61–70 M 13 32 2 Pos N 2.7 2.7 0 6 51–60 M 8 10 - Neg DN 2.3 2.3 0 1.5 61–70 M 14 12 - Neg DP 2.8 2.8 0 1.5 51–60 M 7 6 - Neg DM 1.3 1.4 -0.1 1.5 41–50 F 13 12 - Neg AJ 2.5 2.6 -0.1 3 51–60 F 13 16 - Neg DX 2.9 3.1 -0.2 1.5 61–70 F 10 6 - Neg CT 2.8 3 -0.2 1.5 71–80 M 15 23 2 Pos AR 2.3 2.6 -0.3 3 51–60 M 13 22 2 Pos DA 2.4 2.8 -0.4 1.5 51–60 F 36 22 - Baseline pos AX 2.8 3.3 -0.5 3 51–60 M 20 19 - Baseline pos AB 2.3 3.4 -1.1 3 41–50 M 9 9 - Neg NOTE. Parameters associated with the gluten challenge included the amount of gluten ingested daily for 42 days, the age range (median, 59 y; range, 23–74 y), antisera directed against TG2-IgA (anti-TG2-IgA) expressed in intensity units for both baseline and 42 days, and anti–TG2-IgA that was positive (pos) or negative (neg) above threshold (20 intensity units) at 42 days. Several patients were above threshold for anti–TG2-IgA at baseline (baseline pos). Anti–TG2-IgA fold change was expressed in intensity units, which may or may not reflect linearity pending assay validation. F, female; M, male. 6 Garber et al Cellular and Molecular Gastroenterology and Hepatology Vol. 4, No. 1 T-cell genes were expressed relatively strongly in patients with large DVh:Cd and relatively poorly in patients with little to no DVh:Cd. Other T-cell gene lists defined by Bindea et al 33 showed little to no correlation, including T CD8þ,T helper, T, T central memory, T effector memory, T follicular helper, T gd, and Th17. The T-cell gene list from Newman et al 34 also showed no correlation to DVh:Cd (corr. 0.10; P¼3.92E-01). Not surprisingly, gene lists corresponding to other leukocytes showed little to no correlation because B and T cells were the only cells analyzed in this study. The University of North Carolina IgG gene signature, 35 which contains several known B-cell genes that predict favorable outcomes in breast cancer patients, showed a moderate negative correlation with DVh:Cd. Nearly identical results were obtained irrespective of whether multiple microarray probes corresponding to one gene were reduced to a single probe (Figure 2Aand B)or whether all probes were analyzed (Figure 2Cand D) (see the Materials and Methods section). In addition to DVh:Cd, we correlated gene expression to baseline and end-of-study (week 6) Vh:Cd for all 73 patients. Results showed that the net change in expression of B-cell genes from Bindea et al 33 (corr. 0.49; P¼.1.16E-05) and Newman et al 34 (corr. 0.46; P¼4.71E-05) correlated with end-of-study Vh:Cd (Figure 2Eand F) and was only slightly weaker than that observed for DVh:Cd. The correlation was positive, which suggests that a reduction in B-cell gene expression over the 6-week gluten challenge (end-ofstudy minus baseline) was associated with a smaller end-ofstudy Vh:Cd. Th1 (corr. -0.31; P¼.0084) and Th2 (corr. -0.30; P¼.011) gene expression was correlated negatively with the end-of-study Vh:Cd, albeit modestly, which is consistent with results obtained from the DVh:Cd analysis. There was no correlation of gene expression with baseline Vh:Cd (Figure 2Gand H). Serum Anti-TG2 Correlates With DVh:Cd and the Core B-Cell Gene Module Sixty-two of 73 patients (85%) were negative at baseline for serum antibodies directed against TG2 (anti–TG2-IgA) (Table 1, threshold for positivity was 20 intensity units). For those 62 patients who were negative at baseline, 42% seroconverted to a positive value over the course of a 6-week gluten challenge. By using a Student ttest (unpaired, 2-sided) to compare means, and excluding baseline-positive patients, we determined that anti–TG2-IgA positivity correlated with DVh:Cd (P¼.0029) and the core B-cell gene module (P¼.0022). These results showed that anti-TG2 correlated with B-cell gene expression as well as it did with DVh:Cd. A positive anti-TG2 was associated with reduced B-cell gene expression and increased intestinal damage over the course of 6 weeks. A Subset of B-Cell Genes Drives Correlation to DVh:Cd In the correlation analyses, probes were averaged to generate a mean expression