This study was conducted in accordance with French guidelines on animal experimentation and validated by the Ethics Committee in Animal Experiment of INRA Jouy-en-Josas (Comethea, registration number: APAFIS#1620-2015102618572930v2). Male C57BL/6NRj mice (7 weeks old, Janvier Labs, St Bertevin, France) were individually housed and maintained in 12-h light/12-h dark cycle. Mice received, before and during the experimental procedure, a chow diet meeting their complete need (Ssniff, Soest, Germany). After 2 weeks of adaptation, the mice received as drinking solution either water (Control), 15% glucose (Glucose), or 15% fructose (Fructose) for 9 weeks. The duration and sugar concentration were based on previous studies showing modest metabolic outcomes in response to 15% fructose (38–40). Food and water were provided ad libitum. Body weight, drink, and food intakes were measured three times per week. Caloric intake measurement was calculated based on the daily average intake (in grams and milliliters) of chow (3.27 kcal/g of chow) and drinking solution (0.6 kcal/ml of glucose or fructose solution), respectively. After 9 weeks, non-starved mice were euthanized. Cardiac blood was collected under anesthesia into EDTA tubes, centrifuged (4°C, 2,000g, 15 min), and stored at −80°C for cytokine analysis. Non-flushed 2-cm sections of jejunum, cecum, and colon were sampled and stored in Krebs buffer for instant intestinal permeability measurement in Ussing chambers. Flushed sections of jejunum, cecum, and colon were sampled, flash frozen in liquid nitrogen, and stored at −80°C for protein and mRNA analysis. The cecal content was sampled and stored at −80°C for bacterial DNA analysis. Liver was sampled and weighed.

Oral glucose tolerance test (OGTT) was performed after 8 weeks of exposure to 15% glucose or 15% fructose beverage. Glucose was orally administered (2 g/kg body weight) to the mice after 12 h fasting, during which food as well as fructose- or glucose-drinking solutions were removed and replaced with water.

Shortly after sampling, intestinal sections of jejunum, cecum, and colon were opened along the mesenteric border and mounted in Ussing chambers (Physiologic Instruments, San Diego, USA) using an appropriate slider area size (0.2 cm² for jejunum and cecum, 0.1 cm² for colon). The mucosal side of the tissue was exposed to oxygenated Krebs-mannitol (10 mM) buffer, while the serosal side was exposed to oxygenated Krebs-glucose (10 mM) buffer. All buffers were maintained at 37°C. Fluorescein isothiocyanate-sulfuric acid (FITC-SA; Thermo Fisher Scientific, France) was used as a marker of TJ paracellular permeability. After adding FITC-SA (40 µg/ml) to the mucosal chamber, 100 µl was collected in duplicate from the serosal chamber every 15 min for 2 h. The concentration of FITC-SA was measured via fluorescence at excitation 485 nm and emission 538 nm. Paracellular permeability was measured by the flux of FITC-SA expressed as ng/cm²/h.

The total mRNA from jejunum, cecum, and colon sections were extracted using MirVana miRNA isolation kit (Thermo Fisher Scientific, France). Two micrograms of mRNA was reverse transcribed using the High Capacity Complementary DNA Reverse Transcription Kit (Thermo Fisher Scientific, France) as described previously (41). Quantitative real-time PCR (qRT-PCR) was performed using SYBR-Green^® PCR 2X Master Mix (Applied-Biosystems, France) to quantify the gene expression of TJ proteins and inflammatory markers. β-actin was used as a housekeeping gene to normalize the mRNA abundance of each target gene (primers are listed in Table S1 ). Expression values of target genes were calculated based on the comparative threshold cycle (Ct) method to generate ΔCt values. The relative abundance of each mRNA in each sample was then normalized according to the equation: Relative Quantity RQ = 2^−ΔΔCt.

