Dysbiosis of the gut microbiota has been described in multiple inflammatory and immune-mediated diseases including rheumatoid arthritis, ulcerative colitis, Crohn’s disease ankylosing spondylitis, and type 1 diabetes (T1D). The microbiota is established in the early part of life, and its composition is shaped by deterministic niche-related factors (host environment), historical factors (pre-exposure to certain microbes), and environmental drivers (e.g., diet) [1]. This complexity suggests that the causes of dysbiosis in diseases such as T1D are likely to be multifactorial. As the incidence of autoimmune diseases including T1D has been steadily increasing over the past 50 years [2], it is speculated that lifestyle changes are causing changes in the gut microbiota leading to alterations in the development of the immune system [3]. However, it is unknown whether genetic susceptibility to T1D also contributes to shaping the microbial communities of high-risk individuals, and whether the enteric dysbiosis seen in T1D is a cause or effect of the disease, and the immune dysregulation associated with its development.

A role for the microbiota in T1D etiology is suggested by studies that demonstrate changes in the microbial communities prior to clinical disease onset. Early studies suggested that high-risk, islet-autoantibody-positive children had altered bacterial diversity when compared with low-risk autoantibody-negative children [4, 5]. A recent longitudinal study of Finnish children at risk of T1D found a drop in bacterial diversity prior to the development of clinical disease [6]. Similarly, a cross-sectional study from a US-based cohort found differences in the bacterial composition between new-onset patients, high- and low-risk first-degree relatives, and healthy, unrelated subjects [7]. These data suggest that reduced diversity and/or a change in metabolic function of the gut microbiota may be associated with disease progression in T1D.

In the inbred non-obese diabetic (NOD) mouse model of T1D, susceptible NOD mice were shown to harbor a distinct gut microbiota to a variety of T1D-protected inbred strains [8]. T1D incidence is higher in specific pathogen-free (SPF) or “clean” housing facilities, and many interventions that alter the microbiota (e.g., diet changes) result in disease protection [9]. Exposure to antibiotic regimens early in development both dramatically shifted the bacterial abundance and led to accelerated disease [10]. These studies indicate that in the steady state, NOD mice harbor a “diabetes-permissive” microbiota, and major perturbations in the microbiota can both reduce and accelerate disease.

In other autoimmune diseases such as Crohn’s disease, specific genetic loci associated with disease susceptibility have been shown to influence the gut environment and lead to shifts in the resident microbial populations [11]. To date, such associations have not been investigated in T1D. Genome-wide association studies have identified 58 genetic associations with T1D [12]. These include many genes involved in both adaptive and innate immune pathways [12]. The human leukocyte antigen (HLA)-DQ and HLA-DRB1 major histocompatibility complex (MHC) class II region genes have the strongest association with disease susceptibility [13]. The interleukin-2 (IL-2) signaling pathway, which is linked to regulatory T cell (Treg) function, is strongly implicated in disease pathogenesis due to both a cluster of associated genes and a suite of mechanistic studies in mice [14]. Treg residents in the gut contribute to controlling tolerance to the commensal microbiota [15]; hence, T1D-associated variants in IL-2 pathway genes which reduce Treg function could potentially lead to changes in the microbiota.

In this study, we have investigated whether the gut microbiota of the NOD strain is genetically determined. We utilized NOD mice from multiple animal facilities and congenic NOD mice that carry protective alleles of T1D susceptibility alleles to investigate whether T1D predisposing host genetics modifies the gut microbiota through alterations in the gut immune environment. The regions we investigated included the MHC and the Idd3 and Idd5 loci which are linked to the IL-2 pathway and Treg function [16–18]. We also investigate whether therapeutic treatment of NOD mice with IL-2, which improves Treg function and mimics the genetic effect of protective Idd3 alleles [17], reduces inflammation in the gut and restores the microbiota to that found in disease-protected mice. To determine whether T1D-associated risk alleles are associated with alterations in the gut microbiota in human subjects, we utilized the TwinsUK cohort of individuals with matching data on genetic variation and microbiota composition. Our findings for the first time demonstrate a role for T1D susceptibility alleles in shaping the gut microbiota, independently of the presence of T1D, consistent with the gut microbiota having a causative role in the development of T1D.

Colonization of the gastrointestinal tract soon after birth is determined by a complex interplay between environmental modifiers and the host metabolic state and intestinal immune response, factors significantly affected by host genetics. Human twin studies and studies of intercrossed mice bred in highly controlled environments have enabled investigations into the contribution of host genetic factors in shaping the gut microbiota (reviewed in [28]). These studies have identified genes involved in metabolic function and innate immune activation and regulation. Our findings that protective Idd3/5 alleles, which are associated with enhanced IL-2 production and regulatory T cell function, give rise to notable alterations in the microbiota fit well with the premise that regulatory immune cell populations of the lamina propria are able to modulate the gut microbial community [29].

