A Maternal Gluten-Free Diet Reduces Inflammation and Diabetes Incidence in the Offspring of NOD Mice

Gluten has previously been shown to affect the development of type 1 diabetes (T1D) in animal models. A gluten-free (GF) diet decreased the incidence of diabetes from 64% to 15% when nonobese diabetic (NOD) mice were fed a GF diet after weaning (1), and eating a GF diet decreased the incidence of diabetes to just 6% in the offspring in two generations, which indicates that the interplay between gut antigens and immune pathways leading to diabetes is particularly important in the preweaning period when insulitis starts to progress (2).

Accumulating evidence suggests that gut immune reactivity is skewed in human and murine diabetic patients. Studies in young human patients with T1D have demonstrated increased numbers of interferon-γ (IFN-γ)–producing, interleukin (IL)-1α–producing, and IL-4–producing cells in the small intestinal lamina propria, reflecting T1D preceded by intestinal immune activation (3). Similarly in NOD mice, a diabetes-promoting diet induced proinflammatory cytokines IFN-γ and tumor necrosis factor-α in the small intestinal lamina propria (4), and an antidiabetogenic diet decreased the high numbers of CD11b⁺CD11c⁺ dendritic cells (DCs) found in the colon lamina propria (5). Under germ-free conditions, reduced expression of forkhead box P3 (FoxP3) in the ileum, colon, and the draining lymph node was associated with accelerated development of insulitis in NOD mice (6), and, likewise in humans, Badami et al. (7) found that jejunal biopsy samples from T1D patients showed reduced frequency of CD4⁺CD25⁺FoxP3⁺CD127⁻ regulatory T cells (Tregs). The link between the gut and pancreas has also been emphasized in studies demonstrating that pancreatic islet T cells express gut homing receptor α4β7 integrin, which recognizes mucosal addressin cell adhesion molecule-1 in the pancreas (8,9).

Failure in immune tolerance leading to pancreatic β-cell depletion has been suggested to be regulated in part by gliadin-induced intestinal enteropathy and innate immune responses (10,11). However, altered gut microbiota, previously demonstrated in GF-fed versus gluten-fed mice (12), might also contribute to modify intestinal inflammation and development of autoimmune diabetes. In support of this, impaired oral tolerance to intestinal microbes was demonstrated in NOD mice (5), and the impact of microbes has also been verified in germ-free (13), antibiotic-treated (14,15), and probiotic-treated diabetes-prone rodent models (16). It seems reasonable to assume that gluten and certain microbes have a synergistic effect on the development of T1D, as was also recently suggested by Patrick et al. (17). Cytokine profiles of gut-associated lymphoid tissue have revealed a strong association between intestinal IFN-γ production and the incidence of diabetes, especially in several gluten intervention studies (4,11,17–20). Also, type 1 T-helper cells proliferated specifically in the mesenteric lymph node (MLN) in response to wheat protein antigens (19). A GF diet was furthermore shown to reverse this shift in gut homeostasis toward an anti-inflammatory state with more transforming growth factor-β (TGF-β)–producing T cells (18).

As early-life interventions in the intestinal environment can influence the incidence of diabetes, we hypothesized that a GF diet exclusively fed to mice during gestation and lactation would be sufficient to protect the offspring from the development of diabetes even though they were weaned to a standard gluten-containing (STD) diet. We hypothesized that the dietary protective effect would be partly mediated by a shift in the gut microbiota, and that this shift is of imperative importance in the first period of life during which the immune system develops.

The experiment was performed in accordance with the Council of Europe Convention European Treaty Series 123 on the Protection of Vertebrate Animals used for Experimental and Other Scientific Purposes, and the Danish Animal Experimentation Act (LBK 1306 from 23 November 2007). The study was approved by the Animal Experiments Inspectorate, Ministry of Justice, Denmark.