Total protein from cecum sections was extracted using RIPA buffer (Sigma-Aldrich, France) with 4% and 10% protease inhibitor (Roche, Sigma-Aldrich, France) and phosphatase (Roche, Sigma-Aldrich, France), respectively, and quantified using the RC DC Protein Assay kit (Biorad, France). Western blot analyses were performed using 50 μg of intestine and membranes were probed with polyclonal antibodies against myosin light-chain kinase (MLCK) (Sigma-Aldrich, France) diluted 1:2,000 and then stripped and reprobed with mouse anti-GAPDH (glyceraldehyde 3-phosphate dehydrogenase) antibody (Sc-365062 Santa Cruz, USA). The blots were revealed using enhanced chemiluminescent horseradish peroxidase substrate (ECL-HRP Thermo Fisher Scientific, France) and visualized using ChemiDoc MP imaging system (Bio-Rad, France). All bands were digitally captured and densitometry analysis was performed using image Lab 5.0 software (Bio-Rad, France).

Three hundred micrograms of cecum protein extract was incubated for 2 h at RT and then overnight at 4°C with rabbit monoclonal anti-GFAP antibody (1:40, Ab68428, Abcam, France). The immune complexes were captured with 50% Protein A Agarose beads (Cell Signaling, Saint Quentin Yvelines, France) on a rotator for 2 h at RT. After centrifugation (4°C, 20,000g, 5 min), the precipitates were washed three times with RIPA buffer. The final pellets were suspended in 30 µl of 2× loading buffer and incubated at 95°C for 15 min. The entire suspension was loaded onto a SDS-PAGE gel, and the Western blot for GFAP (1:10,000) was performed as described above and using peroxidase-labeled secondary anti-rabbit antibody (1:400, Abcam, France). Blot image analysis was performed as described above.

IL6, TNFα, and IFNγ levels were determined by ELISA (Peprotech, France) using 50 µl of plasma diluted 1:2.

Caco-2BBe cells (kindly donated by JR. Turner, Harvard Medical School) and cultured in high-glucose (25 mM) Dulbecco’s modified Eagle’s medium (DMEM, Gibco, Thermo Fisher Scientific, France) with 10% fetal bovine serum (FBS, Sigma-Aldrich, France) and 10 ng/ml human transferrin. Caco-2BBe cells were seeded at 2 × 10⁵ cells/well and differentiated as monolayers on collagen-coated polycarbonate membrane transwells with 0.3-μm pores (Thermo Fisher Scientific, France) for 15–18 days after confluence. DMEM medium was renewed every 3 days. Transepithelial resistance (TEER) was measured every 3 days with Voltohmmeter instrument (Millicell-ERS, Merck Millipore, France). In experiment 1, after 15–18 days (when TEER was about 250 Ω.cm² and stable), the culture medium was changed to 25 mM glucose or 25 mM fructose medium with 10% FBS and 10 ng/ml transferrin in both basal and apical chambers 1-day prior to cytokine treatment (D0). In experiment 2, to determine whether glucose and fructose were necessary for IFN-γ/TNF-α-induced changes in TEER, at D0, the 25 mM glucose medium was changed to 1 mM glucose or 1 mM fructose medium. At D1, recombinant human IFN-γ (Peprotech, France) was added into glucose or fructose medium (10 ng/ml concentration) in the basal chamber for 24 h. At D2, new medium containing 50 ng/ml TNF-α (Peprotech, France) was placed in the basal chamber. Both IFN-γ and TNF-α treatments were done without manipulating the apical chamber. The TEER was measured before adding TNF-α and every hour afterward for 6 h. Results were expressed as percent of TEER of control monolayers (cells exposed to glucose or fructose for 48h without IFN-γ and TNF-α) of the same experiment.

