The incidence of type 1 diabetes in many developed countries has been increasing at rates faster than can be explained by the known genetic propensity towards the disease [
1]. The environmental triggers of this disease have not yet been identified, despite the many efforts to associate diet, vitamin D, viruses and other factors with disease [
1]. A leaky gut has been correlated with type 1 diabetes [
2], and an aberrant gut microbiome was proposed as the factor that results in a leaky gut followed by altered immune responses leading to disease [
3]. This review describes a model for the role of the gut microbiome in type 1 diabetes based on the latest results.
Animal models
Past work with non-obese diabetic (NOD) mice and BioBreeding diabetes-prone (BB-DP) rats provided the first evidence that bacteria may play an important role in the onset of type 1 diabetes. BB-DP rats fed sulfamethoxazole, trimethoprim and colistin sulphate demonstrate a significantly decreased incidence of diabetes compared with controls [
4]. A similar increase in disease-free animals vs controls was observed in NOD mice after administration of doxycycline or vancomycin [
5,
6].
Bacteroides were higher in stools from the BB-DP rat while
Lactobacillus was higher in the diabetes-resistant rat (BB-DR) [
7]. Strains of
Lactobacillus johnsonii and
Lactobacillus reuteri isolated from BB-DP rats prevented and promoted diabetes in BB-DP rats, respectively [
8]. The
L. johnsonii strain induces a T helper 17 cell bias in the mesenteric lymph nodes of C57BL/6 mice while the
L. reuteri strain does not [
9].
Although the incidence of insulitis is not significantly different between germ-free and specific-pathogen-free (SPF) NOD mice, germ-free mice develop insulitis earlier than SPF mice [
10,
11]. The presence of segmented filamentous bacteria (SFB) in female NOD mice is correlated with decreased insulitis incidence and with the percentage of IL-17-expressing lymphocytes [
12].
Diet is known to significantly affect diabetes incidence in NOD mice. For example, a high-cereal diet increases diabetes incidence in NOD mice, while a high-protein diet reduces risk [
13]. Recent studies observed that acidic drinking water also increases diabetes incidence in NOD mice, and this is prevented by inoculating mice with SFB [
14].
Humans
The results of the murine experiments encouraged efforts to determine whether associations between gut bacteria and type 1 diabetes could be discovered in humans. Rodent model investigations suggested that healthy children may have high populations of probiotic-like bacteria such as
Lactobacillus, while the gut microbiome of unhealthy children may be dominated by
Bacteroides. Perhaps antibiotic or probiotic use early in life could prevent type 1 diabetes, but this requires more study. Analysis of a Finnish dataset revealed no connection between antibiotic use and autoimmunity for type 1 diabetes [
15], but larger cohorts may be needed to observe an antibiotic influence on this disease. Early on, a simple picture was expected to emerge by simply studying the stool content of a few human cohorts with individuals at high genetic risk of type 1 diabetes.
At first, with very small human cohorts with only four healthy children and four children with type 1 diabetes autoantibodies from the Diabetes Prediction and Prevention (DIPP) study in Finland, the human situation did appear to be very similar to the murine one [
16,
17]. These studies showed highly significant taxonomic and functional differences between cases and controls prior to autoimmunity for type 1 diabetes. Levels of
Bacteroides were higher in children with autoimmunity, and levels of seemingly protective unclassified Firmicutes were higher in healthy children. These results match the findings of the murine experiments except for the fact that the human samples had negligible amounts of
Lactobacillus.
Contrasting results were obtained from an investigation of 298 stool samples collected during the first 3 years of age from 22 case and 22 control children enrolled in the German BABYDIET study [
18]. Unlike the small Finnish cohort, no significant differences between taxa were observed between cases and controls after correction for false discovery rate. Instead, network analysis showed that bacterial communities in cases were far less strongly correlated with each other at <6 months of age and again at about 2 years of age than in controls. However, the results from the Finnish and German cohorts were similar in that the autoimmune bacterial communities were less stable than the healthy communities. The cause for this instability in autoimmune samples remains unknown.