NOD/BomTac mice (Taconic, Hudson, NY) were fed ad libitum either a GF modified Altromin diet or an STD Altromin diet (Altromin, Lage, Germany), as described by Funda et al. (1). The two groups were mated separately, and their female offspring were group-housed (five mice/cage) in our barrier-protected rodent facility (Faculty of Health and Medical Sciences, University of Copenhagen, Frederiksberg, Denmark) under standard conditions in open cages without filter lids. All offspring were weaned at 4 weeks of age to the STD diet. Ten pups were killed from each group at 4 weeks of age, and 10 mice from each group were killed at 10 weeks of age. The remaining 30 mice in each group were killed when a diagnosis of diabetes was made or at 30 weeks of age, when the study ended. Measurements of tail blood glucose levels were made twice a week from 10 weeks of age, and a mouse was considered to be diabetic when blood glucose levels exceeded 12 mmol/L on 2 consecutive days. The body weight of all offspring was monitored once a week.

Hematoxylin-eosin–stained pancreas sections were evaluated for insulitis score in a blinded fashion by two persons. Lymphocytic infiltration was graded as follows: 0, no infiltration; 1, intact islets but with few mononuclear cells surrounding the islets; 2, peri-insulitis; 3, islet infiltration <50%; and 4, islet infiltration >50%. Twenty-five islets were scored for each nondiabetic mouse killed at 10 and 30 weeks of age.

Feces samples aseptically obtained from the mothers during pregnancy and from the offspring at 4 and 10 weeks of age when they were killed were analyzed by PCR amplification of the V3 region of the 16S rRNA gene followed by denaturing gradient gel electrophoresis (DGGE) as described previously (21). The resulting DGGE profiles were analyzed using BioNumerics version 4.5 (Applied Maths, Sint-Martens-Latem, Belgium). The composition of the prokaryotic community of feces samples from the mothers and the 4-week-old pups was determined using tag-encoded 454/FLX Titanium (Roche) pyrosequencing of the V3 and V4 regions of the 16S rRNA gene by the National High Throughput DNA Sequencing Centre, University of Copenhagen, Copenhagen, Denmark (22), and was analyzed as described by Krych et al. (23). An open source software package, Quantitative Insight Into Microbial Ecology (QIIME version 1.7.0) was used to analyze the pyrosequencing data (National Center for Biotechnology Information database accession #PRJNA215143). Principal coordinate analysis (PCoA) was made using the jackknife_beta_diversity.py workflow (the –e value: 2,000 sequences). The PCoA plot including bacterial taxa was drawn out with the make_3d_plots.py script based on the summary information of bacterial phyla, and the differences in the taxa relative distribution between categories were tested with Metastats (24) independently for both the phylum-level and the genus-level summarized taxa. The P value was calculated based on 1,000 permutations.

Single-cell suspensions from spleen, MLN, and pancreatic lymph node (PLN) isolated from 4- and 10-week-old offspring immediately upon their being killed, and flow cytometric analyses of CD11b⁺CD11c⁺ DCs and T-cell populations, including FoxP3⁺ Tregs, were performed as previously described (22). All antibodies were purchased from eBiosciences (San Diego, CA). The analyses were performed using an Accuri C6 flow cytometer (Accuri Cytometers Inc., Ann Arbor, MI).

Immediately after the mice were killed, 1-cm fragments of the ileum and colon were placed in RNAlater (Ambion, Austin, TX), after all luminal content was scraped out of the gut. Homogenization, RNA isolation with MagMAX-96 RNA Isolation Kit (Ambion), and cDNA synthesis using the High-Capacity cDNA Reverse Transcriptase Kit (Applied Biosystems, Foster City, CA) were performed as described previously (25). An inventoried TaqMan Mouse Immune Array (Appled Biosystems) containing 90 TaqMan gene expression assays of immune-related genes was used to investigate ileal samples isolated at weaning, which were analyzed as described previously (26). Actinβ, Ocln, Tjp1, Cldn8, Cldn15, Muc1, and Muc2 TaqMan gene expression assays (Applied Biosystems) were used for quantitative PCR (qPCR) analyses on ileum and colon cDNA isolated at 4 weeks of age, and the data were analyzed as described previously (25).

cDNA samples from the ileum and colon collected at weaning were further analyzed for the presence of Akkermansia muciniphila, which was quantified in duplicate using the 7500 Fast Real-time PCR System (Applied Biosystems), as previously described (27).

GraphPad Prism version 5.02 (GraphPad Software, San Diego, CA) was used for statistical analysis, and P values <0.05 were considered to be significant. Cumulative diabetes incidence was calculated using the Kaplan-Meier estimation, whereas statistical significance was evaluated by the log-rank test. Other differences were estimated by two-tailed t test or one-way ANOVA test with Tukey post test.

