In Well‐Treated Celiac Patients Low‐Level Mucosal Inflammation Predicts Response to 14‐day Gluten Challenge

Abstract In celiac disease (CeD), gluten activates adaptive immune cells that cause damage to the small intestinal mucosa. Histological evaluation of intestinal biopsies allows for grading of disease severity. CeD can effectively be treated with a life‐long gluten‐free diet. Gluten challenge of treated CeD patients is used to confirm diagnosis and to test drug efficacy in clinical trials, but patients respond with different magnitudes to the same gluten challenge. In this study of 19 well‐treated CeD patients, proteome analysis of total tissue or isolated epithelial cell compartment from formalin‐fixed paraffin embedded biopsies collected before and after 14‐day gluten challenge demonstrates that patients with strong mucosal response to challenge have signs of ongoing tissue inflammation already before challenge. This low‐level tissue inflammation at baseline is paralleled by increased gluten specific CD4+ T‐cell frequencies in the gut and presence of a low‐level blood inflammatory profile. Thus, apparently well‐treated CeD is frequently not entirely quiescent, with presence of low‐grade inflammation and antigluten immunity in the gut mucosa. Histology assessment alone appears insufficient to judge full recovery and gut mucosal healing of CeD patients. The findings raise a concern whether a seemingly proper gluten‐free diet is able to curb gut inflammation in all CeD patients.


Patient cohort and clinical information
The patient cohort subjected to 14-day oral gluten challenge has previously been described by Sarna et al. where full clinical information can be found. [1] For this study we have used material from formalin fixed paraffin embedded (FFPE) small intestinal biopsies from 19 of 20 enrolled patients that completed the challenge. Biopsies were collected at baseline before challenge and at day 14 of challenge. Histomorphometry (Marsh score and villous height to crypt depth (Vh:Cd) ratio) and intraepithelial lymphocyte (IEL) counts was previously reported and were performed on sections from the same FFPE biopsy blocks as used in this study. All patients were upon study initiation considered to be in complete clinical and mucosal remission at baseline before challenge as assessed by Marsh score and serum anti-transglutaminase IgA titers. One patient (P11, CD442) was initially evaluated to be in mucosal remission before onset of gluten challenge, but was later revised to "Marsh 3" after a blinded re-evaluation of all biopsies.
Clinical biochemistry and cytokine measurements were performed on baseline blood samples.
Frequencies of CD4+ gluten specific T cells in blood (baseline for all patients and at day 6 for 15 patients) and gut (baseline and day 14 for 7 patients) have previously been reported. [1,2]

Total tissue sample digestion and processing
To generate total tissue digests, fifteen 5 µm thick sections from each FFPE biopsy block were collected in a tube (Eppendorf, Hamburg, Germany) and dehydrated by 10 min incubations in 80 %, 96 % and 100 % ethanol, respectively, followed by dewaxing by two incubations with xylene (2x3 min at 55 °C) and two incubations in 100 % ethanol (3 min at room temperature). Dewaxed tissue was resuspended in 20 µL 50 mM ammonium bicarbonate with 0.2 % ProteaseMAX surfactant (Trypsin enhancer; Promega, Madison, WI) followed by addition of dithiotreitol (1µL 0.5 M) and ammonium bicarbonate (73.5 µL 50 mM). Formalin crosslinks were cleaved by heating the samples (98°C 90 min) followed by sonication (60 min in a water bath). Protein amount was estimated by DirectDetect (Millipore, Merck, Darmstadt, Germany). Disulfide bonds were reduced with dithiotreitol (1 µL 0.5 M per sample, 20 min incubation with gentle agitation at 56°C) and alkylated with iodoacetamide (2.7 µL 0.55 M per sample, 15 min incubation with gentle agitation at room temperature in the dark). To digest proteins, ProteaseMAX (2 µL 1%) and Trypsin (1 µg) (Sequencing grade, ProMega) was added to each sample followed by incubation in a wet chamber over night at 37°C. Peptides were purified on C18 micro columns as previously described. [3] Purified samples were adjusted to a final volume of 11 µL or 20 µL depending on protein concentration.

