Female NOD and BALB/c mice were purchased from Charles River Laboratories. All mice were maintained under specific pathogen-free conditions in the animal facility at San Raffaele Scientific Institute and all experiments were conducted in accordance with the Institutional Animal Care and Use Committee (IACUC) according with the rules of the Italian Ministry of Health.

Immediately after weaning, 4-weeks old NOD mice were separated in two groups, housed in different cages and fed with either an anti-inflammatory diet (AID) or standard (STD) diet. The AID was designed with the supervision of a nutritionist and purchased from Rieper S.p.A. (Altromin C1000). The AID diet was enriched with fish oil (3,6%), linseed oil (2%) and corn oil (7,7%), for a total of 13,3% of fats (1:3 PUFA ω3:ω6 ratio with specifically 20g/kg ω3 and 60g/kg ω6). The STD diet contained only 5% of fats (100% Corn oil) with 1:190 PUFA ω3:ω6 ratio (0,15g/kg ω3 and 28,5g/kg ω6). AID diet contained 8% of fibers (100% Inulin) while the STD diet with only 5% of fibers (100% cellulose). For more information about composition of the AID see Supplementary Table S1 .

Diabetes was monitored by weekly measurements of blood glucose levels with a GB35 Ascensia Breeze 2 glucometer (Bayer). The animals were considered diabetic after two consecutive blood glucose measurements > 250mg/dl according to standard method (35).

Intestinal permeability was determined by FITC-dextran assay as previously described (36). Briefly, 20 mL/kg of body weight of PBS containing 25 mg/mL FITC-conjugated dextran (FITC-dextran; molecular mass, 4.4 kDa; FD4, Sigma-Aldrich) was administered to each mouse by oral gavage. After 4 h, blood was collected and the concentration of fluorescein was determined by spectrophotofluorometry (Wallac Victor; Perkin-Elmer Life Sciences) with an excitation wavelength of 485 nm and an emission wavelength of 530 nm using serially diluted samples of the FITC-dextran marker as standard.

Intestinal mononuclear cells were isolated from the small and large intestinal lamina propria as previously described (Lefrancois and Lycke, 2001; Weigmann et al., 2007). Briefly, after the removal of Peyer’s patches, small and large intestines were flushed with PBS, opened longitudinally, and incubated twice with 5 mM EDTA and 1 mM DTT for 20 min at 37°C to remove epithelial cells and adipose tissue. Then, the intestines were cut into small pieces and digested in HBSS containing 0.5 mg/ml Collagenase D (Roche Diagnostics), 1 mg/ml Dispase II (Roche Diagnostics) and 5 U/ml DNase I (Sigma-Aldrich) for 20 min at 37°C in a shaking incubator. The digested tissues were washed, resuspended in 5 ml of 40% Percoll (Sigma-Aldrich) and overlaid on 2.5 ml of 80% Percoll in a 15-ml Falcon tube. Percoll gradient separation was performed by centrifugation at 1000 g for 20 min at 20°C. The interface cells were collected and used as intestinal lymphocytes for FACS analysis. Splenocytes and pancreatic lymph node (PLN) cells were isolated by meshing of the tissues. For isolation of intra-islet lymphocytes (IIL), freshly collected pancreata were cut into small pieces with a scissor and digested 3 times for 15min at 37°C with agitation in 3 ml/pancreas of Hank’s Balanced Salt Solution (HBSS) 1X with Ca2⁺ and Mg2⁺ complemented with 1 mg/ml of Collagenase IV (Gibco). The digestion was stopped by extensive washing with HBSS containing 5% FBS and 5-minutes incubation on ice to let clumps of undigested pancreatic tissue to settle. Intra-islet lymphocytes (IIL) were collected from the supernatants of the three digestion/incubation cycles and extensively washed in HBSS 5% FBS before use in FACS analysis.

Single cell suspensions from different organs were resuspended in staining buffer containing PBS, 1% FBS and 0.09% NaN₃ and stained with monoclonal antibodies against surface markers. For intracellular cytokine staining, single cell suspensions isolated from different organs were stimulated for 2.5 hours with the Leukocyte Activation Cocktail (BD Bioscience). Cells were first stained for surface markers, then fixed and permeabilized using the BD Cytofix/Cytoperm kit (BD Bioscience), and finally stained for intracellular cytokines. The following antibodies were used: FITC anti-mouse CD3 (17A2, BD Biosciences), PercCP-cy5.5 anti-mouse CD4 (RM4-5, BD Biosciences), APC anti-mouse INF-γ (XMG1.2 BioLegend), APC-cy7 rat anti-mouse IL-17A (TC11-18H10. BD Bioscience), PE anti-mouse IL-10 (JES5-16E3 eBioscience), eFluor 647 anti-mouse Foxp3 (FJK-16s, eBiosciences) and APC anti-mouse CD25 (PC61, BioLegend). Dead cells were stained with Fixable Viability Dye eFluor 506 (eBioscience) and excluded from the analysis. Flow cytometry was performed using FACSCanto III (BD Biosciences) and data were analyzed with FCS Express V4 software (De Novo Software). See Supplementary Figure 1 for gating strategy.

