Mice

Animal studies were performed according to institutional and national guidelines under UK Home Office Project Licenses 80/2156 and 80/2442. The two NOD sister colonies used in this study were kindly provided by Prof Linda Wicker and bred under specific pathogen–free (SPF) conditions either at the Centre for Biomedical Services, University of Cambridge, named “NOD^low”, or at Biological Services Unit of the Department of Pathology, University of Cambridge, named “NOD^high”. The NOD^low colony has consistently maintained a T1D incidence of ≈ 22% (measured in females up to 30 weeks of age). The NOD^high colony was derived from NOD^low founders around 2002; its T1D incidence rose gradually after the founders were moved to Biological Services Unit of the Department of Pathology, and has subsequently remained high for several years (≥ 90% by 30 weeks of age in females). NOD genotype was authenticated in a breeding trio from each colony using an array of 384 SNPs, spaced ≈ 7 MB apart across the entire genome, which distinguish common inbred mouse strains (Charles River Laboratories). Results from one of the SNPs in the NOD^low analysis were technically uninformative; all other results were consistent with homozygous NOD genotype. For incidence studies, animals were aged to 30 weeks and observed daily for signs of diabetes, such as polyuria, and of other signs of ill health; diabetes was confirmed by measurement of glucosuria, using Diastix (Bayer Diagnostics). Glycosuria is a physiologically relevant endpoint, indicating that blood glucose levels have exceeded the capacity for renal reabsorption. This is thought to occur at ca. 300 mg/dl plasma glucose, i.e., at twice the normal levels [29]. As a humane endpoint, development of diabetes triggered prompt euthanasia by CO₂ anesthesia followed by cervical dislocation. There were no unexplained deaths or other signs of ill health.

Animals were maintained in individually ventilated cages and received water ad libitum. Both colonies were maintained by breeding the first litter in each generation and initially kept on the same diet (RM3 [E] irradiated diet, Special Diet Services, Witham, UK); breeders received the same diet supplemented with 7% oil. In 2012, the supplier for the diet used in the NOD^low colony was changed (no. 105 diet, SAFE SAS, Augy, France) without a change in incidence. Routine veterinary microbiological screening of sentinel mice in both colonies was carried out according to current recommendations [30, 31] of the Federation of European Laboratory Animal Sciences Associations, or better, as discussed in the text (B&K Diagnostics).

Insulitis scores

Pancreata were harvested, preserved in 4% v/v in paraformaldehyde solution, and processed for wax histology. Five-micrometer sections were taken at three levels and stained with hematoxylin and eosin. Total islets per section were counted and the degree of cellular infiltration was scored as follows: no infiltration; peri-insulitis = infiltrate present as a single cluster of immune cells at one pole of the islet; intra-insulitis = diffuse pattern of infiltration within the islet.

PD-L1 inhibition

As described previously [32], mice were injected intra-peritoneally with a PD-L1 blocking monoclonal antibody (clone MIH5) at a dose of 2 mg/mouse. Development of diabetes was followed for 24 days post injection.

Cyclophosphamide treatment

As described previously [33], cyclophosphamide (Endoxana, Baxter Healthcare) was prepared in 0.9% normal saline at 20 mg/ml immediately before intra-peritoneal administration. Development of diabetes was followed for 24 days post injection.

Cell isolation, antibodies and flow cytometry

Lymphocyte cell suspensions were prepared in complete medium RPMI (Lonza) by lymph node disruption using 100 μm nylon mesh (BD Falcon).

For flow cytometry, the following antibodies were purchased from Biolegend: anti CD3-PE-Cy7 (clone 145-2C11; anti CD4-PerCP (clone RM4-5); anti CD69-FITC (clone H1.2F3). Anti B220-eFluor450 (clone RA3-6B2) was from eBioscience. For anti Foxp3-PE (clone FJK-16s) staining, cells were first fixed and permeabilized using the Foxp3 / Transcription Factor Staining Buffer Set from eBioscience following the manufacturer’s instructions. Analytical flow cytometry was conducted using a FACSCanto II (BD Biosciences), and the data were processed using FlowJo software (Tree Star).

Anti-BAFF ELISA

Serum BAFF was measured by standard colorimetric ELISA using rat anti-mouse BAFF (clone 121808) as capture antibody and biotinylated goat-anti mouse BAFF as detection antibody (both R&D Systems). BAFF concentration was determined using a standard curve of recombinant mouse BAFF (R&D Systems).