profileacrosstheentiregene list, which then was correlated to DVh:Cd; individual genes or probes were not analyzed separately. We asked whether all or a fraction of the genes in the two published B-cell gene lists from Bindea et al 33 and Newman et al 34 correlated with DVh:Cd. To define a core set of B-cell genes that drives the correlation, all probes in the two B-cell gene lists were consolidated into a single list and then the probes were ranked separately based on their ability to correlate with DVh:Cd.Therewere63uniqueprobesbetween the two B-cell gene lists, representing 48 unique genes. Of these, 28 probes representing 24 unique genes significantly correlated with DVh:Cd (P<.01). We defined these 28 probes as a core B-cell gene module representing a subset of known B-cell genes (see Table 2 for genes). The gene SPIB, present in both the Bindea et al 33 and Newman et al 34 B-cell gene lists, showed the strongest correlation to DVh:Cd (corr. -0.56; P¼2.5E-07) (Supplementary Table 2). Expression levels for all 28 probes in the core B-cell gene module were averaged across the data set, and the mean expression profile was correlated to DVh:Cd. The core B-cell gene module correlated strongly with DVh:Cd (corr. -0.59; Figure 1. Gluten-dependent intestinal mucosal injury as a clinical end point in a human clinical study. Vh:Cd, a histologic measure of mucosal health, was determined at baseline and at the 6-week time point. DVh:Cd, defined as baseline minus the 6-week Vh:Cd, represents intestinal damage (positive number) or healing (negative number) over the 6-week timeframe. The bar graph shows the number of patients (y-axis) with a given DVh:Cd (x-axis) for a total of 73 patients. The number of patients for a given DVh:Cd was subdivided further to indicate the amount of gluten ingested daily per patient, which was 6 (blue, 20 patients), 3 (yellow, 26 patients), and 1.5 (grey, 27 patients) grams of gluten daily for 6 weeks. July 2017 B Cell Gene Signature in Celiac Disease 7 P¼3.3E-08). As expected, all 28 single probes correlated strongly with the mean expression profile (Supplementary Table 2). In contrast, for the remaining 35 B-cell probes (24 unique genes) that correlated poorly as single probes (P>.01), in the aggregate, the mean expression profile across all 35 probes also correlated poorly (corr. -0.24; P¼ 3.8E-02). We refer to the published B-cell genes that correlated poorly with DVh:Cd as the non-correlating B-cell gene list (see Table 2 for genes). Correlations and Pvalues comparing the relative performance of relevant gene lists are summarized in Table 3. Point plots showed the scatter associated with the relationship between DVh:Cd and the net change in gene expression for the single-probe SPIB (Figure 3A) and the mean expression of all 28 probes in the core B-cell gene module (Figure 3B). Seventy-three patients were scattered Figure 2. Spearman rank correlation of gene signatures with the extent of gluten-induced intestinal injury. Gene lists obtained from three publications corresponded to Band Tcell populations, other leukocytes, cancer cells, and normal mucosa (y-axis). All gene lists were obtained from Bindea et al, 33 except for Band T-cell lists from Newman et al 34 (as indicated) and the University of North Carolina (UNC) IgG cluster from Fan et al. 35 (A) The mean expression profile for a given gene list was correlated with DVh:Cd and the correlation was reported on a scale of 0 to 1(x-axis). Multiple microarray probes corresponding to a single gene were consolidated to a single probe by taking only the probe with the greatest SD of expression across the 73 patients. (B) Significance for each correlation in panel Awas expressed as a Pvalue. Gene signatures also were correlated with (C)DVh:Cd, (E) endof-study Vh:Cd, and (G) baseline Vh:Cd using mean expression profiles and all probes representing a given gene. Significance for each correlation in panels C,E, and Gwas expressed as a Pvalue in panels D,F, and H, respectively. aDC, activated dendritic cell; DC, dendritic cell; iDC, immature dendritic cell; NK, natural killer; Tcm, T central memory; Tem, T effector memory; TFH, T follicular helper; Tgd, T gamma delta. 