Immediately after sacrifice, mice’s colons were flushed with PBS, opened longitudinally and placed in 500μl of TRIzol reagent (Life Technologies). After homogenization with TissueRuptor (QIAGEN), RNA was extracted by adding 100μl of chloroform, precipitating the acqueous phase with 300μl of 70% ethanol and purifying RNA with RNeasy Mini Kit (QIAGEN). RNA was retrotranscribed with SuperScript III First-Strand Synthesis System following manufacturer’s instructions (Life Technologies). Real time qPCR assay was performed with SYBR Select Master Mix (Life Technologies) using primers specific for different tight junction proteins, mucins and cytokines ( Supplementary Table S2 ) on a ViiA 7 Real-Time PCR System (Life Technologies). Transcript expression was normalized against Rpl32 gene (housekeeping gene). See Supplementary Table 2 for the list of primers used in this study.

Segments of the distal colon were fixed in Carnoy’s fixative (60% dry methanol, 30% chloroform, and 10% acetic acid), washed in absolute methanol, ethanol, xylene, embedded in paraffin wax and sectioned at 6 μm. For mucus layer structure evaluation, colon sections were stained with Alcian Blue staining kit, pH 2.5 (Sigma-Aldrich) and images acquired with Axio Imager M2m Light Microscope (ZEISS). The proportion of GC in colon sections was determined within the total number of cells in randomly selected stained sections. To assess histopathological signs of diabetes, pancreata were fixed in formalin, embedded in paraffin, sectioned into 4-μm slices and then stained with hematoxylin–eosin for insulitis score.

Total DNA was isolated from the mucosa and luminal content of the large intestine of NOD mice using PowerFecalTM DNA Isolation kit (MoBio) following the manufacturers’ instructions. Microbiome characterization was performed by amplification of three regions of the 16S rRNA (V3, V4, V5) using universal primer pairs. The analysis of metabolically active microbiota was performed by pyrosequencing of rRNA cDNA 16S (GS Junior, Roche Diagnostics GmbH). Sequences with a high-quality score were used for the taxonomic analysis with QIIME (Quantitative Insights into Microbial Ecology version 1.6).

Fresh snap-frozen feces were collected from mice after 10 weeks of diet and used for metabolomics analysis. Metabolites were extracted three times using the following mixture of solvent, acetonitrile:methanol:water = 4:4:2. The collected supernatants were filtered on 96 well-plate with positive pressure manifold (0.22 μm pore size), dried under gentle stream of nitrogen and recomposed with 100 µl of water containing 0.1% formic acid. A QC sample was prepared by taking 20 µL from each sample. The samples were analysed by LC-MS/MS in 5 technical replicates by injecting 3 µl on the TripleTOF 5600+ mass spectrometer (SCIEX) connected to the UPLC 1290 (Agilent Technologies). The chromatographic separation occurred on a Waters Acquity UPLC HSS T3 column (100x2.1 mm, 1.8 µm) through the following 25 minutes gradient of solvent A (water containing 0.1% formic acid) and solvent B (acetonitrile containing 0.1% formic acid) at flow rate of 600 µl/min: 1 min of equilibration at 2% B; from 2% B to 95% B in 14 min; washing step at 95% B for 5 min, re-equilibration at 2% B for 5 min. The mass spectrometry analysis was performed both in positive and negative polarity, in the range of m/z 50-500 (TOF-MS scan with an accumulation time of 0.15 sec), with a SWATH acquisition comprising 10 windows of 45 Da each (accumulation time of 0.07 sec). The MS-DIAL software version 4.7.0 was used to analyze MS and MS/MS data for the identification of metabolites.

Cumulative diabetes incidence was calculated using the Kaplan-Meier estimation, whereas statistical significance was evaluated by the log-rank test. Statistical significance of the differences between 2 or more samples for qRT-PCR data and immunological data was calculated by unpaired 2-tailed Student’s t-test or ANOVA respectively. Metabolic profile data, normalized to feces weight, were submitted to ANOVA test, with FDR correction. We could observe 17630 features in positive polarity and 10535 features in negative polarity with FDR p-value < 0.05. The PCA analysis was performed using an in-house developed R script, while the heatmap was obtained using the MeV software v. 4.9.0 (2). For statistical analysis of the microbiome, microbial reads were discriminated against human reads with BMTagger (ftp.ncbi.nlm.nih.gov/pub/agarwala/bmtagger/) and mapped to the collection of all available genomes (https://www.ncbi.nlm.nih.gov/genome/) with Kraken2 for exact alignment of k-mers and accurate read classification (37). Relative abundance profiling and differential analysis were performed with DESeq2 upon variance-stabilizing transformation (38). Visualization was performed with ggplot2 (39). Species alpha diversity and dominance indices were calculated with vegan (https://cran.r-project.org/web/packages/vegan). p values <0.05 were considered statistically significant.