Bacterial 16S DNA detection in feces

Fecal samples were collected from mice and stored at -80°C until processing. DNA was isolated using a DNA Qiagen kit following manufacturer’s instructions. Bacterial DNA was detected by PCR analysis for 16S ribosomal RNA genes, using the following primers: for Existing Universal Bacterial (EUB), forward primer CGGCAACGAGCGCAACCC and reverse primer CCATTGTAGCACGTGTGTAGCC were used as described previously [34]. For Helicobacter hepaticus, forward primer ATGGGTAAGAAAATAGCAAAAAGATTGCAA and reverse primer CTATTTCATATCCATAAGCTCTTGAGAATC were used as described previously [35]. For SFB, forward primer GACGCTGAGGCATGAGAGCAT and reverse primer GACGGCACGGATTGTTATTCA were used as described previously [22]. Quantitative PCR was carried out using KAPA SYBER FAST qPCR Kit (KAPABIOSYSTEMS) and following the manufacturer’s instructions on the ABI PRISM 7000 Sequence detection system (Applied Biosystems). PCR signals for species-specific 16S rRNA genes were normalized to EUB. For qualitative analysis, the 705 bp PCR products were run on a 2% agarose (BIOLINE) gel in TEA; the band size was determined against the Gene Ruler 100 bp DNA Ladder (Thermo Scientific).

Co-housing experiment

Three-week-old NOD^low mice were transferred from the Centre for Biomedical Services, University of Cambridge to the Biological Services Unit of the Department of Pathology, University of Cambridge. After resting for 48 hours, mice were co-housed with age- and sex-matched NOD^high mice and scored for T1D up to 30 weeks of age. At 6 weeks of age, some of the co-housed animals of NOD^low and NOD^high origin were separated from each other and set up for breeding. Female offspring from both colonies of origin were then co-housed once more at weaning age (three weeks) and scored for T1D until 30 weeks of age.

Fecal matter transfer experiments

Three-week-old NOD^low females were gavaged with diluted fecal contents from NOD^high female donors. Briefly, fecal matter was collected from three12-week-old NOD^high females and resuspended in 50 volumes of sterile water. After filtering the suspension with a 70 μm nylon strainer, 250 μl of this suspension was given to each recipient by oral gavage using a 24G round tip gavage needle. Recipients received another dose of fecal matter after resting for 48 hours. Fresh fecal samples from individual recipients were collected at various times, starting from two weeks post transfer. Mice were maintained in an isolator dedicated to this study and scored for T1D up to 30 weeks of age.

IonTorrent sequencing and analysis

DNA from fecal samples was extracted using QIAamp FAST DNA Stool Kit (QIAGEN) with additional digestion in the presence of lysozyme (20 mg/ml). Variable V5 and V6 regions of the 16S rRNA gene were amplified from DNA from fecal samples using the barcoded forward fusion primer CCATCTCATCCCTGCGTGTCTCCGACTCAGATTAGATACCCYGGTAGTCC in combination with the reverse fusion primer CCTCTCTATGGGCAGTCGGTGATACGAGCTGACGACARCCATG. The sequences include IonTorrent PGM-specific adaptors (in italics) that are required for high throughput sequencing. The PCR-amplified 16S V5-V6 amplicons were purified from agarose gels using Qiagen Gel extraction kit (according to manufacturer’s instructions) and then prepared for sequencing on the IonTorrent PGM system using Ion PGM^TM Template OT2 400 Kit and Ion PGM^TM Sequencing 400 Kit according to the manufacturer’s instructions (Life Technologies). The number of reads obtained per sample was between 7000 and 65000. Data analysis was performed using the QIIME pipeline version 1.8.0. Operational taxonomic units (OTU) were picked using uclust [36] and the latest greengenes database (http://greengenes.secondgenome.com).

Statistical analysis

Statistical analyses were performed using GraphPad Prism. Kaplan–Meier analysis of cumulative disease incidence was performed, and incidence curves were compared by the log-rank test. Differences in immuno-phenotypes were evaluated by unpaired, two-tailed Student t test, or one- or two-way-ANOVA, as appropriate.

Statistical analysis of metagenomic data using R was performed using the QIIME pipeline [37]. Briefly, the 'phyloseq' and 'DESeq2' packages were used. The function 'DESeq' performs the differential taxon-count analysis based on the negative binomial distribution, and uses the OTU table as its input. It returns the estimated log₂-fold-changes and p values (Wald test) corrected for multiple testing for the comparison of interest (Benjamini & Hochberg correction). Principal coordinates analysis, a standard biostatistics technique [38, 39], was performed in order to represent the dissimilarity between any two microbiomes (measured by their unweighted UniFrac distances) in two dimensions.

Natural history and genetics of NOD^low and NOD^high colonies

We have previously reported that two sister NOD colonies kept in different animal houses develop T1D with low (NOD^low) and high (NOD^high) cumulative incidence in females [28] (shown here again in Fig 1A). The difference spans the range observed in NOD colonies worldwide [19]. In both colonies, male incidence was lower, consistent with findings in many SPF or conventionally-housed NOD colonies [19, 40, 41]. Accordingly, subsequent microbiological, phenotypic and disease incidence studies were carried out on females only. NOD^low mice had been housed in an SPF facility with individually ventilated cages and maintained stable low rates of disease incidence for at least a decade. The NOD^high colony was derived from NOD^low founders around 2002, when founders were transferred to a different facility with open cages; incidence rose gradually over the next two years and subsequently remained stable at a high level.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Characterisation of NOD^low and NOD^high colonies.