In this study, it was demonstrated that feeding a GF diet to pregnant NOD mice significantly reduced the cumulative diabetes incidence (P < 0.01) and increased the onset time (P < 0.01) in their offspring, even though all pups were weaned to an STD diet at 4 weeks of age (Fig. 1A). The diabetes incidence at 210 days was 51% (n = 37) in the offspring of STD diet–fed mice, and 22% (n = 36) in the offspring of GF diet–fed mice. Histological evaluation of insulitis in pancreatic sections from nondiabetic offspring revealed a significantly lower insulitis score in offspring of GF diet–fed mice compared with the offspring of STD diet–fed mice at both 10 weeks (P < 0.05; Fig. 1B and C) and 30 weeks of age (P < 0.05; Fig. 1D and E). No significant difference in body weight gain was observed between the two groups of nondiabetic NOD mice within the observational period (data not shown).

Gut microbiota analysis by DGGE demonstrated a difference in the fecal gut microbiota between the two groups of pregnant NOD mice (Fig. 2A). ANOVA based on the first (X), second (Y), and third (Z) principal component (PC) revealed a significant difference in PC2 values (P < 0.05), and a tendency to cluster in PC1 (P = 0.07) and PC3 values (P = 0.08). The separate clustering on the PC analysis plot was also evident in their offspring at weaning, confirming that parental microbiomes altered by diet are inheritable (28); significant differences in PC1 values (P < 0.05) and in PC3 values (P < 0.001) were evident between the two groups of offspring. However, how much of this difference in microbiota is due to vertical transfer from mothers to pups or to early ingestion of the GF diet by the pups is not known. Gut microbiota analysis of feces from offspring at 10 weeks of age, 6 weeks after weaning to the STD diet, revealed no difference between the two groups of offspring (Fig. 2B). The influence of the diet and the gut microbiota of the mothers on the gut microbiota of the offspring was thus not permanent.

The differences in the fecal microbial composition between the two groups of NOD mice and their offspring at 4 weeks of age were further corroborated by tag-encoded 16S rRNA gene–based pyrosequencing. The raw number of reads generated from all 39 fecal samples scored 1,332,137. Sequences that met all requirements of the quality control (minimum length 300 bp, quality score ≥25) and were free from chimeric reads yielded 848,346, providing an average of 21,752 sequences per sample (minimum 1,178 sequences, maximum 88,674 sequences, SD = 17,103 sequences), with a mean sequence length of 458 bp (minimum 300 bp, maximum 470 bp). One sample was discarded because of the low number of reads (<1,000 reads). PCoA based on weighted UniFrac distance metrics showed a clear separation of the two categories comprising GF diet–fed NOD mice and their offspring, and STD diet–fed NOD mice with their offspring. The proportion of the cumulative information describing the variance using the first two PCs reached 60% (Fig. 2C).

The most abundant phyla in both categories were Firmicutes and Bacteroidetes that constituted 50% and 40%, respectively (Table 1). Metastats analysis revealed that the difference observed between the groups was mainly due to a significantly expanded representation of the bacterial phyla Verrucomicrobia, TM7, and Proteobacteria in the mothers that were eating a GF diet and their pups compared with STD diet–fed mice (Fig. 2D). In addition, the phylum Cyanobacteria was found in approximately half of the offspring of STD diet–fed mice, but in none of the offspring of GF diet–fed mice. The annotation of reads within the Verrucomicrobia represented one species, A. muciniphila, whereas the genus Proteus was responsible for the difference evident in Proteobacteria phyla (Table 2).

qPCR analysis of Akkermansia in the ileum mucosal layer at weaning revealed a higher abundance in the offspring of GF diet–fed mice compared with STD diet–fed mice (Fig. 2E; P < 0.001). A similar tendency was also evident in the colonic mucus layer (Fig. 2F; P = 0.06).

