This study was approved by the Washington University in St. Louis Human Research Protection Office. Informed consent was obtained from parent(s), and assent from children aged 12 years and older. Cases, defined as treatment-naïve (i.e., a gluten-free diet had not yet been started) children ages 0–17.9 years with suspected CD based on an elevated tTG IgA (n = 50) or tTG IgG (n = 4) who were referred for endoscopy, were eligible to participate. Participants were excluded if they had known inflammatory disorders of the gut or biopsy-proven CD. Controls were recruited from children ages 0–17.9 years with normal celiac serology, and no known inflammatory disorders of the gut. Metadata collected consisted of anthropometrics, laboratory results, medications, past medical history, current symptoms, and last oral intake prior to ingesting the sugars.

Because tTG IgA serologies lack universal normalized ratios, values were related to the upper limit of normal (ULN) used by the respective laboratories in which each test was performed. These tests were performed at either St. Louis Children's Hospital (66%) or other hospital or commercial laboratories in our region (34%).

Participants were categorized as cases or controls and then randomized (50:50) to receive either an oral solution consisting of 200 mg of monosaccharide [L-rhamnose (TCI, Portland, OR, USA) or M (Letco, Decatur, AL, USA)] and 1,000 mg of disaccharide [L, Cumberland Pharmaceuticals, Nashville, TN, USA)] in 10 mL of sterile water, prepared by the St. Louis Children's Hospital pharmacy. Fasting participants were provided the solution, and ingestion was monitored and completed within 5 min. Blood was collected when the intravenous line was placed for the anesthesia.

Pre- and 1-h post-dosing urines were collected in a sterile bag or hat depending on the age of the child. If the former was used for collection, a new bag was placed after dosing. If urine was produced <30 min after dosing, an additional bag was placed and this was used for the post-dosing specimen. Start (bag placement) and end (removal) times of bagged specimens and volume of all urines collected were recorded. Urines were kept on ice, volumes determined, and aliquots were frozen (−80°C) until shipped on dry ice to the Mayo Clinic (Rochester, MN) for assay. Sugars were quantified in urine using HPLC MS:MS as previously described (2). These values were used to calculate the concentration of sugars per mL of urine collected, the percent recovery, and L:M and L:R. Percent L was calculated across all participants, and not within groups to which the participants were randomized.

For each participant with suspected CD two and four biopsies were obtained from the bulb and the second portion of the duodenum, respectively.

All pathologists were blinded to the clinical status of the participants. Two readers initially assigned modified Marsh scores (16) for each biopsy (hereafter described as Marsh scores). If the two scores agreed, we considered this reading to be final. If scores differed by one step (e.g., 3a vs. 3b), the more severe value was used. If scores differed by more than one step (e.g., 3a vs. 3c), a third pathologist was enlisted to read the slide. If two of the three scores agreed, this was used. If all three scores varied, the middle score was used.

We used lipopolysaccharide (LPS) from strain F12 (pSK+), a TnphoA mutant of Escherichia coli O157:H7 that cannot express the O157 O-side chain (17) to measure seroreactivity to generic (core) Enterobacteriaceae LPS. LPS was prepared and characterized, and enzyme immunoassays (EIAs) were optimized and performed, as described (18), with the following changes: each human plasma sample was diluted 1:1,000 in 0.5% BSA in phosphate buffered saline containing 0.05% Tween-20 and added to wells in triplicate. Anti-human IgG F(ab')₂-horseradish peroxidase (Millipore, Burlington, MA, USA) at a 1:10,000 dilution served as the second antibody. Plates were developed by adding peroxidase substrate tetramethyl benzidine (ThermoFisher, Waltham, MA, USA) (15 min, room temperature). Reactions were stopped with 2M H₂SO₄, and absorbance was measured at 450 nm. We constructed a standard curve on each plate using serial dilutions (1:2,000–1:64,000) of serum from a patient recently infected with E. coli O157:H7 that was known to be highly reactive to the core LPS. The reactivity of the positive control (1:2,000 dilution) was arbitrarily defined as 1,000 EIA units. All samples were run in triplicate.

We used commercial EIAs to measure concentrations of circulating LPS binding protein (LBP) (HyCult, Wayne, PA, USA) and α-1-acid glycoprotein (α-1AGP) and C-reactive protein (CRP) (R&D Systems, Minneapolis, MN, USA), according to the manufacturers' instructions. Values above the upper limit of quantification for CRP were designated as 5,000 ng/mL.

Group differences were analyzed by the Kruskal-Wallis one-way analysis of variance, with post-hoc testing and correction for multiple comparisons. Correlation analysis was performed using Spearman's r. The significance of differences between continuous and categorical variables were determined by the Mann-Whitney and Fisher's exact test, respectively. Receiver operating characteristic (ROC) curves and area under the curve were calculated to determine the ability of L:M and L:R to distinguish controls from those with Marsh scores of 3a or higher. All p-values were two sided, and α-value <0.05 was considered significant. Analyses were performed with Prism 9.3.1 (GraphPad).

We enrolled 54 cases (50 and four with elevated tTG IgA or tTG IgG, respectively), and 10 controls (Table 1). Endoscopies were performed on controls because of chronic abdominal pain, nausea, and/or vomiting.

In one control participant, the normal endoscopy was grossly normal, and the rapid urease test was negative, but Helicobacter pylori infection of the stomach was diagnosed on immunostaining.