Total DNA was extracted from cecal content samples using GNOME DNA isolation kit (MP Biomedicals, Strasbourg, France). The V3–V4 hypervariable region of the bacterial 16S rDNA was amplified by PCR (forward primer: 5’-CTTTCCCTACACGACGCTCTTCCGATCTACGGRAGGCAGCAG-3’, reverse primer: 5’-GGAGTTCAGACGTGTGCTCTTCCGATCTTACCAGGGTATCTAATCCT-3’; 94°C for 1 min and then 30 cycles at 94°C for 1 min, 65°C for 1 min, and 72°C for 1 min before a final step at 72°C for 10 min) (42). After quality checking by electrophoretic 2% agarose gel migration, obtained amplicons were sequenced using Illumina MiSeq technology (GenoToul platform, Toulouse, France). Paired-end reads obtained from MiSeq sequencing were analyzed using the Galaxy-supported pipeline named FROGS [Find, Rapidly, OTUs (operational taxonomic units) with Galaxy Solution] (43). For the preprocessing, reads with length ≥ 380 bp were kept. The clustering and chimera removal tools followed the guidelines of FROGS (43). Assignment was performed using SILVA 16 S. OTUs with abundances lower than 0.005% were removed from the analysis (44). Real-time PCR analyses were performed to quantify Desulfovibrio genus abundance. Quantitative PCR reactions were conducted in a final volume of 25 μl using SYBR-Green^® PCR 2X Master Mix (Applied-Biosystems, France) and with 0.2 μM final concentration of each primer (DSV691-F CCGTAGATATCTGGAGGAACATCAG and DSV826-R: ACATCTAGCATCCATCGTTTACAGC) and 5 μl of DNA samples (45).

Bacteroides vulgatus (ATCC 8482) and Lactobacillus johnsonii (strain CIP 103620) were grown in DP2 growth medium enriched with 5 g/L of glucose or 5 g/L of fructose (46). Desulfovibrio vulgaris subsp. vulgaris (DSM 644) was grown in appropriate Medium 63 containing either glucose (5 g/L) or fructose (5 g/L). All bacteria were cultured in anaerobic condition at 37°C using the Hungate culture method. For each bacterial species, 24 h growth media (containing no fructose) were inoculated with a 1:9 dilution to a duplicate set of young culture containing glucose or fructose. The growth of B. vulgatus and L. johnsonii was measured every 30 min or 1 h by absorbance (A660) for 8–10 h. For D. vulgaris after 4 h and 9 h, the DNA was extracted using GNOME DNA isolation kit (MP Biomedicals, Strasbourg, France) and the growth was measured by qPCR using the primers DSV691-F CCGTAGATATCTGGAGGAACATCAG and DSV826-R: ACATCTAGCATCCATCGTTTACAGC. The total number of bacteria was inferred from averaged standard curves as previously described (47).

Statistical analysis was performed using GraphPad Prism software (v7; San Diego, CA). All data were analyzed using Kruskal–Wallis test followed by a Dunn’s multiple comparison test. 16S rDNA sequencing data were analyzed using the Phyloseq package in R and custom scripts as described previously (48). Briefly, 16S sequencing data were analyzed using the Phyloseq (49), ggplot2 (50), DESeq2 (51), and custom scripts. Samples were rarefied to even sampling depths before α-diversities (observed richness and Inverse Simpson) and β-diversity (Bray–Curtis) analysis. Principal coordinates analysis (PCoA) was performed on Bray–Curtis dissimilarities. α-diversity data were analyzed using Kruskal–Wallis test. Permutational multivariate analysis of variance (PERMANOVA) test was performed on the Bray–Curtis matrices using 999 random permutations and at a significance level of 0.01. Phylum and family abundances data were compared using Kruskal–Wallis test. Raw, unrarefied OTU (also named clusters) counts were used to produce relative abundance graphs and to find taxa with significantly different abundances between two groups. DESeq2 was used to estimate abundance log-fold changes (logfc) between glucose/control, fructose/control, or glucose/fructose groups. Clusters were selected based on effect size (fold change (FC) > 1.5 or FC < 1/1.5) and adjusted p-value (<0.05). Pro-inflammatory cytokine gene expression levels (TNFα, IL13, Il22, and IFNγ) and normalized bacterial abundance (52) were integrated and analyzed using Data Integration Analysis for Biomarker discovery using a Latent cOmponents (DIABLO) method implemented in R mixOmics package (53). Correlation analysis between cytokines gene expression and bacterial species abondance was performed using the multiblock analysis DIABLO (block.splsda function) (54). Horizontal sparse partial least squares-discriminant analysis (sPLS-DA) was used to integrate the relative abundances of OTU (clusters) with the cytokine gene expression levels. To perform the analysis, 5% of the OTU (which were the most involved in the group discrimination) were selected. The level of correlation between the OTU and the cytokine gene expression level was determined and the resulting correlation network showing only positive (>0.83) and negative (<−0.83) correlations between the selected variables was built in DIABLO.