NOD mice that spontaneously develop autoimmune diabetes have a defect of the GB that primarily affects the mucus layer structure and composition (10), however a causal relationship between loss of gut barrier integrity and autoimmunity in this preclinical model of T1D is yet to be determined. Here, we restored GB integrity in NOD mice and test the effect on T1D occurrence. We decided to use a dietary approach containing nutrients known to reduce intestinal inflammation and promote GB function. To this aim, we elaborated a specific anti-inflammatory diet (AID) that includes omega-3 PUFA capable to reduce gut inflammation and promote immune tolerance, i.e., FoxP3⁺ Treg and IL-10⁺Tr1 cell differentiation (21–24), and inulin, a soluble fiber, that plays a direct beneficial effect on the GB and mucus layer integrity and promotes FoxP3⁺ Treg cell differentiation (26, 28–31) ( Figure 1A and Supplementary Table S1 for diet composition). Female NOD mice were either fed with standard (STD) diet or with the AID starting at 4 weeks of age just after weaning and before onset of beta cell autoimmunity and first signs of GB damage occur (10). In our animal facility female NOD mice normally develop spontaneous autoimmune diabetes with an incidence of 80-90% between 16 and 30 weeks of age. Our data reveal that the administration of the AID protected female NOD mice from occurrence of T1D ( Figure 1B ) with 40% of diabetic mice in the AID-fed group compared to 80% in the STD diet-fed counterparts at 32 weeks of age (end of experiment) (p< 0.05). The protective effect of the AID was confirmed at the histological level and NOD mice fed with the AID showed reduced degree of insulitis and increased number of intact islets (70%) compared to STD diet-fed mice (15%) ( Figure 1C ).

The GB alterations in the NOD mice are detectable at 10-12 weeks of age concomitantly with onset of beta cell autoimmunity and thus suggesting that they may have a triggering effect on T1D pathogenesis (10). In order to assess whether prevention of T1D in AID-fed NOD mice is linked to restoration of GB integrity, we performed a gut permeability test (FITC-dextran) and measured by RT-qPCR the expression of mRNA transcripts of tight junction proteins and analyzed different biomarkers of mucus layer integrity (mucus layer thickness, percentage of GC, mRNA expression levels of mucins) in the intestine of AID vs STD diet-fed NOD mice. Our FITC-dextran test confirmed previous findings of increased gut permeability in NOD mice compared to non-autoimmune mice ( Supplementary Figure 2 ) and a decrease of gut permeability in AID-fed NOD mice compared to STD diet-fed counterparts at 24 weeks of age but not earlier in the pre-diabetic phase (14 weeks of age) ( Supplementary Figure 2 ). Importantly, we observed that AID-fed NOD mice had increased mRNA expression levels of claudin 1 (CLDN1), one structural protein of the intestinal epithelial barrier (IEB) ( Figure 2A ; p<0.0001). Importantly, mRNA transcripts relatives to several mucins including the mucus structural mucin Muc2 (p<0.0001) and immune-regulatory mucins Muc1 (p<0.0001) and Muc3 (p<0.01) were also up-regulated in the intestinal tissue of AID-fed NOD mice ( Figure 2B ), while Muc4, a pro-inflammatory mucin (40), is decreased in AID-fed NOD mice ( Figure 2B ; p<0.01). Restoration of the mucus layer integrity in AID-fed NOD mice was confirmed at the histological level with an almost complete absence of mucus layer in STD-fed NOD mice that was corrected by administration of AID ( Figure 2C ). In line with this observation, we measured a significantly higher percentage of GC in the large intestine of AID-fed NOD mice compared to their STD diet-fed counterparts ( Figure 2C ; p<0.05).