LCM sample collection and processing
Eight µm tissue sections were adhered to PEN-covered slides (Zeiss) and dried at 37°C. Dry sections were dewaxed in xylene (3 min + 2 min) followed by 1 min in 100 % ethanol, 95 % ethanol and 70 % ethanol, respectively followed by 2 x 1 min in water. Tissue was visualized by staining with Mayer's hematoxylin solution (Sigma) for ~30 s followed by rinsing in tap water.
Stained sections were air-dried and stored dry until cutting. Samples were collected using a PALM MicroBeam laser capture microdissection system (Carl Zeiss MicroImaging, Munich, Germany), and isolated tissue collected in 0.5 mL opaque adhesive cap tubes (Zeiss). To validate our LCM approach and confirm that we could obtain region specific protein expression data we collected samples from three distinct tissue regions for comparative analysis: Two samples were collected per region from lamina propria, villus and crypt epithelium. For total epithelial cell layer analysis on average 250 000 µm 2 tissue was collected per sample, while for apical epithelial tissue from baseline biopsies, on average 150 000 µm 2 tissue was collected per sample.
Two cohorts of total epithelial cell layer samples spanning the entire crypt villus axis were collected (LCM1; 17 samples from 13 biopsies; LCM2: 25 samples from 20 biopsies). For apical epithelial cell layer analysis, 24 samples from 12 biopsies before challenge were collected.
Dissected tissue was retrieved from adhesive caps using ammonium bicarbonate (10 µL 50 mM) with ProteaseMax Surfactant (0.2 %) followed by ammonium bicarbonate (10 µL 50 mM) and transferred to 0.5 mL Low-Bind tubes (Eppendorf, Hamburg, Germany). Samples were heated to 98°C for 90 min followed by sonication in water bath for 60 min. Disulfide bridges were reduced by addition of dithiotreitol (2 µL 0.1 M) followed by incubation for 20 min at 56°C and alkylated by addition of iodoacetamide (2 µl 55 mM) followed by incubation for 15 min in the dark at room temperature. Samples were digested by addition of trypsin (1.5 µL 0.01 g L -1 ) and incubation in wet chamber over night at 37°C. Peptides were purified on C18 micro columns and eluted samples were adjusted to a final of 7 µL with 0.1 % formic acid.

Mass spectrometry analysis
Three µL digest was injected per run for all samples. Total tissue digests were analyzed with two technical replicates. Digested LCM1 samples were analyzed together (LCM cohort 1, 17 samples from 13 biopsies). Digested LCM2 samples were run together with 13 samples from LCM1 resulting in total 38 samples from 21 biopsies in the LCM2 cohort 2 dataset (LCM cohort 2).
For total tissue samples, the solvent gradient was 2% to 5 % in 10 min, to 19% in 170 min and then to 35% B in 60 min followed by a wash with 90% B for 20min. For LCM isolated sample, the solvent gradient was 2 -7% B in 10 min, then to 30 % B in 55 min and finally a wash with 90 % B in 20 min. Solvent A was aqueous 0.1 % formic acid, whereas solvent B was 100 % acetonitrile in 0.1 % formic acid. Column temperature was kept at 60 o C.
The mass spectrometer was operated in the data-dependent mode to automatically switch between MS and MS/MS acquisition. Survey full scan MS spectra (total tissue: m/z 300 to 1,500; LCM samples: m/z 400 to 1,200) were acquired in the Orbitrap with resolution R = 70,000 at m/z 200 (after accumulation to a target of 3,000,000 ions in the quadruple). The method used allowed sequential isolation of the most intense multiply-charged ions, up to ten, depending on signal intensity, for fragmentation on the HCD cell using high-energy collision dissociation at a target value of 100,000 charges or maximum acquisition time of 100 ms. MS/MS scans were collected at 17,500 resolution at the Orbitrap cell. Target ions already selected for MS/MS were dynamically excluded for 30 seconds. General mass spectrometry conditions were: electrospray voltage, 2.1 kV; no sheath and auxiliary gas flow, heated capillary temperature of 250°C, normalized HCD collision energy 25%.