The Marsh scores assigned by two pathologists agreed perfectly in 27 of the 54 cases, and differed by one step in 19. A third pathologist resolved the scores with greater discrepancies without knowing the values of the first pair of readings. If two of the three scores agreed, that value was used (n = 6) (with agreement defined as being identical). If three different values were generated, we used the intermediate score (n = 2).

Of the 50 participants with elevated tTG IgA concentrations, 11 (22%) had Marsh 0 and two (4%) had Marsh 1 histology. The remaining Marsh scores were ≥3a. Each of the 4 case participants with abnormal tTG IgG concentrations (but normal tTG IgA) had Marsh 0 (3) or 1 (1) histology. Marsh scores did not vary according to interval between diagnostic serology and biopsies (Kruskal-Wallis p = 0.55; Supplementary Table 1). The median (IQR) intervals between obtaining the tTG IgA and biopsy did not significantly differ for children with and without type 1 diabetes [34 (17–62) vs. 42 (40–54) days].

Overall, the circulating tTG IgA concentration significantly varied with Marsh scores (Figure 1) among cases (p = 0.03). Ancillary celiac screening tests were obtained in a subset of participants at the discretion of the treating physician (Supplementary Table 2). Anti tTG IgG values were normal in four of 14 cases tested. Anti-endomysial antibody was negative in one case participant whose Marsh score was 3b. Anti-deamidated gliadin peptide (DGP) IgA was negative in two of the eight cases in which it was measured. Anti-DGP IgG was elevated in each of the four cases in whom a circulating concentration was determined. The median (IQR) time between type 1 diabetes diagnosis and elevated serology was 2 (1–219) days.

Six controls and 26 seropositive children were randomized to L and M, while four controls and 28 seropositive cases were randomized to L and R (Supplementary Table 3). One 2 year-old randomized to L:M and one 12 year-old randomized to L:R did not void for over 5 h after ingestion. One 3 year-old randomized to L:R ingested only half of the L:R solution. These three participants were removed from analysis of the sugar probes. Urines were obtained at medians of 65 min (IQR: 60–80) after ingesting sugar in the L:R group, and 67 min (IQR = 60–73) in the L:M group (Mann-Whitney, p = ns).

L:R (Kruskal-Wallis p = 0.01) but not L:M (Kruskal-Wallis p = 0.08) significantly differed according to the Marsh scores (Figure 2). We also combined cases with Marsh 2–3C scores and compared those to controls, and again found no significant association with L:M (adjusted post-hoc Dunn test p = 0.05), but did find an association with L:R (adjusted post-hoc Dunn test p = 0.01). Additionally, we calculated the area under the curve when comparing controls vs. those with Marsh 3a or higher for both L:M and L:R. The area under the ROC curve (AUC) for L:R was greater than for L:M (0.89 vs. 0.78). We also evaluated the sensitivity and specificity of L:M and L:R at various cutoffs to distinguish controls from those with Marsh scores of 3a or greater. For L:R the following cutoffs give the corresponding sensitivity and sensitivity: 0.098 (sensitivity 94%, specificity 50%), 0.148 (sensitivity 89%, specificity 75%), and 0.262 (sensitivity 78%, specificity 100%). For L:M the following cutoffs give the corresponding sensitivity and sensitivity: 0.059 (sensitivity 100%, specificity 50%), 0.088 (sensitivity 88%, specificity 67%), and 0.232 (sensitivity 38%, specificity 100%). TTG IgA concentrations correlated better with L:R (spearman r = 0.409, p = 0.03) than with L:M (spearman r = 0.382, p = 0.04). However, these represent a moderate and weak correlation, respectively.

We next determined if excretion of single sugars differed according to Marsh score and found no significant association (Supplementary Figure 1). Each of the 29 children tested with L and M who voided before the sugars were administered had detectable M in the urine. Of the 29 children who received L and R and produced pre-dosing urine, 13 (45%) had detectable R (Fisher's exact test p < 0.0001). The median (interquartile range) urinary sugar concentrations were considerably greater for M than for the 13 children in whose urines we detected R [14 μg/mL (IQR 9.3–27) vs. 0.63 μg /mL (IQR 0.44–1.0), respectively, p < 0.0001; Figure 3].

We asked if subtracting pre-administration urinary M or R improved L:M or L:R association with Marsh scores. Again, there was no significant association with L:M (Kruskal-Wallis p = 0.79), but the L:R ratios (Kruskal-Wallis p = 0.015) significantly associated (Supplementary Figure 2). We also had the opportunity to estimate the rate of clearance of baseline M by calculating the M in the urine in the post-dosing urine in the children who were tested with L:R, i.e., those who received no M. The median clearance was −6.17 μg/min (IQR −15.3 to −1.6). In two children who received R, the concentration of M actually increased between the pre- and the post-testing urines. This might be attributed to incomplete voiding in the pre-dosing determination, or surreptitious ingestion of a substance containing M soon before the pre-dosing sample was obtained. Using the median M clearance rate, we calculated a corrected cleared M concentration in the children tested with L:M. Again, we found no significant association with this corrected L:M and Marsh score (Kruskal-Wallis p = 0.17; Supplementary Figure 3).

Blood was obtained at the time of intravenous line placement for upper endoscopy in nine of the 10 controls and 44 of the 54 cases. There were no significant associations between circulating E. coli anticore LPS IgG, α-1AGP, LBP, or CRP concentrations and Marsh score or tTG IgA concentration.