8 Garber et al Cellular and Molecular Gastroenterology and Hepatology Vol. 4, No. 1 fairly uniformly on both sides of the regression line over the entire spectrum of intestinal damage. The Extended Core B-Cell Gene Module Genes representing the core B-cell gene module and the non-correlating B-cell gene list were obtained exclusively from the Bindea et al 33 and Newman et al 34 curated gene lists. We asked whether the core B-cell gene module could be used to discover additional disease-relevant genes that were not included in the Bindea et al 33 and Newman et al 34 published gene lists. To this end, we took the Spearman correlation between the mean expression profile of the core B-cell gene module and all 20,624 probes in the CeD data Figure 2. (continued). July 2017 B Cell Gene Signature in Celiac Disease 9 References 1. Pabst O, Mowat AM. Oral tolerance to food protein. Mucosal Immunol 2012;5:232–239. 2. Ribeiro M, Nunes-Miranda JD, Branlard G, et al. One hundred years of grain omics: identifying the glutens that feed the world. J Proteome Res 2013;12: 4702–4716. 3. Kuitunen P, Kosnai I, Savilahti E. Morphometric study of the jejunal mucosa in various childhood enteropathies with special reference to intraepithelial lymphocytes. J Pediatr Gastroenterol Nutr 1982; 1:525–531. 4. Kurppa K, Collin P, Viljamaa M, et al. Diagnosing mild enteropathy celiac disease: a randomized, controlled clinical study. Gastroenterology 2009;136:816–823. 5. Taavela J, Koskinen O, Huhtala H, et al. Validation of morphometric analyses of small-intestinal biopsy readouts in celiac disease. PLoS One 2013;8:e76163. 6. Leffler D, Schuppan D, Pallav K, et al. Kinetics of the histological, serological and symptomatic responses to gluten challenge in adults with coeliac disease. Gut 2013; 62:996–1004. 7. Leffler D, Kupfer SS, Lebwohl B, et al. Development of celiac disease therapeutics: report of the Third Gastroenterology Regulatory Endpoints and Advancement of Therapeutics Workshop. Gastroenterology 2016; 151:407–411. 8. Hindryckx P, Levesque BG, Holvoet T, et al. Disease activity indices in coeliac disease: systematic review and recommendations for clinical trials. Gut 2016. Epub ahead of print. 9. Thorsby E. Invited anniversary review: HLA associated diseases. Hum Immunol 1997;53:1–11. 10. Nepom GT, Erlich H. MHC class-II molecules and autoimmunity. Annu Rev Immunol 1991;9:493–525. 11. Cosnes J, Cellier C, Viola S, et al. Incidence of autoimmune diseases in celiac disease: protective effect of the gluten-free diet. Clin Gastroenterol Hepatol 2008; 6:753–758. 12. Green PH, Cellier C. Celiac disease. N Engl J Med 2007; 357:1731–1743. 13. Sollid LM, Lie BA. Celiac disease genetics: current concepts and practical applications. Clin Gastroenterol Hepatol 2005;3:843–851. 14. Tollefsen S, Arentz-Hansen H, Fleckenstein B, et al. HLADQ2 and -DQ8 signatures of gluten T cell epitopes in celiac disease. J Clin Invest 2006;116:2226–2236. 15. Stamnaes J, Sollid LM. Celiac disease: autoimmunity in response to food antigen. Semin Immunol 2015; 27:343–352. 16. Korponay-Szabó IR, Halttunen T, Szalai Z, et al. In vivo targeting of intestinal and extraintestinal transglutaminase 2 by coeliac autoantibodies. Gut 2004; 53:641–648. 17. Kaukinen K, Peraaho M, Collin P, et al. Small-bowel mucosal transglutaminase 2-specific IgA deposits in coeliac disease without villous atrophy: a prospective and randomized clinical study. Scand J Gastroenterol 2005;40:564–572. 18. 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. 19. Tosco A, Maglio M, Paparo F, et al. Immunoglobulin A anti-tissue transglutaminase antibody deposits in the small intestinal mucosa of children with no villous atrophy. J Pediatr Gastroenterol Nutr 2008;47:293–298. 20. Salmi TT, Collin P, Jarvinen 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. 21. Mesin L, Sollid LM, Di Niro R. The intestinal B-cell response in celiac disease. Front Immunol 2012; 3:313. 