Both intestinal and pancreatic DCs have been reported to play important roles in T1D, and especially the CD11b⁺ subset has been associated with both pathogenic and tolerogenic immunity to pancreatic islets. In this study, the CD11b⁺CD11c⁺ DCs were less abundant among pancreatic (P < 0.01), intestinal (P < 0.001), and systemic (P < 0.05) lymphocyte populations in the offspring of GF diet–fed mice compared with STD diet–fed offspring at the time of weaning (Fig. 3A). No differences in the overall proportion of CD11c⁺ DCs were observed, and no differences were observed between the groups at 10 weeks of age.

The proportions of Tregs (CD4⁺FoxP3⁺) in the MLN and spleen were similar in the offspring of GF diet–fed and STD diet–fed mice, both at weaning and at 10 weeks of age, but, interestingly, a significantly higher proportion (P < 0.001) of these cells was evident in the PLN in the offspring of GF diet–fed mice compared with offspring of STD diet–fed mice at weaning, indicating a more anti-inflammatory local immune system due to an early GF environment (Fig. 3B). The proportions of CD4 and CD8 T cells in the MLN, PLN, and spleen were not different between the two groups either at weaning or at 10 weeks of age, except for a higher proportion (P < 0.001) of CD8 T cells in the PLN at weaning in the offspring of GF diet–fed mice compared with offspring of STD diet–fed mice (Fig. 3G and H). Furthermore, these CD8 T cells were marked with gut-homing receptor α4β7 integrin, which is induced when T cells are activated in the intestinal environment (Fig. 3I). Also on the CD4 T cells in PLN a higher proportion of cells (P < 0.01) was marked with α4β7 in the offspring of GF diet–fed mice compared with offspring of STD diet–fed mice (Fig. 3J). However, this difference in α4β7 integrin was evident only at weaning and not at 10 weeks of age.

To further explore how the early-life GF environment affected gut homeostasis and the establishment of regulatory immunological activity, genes implicated in immune cell migration, microbial recognition, and response, as well as T-cell activation and signaling in the gut were analyzed at weaning. Thirty-one of the 90 gene expressions analyzed were significantly altered between the two groups of offspring (Table 3). Of importance, the early GF environment caused significantly less expression of several inflammatory mediators of insulitis such as Ifng, Il12b, and Il18 genes, and Prf1 and Gzmb genes for the cytolytic enzymes perforin-1 and granzyme B, respectively. Interestingly, the Cd68 macrophage marker, the Hmox1 gene for the heme oxygenase-1 enzyme, and the Stat6 gene involved in exerting IL-4 were more profoundly expressed in the offspring of GF diet–fed mice compared with STD diet–fed offspring, and they are all characteristics of immunosuppressive M2 macrophages, whereas the expression of the Nos2 (iNos) gene marker of M1 macrophages was low. However, gene expressions of intestinal cytokines Il10 and Il17, which have previously been associated with diabetes protection, were lower in these mice similar to Socs1, an inhibitor of IFN-γ signaling and Smad7 and Ski, which are inhibitors of the regulatory cytokine TGF-β signaling.

Expressions of T-cell marker genes such as Cd3e, Cd8a, and Cd38 were also downregulated in the ileum, including Ccr4 expression, which has been demonstrated on pathogenic autoimmune T cells in NOD mice. Supportive of the flow cytometry results, the CD34 gene expression, which encodes a key molecule in T-cell trafficking to lymph nodes, was higher in the offspring of GF diet–fed mice compared with STD diet–fed offspring.

To investigate whether the more anti-inflammatory gene expression profile in the gut was associated with an improved intestinal barrier at weaning, gene expressions of tight junction and mucus components were analyzed in the ileum and colon. The following tight junction component genes have been shown to be regulated in association with improved intestinal permeability assay results (29). Gene expressions of Ocln, which encodes occludin (P < 0.01; Fig. 4A), Tjp1, which encodes tight junction protein 1 (P < 0.01; Fig. 4B), and Cldn15, which encodes claudin-15 (P < 0.05; Fig. 4D), were elevated in ilea from the offspring of GF diet–fed mice compared with those of STD diet–fed mice. Ocln (P < 0.05) gene expression was also elevated in the colon together with a tendency for higher expression of Muc1, which encodes a protein that represents membrane-associated mucin (P = 0.06; Fig. 4E). Conversely, colonic Muc2 expression representing secreted mucin was lower (P < 0.05; Fig. 4F).