Next, we asked whether restoration of the GB integrity (IEB and mucus layer) and protection from T1D in AID-fed NOD mice were associated with reduction of inflammation and induction of tolerogenic mechanisms in the gut mucosa. Specifically, since T1D occurrence is associated with acquisition of an effector phenotype by diabetogenic T cells in the gut (10), we asked whether T1D protection in AID-fed NOD mice was linked to a shift of functional phenotype of intestinal T cells from an effector Th17/Th1 type to a regulatory type (FoxP3⁺ Treg and IL-10⁺ Tr1 cells). First, we analyzed gut inflammation by measuring mRNA expression levels of inflammatory cytokines IL-1β, IL-23 and IL-17 in the intestinal tissues and found a statistically significant reduction of those inflammatory cytokines in AID-fed NOD mice ( Figure 3A ; p<0.001 for IL-1β and IL-17 and p<0.0001 for IL-23). Furthermore, our FACS analysis of the functional phenotype of intestinal T cells revealed that AID feeding in NOD mice significantly reduced the relative percentages of effector Th1 and Th17 particularly in the small intestine ( Figure 3B ; p<0.05 for Th17 and p<0.01 for Th1 cells) and to a lesser extent in the large intestine ( Supplementary Figure 3 ) while simultaneously promoting expansion/differentiation of FoxP3⁺ Treg cells in the small and large intestine ( Figure 3C ; p<0.01 and Supplementary Figure 3 ) and IL-10⁺ Tr1 cells in the small intestine ( Figure 3C ; p<0.01). Importantly, improved immune tolerance mechanisms, i.e., increased FoxP3⁺ Treg and IL-10⁺ Tr1 cells, and decrease of Teff cell differentiation found in the intestinal mucosa of AID-fed mice were spread to pancreatic lymph nodes (PLN) and islets. In fact, we detected a decrease of effector Th17 cells in the PLN ( Figure 4A ; p<0.05) and, most importantly, a decrease of effector Th17 and Th1 cells within the intra-islet lymphocytes (IIL) of AID-fed NOD mice ( Figure 4B ; p=0.05 for Th17 cells and p<0.01 for Th1 cells). Simultaneously, we measured an increase of the relative percentages of both FoxP3⁺ Treg cells and IL-10⁺ Tr1 cells in the PLN ( Figure 4C ; p<0.01 for FoxP3⁺Treg cells and p<0.05 for Tr1 cells) and IIL ( Figure 4D ; p<0.05 for FoxP3⁺Treg cells and p<0.01 for Tr1 cells) of AID-fed NOD mice compared to their STD diet-fed counterparts.

To clarify the mechanisms responsible for restoration of GB integrity and immune homeostasis in AID-fed NOD mice, we analyzed the microbial and the metabolic profile at the intestinal level. Our 16s rRNA analysis of the gut microbiota profiles (from the intestinal mucosa and luminal content) showed important modifications in AID-fed NOD mice compared to STD diet-fed controls. Specifically, we detected an increased alpha-diversity (Shannon index) in the AID-fed NOD mice ( Figure 5A ), a characteristic of the gut microbiota profile that is normally associated with reduced intestinal inflammation and protection from T1D. We also measured statistically significant differences in the relative abundance of 86 bacterial taxa ( Figures 5B, C and Supplementary Table 3 ). Notably, we found a very high increase in AID-fed NOD mice of the relative abundance of two Akkermansia species (log2fold change of 8.7 for Akkermansia muciniphila and 2.1 for Akkermansia glycaniphila) ( Figure 5D and Supplementary Table 3 ; p=6.9X10^-14 and p<0.01 respectively), two bacterial species associated with mucus-regeneration (41–44). We also detected a statistically significant augment of relative abundance of beneficial species previously linked with dietary administration of omega-3 PUFA and soluble fibers such as bifidobacteria (Bifidobacterium pseudolongum and Bifidobacterium animalis) ( 21, 22, 26 ): ( Figure 5D and Supplementary Table 3 ; p=4X10^-5 and p<0.05 respectively). Conversely, some pro-inflammatory bacterial strains such as Bacteroides intestinalis and Streptococcus sp. KS 6 were decreased in AID-fed NOD mice ( Figure 5D and Supplementary Table 3 ; p=4X10^-4 and p<0.01 respectively). Next, we asked whether the AID modified the metabolic environment in the intestine of NOD mice. Our untargeted metabolic analysis with high performance liquid chromatography tandem-mass spectrometry (LC-MS/MS) revealed a completely different metabolic profile in AID-fed NOD mice compared to NOD mice fed with STD diet ( Figure 6A ). Importantly, the metabolic profile of AID-fed NOD mice was much closer to that of a control non-autoimmune mouse strain (Balb/c mice) ( Figure 6A ), even if the Balb/c mice clustered in two groups possibly due to different environmental conditions ( Figure 6A ). Numerous metabolites were differentially represented in NOD mice fed with AID compared with STD diet ( Figure 6B ) including several metabolites involved in the ω6/ω3 PUFA metabolic pathway that perform pro-resolving actions in experimental models of colitis such as Docosapentaenoic acid (DPA), Docesahexaenoic acid (DHA) and Eicosapentaenoic acid (EPA) (45–47) and in the arachidonic acid metabolic pathway (Eicosanoic acid/Arachidic acid) involved in regulation of mucosal immunity and gut epithelial barrier function (48) ( Figures 6B, C and Supplementary Table 4 ; p<0.05 for DHA and DPA, p<0.001 for EPA and p<0.01 for Eicosanoic/Arachidic acid).