(A) Kaplan-Meier survival curves showing diabetes-free survival up 30 weeks of age in 85 NOD^low (black circles) and 45 NOD^high (white circles) female mice. Incidence curves were compared by the log-rank test. From [28] with permission; Copyright 2013, The American Association of Immunologists, Inc. (B) and (C) Percent of islets exhibiting no insulitis, peri- or intra-insulitis in six NOD^low vs. eight NOD^high 6-week-old female mice (B), or in seven NOD^low vs. five NOD^high 30-week-old female mice (C). Each bar represents analysis of three pancreatic sections from an individual mouse. (D) Examples of H&E stained pancreatic sections from 30-week-old NOD^low females showing, on the left, two islets without insulitis and, on the right, one islet each with peri-insulitis (*) and intra-insulitis (**).

https://doi.org/10.1371/journal.pone.0181964.g001

In order to exclude the possibility of genetic contamination with other common inbred strains, genotyping was performed in a breeding trio from each colony by screening 364 SNPs spread across the entire genome, spaced ≈ 7 MB apart on average, and representing common inbred mouse strains. All informative SNPs were consistent with homozygosity for the NOD genotype.

To assess whether the difference in development of T1D in the two colonies reflected different degrees of insulitis, we examined H&E-stained pancreatic sections. The level of infiltration in the pancreatic islets was similar between NOD^high and NOD^low mice, both at 6 weeks (Fig 1B) and at 30 weeks of age (Fig 1C and 1D), but overall there was more insulitis in the aged mice. In summary, the difference in diabetes development was not due to genetic contamination, nor was it related to differences in the degree of insulitis between the two colonies.

Low T1D incidence in NOD^low mice is unaffected by cyclophosphamide treatment but depends on PD-L1

Next, we examined mechanisms that could prevent progression to T1D in NOD^low mice. In some settings, cyclophosphamide has been shown to promote the rapid progression to T1D in pre-diabetic NOD mice by depleting regulatory T cells [33, 42]. We confirmed our previous findings [33] that most cyclophosphamide-treated pre-diabetic NOD^high females rapidly progressed to overt T1D within two weeks (Fig 2A). This contrasted with the variable age of onset of T1D in untreated NOD^high females, between 12 and 30 weeks of age (cf. Fig 1A). Surprisingly, however, the same treatment failed to trigger T1D in pre-diabetic NOD^low females (Fig 2A). Moreover, by flow cytometry we established that percentages and absolute numbers of CD4⁺ Foxp3⁺ T cells were similar in NOD^low and NOD^high mice (Fig 2B and 2C). Together, these findings suggest that the low rate of T1D development in NOD^low mice does not depend on CD4⁺ T_reg cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. T1D is induced in NOD^low mice by anti-PD-L1 antibody, but not by cyclophosphamide.

(A) Normoglycaemic (pre-diabetic) NOD^low (black circles) and NOD^high (white circles) female mice were treated at 16–18 weeks of age with cyclophosphamide, and T1D development was monitored daily up to 25 days post intra-peritoneal injection. (B) Representative FACS plots for CD4 (x-axis) vs. FoxP3 (y-axis) on gated CD4⁺ T cells from NOD^low (left) and NOD^high (right) spleens. (C) Percentages (left) and absolute numbers (right) of FoxP3⁺ CD4 T cells from pancreatic lymph nodes of NOD^low (full circles) and NOD^high (white circles) mice. Statistical analysis was performed using Student’s t test. (D) Normoglycaemic (pre-diabetic) NOD^low females were treated with anti- (α-) PD-L1 (clone MIH5, grey circles) or PBS (black circles) at 12–14 weeks of age, and T1D development was monitored daily up to 25 days post intra-peritoneal injection. In (A) and (D), diabetes-free survival was compared between groups by Kaplan-Meier analysis and log rank test.

https://doi.org/10.1371/journal.pone.0181964.g002

Anti-PD-L1 antibody treatment affords another means of overcoming resistance to autoimmunity [43, 44] by disrupting inhibitory signals that prevent activation and effector functions of pathogenic T cells [45, 46]. Blockade of this pathway by intra-peritoneal injection of anti-PD-L1 antibody rapidly precipitated T1D in old pre-diabetic NOD^low mice (Fig 2D). The comparison to a small control group (n = 4) in Fig 2D did not reach statistical significance (p = 0.09, log rank test), but the low incidence of T1D in the control group was in line with incidence in the NOD^low colony (see Fig 1A), whereas the rapid rise in T1D incidence (to 5/6 animals = 83%) within 3 weeks of anti-PD-L1 treatment was clearly different. We concluded that in NOD^low mice autoimmunity is maintained in a latent, sub-clinical state by PD-1/PD-L1 signalling.

Environmental differences between NOD^high and NOD^low colonies

Next, we examined environmental differences that might explain the colony difference in T1D development. Both colonies were maintained on the same 12h/12h light/dark cycle, and kept on the same diet initially; the diet was modified in the NOD^low colony in 2012 without incidence being affected. The drinking water in both animal facilities was neutral.

