Grain of treated wheat (according to the patented method PCT n. PCT/IB2013/000797) was milled commercially and kindly supplied by Casillo Group S.p.A¹. The treatment was performed as follows: 100 g of cleaned wheat grains was dampened until reaching 15–18% humidity, which was measured by a halogen thermal balance (Mettler Toledo, HB43-S, Switzerland), and subjected to rapid heating via microwaves (Delonghi, Italy; approximately 1 min between 1000 and 750 watts), followed by slow evaporation of the water. The rapid heating and subsequent slow evaporation of the water was repeated until reaching a final temperature of 80–90°C, which was measured by a thermal camera (Fluke, i20 Model, Italy), and a moisture degree of 13–13.5% in the wheat grains.

After microwave treatment, the wheat kernels were cooled and dried at room temperature (24°C) for 12–24 h and then ground using an automatic laboratory mill MCKA (Bühler AG, Uzwil, Switzerland; diameter of grid 118–180 μm) (Bevilacqua et al., 2016).

Gluten Friendly bread was prepared according to the bread-making process (100 g wheat flour, 66 mL water, 1.33 g yeast, 1 g salt) to the Chorleywood Bread Process (Bevilacqua et al., 2016).

Prior to being added to the in vitro colonic system, GFB was digested in vitro, under appropriate conditions according to the procedures described by Maccaferri et al. (2012) to mimic mouth, stomach, and intestine’s condition.

Fecal samples were obtained from two healthy and two celiac donors (male and female aged 30–50 years old) who were free of any metabolic and gastrointestinal diseases, were not taking probiotic or prebiotic supplements, and had not taken antibiotics 6 months before fecal sample donation. All donors have provided written informed consent and filled in a standard questionnaire to collect information regarding health status, drug use, clinical anamnesis, and lifestyle factors. The University of Reading Research Ethics Committee (UREC 15/20) approved this study in accordance with the Declaration of Helsinki. Sample size was in accordance with previous studies (Bevilacqua et al., 2016). Fecal samples were placed in an anaerobic jar (AnaeroJar^TM 2.5 L, Oxoid Ltd.) including a Gas-Generating Kit (AnaeroGen^TM, Oxoid Ltd.) in order to reproduce anaerobic conditions inside the chamber. An aliquot of 20 g of samples was diluted in 100 mL anaerobic PBS (0.1 mol/L phosphate buffer solution, pH 7.4, w/w) and homogenized (Stomacher 400; Seward, West Sussex, UK) for 2 min at 240 paddle beats per minute. Samples were added to anaerobic fermenters within 15 min of voiding.

Physicochemical conditions in the colon were replicated in a three-stage continuous system, comprised of a cascade of three glass fermenters of increasing working volume connected in series. A small-scale version of the validated system by Macfarlane et al. (1998) has been used in this study, as suggested by Grimaldi et al. (2017). The gut model is often used to resemble the complexity and diversity of the intestinal microbiota and simulates the proximal [vessel 1 (V1), 80 mL, pH 5.5], transverse [vessel 2 (V2), 100 mL, pH 6.2], and distal colon [vessel 3 (V3), 120 mL, pH 6.8] (Figure 1); the vessels were filled with fecal homogenate from healthy and celiac volunteers (20%) and a complex colonic growth medium (80%). The growth medium contained the following ingredients: starch, 5 g/L; mucin, 4 g/L; casein, 3 g/L; peptone water, 5 g/L; tryptone water, 5 g/L; bile salts, 0.4 g/L; yeast exact, 4.5 g/L; FeSO₄, 0.005 g/L; NaCl, 4.5 g/L; KCl, 4.5 g/L; KH₂PO₄, 0.5 g/L; MgSO₄ × 7H₂O, 1.25 g/L; CaCl₂ × 6H₂O, 0.15 g/L; NaHCO₃, 1.5 g/L; Tween 80, 1 mL; hemin, 0.05 g/L; and cysteine HCl, 0.8 g/L. Following inoculation, the colonic model was run as a batch culture for 24 h to stabilize the bacterial populations prior to the initiation of medium flow. After 24 h (T₀), the medium flow was initiated and the system was ran for eight full volume turnovers to allow for steady state to be achieved (SS1) assessed through stabilization of the SCFA profiles (±5%); the duration of each turnover was 48 h, thus this first step was done for 384 h. The flow rate of the system was 6.25 mL/h. The system was run for further eight volume turnovers upon which steady-state 2 (SS2) was achieved; daily 3.75 mL of digested GFB was added. Each steady state was confirmed by stabilization of SCFAs profiles over 3 consecutive days. SS2 was achieved after 384 h; the whole experiments took place for 792 h. Each vessel was magnetically stirred and continually sparged with oxygen free nitrogen gas. Temperature (37°C) was maintained by a water-cooling system and culture pH was controlled automatically through the addition of 1 N NaOH or HCl. Aliquots of 4.5 mL were removed at SS1 and SS2 from each vessel (V1, V2, and V3).