A still larger set of 947 stool samples collected during the DIPP study from 29 cases and 47 controls during the first 4 years of life was examined [
15]. All of the children enrolled in the DIPP study are genetically at high risk for type 1 diabetes. Analysis of these samples revealed a very striking result: approximately 8 months prior to the average time of appearance of the first autoantibody in cases, a large increase in one bacterial species,
Bacteroides dorei, was observed. In addition,
B. dorei was by far the most abundant species in autoimmune prone children, with over 20% of the population represented by this one species—a rate that was more than double the relative abundance of this species in healthy children. This difference was highly statistically significant even after correcting for false discovery rate and was found to occur at about the same time as the introduction of solid food. A few other minor taxonomic differences were also observed, but nothing as striking as the
B. dorei result.
So why was the
B. dorei result seen in the large Finnish DIPP cohort but not in the German BABYDIET study? At first glance, it may appear that since the
B. dorei result in Finland was not reproduced in Germany, it must then be an anomaly. However, there is one very important difference between the designs of these two cohorts that may explain this difference. The samples used in the DIPP study were all from children born in the same hospital, Turku University Hospital [
15]. All of the children lived within 80 km of each other. Thus, many of the known confounders of the microbiome community were controlled; that is, climate, diet, culture, water supply, pollution levels, air quality and medical practices were all very similar for these children.
In contrast, the children in the BABYDIET cohort were from all over Germany, allowing enough variability in the microbiome confounders to affect microbiome composition. In addition, all of the children in the BABYDIET study had first-degree relatives with type 1 diabetes, while in DIPP the children were chosen for the study based on their high-risk HLA genotype at birth. These confounders add to the noise of the data and may explain why no significant taxonomic differences were observed between cases and controls. These confounders may have been sufficient to mask any taxonomic differences that might otherwise have been detected in the BABYDIET study but too weak to mask the network differences seen in this cohort [
18].
Hence, a working hypothesis is that geography plays a very large role in whether it is possible to observe significant case–control differences in any cohort designed to examine associations between the microbiome and disease. From this perspective, the microbiome results from the BABYDIET and DIPP cohorts do not contradict each other but rather help us understand how best to design cohorts for future disease-related microbiome studies.
Two other recent studies from Europe and Mexico showed taxonomic differences in the gut microbiome between individuals with type 1 diabetes and individuals without the disease [
19,
20]. In both studies, levels of
Bacteroides were significantly higher in those children with type 1 diabetes compared with healthy children. These studies were intended to show the effect of the disease itself as opposed to identifying a trigger for autoimmunity as in the previous studies. The Mexican study was restricted to Sonora state, which borders Arizona, and was limited to 29 children, including eight controls. Based on the findings of these studies and the pre-autoimmunity cohorts described above,
Bacteroides appears to be a major contributor to microbiome dysbiosis both prior to the development of autoimmunity and after disease diagnosis.
There is evidence that the human genome has some control over the taxonomic composition of bacteria in the gut. However, it is not known whether a genetic propensity to type 1 diabetes, as manifested by HLA genotype, affects bacterial gut composition [
21]. The studies conducted to date on the relationship between type 1 diabetes autoimmunity and the microbiome have not seen an HLA genotype effect on the microbiome, but these studies did not have enough participants to observe such an effect [
15,
18].
The hygiene hypothesis suggests that a lack of exposure to microbes early in life under hygiene conditions in the developed world contributes to a weakened immune system incapable of warding off the effects of detrimental bacteria in the gut [
22]. Efforts are underway to examine this issue carefully between Finland and neighbouring Estonia and Russian Karelia [
22]. Testing this hypothesis will be difficult in any study protocol. Our view that separating signal from noise is best done by careful cohort design intended to reduce the confounders of the microbiome.