The importance of gluten for the development of autoimmune diabetes was previously demonstrated in both NOD mice and BB rats, and it has become clear that disease-modulating interventions, including dietary or microbial antigen treatments, in these animal models are particularly imperative in early life. To clarify whether the effect of gluten could be prevented exclusively by limiting its exposure in the postnatal period, a two-generation approach was used. Most importantly, the diabetes incidence was significantly reduced in the offspring of GF diet–fed NOD mice even though they were weaned to an STD diet. Thus, a GF diet during gestation and lactation was protective later in life. The fact that we, despite the low incidence of T1D in our facility, see a significant difference in the GF group further substantiates the strong effect of this diet. The 30-week-old nondiabetic offspring of GF diet–fed mice had a lower incidence of insulitis than the nondiabetic STD mice, which indicates that the GF group is not just delayed in diabetes development, but is as far from diabetic as the nondiabetic control mice that usually do not develop diabetes; however, a longer observational period would be necessary to fully clarify whether the mice are completely protected against diabetes.

In agreement with this result, the ability of a low-protein diet to modify diabetes incidence was reported to cause a significant drop in diabetes incidence from 86% in control NOD mice to 63% in NOD mice when given only during pregnancy and lactation (30). In contrast, wheat and barley protein deprivation only until weaning was demonstrated not to be sufficient to significantly delay diabetes development. However, in the gestational and preweaning period this diet was also supplemented with fish oil and Vit D3, which in the same study (31) were demonstrated to abrogate the protective effect of a wheat and barley protein-free diet. Interestingly, it was also shown that accelerated diabetes was not completely reconstituted by supplementing the wheat and barley protein-free diet with gliadin, which indicates that other dietary or microbial antigens also mediate the protective effect. For example, the gut microbiota in GF diet–fed mice, which may not be altered by a pure gliadin supplement, is different from that in gluten-fed mice (12). A recent article (17) proposed cereal dietary antigens as a stronger T1D inducer than microbes. The authors reported a similar protective effect on diabetes incidence in BB rats that were fed a low-antigen hydrolyzed casein (HC) diet in both germ-free and specific pathogen–free (SPF) conditions. However, the HC diet was more protective in the germ-free than in the SPF condition, also indicating a diabetes-promoting effect of the microbes. This was further supported by a low β-cell mass only in the cereal-fed BB rats compared with the germ-free and HC-fed BB rats.

In the current study, a distinct bacterial profile enriched in especially Akkermansia, TM7, and Proteobacteria was evident in both NOD mice fed a GF diet and in their offspring at weaning. These taxonomic groups were also previously associated with protection against the development of autoimmune diabetes in NOD mice only when present before weaning (22), which is interesting as Akkermansia has been demonstrated to modulate host immune responses in monocolonized mice (32). In addition, taxonomic differences between the gut microbiomes of healthy and diabetic children were characterized by mucin-degrading Akkermansia, which was more abundant in control subjects than in case patients (33). As gluten has potential irritating effects in the small intestine where it is degraded, this might affect mucus production and barrier function. By leaving out gluten from the diet, the mucus production may increase, which in turn can lead to an increase in the presence and metabolic activity of specific bacterial strains, not least Akkermansia, which previously has been shown to grow on mucus proteins (34). Minor fold-change differences were seen in tight junction component gene expressions, which indicated that Akkermansia might be associated with an improved intestinal barrier in both the ileum and colon, although this link is purely speculative. The restoration of impaired intestinal barrier and alleviated signs of gut epithelial irritation such as colonic crypt hyperplasia have also previously been shown in response to antidiabetogenic diets in T1D rodent models (5,35).

Not much is known about the role of TM7 in mammalian health and disease, but in humans it is mainly associated with the oral cavity, where it has been associated with periodontitis (36,37). Furthermore, Kuehbacher et al. (38) suggested that TM7 members are involved in the ethology of Crohn disease, although the mechanism remains unknown. A detailed examination of the influence of the microbes from GF diet–fed NOD mice by transferring the microbiota to germ-free mice would be informative. It is striking that the effect on the microbial composition is only present as long as the mice are fed a GF diet. This indicates that the effect of the microbiota is dependent on the diet and that this effect is especially important during the development of the immune system. Even more striking is the fact that even though the microbiota reverses after the introduction of gluten in the diet at weaning, the mice born by GF mothers are protected against the development of T1D later in life. Thus, its immune regulatory effect on, for example, Tregs at weaning in the prediabetic stage seems to have a long-lasting impact on the capability of the immune system to protect against autoimmune attack on the β-cells.