Mice-fed glucose solution gained 15% more weight than the fructose- and water-drinking mice after 9 weeks of experiment ( Figure 1A ). This likely resulted from a significantly higher total daily calorie intake by the glucose-drinking mice when compared to mice receiving fructose or water solutions ( Figure 1B ). Eight weeks of chronic intake of glucose impaired glucose tolerance both in the individual points of the curve ( Figure 1C ) and in the AUC ( Figure 1D ) when compared to control and fructose groups. Nine weeks of chronic intake of fructose was associated with a significant increase in plasmatic levels of IL6 when compared to the control group ( Figure 1E ) while the plasmatic levels of TNFα and IFNγ remained unchanged among the three groups ( Figures 1F, G ). After 9 weeks, glucose-drinking mice also displayed a significantly higher visceral adiposity (2.94% ± 0.19) than the fructose-drinking (1.66% ± 0.22) or control (1.52% ± 0.14) groups. There was no difference in liver weight among the three groups (data not shown). Despite a major effect of glucose intake on body weight gain and glucose tolerance, only fructose initiated an increase in circulating levels of pro-inflammatory cytokines.

Earlier studies suggested that barrier defects contribute to the establishment of dietary-associated chronic low-grade inflammation (55, 56). To determine whether fructose-induced systemic inflammation was associated with alteration of intestinal permeability, we evaluated this parameter in the small (jejunum) and large intestine (cecum and proximal colon) ex vivo using the Ussing chamber system. Nine weeks of glucose consumption significantly increased paracellular permeability to FITC-SA in both jejunum and cecum when compared to the fructose groups ( Figures 2A1, A2 ). Glucose intake also lead to a trend toward increased permeability in jejunum and cecum when compared to the control group. In the colon, neither glucose nor fructose changed the paracellular permeability ( Figure 2A3 ). In the jejunum, glucose-induced changes in permeability were not associated with changes in Mlck and TJ protein expression levels ( Figure 2B1 ). Of the pro-inflammatory cytokines tested, only Il1β expression was significantly upregulated in the jejunum of the mice consuming glucose when compared to the control and fructose groups ( Figure 2C1 ). In the cecum, glucose-induced increase in permeability was associated with a significant increase in Mlck expression when compared to the control group ( Figure 2B2 ). In contrast, fructose-drinking mice displayed a significant decrease in expression of claudin2 and occludin relative to the control and glucose groups respectively. In cecum, both glucose and fructose significantly increased the expression of Tnfα, Il13, Ifnγ, and Il10 when compared to the control group ( Figure 2C2 ). In addition, glucose intake was also associated with a significant increase in Il22 expression in the cecum when compared to the mice of the control group. Only Il1β expression remained unchanged among the three groups. In the colon, fructose moderately increased claudin2 expression while the expression of the other TJ proteins remained unchanged in response to glucose or fructose ( Figure 2B3 ). Likewise, none of the pro-inflammatory cytokines tested displayed any changes in expression among the three groups ( Figure 2C3 ). Il6 expression levels remain unchanged among the three groups in the jejunum, cecum, and ileum (data not shown). Western blot showed that in the cecum, MLCK protein level increased significantly in response to glucose and fructose ( Figure 3 ). Thus, in the cecum, both glucose and fructose caused similar upregulation of pro-inflammatory cytokine expression and MLCK protein levels, but only glucose was associated with increased permeability.