The production of acetic, propionic, and butyric acids in the fermentations was determined by HPLC (Merck, NJ, United States) equipped with RI detection. The column used was an ion-exclusion REZEX-ROA organic acid column (Phenomenex Inc., United Kingdom) and temperature maintained at 84°C. Sulfuric acid in HPLC-grade H₂O (0.0025 mmol/L) was used as the eluent, and the flow rate was maintained at 0.5 mL/min. Aliquots of 1 mL collected from each vessel in microcentrifuge tubes were centrifuged at 1136 × g for 10 min to remove all particulate matter. The supernatants were then filtered using 0.22 μm low protein binding Durapore polyvinylidene fluoride (PVDF) membranes (Millex; EMD Millipore, Billerica, MA, United States). Twenty microliters of each sample was injected with a run of 45 min. Peaks were integrated using the Atlas Lab Managing Software (Thermo Lab Systems, Mainz, Germany). Quantification of the samples was obtained through calibration curves of acetic, propionic, and butyric acids in concentrations ranging between 12.5 and 100 mM.

Fluorescence in situ hybridization analysis was performed as described by Costabile et al. (2014). Briefly, aliquots (375 μL) of gut model samples were fixed in three volumes of ice-cold 4% (w/v) paraformaldehyde for 4 h at 4°C. They were then centrifuged at 13,000 × g for 5 min and washed twice in 1 mL of sterile PBS. The cells were again pelleted by centrifugation and re-suspended in 150 μL of sterile PBS, to which 150 μL of ethanol was added. Samples were then vortexed and stored at -20°C until used in hybridizations. For the hybridizations, all probes were commercially synthesized and 5′-labeled with the fluorescent dye (Sigma-Aldrich, St. Louis, MO, United States). The probes used in this study are detailed in Table 1.

The analyses were performed on two independent batches (two healthy donors and two celiac subjects); for each batch, the experiments were done in triplicate. Bacterial counts and SCFAs were statistically analyzed using paired t-tests to assess the effect of the same treatment at different time points. Significant differences were defined at p < 0.05. All analyses were performed using GraphPad Prism 6.0.

Furthermore, FISH and SCFAs data were standardized and reported as increase/decrease of SS2 (steady state at the end of the assay) relative to SS1 (steady state before the addition of GFB); each donor was treated as a separate “sample.”

First, a principal component analysis was run by using all SCFA and microbial groups as leading variables. The analysis was done following the traditional approach of PCA to qualitatively study the effect of the supplementation of GFB on each donor in each vessel.

Thereafter, the PCA was modified and run through the approach of the Multivariate Statistical Process Control/NIPALS algorithm (non-linear iterative partial least square); the number of interactions was set to 50. For this second analysis, the leading variables were either standardized SCFA or standardized microbial counts. The main outputs of this second approach are the standardized plots, showing the cases (i.e., each donor per each vessel), the line of the leading variables (the microbial groups for the first analysis and the SCFAs for the second analysis) and for each variable the minimum and maximum values. The benefit of this approach is that it allows the reader to point out both a qualitatively and a quantitatively grouping of cases. PCA and modified PCA were done through the software Statistica for Windows, ver. 12.0 (Statsoft, Tulsa, OK, United States).

Changes in the bacterial composition are in Figure 2; these results focused on the global effect of GFB, without considering a possible variability linked to donors. A significant increase in numbers of bacteria within the Bifidobacterium genus was found; fecal microbiota of celiac donors showed a significant increase in this population from 8.42 to 8.90 log CFU/mL (p < 0.05) and from 8.60 to 9.20 log CFU/mL (p < 0.05) in V2 and V3, respectively, transverse and distal colon. In the fecal samples of healthy volunteers, there was a significant increase in numbers of bifidobacteria (BIF) in V3, from 7.90 to 8.40 log CFU/mL (p < 0.05). Significant increases were also found in numbers of bacteria in all vessels for celiac volunteers in Clostridium cluster population, from 8.85 to 9.50 (V1), from 9.1 to 9.60 (V2), and from 9.00 to 9.50 log CFU/mL (V3) (p < 0.05) (ERE).