22. Hobeika E, Maity PC, Jumaa H. Control of B cell responsiveness by isotype and structural elements of the antigen receptor. Trends Immunol 2016;37:310–320. 23. Lanzavecchia A. Antigen-specific interaction between T and B cells. Nature 1985;314:537–539. 24. du Pre MF, Sollid LM. T-cell and B-cell immunity in celiac disease. Best Pract Res Clin Gastroenterol 2015; 29:413–423. 25. Leffler DA, Green PHR, Fasano A. Extraintestinal manifestations of coeliac disease. Nat Rev Gastroenterol Hepatol 2015;12:561–571. 26. Heikkilä K, Pearce J, Mäki M, et al. Celiac disease and bone fractures: a systematic review and meta-analysis. J Clin Endocrinol Metab 2015;100:25–34. 27. Hadjivassiliou M, Mäki M, Sanders DS, et al. Autoantibody targeting of brain and intestinal transglutaminase in gluten ataxia. Neurology 2006;66:373–377. 28. Korpimäki S, Kaukinen K, Collin P, et al. Gluten-sensitive hypertransaminasemia in celiac disease: an infrequent and often subclinical finding. Am J Gastroenterol 2011; 106:1689–1696. 29. Kaukinen K, Halme L, Collin P, et al. Celiac disease in patients with severe liver disease: gluten-free diet may reverse hepatic failure. Gastroenterology 2002; 122:881–888. 30. Anderson RP, Degano P, Godkin AJ, et al. In vivo antigen challenge in celiac disease identifies a single transglutaminase-modified peptide as the dominant A-gliadin T-cell epitope. Nat Med 2000;6:337–342. 31. Ráki M, Fallang L-E, Brottveit M, et al. Tetramer visualization of gut-homing gluten-specific T cells in the peripheral blood of celiac disease patients. Proc Natl Acad Sci U S A 2007;104:2831–2836. 32. Lahdeaho ML, Kaukinen K, Laurila K, et al. Glutenase ALV003 attenuates gluten-induced mucosal injury in patients with celiac disease. Gastroenterology 2014; 146:1649–1658. 33. Bindea G, Mlecnik B, Tosolini M, et al. Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer. Immunity 2013; 39:782–795. 34. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 2015;12:453–457. 16 Garber et al Cellular and Molecular Gastroenterology and Hepatology Vol. 4, No. 1 35. Fan C, Prat A, Parker JS, et al. Building prognostic models for breast cancer patients using clinical variables and hundreds of gene expression signatures. BMC Med Genomics 2011;4:3. 36. Rosser EC, Mauri C. Regulatory B cells: origin, phenotype, and function. Immunity 2015;42:607–612. 37. Menon M, Blair PA, Isenberg DA, et al. A regulatory feedback between plasmacytoid dendritic cells and regulatory B cells is aberrant in systemic lupus erythematosus. Immunity 2016;44:683–697. 38. Miyagaki T, Fujimoto M, Sato S. Regulatory B cells in human inflammatory and autoimmune diseases: from mouse models to clinical research. Int Immunol 2015;27:495–504. 39. Mishima Y, Liu B, Hansen JJ, et al. Resident bacteriastimulated IL-10-secreting B cells ameliorate T cellmediated colitis by inducing Tr-1 cells that require IL-27-signaling. Cell Mol Gastroenterol Hepatol 2015; 1:295–310. 40. Geem D, Ngo V, Harusato A, et al. Contribution of mesenteric lymph nodes and GALT to the intestinal Foxp3þregulatory T-cell compartment. Cell Mol Gastroenterol Hepatol 2016;2:274–280. 41. Lähdeaho M-L, Mäki M, Laurila K, et al. Small-bowel mucosal changes and antibody responses after lowand moderate-dose gluten challenge in celiac disease. BMC Gastroenterol 2011;11:129. 42. Mäki M, Mustalahti K, Kokkonen J, et al. Prevalence of celiac disease among children in Finland. N Engl J Med 2003;348:2517–2524. 43. Lohi S, Mustalahti K, Kaukinen K, et al. Increasing prevalence of coeliac disease over time. Aliment Pharmacol Ther 2007;26:1217–1225. 44. Vilppula A, Kaukinen K, Luostarinen L, et al. Increasing prevalence and high incidence of celiac disease in elderly people: a population-based study. BMC Gastroenterol 2009;9:49. 45. Mäki M, Holm K, Koskimies S, et al. Normal small bowel biopsy followed by coeliac disease. Arch Dis Child 1990; 65:1137–1141. 46. Mäki M, Lähdeaho ML, Hällström O, et al. Postpubertal gluten challenge in coeliac disease. Arch Dis Child 1989; 64:1604–1607. 