In addition to the above gut homeostasis effects, glucose and fructose intake significantly upregulated in jejunum the expression of the apical sodium/glucose cotransporter, Sglt1, and of the basolateral glucose transporter, Glut2, but only fructose drastically upregulated the expression of its apical transporter, Glut5 ( Figure S1A ). In the cecum, glucose intake was associated with an increase in expression levels of both glucose transporters, Sglt1 and Glut2, while the fructose intake did not change Glut5, Sglt1, or Glut2 mRNA levels ( Figure S1B ). These data suggest that glucose and fructose transport levels are likely different between the jejunum and cecum.

Usually, pro-inflammatory cytokines such as IL13, TNFα, and IFNγ are associated with increased paracellular permeability (22, 57). Yet, only glucose, but not fructose increased the permeability in the cecum. To clarify this surprising observation, we investigated the impact of fructose and glucose on the EGCs of the ENS, which has been described as contributing to the maintenance of epithelial barrier function (36). GFAP protein levels were quantified in full thickness preparation from cecum segment. GFAP protein expression level significantly increased in cecum of mice drinking fructose when compared to the cecum of control mice ( Figures 4A, B ), suggesting active gliosis in response to fructose.

To investigate further the differential effects of glucose and fructose on intestinal permeability and to determine whether the two monosaccharides act directly on enterocytes, we used the Caco-2 cell line since these cells grow as polarized monolayers and express sugar transporters at the apical membrane (58). They have been extensively used as a model to study TJ regulation in response to pro-inflammatory cytokines (22, 59, 60). The permeability of tight junctions was measured by monitoring TEER. After differentiation of the Caco-2, enterocyte monolayers were cultured in the presence of 25 mM glucose or fructose for 48 h. Then, the TEER was measured in response to IFNγ priming followed by TNFα treatment. In the absence of pro-inflammatory challenge, 25 mM fructose did not affect the TEER when compared to 25 mM glucose ( Figure S2 ). In both glucose and fructose culture conditions, TNFα/IFNγ caused a time-dependent decrease in Caco-2 TEER ( Figure 5 ). However, when compared to glucose, fructose significantly attenuated the TEER decrease in response to cytokine stimulation after 5 h. The lack of effect of TNFα/IFNγ treatment on TEER when the glucose or fructose concentrations were reduced to 1 mM suggests that the sugar metabolism is a key factor enabling pro-inflammatory cytokines to reduce TEER ( Figure S2 ).

Several studies have reported that diet-induced changes in gut microbiota composition contribute to the intestinal inflammation and alteration of the gut permeability (37). Therefore, we investigated the effects of glucose- and fructose-drinking treatment on the cecal ecosystem composition using microbial 16S rRNA sequencing. The α-diversity was measured at the OTU level using the richness (observed species) and inverse Simpson index. Both parameters showed no significant change in richness and α-diversity of the microbiota in response to glucose or fructose intake ( Figures 6A, B ). However, PCoA of Bray–Curtis compositional dissimilarity between cecum samples from each group showed a significant separation between the communities in response to glucose and fructose when compared to the control group (p = 0.003 and p = 0.004, respectively) as well as between glucose and fructose (p = 0.007) ( Figure 6C ). Glucose-drinking mice demonstrated a more significant shift in cecal microbial composition than fructose when compared to the control mice. Firmicutes relative abundance decreased in the glucose group when compared to control while the relative abundance of Proteobacteria increased significantly in the cecal community of the glucose group when compared to control ( Figure 6D ). At a lower taxonomic level, within the Proteobacteria phylum, the relative abundance of Desulfovibrionaceae significantly increased in the cecal content of the glucose group when compared to control groups ( Figure 6E ). Within the Firmicutes phylum, the fructose intake increased the relative abundance of Lactobacillaceae when compared to the glucose group. Within the Bacteroidetes phylum, fructose modestly decreased the relative abundance of Bacteroidaceae when compared to the control group. The increase in Desulfovibrionaceae indicates that glucose consumption modified microbiota composition towards pro-inflammatory profile while the higher abundance in Lactobacillaceae suggests that beneficial microbiota environment was associated with fructose chronic intake.