The results of SCFAs are in Figure 3. The fermentation of GFB by the fecal microbiota of healthy donors showed significant differences of acetic acid in all vessels, from 28.80 to 22.10 (V1) (p < 0.01), from 44.40 to 56.94 (V2) (p < 0.01), and from 46.00 to 76.50 mM (V3) (p < 0.001), respectively; butyrate also increased. Regarding the fecal microbiota of celiac volunteers, significant increases of propionic production were found in V1 and V2, from 45.10 to 69.20 (p < 0.01) and from 50.80 to 70.20 mM (p < 0.05), respectively; also, acetic acid showed a significant increase in V1, from 41.20 to 89.00 mM (p < 0.01).

The last approach was a focus on the multivariate differences induced by the supplementation of GFB. The results were reported as increase/decrease in FISH, and SCFAs (difference SS2-SS1). Figure 4 shows the Principal Component Analysis on FISH and SCFA variations; all acids and microbial counts were used as leading variables. A difference from the previous approach (Figures 2, 3) was the use of each donor as a separate sample to highlight the differences due to the different fecal microbiota (effect of “donor”). Figure 4A shows the projection of the leading variables. Bifidobacteria (BIF) and Bacteroides/Prevotella (BAC) were mostly related to the factor 2 (the correlation coefficients were 0.918 and 0.874, respectively), whereas Lactobacillus/Enterococcus (LAB), Clostridium (EREC), and propionate (PRO) were related to the factor 1 (correlation coefficients ranging from 0.804 to 0.934). This first PCA suggests an effect of GFB on the whole ecosystem (fecal microbiota and produced acids); however, this effect could be also affected by the initial situation of the system, in this analysis expressed as “the donor.” GFB exerted a strong effect on the fecal microbiota of the celiac donor 2, as suggested by the shift from c2–v1 (acids and fecal microbiota of the celiac donor 2 in the first vessel) to c2–v2 and c2–v3. The increase of BIF played a major role for this increase. On the other hand, the same effect was not found in the vessels inoculated with the fecal microbiota of the celiac donor 1. A modification of the system was also found in the vessels inoculated with the fecal microbiota of healthy donor 1, as a slight shift toward right was found (from v1 to v2–v3).

This first approach suggests that GFB could induce a shift and a modification of the ecosystem; however, due to the high number of the leading variables (nine), the accounted variability explained by the statistic is quite low (ca. 57%). Therefore, a second approach was used by using either the data of FISH and on SCFA; in addition, PCA was modified with NIPALS algorithm to achieve quantitative data for each vessel and see what is the effect of the supplementation on the evolution from V1 to V2 and V3 and simulate what could happen from a region to another of the colon.

Figure 5A shows the results for the FISH. In the fecal microbiota of the healthy donor 1, the effect of the supplementation could be seen in the vessel 1, without differences from the vessels 2 and 3. On the other hand, the fecal microbiota of the healthy donor 2 experienced some differences, as from V1 to V2–V3 there was an effect on BIF, with a slight increase (ca. 0.4 log CFU/mL).

The most important differences were found for the fecal microbiota of the celiac donors: in the donor 2, there was a shift from vessel 1 to vessels 2 and 3, probably due to BIF. In the vessel 1, there was a decrease of BIF (ca. 1 log CFU/mL), while in the vessels 2 and 3 BIF were restored with a mean decrease of 0.0–0.2 log CFU/mL.

The effect on the fecal microbiota of the celiac donor 1 was different. A slight effect on BIF was found in the vessel 1 and maintained in the vessels 2 and 3 (0.5–0.8 log CFU/mL), but a shift was found due to the different effect on Lactobacillus/Enterococcus and Clostridium cluster XIVa+b, as in the vessel 1 the increase was not significant (0.2 log CFU/mL), but in the vessel 3 the mean increase was 0.6–0.8 log CFU/mL. The difference in lactobacilli population was not significant when the mean between the two donors was done; on the other hand, a preliminary t-test revealed that on the donor 1 it was significant (p < 0.01).

Figure 5B shows the NIPALS approach for the SCFA; the main difference was in c2–v2 (vessel 2 inoculated with the fecal microbiota of the celiac donor 2). In this vessel, there was a strong increase of butyrate (ca. 45–50 mmol/L).