Given the importance of microbes in regulating autoimmunity, we reviewed veterinary screening reports of the two colonies. Differences in microbial colonisation and persistent viral infection were apparent (summarized in Fig 3A). Helicobacter (H.) hepaticus and H. typhlonius were detected in NOD^high, but not in NOD^low mice. The difference in H. hepaticus infection was confirmed by PCR analysis of feces from the two colonies: 5/6 NOD^high animals tested positive for H. hepaticus-specific 16S RNA genes, but none of the NOD^low animals did (Fig 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Microbiological characterisation of female NOD^low and NOD^high mice.

(A) Venn diagram depicting microorganisms consistently detected by routine health screens in sentinel mice from the NOD^low and NOD^high colonies (also includes the results of screening for SFB). None of the other viruses, bacteria or parasites that are routinely tested under FELASA guidelines [30, 31] were detected in either colony. (B and C) Quantification by qPCR (normalized to EUB) of H. hepaticus (B) and SFB (C) in the feces of individual NOD^low (black circles) and NOD^high (white circles) mice; horizontal bars represent means. Species-specific primers for the 16S RNA gene were used. (D) Metagenomic analysis of bacterial 16S rRNA genes from NOD^low and NOD^high females at 5 weeks of age. Different bacterial clades are color-coded, and log₂-fold mean differences between the detection frequencies in NOD^high and NOD^low mice shown on the y-axis; positive values show over-representation in the NOD^high colony. The dotted horizontal lines represent tenfold differences in either direction; significant colony differences (p < 0.05 after correcting for multiple comparisons) of tenfold or greater are shown as large squares with their names given. (E) Weights of individual age-matched NOD^low and NOD^high females (symbols as in B/C); means (horizontal lines) were compared by Student’s t test.

https://doi.org/10.1371/journal.pone.0181964.g003

Veterinary screening for murine norovirus (MNV), in accordance with the 2014 revised FELASA guidelines [31], was positive in the NOD^high colony, as well (see Supporting Information, S1 Appendix). MNV was known to have been introduced into this colony years after the animals had already acquired their high cumulative T1D incidence. Therefore, MNV could not have been the cause of the rise in incidence. We nonetheless addressed the formal possibility that MNV might be diabetogenic, but observed no rise in T1D after infecting NOD^low mice with MNV strain 3 (see Supporting Information, S1 Appendix). Thus, neither the natural history of the NOD^high colony nor the outcomes of deliberate infection suggest any diabetogenic role for MNV in this system.

Even though they are not included in routine veterinary screening, we used PCR to test for segmented filamentous bacteria (SFB), whose presence in NOD intestines reportedly correlates with protection from T1D [22]. However, feces from both colonies harbored similar levels of SFB, as judged by quantitative PCR analysis (Fig 3C). This finding ruled out differences in SFB colonisation as the explanation for the difference in diabetes incidence.

A more comprehensive survey of microbial differences between the two NOD colonies was then performed. High-throughput sequencing of 16S rRNA genes confirmed that NOD^high but not NOD^low mice were colonized by Helicobacter spp. (Fig 3D and S1 Fig). This analysis also detected differences in the abundance of several other bacterial taxa, some of which appeared highly significant (Paraprevotellaceae, Anaerostipes), whereas others were more modest (Prevotella copri). Most represented gains by the NOD^high colony, but there were a few losses, as well (Fig 3D). The overall abundance of major clades barely differed between the colonies (horizontal lines in Fig 3D).

Helicobacter spp. and MNV have been associated with intestinal inflammation in other model systems (cf. Discussion), but in the NOD^high colony, we found no inflammation of the intestinal lining upon histological examination. Consistent with this, young animals in this colony exhibited no failure to thrive; indeed, despite being on the same diet, they gained slightly more weight than age-matched NOD^low animals (Fig 3E).

In summary, the microbiota of NOD^low mice differed markedly from those of the NOD^high colony. Compared to NOD^low mice, NOD^high mice had lost a few species while gaining several others, including the known pathobionts, MNV and two Helicobacter species. The presence of SFB in the NOD^high colony raised the possibility that any suppressive influence of SFB on autoimmunity was overcome by other, diabetes-promoting factors in its environment.

Increased cellularity and B-cell activation in mesenteric lymph nodes from NOD^high mice

We hypothesized that persistence of several pathobionts in the NOD^high colony might be associated with increased immune activation in gut-related lymphoid organs. Indeed, young NOD^high females had greater numbers of mesenteric lymph node (MLN) cells than NOD^low mice (Fig 4A). In contrast, no colony differences were observed in the cell counts of inguinal, axillary, or pancreatic lymph nodes. Interestingly, the percentage of B cells expressing the early activation marker CD69 was increased in MLN from NOD^high mice, but not spleen (Fig 4B and 4C). Therefore, our data suggested that B cells in the MLNs of young female NOD^high mice respond more actively to local stimulation, well before the onset of overt autoimmunity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Increased B-cell activation in mesenteric lymph nodes of NOD^high mice.