Differential abundance analysis identified 34 significant (p < 0.05) species or clusters that were positively (25 clusters) or negatively (9 clusters) differentially abundant between glucose and control groups ( Figure 7A ). Among them, 7 clusters exhibited a fold change > 32 and 11 of them belong to Lachnospiraceae family. This differential analysis also highlighted cluster_4 belonging to the Desulfovibrio genus. This cluster related to the most abundant species identified by DESeq2 (base mean = 719, Table S2 ), suggesting that the increase in Desulfovibrionaceae abundance in response to glucose identified previously (and confirmed by qRT-PCR) ( Figure S3 ) was likely related to the increase in cluster_4 abundance ( Figures 7A, B ). Similarly, 46 species or clusters were positively (33 clusters) or negatively (13 clusters) differentially abundant between fructose and control groups and only 15 of them were similar to those identified as differentially abundant in response to glucose ( Figure 7C ). Among the species showing increased differential abundance in response to fructose, three species (cluster_253, cluster_208, and cluster_217) belong to the Lactobacillus family, which appears as a signature of the fructose intake ( Figures 7B, C ). Six species were positively (among which five species belong to Lachnospiracae family and one, cluster_4, to Desulfovibrionaceae family) and seven were negatively (among which the clutser_217 was from Lactobacillaceae family) differentially abundant between glucose and fructose groups ( Figure 7B ).

When ingested as liquid, glucose and fructose are not fully absorbed by the enterocytes and reach the lumen of the lower intestine (37, 61). In vivo, changes in the abundance of specific bacterial species under glucose or fructose suggest different abilities of the bacteria to use directly these two monosaccharides to promote their growth. We tested this hypothesis for the bacteria that showed differential abundance in vivo and that can be grown in culture by assessing their growth in vitro in the presence of either glucose or fructose. These bacteria were (1) either representative of bacteria that had an increased relative abundance in response to both glucose and fructose intake in comparison to the control group (cluster_191 identified as B. vulgatus, Table S2 ), (2) bacteria that displayed an enrichment specifically in response to fructose (L. johnsonii characteristic of Lactobacillus genus and corresponding to cluster_253), or (3) bacteria that displayed an enrichment specifically in response to glucose (as D. vulgaris characteristic of Desulfovibrio genus and corresponding likely to cluster_4). B. vulgatus, L. johnsonii, and D. vulgaris were all able to grow on either glucose- or fructose-containing media ( Figure S4 ) and for all three species, growth was promoted by glucose in microbial culture media when compared to fructose.

Multi-group sPLS-DA was performed to identify the OTU that discriminated for components 1 and 2 between glucose, fructose, and control groups ( Figure S5 ). Using a cutoff of ≥0.6, 25 bacterial species (clusters) were correlated with four pro-inflammatory cytokines’ (IL22, IL13, TNFα, and IFNγ) gene expression levels ( Figure 8 ). Two clusters belonging to the Clostridiales order were positively linked to three cytokines: cluster_66 (Lachnospiraceae family) correlated to IL22, IFNγ, and IL13 and cluster_372 (Ruminococcaceae family) correlated with IL22, TNFα, and IL13. Four other clusters, all members of the Clostridiales order, were correlating positively with IL22 and IL13 (cluster_291, cluster_637, cluster_406, and cluster_821) while cluster_95 (Oscillibacter genus) was strongly negatively correlated to those. Cluster_4 (Desulfovibrio genus) was linked with Il22 and IFNγ while cluster_568 (Clostridiales vadinBB60 family) was associated with TFNα and IL13. In addition to cluster_95, two other members of the Bacteroidales S24-7 group were negatively correlated to IL13 (cluster_490 and _360). Four clusters were also negatively associated to TNFα, among which two, cluster_240 (Ruminococcaceae family) and cluster_389 (Lachnospiraceae family), display a negative fold change in response to fructose intake ( Figure 7C ). Interestingly cluster_4, cluster_66, cluster_637, cluster_40 (Lachnospiraceae family), and cluster_80 (Bacteroidales S24-7 group family) were among the bacterial species significantly upregulated in response to glucose ( Figures 7A, B ), highlighting the potential role of glucose-induced gut microbiota changes in the development of inflammation in the epithelium of the lower GI tract.