Recently, genome-wide association studies (GWASs) of the gut microbiota in humans have also attempted to identify genetic loci or pathways associated with changes in the abundance of individual microbial taxa [26, 30–32]. Although these studies all concluded that host genetic factors did contribute to the variation in the microbiota, effect sizes of individual loci were small and varied between studies. The loci identified included SNPs in immune response genes linked to autoimmune diseases including Crohn’s disease, ulcerative colitis, multiple sclerosis, and T1D (CD5, CCL2, IL23R, CD86, HLA, C11ORF30-LRRC32, CLEC16A) [30–32]. Rather than looking for associations with individual SNPs, we performed a pathway analysis to look for associations with genes linked to IL-2 signaling and Treg function and the microbiota composition. Suggestive associations were observed between a low disease risk score and the abundance of members of the Clostridiales, Bacteroides, Lachnospiraceae, and Ruminococcaceae families. Although our analysis utilized data from 1392 individuals, a substantially larger GWAS scale dataset will be needed to provide sufficient statistical power to confirm these findings. Nevertheless, our study supports a role for natural variation in immune response genes influencing the gut microbiota, although high variability between populations originating from diverse geographical locations clearly is an impediment to the replicability of these findings.

GWAS studies of the microbiota in both mouse and humans have shown that certain taxa have a higher heritability than others [26]. This may be due to the relative sensitivity or lack of adaptability of these bacteria to changes in their preferred environment. For example, Akkermansia, which is associated with the mucin layer, was associated with SIGLEC15 gene expression [26]. SIGLEC15 is associated with expression of a carbohydrate found bound to mucin. Akkermansia, which was increased in Idd3/5 mice, has been associated with a healthy microbiota via improved epithelial integrity [33]. Other mucinophiles such as Bacteroidales members including Parabacteriodes were significantly increased in protected Idd3/5 mice. Parabacteriodes distasonis abundance was decreased in inflamed tissue biopsies of IBD patients [34], and P. distasonis administration reduced the severity of DSS-induced colitis suggesting a protective role in the colon [35]. Another mucin utilizer, Aldercruetzia from the Actinobacteria phyla, was also increased in Idd3/5 mice. Aldercruetzia has been associated with health as a reduction in Aldercruetzia was observed in patients with multiple sclerosis compared to healthy individuals [36]. The increase in mucinophiles observed in Idd3/5 mice may be related to the increase in goblet cell mucous production observed in this strain.

Taxa that have been previously reported as heritable include many of those found with differential abundance between either the NOD, Idd3/5, or B6 core microbiomes for example Lactobacillus [32], S24-7 [37], Ruminoccoccus [31], Oscillospira [37], Bacteriodes [38], Ruminococcaceae [26, 32, 37], Lachnospiraceae, [37], and Prevotella [31]. Several of these taxa also had a suggestive association with T1D genetic risk in our human association study (e.g., Ruminococcaceae, Lachnospiraceae, Bacteroides, and Oscillospira). This supports our findings that the NOD microbiome is maintained by host genetics.

Several studies have investigated alterations in the microbiota in human subjects with a high genetic risk of developing T1D either with or without islet autoimmunity (defined by the presence of autoantibodies) or in longitudinal study designs monitoring individuals before and after disease onset [5–7]. Although differences have been found between these studies, which could be related to sample size, differences in study design, and geographic locations, several interesting findings have been noted. Kostic et al. performed a longitudinal study of a Finnish cohort and found that alpha diversity dropped in T1D cases after seroconversion but prior to clinical diagnosis of disease [6]. This was driven by a relative overabundance of several groups including Ruminococcus, Blautia, and Streptococcus while there was a relative underabundance of Veillonellaceae and Lachnospiraceae in T1D converters. Consistent with this, Lachnospiraceae was reduced in the NOD compared with the Idd3/5 core microbiome and also associated with protective genotypes in the human analysis. In a cross-sectional study, Alkanani and colleagues found a trend towards a reduced abundance of the genera Prevotella in individuals with a higher risk of disease progression [7], and increased Prevotella in healthy children compared to those with T1D was seen in another study [39]. These findings are similar to the increase in Prevotella that we found in Idd5 and Idd3/5 mice compared with NOD mice. One early study found an increase in Faecalibacterium prausnitzii in control children compared with those with islet autoimmunity [4]. F.prausnitzii is a butyrate producer associated with gut health that is depleted in Crohn’s disease [40] that was associated with a T1D-protective genotype in our risk set analysis. While F.prausnitzii are not members of the mouse gut microbiome, other butyrate-producing bacteria such as Coprococcus from the Lachnospiraceae family fill a similar niche [41], and abundances of these were increased significantly in Idd3/5 compared with NOD mice. Thus, although there are major differences between the murine and human microbiota, some commonalities emerge, suggesting that both in mice and humans, changes in some T1D-associated microbes are caused by host immunogenetic factors rather than just the disease itself.