(A) Total cell counts in the indicated lymphoid organs of NOD^low (closed circles) vs. NOD^high (open circles) mice: MLN = mesenteric lymph nodes, PLN = pancreatic lymph nodes, ILN = inguinal lymph nodes. Analysis by 2-way ANOVA with p values given for significant colony differences. (B) Representative flow cytometry dot plots of CD69^high B cells isolated from mesenteric lymph nodes of NOD^low (left) and NOD^high (right) females at six weeks of age. (C) Frequencies of CD69^high cells in gated B220⁺ B cells from mesenteric lymph nodes (MLN) or spleens (SP) of NOD^low (closed circles) vs. NOD^high (open circles) female mice. Individual mice, means and significant p values (p < 0.05, by 2-way ANOVA) are shown.

https://doi.org/10.1371/journal.pone.0181964.g004

Even though the activation phenotype appeared to be restricted to MLN, it was possible that B cell immune homeostasis was perturbed more broadly in NOD^high mice. To address this possibility, we measured the concentration of serum B cell activating factor (BAFF), a cytokine that plays a key role in B cell activation as well as maturation and is elevated in autoimmune conditions [47, 48]. The level of BAFF was increased in NOD^high compared to NOD^low females at 6 weeks of age (S2 Fig), consistent with altered B-cell homeostasis as well as with B-cell activation.

Exposure of NOD^low mice to NOD^high mice after weaning does not increase T1D incidence

To examine the transmissibility of colony differences in T1D development, 3-week-old female NOD^low weanlings were co-housed with sex- and age-matched NOD^high mice, and both groups were monitored for T1D up to 30 weeks of age (Fig 5A). Co-housed NOD^low and NOD^high mice developed T1D with the same cumulative incidence as their parental colonies (Fig 5B), indicating that exposure to the NOD^high environment after the age of weaning had no effect on the course of the disease, and vice versa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Exposure to a diabetogenic environment after the age of weaning does not modify T1D development in NOD^low mice.

(A) Schematic illustration of continuous co-housing of NOD^low (grey outline) with NOD^high (black outline) mice from 3 weeks of age onwards. (B) Kaplan-Meier analysis of diabetes-free survival in females up to 30 weeks of age, comparing the original NOD^low (black circles, n = 85) and NOD^high (white circles, n = 45) colonies (cf. Fig 1A) with NOD^low (full grey triangles) and NOD^high females (empty grey triangles) that were co-housed from three weeks of age. Despite being in a shared environment, the animals retained the disease incidence curves of their colonies of origin. (C) Diabetes-free survival in females from the original colonies (symbols as in B) was compared with that of NOD^low mice orally gavaged with fecal matter collected from 12-week-old female pre-diabetic NOD^high mice (closed grey diamonds) at three weeks of age and maintained in isolators. Fecal gavage did not raise T1D incidence over that in the NOD^low parental colony (p > 0.05, log rank test), despite successful transmission of H. hepaticus (cf. S3 Fig).

https://doi.org/10.1371/journal.pone.0181964.g005

This experiment did not rule out the possibility that the relevant environmental factor was difficult to transmit by co-housing. In order to examine another method of exposure, NOD^low weanlings were orally gavaged with faecal matter obtained from 12-week-old pre-diabetic NOD^high females and were kept in dedicated isolators until they reached 30 weeks of age. Acquisition of H. hepaticus was verified by PCR (S3 Fig). Nonetheless, these mice developed T1D with the same low rate of incidence shown by the NOD^low colony (Fig 5C).

These data show that, after the age of weaning, exposure to transmissible factors in the NOD^high environment, including some known pathobionts, does not raise T1D incidence in NOD^low mice. This was surprising because post-weaning exposure to various microbial or viral species has previously been shown to alter T1D incidence in NOD mice [21, 23, 24, 49].

Offspring of co-housed NOD^low mice acquire increased T1D incidence, microbiota and immune phenotypes similar to NOD^high mice

We hypothesized that, in order to affect T1D development, exposure to the diabetogenic environmental factor(s) has to occur prior to weaning. To test this hypothesis, sex- and age-matched NOD^low and NOD^high mice were co-housed in the NOD^high animal facility, beginning at three weeks of age. The co-housed mice were separated at 6–7 weeks for breeding. Their offspring were then co-housed again, starting at three weeks of age, and T1D development was monitored up to 30 weeks of age (Fig 6A). While the offspring of co-housed NOD^high mice maintained high rates of T1D incidence, the offspring of co-housed NOD^low mice showed significantly increased T1D incidence, compared to the parental NOD^low colony (Fig 6B). The cumulative disease incidence was not as high as that observed in the established NOD^high colony; nonetheless, the data suggested that exposure to the NOD^high diabetogenic environment throughout the animals’ life span, including the pre-weaning period, had a dominant effect on increasing T1D development in these mice.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Increased T1D incidence and acquisition of NOD^high-like B-cell phenotypes in female offspring of co-housed NOD^low mice.