47. Kurppa K, Koskinen O, Collin P, et al. Changing phenotype of celiac disease after long-term gluten exposure. J Pediatr Gastroenterol Nutr 2008;47:500–503. 48. Kim KS, Hong S-W, Han D, et al. Dietary antigens limit mucosal immunity by inducing regulatory T cells in the small intestine. Science 2016;351:858–863. 49. Du Pré MF, Kozijn AE, van Berkel LA, et al. Tolerance to ingested deamidated gliadin in mice is maintained by splenic, type 1 regulatory T cells. Gastroenterology 2011; 141:610–622. 50. van Leeuwen MA, Costes LMM, van Berkel LA, et al. Macrophage-mediated gliadin degradation and concomitant IL-27 production drive IL-10and IFNg-secreting Tr1-like-cell differentiation in a murine model for gluten tolerance. Mucosal Immunol 2016. Epub ahead of print. 51. Adelman D, Essenmacher K, Garber M, et al. sa1280 the gluten-free diet alone does not control symptoms and mucosal injury in many patients with celiac disease. Gastroenterology 2015;148:S-280. 52. Fu H, Ward EJ, Marelli-Berg FM. Mechanisms of T cell organotropism. Cell Mol Life Sci 2016;73: 3009–3033. 53. Brand A, Allen L, Altman M, et al. Beyond authorship: attribution, contribution, collaboration, and credit. Learn Publ 2015;28:151–155. Received October 24, 2016. Accepted January 20, 2017. Correspondence Address correspondence to: Mitchell E. Garber, PhD. e-mail: [email protected]om. Acknowledgments This work benefitted from data assembled by the Immgen consortium. Conflicts of interest These authors disclose the following: Markku Mäki is a scientific advisor to ImmusanT, Inc, and Celimmune LLC; Daniel Adelman is a scientific advisor to ImmunogenX, Inc. The remaining authors disclose no conflicts. Mitchell E. Garber conceptualized the study; Mitchell E. Garber, Alok Saldanha, Joel S. Parker, and Markku Mäki were responsible for the methodology; Alok Saldanha and Joel S. Parker were responsible for the software; Alok Saldanha validated the study; Mitchell E. Garber, Alok Saldanha, Joel S. Parker, and Wendell D. Jones performed the formal analysis; Mitchell E. Garber, Kaija Laurila, and Katri Kaukinen performed the investigation; Mitchell E. Garber, Alok Saldanha, Joel S. Parker, Marja-Leena Lähdeaho, Daniel C. Adelman, and Markku Mäki provided resources; Mitchell E. Garber, Alok Saldanha, Joel S. Parker, Wendell D. Jones, and Kaija Laurila curated data; Mitchell E. Garber wrote the original draft; Mitchell E. Garber, Wendell D. Jones, Purvesh Khatri, Chaitan Khosla, Daniel C. Adelman, and Markku Mäki reviewed the writing and edited the paper; Mitchell E. Garber, Alok Saldanha, Joel S. Parker, and Wendell D. Jones were responsible for visualization; Mitchell E. Garber, Chaitan Khosla, Daniel C. Adelman, and Markku Mäki supervised the study; Mitchell E. Garber, Daniel C. Adelman, and Markku Mäki were responsible for project administration; and Mitchell E. Garber, Alok Saldanha, Joel S. Parker, Chaitan Khosla, Daniel C. Adelman and Markku Mäki acquired funding. Categories were defined by CRediT. 53 Funding This study was funded by Alvine Pharmaceuticals, Inc, and grants from the American Recovery and Reinvestment Act (to Alvine Pharmaceuticals), from the Competitive State Research Financing of the Tampere University Hospital (M.M. and K.K.), from the Academy of Finland Research Council for Health (K.K.), and in part from the National Institutes of Health (R01 DK063158 to C.K.). M.E.G. was supported part time during the writing of this manuscript by the National Institutes of Health (R01 DK063158 to C.K.). All authors currently have no financial or nonfinancial competing interests in Alvine Pharmaceuticals, which had a role in the clinical study design and collection of biopsy and blood samples, but no role in data collection, analysis, or interpretation associated with the biopsies, blood samples, and genomic data used in this study. Alvine Pharmaceuticals had no role in writing this report, or the decision to submit the report for publication. July 2017 B Cell Gene Signature in Celiac Disease 17