Treatment with IL-2 resulted in highly significant effects on the intestinal phenotype including reduced immune infiltration and increased goblet cell mucous production. This was associated with alterations in the microbiota including decreased Oscillospira and Bacteriodales. Although Oscillospira was not significantly altered by Idd3 or Idd3/5 in comparison to control NOD/Mrk mice, the overall abundance of Oscillospira in the combined NOD colonies was significantly higher than both the Idd3/5 and B6 cohorts (Additional file 1: Figure S3). Similarly, Oscillospira_839964 trended to be increased by the human IL-2 risk alleles, suggesting a common IL-2-dependent mechanism may drive down Oscillospira abundance in humans and mice. In contrast, Bacteriodales was reduced by IL-2c treatment but was increased in Idd3 mice although remained low in Idd3/5 mice. This suggests that Bacteriodales abundance is influenced by other factors in addition to IL-2.

Despite being profoundly protected from T1D, we found only subtle changes in the fecal microbiota of NOD.H-2^b mice in comparison with control NOD mice with only Lachnospiraceae [Ruminococcus] significantly different after adjustment for multiple testing. This was despite a reduction in immune infiltration in the colon and ileum of NOD.H-2^b mice and improved Paneth cell morphology. This suggests that the phenotypic differences found between NOD and NOD.H-2^b guts were primarily driven by host genetics although feedback from the microbiota may exacerbate them. A recent study described differences in the cecal microbiota between NOD mice transgenic for the MHC-II Eα gene and control NOD mice [42]. Wildtype NOD but not B6 mice lack the E complex gene, and Eα16/NOD transgenic mice are highly protected from diabetes. Silverman and colleagues found differences in the overall diversity of the microbiota between these strains although only the order Clostridales and the genus Blautia were individually significant. As this study sampled the cecal contents rather than the feces, it is possible that the impact of the Eα gene was greater in the cecum than in the colon. In line with this, the only immune cell difference detected in Eα16/NOD mice was in Treg abundance within the cecum but not the small or large intestine.

Our analysis of immune cell populations found within the NOD intestine pointed to an increase in inflammatory populations. Although neither NOD mice nor patients with T1D typically present with the overt gut disease, previous studies have reported intestinal inflammation or immune activation using biopsy tissue from the duodenum of T1D patients. In comparison with healthy individuals, T1D patients had increased HLA-class II, IL-1α, and IL-4 expression in the jejunum [43]. Additionally, T1D patients had significantly fewer Foxp3⁺ Tregs in the duodenum, and lamina propria dendritic cells isolated from the duodenum were not able to induce Treg differentiation [44]. NOD mice were reported to have increased proportions of T cells with an activated phenotype and producing IL-17 in the colon [45]. Together with our findings, these data suggest that increased intestinal immune cell activation and infiltration are associated with T1D and are likely to at least partly underlie changes in the microbiota associated with T1D.

Paneth cell dysfunction is associated with Crohn’s disease and is linked to the presence of risk alleles for an autophagy gene ATG16L1 [46]. Paneth cells produce antimicrobial peptides, which are released into the crypt lumen where they protect against infection by enteric pathogens and help to shape the commensal microbiota. Paneth cell degranulation coupled with the death of the Paneth cells can be induced by IFNγ exposure [47, 48]. Paneth cell dysfunction can also be induced by unfolded protein response-induced ER stress [49], which can be exacerbated by IL-10 deficiency [21]. As such, a combination of increased infiltration of inflammatory cells along with impaired regulatory T cell responses may contribute to Paneth cell dysfunction in NOD mice.

Another factor known to influence the composition of the microbiota in NOD mice is gender [50, 51]. In this case, testosterone produced in males is thought to be metabolized by certain bacterial taxa influencing their activity and community composition. This results in a feedback loop whereby androgen levels influence the microbiota which impact inflammation and the immune response giving rise to disease protection in males. This is therefore another example whereby host genetics (male sex) influences the microbiota and diabetes incidence.

One limitation of our analyses was the relatively low sequencing depth used after rarefaction of the data. Rarefaction is a useful means of normalizing microbiome data to account for varied ability to detect rare taxa depending on sequencing depth. However, this will have limited our ability to detect significant alterations in low abundance taxa. Another limitation of our cohousing study was that the co-housed NOD pups were still nursed by their own NOD mothers despite the presence of B6 females within their cages. The lack of B6-derived maternal factors (e.g., maternal antibodies) passed to the NOD pups may have reduced the efficiency of colonization by B6-derived taxa as described by others [52, 53]. We also note that our studies of the microbiota of the various congenic NOD strains were not controlled for with non-congenic littermate animals. This would additionally control for differences in exposure to various microbial taxa due to founder effects that may have occurred when each line was established.

In conclusion, we show that natural variation in genetic loci associated with risk of T1D influence the structure of the gut microbiota. Pathways linked to IL-2 signaling and immune regulation were associated with microbial shifts in both mouse and human with higher disease risk associated with a decrease in the presence of Ruminococcus, Lachnospiraceae, and Clostridiales. These taxa have previously been reported to have a high heritability estimate in human and mouse studies. Mechanistically, we found that the IL-2 pathway profoundly impacted inflammation present in the colon and small intestine with IL-2 therapy increasing Tregs and immunoregulatory cytokines promoting increased goblet cell mucous production. This suggests that immunotherapy strategies currently aimed at supplementing IL-2 and boosting Treg function in T1D patients may potentially influence the microbiota in these patients and microbiota monitoring strategies could be investigated for determining response to therapy.