(A) Scheme illustrating co-housing of NOD^low with NOD^high mice from 3 weeks of age, followed by breeding and continued co-housing of offspring from their age of weaning. (B) Kaplan-Meier analysis, comparing diabetes-free survival in females up to 30 weeks of age between the original NOD^low (black circles, n = 85) and NOD^high (white circles, n = 45) colonies (cf. Fig 1A), and between the offspring of co-housed NOD^low (closed grey squares) and co-housed NOD^high (open grey squares) mice. Survival curves for the offspring of co-housed animals were compared to the appropriate colony of origin by log rank test. (C) Intestinal microbiota (by metagenomic analysis) at 5 weeks of age in the offspring of co-housed NOD^low mice (filled purple circles), compared with those in the original NOD^low (filled orange triangles) and NOD^high (open green squares) colonies. Principal Coordinates Analysis was performed, with the top two components displayed. Each group of mice clusters separately, with the offspring of co-housed NOD^low animals intermediate between the two original colonies. See also S1 Fig. (D) CD69 expression in B cells from mesenteric lymph nodes (MLN), pancreatic lymph nodes (PLN), and spleens (SP) obtained from 6-week-old NOD^low females (grey full squares) and NOD^high females (grey empty squares). Group means were compared by 2-way ANOVA with post-test for colony differences.

https://doi.org/10.1371/journal.pone.0181964.g006

In order to examine whether the offspring of co-housed NOD^low mice had acquired NOD^high-like gut microbiota, intestinal 16S rRNA gene sequences from these mice were compared with animals from the original NOD^high and NOD^low colonies at 5 weeks of age (Fig 6C and S1 Fig). Fig 6C shows a Principal Coordinates Analysis, a statistical technique that represents differences in microbiota between any pair of mice as a two-dimensional map (see Materials and Methods and [38, 39]). This analysis clearly showed that the gut microbiota of the offspring of co-housed NOD^low mice was markedly different from the original NOD^low colony, approaching that of NOD^high mice (Fig 6C). Examination of the statistically significant species differences showed complete or partial acquisition of most of the species prevalent in the established NOD^high colony but not in the original NOD^low colony, including Helicobacter spp. (S1 Fig). Of the two genera present in the original NOD^low but not in the NOD^high colony, Anaerostipes was displaced in the offspring of co-housed NOD^low animals, whereas Anaeroplasma was retained (S1 Fig).

The MLNs in the offspring of co-housed NOD^low animals also seemed to approach the more activated phenotype of the NOD^high colony. CD69 expression in mesenteric lymph node B cells from second-generation co-housed NOD^low mice was similar to that detected in established NOD^high mice (Fig 6D).

In summary, NOD^low animals co-housed with NOD^high animals (or successfully infected with at least some of their pathobionts by other routes) from weaning age exhibited no change in their T1D incidence in the first generation. In the second generation of co-housing, however, the NOD^low animals partially acquired the increased T1D incidence and altered intestinal microbiota of the established NOD^high colony, correlating with increased MLN B-cell activation.

A dominant diabetogenic transmissible factor in NOD^high mice

The natural history of the two colonies suggested an environmental influence that, over several generations of breeding, raised the cumulative T1D incidence in NOD^low-derived mice to the high incidence (≈ 90% by 30 weeks of age) found in the established NOD^high colony. Both colonies had comparable levels of insulitis, and PD-L1 blockade unmasked latent autoimmunity in NOD^low females, suggesting that both colonies had autoimmune potential. Transmissibility of the differences in T1D incidence was examined co-housing NOD^low and NOD^high animals. Female offspring of co-housed NOD^low animals, exposed to NOD^high-derived environmental factors throughout their life span, developed T1D more frequently (≈ 50% by 30 weeks of age), than their dams or the original NOD^low colony (≈ 20%). In contrast, co-housed offspring of NOD^high animals retained their high (90%) cumulative T1D incidence. Thus, the increased T1D incidence in 2^nd-generation co-housed NOD^low mice was a dominant phenotype, caused by acquisition of a transmissible factor (or factors) from the NOD^high colony.

These findings suggested an infectious cause. Indeed, the NOD^high colony harbored several microbes absent from the NOD^low colony. This included Helicobacter spp. (hepaticus and typhlonius), which establish persistent, asymptomatic infection in immunocompetent mice [50], as seen here, but trigger colitis in immunodeficient models [51–54]. Prevotella copri was ≈ 10× more abundant in the NOD^high colony; this bacterium is also over-represented in patients with recent-onset human rheumatoid arthritis [55]. These species may be diabetogenic in our system. As for viruses, MNV was present only in the NOD^high colony, but there was no evidence that it raised T1D incidence in NOD^high females (S1 Appendix). Conversely, two bacterial genera were present in NOD^low mice but absent from the NOD^high colony. Of these, Anaeroplasma has been associated with protection of mice against a sex-dependent autoimmune demyelinating disease [56]. Thus, displacement of diabetes-protective species might also contribute to the increased T1D development in the NOD^high environment.