Protein identification, quantification and data analysis
MS raw files were processed in the MaxQuant environment [5] (version 1.6.1.0) with the integrated Andromeda search engine [6] for peptide and protein identification, with a FDR threshold of 0.01 for peptide and for protein identification. The human UniProtKB FASTA database (September 2018) was used as forward database for protein identification. Match between runs was enabled and label-free protein quantification (LFQ) was performed using the MaxQuant's Label Free Quantification algorithm with a minimum ratio count of one. [7] Methionine oxidation and N-terminal acetylation was used as variable modification and carbamidomethyl cysteine as fixed modification. We performed separate MaxQuant searches for MS data from total tissue, total epithelial LCM cohort 1, total epithelial LCM cohort 2 (including compartment samples) and apical epithelial LCM samples. MaxQuant results were processed in Perseus (version 1.6.2.2). [8] For all datasets, proteins matched to the reverse decoy database, identified by site or identified as potential contaminant were removed. Poor quality samples were removed based on number of LFQ valid values as shown in Figure S1, Figure S3 and Figure S5.

2D categorical enrichment of biological pathways
Biological pathway enrichment analysis was performed in Perseus (2D enrichment based on student t-tests fold difference) [9] . Enrichment data for Gene Ontology pathways (GO Biological Process, GO Cellular Compartment, GO Molecular Function) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were exported and visualized in R. Enriched pathways were filtered for pathways with <100 genes and Student t-test fold difference (<-0.25 or >0.25) for both "responders" after vs. before challenge and "responders" vs. "non-responders" before challenge (Shared Down and Shared Up). Shared Up were 38 pathways of which 23 were GO Biological Processes (from 410 unique proteins). Shared Down consisted of 77 pathways of which 42 were GO Biological Processes (from 573 unique proteins). Expression of proteins that mapped to GO Biological Processes in Shared Up or Down were compared between responder groups (ANOVA and Tukey's honest significance test, FDR = 0.05) ( Table S3). Z-scored protein expression (median expression per responder group) was visualized for selected pathways.

Cell-type gene set annotation
Small intestinal epithelial cell-type gene-sets were retrieved from [10] . Mouse genes were converted to human gene orthologs (www.ensembl.org/biomart) followed by manual curation of the lists. From the "mature enterocyte" protein list (n =497) 222 proteins were present in the total tissue dataset and 181 proteins were present in the epithelial dataset. From the "goblet cell" protein list (n = 401) we found 157 proteins in the total tissue dataset and 119 proteins in the epithelial dataset (Table S4).

Statistical analysis
Protein LFQ values were log2 normalized before analysis. Details on quality control of mass spectrometry datasets, filtering of low quality samples, filtering of proteins and imputation of missing values is explained in the section "Protein identification, quantification and data analysis". All statistical analysis and data visualization was performed in Perseus [11] (version 1.6.2.2) or the R framework (R version 3.6.1, https://www.r-project.org/). To address differential protein expression, technical replicates (total tissue) or biological replicates (LCM samples) were averaged to give median protein expression per biopsy before comparison by two-sample Student t-test (FDR = 0.05, Benjamini Hochberg adjustment for multiple testing) ( Table S2).
Clinical and serological variables were compared by Mann-Whitney U test with no correction for multiple testing. Expression of individual proteins or proteins mapped to biological pathway were compared between patient groups by Welch t-test or ANOVA and Tukey's honest significance test (FDR = 0.05). Pearson correlation was used to compare cell-type protein expression (median z-scored expression per biopsy) with Vh:Cd ratio or gluten-specific CD4+ T cell frequencies.         a) The clinical variables have previously been published by Sarna et al. [1] ; b) Responders are defined from tissue proteome analysis ( Figure 1B); c) Number of HLA-DQ:gluten tetramerbinding effector-memory gut-homing CD4+ T cells per million CD4+ T cells in blood from [1] ; d) Number of HLA-DQ:gluten tetramer-binding CD4+ T cells per million CD4+ T cells in gut biopsies [2] ; e) Serum biochemistry measured at baseline before gluten challenge. TNF-α was measured in plasma as part of the Bio-Plex Pro Human Cytokine 27-plex Assay as reported in [1] . f) CRP: LOD = 0.6. Values <0.6 are shown as 0.5 as this value was used for calculations in Figure 4 n.d.= not done