Interestingly, an early article (39) demonstrated that adoptively transferred T cells from NOD mice on an HC diet were unable to change the incidence of diabetes and were presented with similar T-cell receptor–mediated proliferative responses compared with controls. Considering this, it was hypothesized that immune regulatory mechanisms in the pancreatic environment at weaning downregulate otherwise fully functional diabetogenic T-cell response in GF diet–fed NOD mice. In the current study, this hypothesis was supported by increased proportions of FoxP3⁺ Tregs in PLN and fewer CD11b⁺ DCs at weaning, which also previously were modulated by an antidiabetogenic diet (5). A similar anti-inflammatory immune profile has been observed in BALB/c mice receiving a GF diet in both pancreatic and gut-associated lymphoid tissue (18). The changes in the immune system were only found at weaning and not at 10 weeks of age. Thus, it seems that changes in immunity later in life when insulitis is more progressive are not as critical as in postnatal life for the development of autoimmune diabetes, during which the insulitis process begins. This further indicates that the GF diet, possibly through a change in the gut microbiota, may have delayed the development of the adaptive immunity, but that the mice, independently of whether they develop T1D or not, eventually develop a mature immune system.

Lower intestinal gene expression of proinflammatory cytokines and higher expression of anti-inflammatory M2 macrophage markers, together with increased gut Cd3e, Cd8a expression were found in the offspring of GF diet–fed mice. It is interesting that these findings were also seen in SPF BB rats fed an HC diet but not to the same extent as in the germ-free HC-fed BB rats (17). Thus, it required the presence of microbes. M2 macrophages were furthermore found to be associated with the microbiota-dependent sex difference observed in T1D development in NOD mice (40). The intestinal alterations found in the offspring of GF diet–fed mice may therefore, also in this experiment be a consequence of the altered microbiota in early life rather than the presence of cereal antigens. Although these data cannot be causally linked to the pancreas, it was demonstrated that an early GF environment increased the trafficking of T cells with a mucosal phenotype to PLN. It would be interesting to further investigate the regulatory properties of these T cells because a high intestinal IL-15 level, as seen in the GF group, promotes intestinal epithelial cell activation of noncytotoxic CD8 T cells with suppressor function (41). These cells have nondetectable granzyme B, which was also expressed significantly lower in the GF group. This is further consistent with the prevalence of M2 anti-inflammatory macrophages as these and Tregs seem to have a mutual ability to promote the differentiation of one another, at least partly through the TGF-β pathway (42,43). It is possible that any effect of bacterial or dietary antigens on pancreatic immune homeostasis would be mediated in part by the migration of these immune cells that are activated in the tolerogenic gut environment.

Whether the early ingestion of cereal antigens before weaning or the altered microbiota exerts its effect separately or synergistically is not known. Most importantly, the early GF environment clearly attenuated diabetes development in the NOD mice even though they were weaned to a gluten-containing diet, but the changes in gut microbiota and immune system were no longer evident later in life, from which we can conclude that the time and the diet before weaning are of imperative importance for protection against diabetes development.

Funding. This work was carried out as part of the 3G Center—Gut, Grain & Greens; the 3G Center is supported by the Danish Council for Strategic Research. This work was further funded by CHANCE (Chemometric Analysis Centre at the University of Copenhagen), the Center for Applied Laboratory Animal Research, and the Beckett Foundation.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. C.H.F.H. conceived and designed the study, analyzed histology, performed flow cytometry, contributed to the discussion, and wrote the manuscript. Ł.K. analyzed the data, contributed to the discussion, and wrote the manuscript. K.B. analyzed histology and contributed to the discussion. S.B.M., C.N., and H.F. were involved in the analysis of gene expression data and contributed to the discussion. L.H.H. performed pyrosequencing and contributed to the discussion. D.S.N. analyzed the data and contributed to the discussion. S.S. performed flow cytometry and contributed to the discussion. A.K.H. conceived and designed the study, contributed to the discussion, and wrote the manuscript. C.H.F.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.