The 2^nd-generation co-housed NOD^low animals did not reach the very high cumulative disease incidence of female NOD^high mice (50% vs. 90% at 30 weeks). This was reminiscent of the gradual rise in disease incidence over generations after the NOD^high colony was first established. These animals acquired only some of the NOD^high–specific microbiota. Thus, multiple diabetes-promoting and dominant-acting changes in microbial composition may be required for the high T1D incidence of established NOD^high mice, some of which may require multiple generations of exposure.

Is early exposure to the NOD^high-derived transmissible factor critical?

Surprisingly, we were repeatedly unable to increase T1D incidence by transferring microbiota to NOD^low mice after the age of weaning, whether by oral gavage or co-housing. Moreover, both Helicobacter spp. and MNV were consistently found in the recipients’ feces shortly after exposure. Thus, at least the known pathobionts were readily transmissible, presumbably by coprophagy [57–59], and survive gastric passage to establish persistent infection.

In contrast, an increase in the frequency of T1D development was seen in the offspring of co-housed NOD^low mice. These animals experienced the diabetogenic environment throughout their entire life span, including during gestation and/or during suckling. Thus, exposure to the NOD^high environment may promote autoimmunity only during a critical window of susceptibility early in immune development. Thereafter, NOD^low mice seem to resist the diabetogenic influences within the NOD^high colony.

The need for pre-weaning exposure to raise T1D incidence in our system is novel. Many previous studies showed that microbial exposures in adulthood, well after the age of weaning, are sufficient to affect T1D development in NOD mice [23, 24, 26, 27, 49, 60–62]. Antibiotic treatment of pregnant NOD dams creates a simplified microbiome and increases T1D development in their offspring [27], but that study did not examine the effects of post-weaning exposure.

Other explanations of our data are possible. For example, the diabetogenic factor(s) might be poorly transmissible after weaning, or slow to reach pathogenic levels in the gastrointestinal tract, so that the exposure in the first generation of co-housing could be insufficient to raise T1D incidence. Alternatively, increased disease incidence may require vertical transmission of the responsible infectious agent(s) or co-transfer of host factors (such as autoantibodies) from exposed parents to offspring, or dilution of epigenetic marks carried over from the first generation. None of these possibilities contradict the need for an early-acting environmental factor. Moreover, as discussed further below, the immunophenotypic analysis was consistent with a window of susceptibility early in immune development.

B-cell activation in MLN correlates with high T1D incidence

The pathogenesis of T1D involves complex interactions of many immune components, both in rodent models and in humans [23, 63]. Thus, we explored immune correlates of the environmental differences in T1D incidence in our system.

Classically, T1D in NOD mice and BB-DP rats has been considered to be a T cell-driven disease [64, 65]. Many previously-described microbial modulators of NOD autoimmunity act via T_reg cells or regulatory cytokines [23]. In this study, however, we found no evidence for major differences in T-cell phenotypes correlating with the observed colony difference in cumulative T1D incidence. Splenic T-cell percentages and counts were similar between the NOD^high and NOD^low colonies (S4 Fig). Moreover, T_reg cells were not responsible for the colony difference in T1D development, as T_reg cell frequencies did not differ between the colonies, and the difference in T1D incidence persisted when T_reg cells were depleted by cyclophosphamide [33]. In contrast, we observed a clear colony difference in mesenteric lymph nodes, especially in B cells. MLN cell counts were increased in NOD^high mice, with increased CD69 expression in B cells, consistent with greater activation by intestinal pathobionts. Interestingly, this phenotype also correlated with increased T1D incidence in the offspring of cohoused NOD^low mice. The difference was clearly established by 6 weeks of age, consistent with an early-acting environmental factor being responsible.

A mechanistic link between increased MLN B-cell activation and increased rates of T1D development is plausible, because B cells are key mediators and regulators of autoimmunity in general [66], and this extends to NOD mice. Genetic ablation of B cells prevents T1D [67]; antibody-mediated depletion of B cells ameliorates the disease in mice [68, 69] and delays T1D progression in human prediabetic children [70]. Autoreactive B cells may present islet antigens to self-reactive T cells, whereas mechanisms depending on autoantibody secretion are thought to be less important [71–73]. NOD mice have impaired peripheral B cell tolerance at the transitional check point [74], which involves BAFF (also known as BLyS), a B-cell stimulating cytokine elevated in autoimmune conditions [75]. Moreover, genetic and phenotypic studies implicate TACI, one of the BAFF receptors, in type 1 diabetes development [76]. Given the elevated BAFF levels in NOD^high sera, such mechanisms may raise T1D incidence in the NOD^high environment. Interestingly, B cells have recently been implicated in rotavirus acceleration of T1D in NOD mice [60], though in contrast to the present study, rotavirus is diabetogenic in adult animals.

We considered the further possibility that immune regulation by B cells might be impaired in NOD^high mice. In the NOD strain, the characterisation of regulatory B cells (B_reg cells) has been difficult due to a paucity of informative markers [77], but IL-10-producing B cells enriched from long-term normoglycaemic NOD mice exhibit immunoregulatory properties [78]. To address a possible contribution of this mechanism in our system, we measured IL-10 secretion by LPS-stimulated B cells in vitro, but found no functional impairment of B_reg cells from NOD^high mice on a per-cell basis (S5 Fig).

Signalling by PD-1/PD-L1, an immunoregulatory ligand/receptor pair [79], could also be involved. Both the ligand and the receptor are expressed on T cells and, inducibly, on B cells [79]. We found by antibody blocking that PD-L1 signalling maintains low rates of T1D development in NOD^low mice. Similarly, previous work had shown that PD-L1 blockade unmasks T1D in pre-diabetic NOD^high mice and reverses suppression of T1D by Salmonella typhimurium [32]. PD-L1 mediates local immune regulation in pancreatic lymph nodes at a young age, and systemic regulation in older prediabetic NOD mice [46]. Prevention of T1D in NOD mice by PD-L1 signalling does not require B cells [46], but this does not argue against a role for B-cell activation in our NOD^high colony.

In conclusion, the activation of B cells in NOD^high MLNs, possibly driven by interaction with the diabetogenic factor(s) in this colony, correlates closely with increased T1D development due to the NOD^high-derived diabetogenic factor(s). This phenotype emerges at a young age, consistent with the early window of susceptibility, and may be related to increased BAFF levels in NOD^high sera, PD-1/PD-L1 signalling, or both. Further work is required to address these possibilities.

A hierarchy of microbial influences on progression to T1D?

This study revealed unexpected complexity in the environmental regulation of T1D. First, our observations do not fit well with the Hygiene Hypothesis (cf. Introduction): NOD^high mice exhibit high T1D incidence despite harbouring a somewhat more complex microbiome (including several pathobionts) than NOD^low mice. Rather, the NOD^high microbiome includes a novel, dominant-acting transmissible factor(s), raising T1D incidence. This adds to the many known microbial and viral influences that promote or inhibit autoimmunity (cf. Introduction).

Surprisingly, NOD^high mice exhibited their high T1D incidence despite harbouring SFB. Previous work had correlated SFB colonisation with low rates of T1D development [22], suggesting a dominant suppressive effect of SFB. The suppressive effect in T1D contrasts with the ability of SFB to enhance Th17-driven autoimmunity in other models [80], and has been linked to altered Th1/Th17 balance in the intestine. In our studies, SFB suppression of T1D in NOD^low mice seems to be overcome by the diabetogenic NOD^high-derived environmental factor(s).

Taken together with prior literature, our findings suggest a hierarchy of diabetes-promoting and suppressive influences, each acting via distinct immune mechanisms. In order from low to high microbial complexity and dominance, these are, at a minimum: sex-independent high incidence in germ-free mice [24]; sex-dependent metabolomic influences of microbiota, suppressing T1D in male mice [24]; Th1-to-Th17 cytokine deviation by SFB, suppressing T1D in females [22]; the NOD^high-derived environmental factor(s), which stimulate(s) mesenteric LN B cells and raises female T1D incidence (this study); and the T1D-suppressive influences of IL-10 production stimulated by parasites, or of PD-L1 signalling stimulated by Salmonella in NOD^high mice [20, 21]. This mechanistic diversity is further supported by the different developmental windows of susceptibility to diabetes-promoting and -suppressive influences, discussed above.

Translational implications

Environmental influences on autoimmunity differ between humans and mice, but parallels may exist. As in mice, human T1D is associated with altered intestinal microbiota (cf. Introduction), and infections influence T1D risk (reviewed in [23]). As in our present study of NOD mice, dominance hierarchies may also characterize environmental influences on human autoimmunity, acting via different immune mechanisms, with some only being effective during a particular developmental window of susceptibility. This may complicate health prognoses based on the composition of intestinal microbiota: if the presence of some microbial species may exert dominant effects, additive models of these influences are insufficient. The possible role of local B-cell activation in the environmental regulation of T1D development merits further study, both in mice and humans. Importantly, our data show that some diabetogenic environmental factors may be acquired asymptomatically by older recipients, without altering their own disease outcomes, yet may increase the potential for autoimmunity if acquired early enough in life and/or transmitted vertically to future generations. If the same applies to humans, this would complicate the identification of viral or microbial species that regulate autoimmunity. It also would raise the possibility of unforeseen autoimmune complications in the children of patients who have received fecal transplants, an emerging therapy for intractable infections, such as antibiotic-resistant Clostridium difficile, and for other inflammatory diseases [81].

Concluding remarks

The ‘Hygiene Hypothesis’ no longer seems sufficient to account for the complex hierarchy of immune mechanisms by which multiple microbial and viral factors influence T1D development. Developmental timing and the role of B-cell activation must be considered when considering microbial regulation of T1D. Potentially, human parents could transmit an increased risk of autoimmunity to their offspring upon vertical transmission of infectious agents acquired asymptomatically.