For metabolic cage measurements, eight mice on each of the four diets as described above were placed at eleven weeks of age into training cages for one week of acclimation. Thus, they were twelve weeks of age upon entry into the metabolic cage (Promethion Metabolic Screening Cages, Sable Systems International, Las Vegas, NV) for continuous measurements. Corn cob bedding was included and there was an intake manifold (a small metal tube that runs along the perimeter of the cage to pull air). The training cage was exactly the same (bedding, dimensions, etc.) as the testing cage, minus the manifold on the perimeter of the cage. Mice were single housed in the training cages and also in the metabolic cages during measurements.

Serum insulin was measured using the Mouse Insulin ELISA kit from Mercodia (Uppsala, Sweden) according to the manufacturer’s instructions. RNA was isolated from liver using RNeasy extraction kits (Qiagen) and cDNA synthesis was carried out using iScript (Bio-Rad). Real-time PCR was conducted using SYBR Green (Bio-Rad) run on a CFX Opus instrument (Bio-Rad) with gene specific primers (available upon request). Gene expression was normalized to the Ribosomal Protein S9 (Rs9) gene using the delta-delta Ct method.

Fresh stool samples were collected from each mouse in all groups, unrefined (13%), unrefined (26%), refined (10%), and refined (45%) fed mice, at 16 weeks of age into a 2.0 mL microcentrifuge tube and placed on ice prior to DNA isolation. DNA was prepared from fecal samples via bead beating and subsequent isolation using the QIAamp DNA Stool Mini Kit (Qiagen). The V4 region of the 16S rRNA gene was amplified via PCR using barcoded primer sequences. Amplicons were sequenced using the Illumina MiSeq platform (250-bp/paired-end reads). To reduce technical confounding from batch effects, all samples were sequenced at the same time and the sequencing order was randomized.

Amplicons were processed via Mothur v. 1.48.0 using default analysis settings ²¹. Operational taxonomic units (OTUs) delimited at 97% identity were identified and classified according to SILVA reference files release 132 ²². After application of standard quality control measures, the library size per sample ranged from 14827 to 1077789. A total of 2123 unique OTUs were observed demonstrating high-quality clustering. The mean sequencing error rate was very close to 0% based on parallel sequencing of a mock community. Estimates of alpha and beta diversities (intra- and inter-sample diversities, respectively) were obtained with Marker Data Profiling module of MicrobiomAnalyst 2.0 ²³. The alpha diversity was primarily estimated based on the observed OTU richness. Weighted and unweighted UniFrac distances were used to estimate beta diversity. Diversity metrics were calculated based on sequence counts scaled with cumulative sum scaling normalization.

Embedding, sectioning, and staining of formalin-fixed paraffin-embedded (FFPE) tissues was conducted as described previously ^24,25. Five micron sections of FFPE tissue was cut onto slides for immunofluorescence staining assays and incubated overnight at room temperature with rat anti-Foxp3 (1:100, Invitrogen 14–5773-82). After three washes at five minutes each with TBST, slides were incubated with Vector Goat anti-Rat HRP polymer for 30 min then washed with TBST, followed by exposure to Biotium CF488 Tyramide (1 uM) in Biotium Amplification Buffer Plus for 10 minutes. Additional primary antibodies were Rabbit anti-CD3 (1:500; ab16669) used at room temperature for 1.5 hours and Rabbit anti-Iba1 (1:1000; Wako 019–19741) and Guinea Pig anti-Insulin (1:500; ab7842) used overnight at four degrees. After counterstain with Hoechst, the slides were mounted in Vectashield Vibrance. Insulitis scoring was conducted as previously described ²⁴. We used the following classification scheme to allow for a semiquantitative analysis: 0 indicates no visible infiltration; 1, visible peri-insulitis with <10% of islet occluded; 2, visible peri-insulitis with partially or complete immune cell encircling the islet but affecting less than half of the islet area; 3, invasive insulitis, defined as comprising 50% or more of the islet area being occluded with leukocytic infiltration. The number of islets quantified for each dietary group are represented as follows: total number of islets/total number of sections analyzed where each section is from an individual mouse. 13% unrefined (73/8), 26% unrefined (102/8), 10% refined (175/13), and 45% refined (182/11).

Statistical analysis was performed using GraphPad Prism 10.1.2 (GraphPad Software, La Jolla, CA). Outliers were detected using the ROUT method with the standard settings (Q = 1%). Otherwise, all data were analyzed by two-tailed Student’s t-test, one-way analysis of variance (ANOVA) using a Tukey’s post hoc, Chi Square analysis (Kaplan-Meier curves), Kruskal-Wallis (microbiome distribution) or repeated-measures ANOVA (for longitudinal measures of body weight and body composition). Data are represented as means ± SEM.

Female NOD mice were started on either unrefined (13% or 26% fat kcal) or refined (10% or 45% fat kcal) diets. A glucose tolerance test reveals a trend towards slower clearance of glucose in the unrefined (26%) relative to the unrefined (13%; Figure 1A). When NOD mice are fed refined diets, there is reduced glucose tolerance in the high-fat (45%) relative to the low-fat refined (10%) group (Figure 1B). When all mice were monitored for diabetes onset, those consuming the unrefined diets developed diabetes at similar frequencies (Figure 1C; compare black and red lines; p = 0.20, designated as n.s.). However, mice on refined diets reveal that high-fat feeding accelerated diabetes onset at a greater incidence when compared with matched low-fat controls (Figure 1C; compare green and blue lines; p = 0.07, designated by #). Thus, type of diet (unrefined versus refined) influences the time to hyperglycemia onset in female NOD mice.

Body mass was similar between each group on unrefined (Supplementary Figure 1A) and refined diets (Supplementary Figure 1B). Fat mass was also comparable between mice within each respective diet group (Supplementary Figure 1C and Figure 1D). There was no difference in lean mass in each dietary group diet (compare Supplementary Figures 1E with 1F).

Mice on both unrefined (Supplementary Figure 2A) and refined (Supplementary Figure 2B) diets displayed similar energy expenditure across one week of measurements. When examining spontaneous physical activity, mice on unrefined diets (Supplementary Figure 2C) were not different. In addition, there were no changes in food intake (Supplementary Figure 2D) or water intake (Supplementary Figure 2E) in mice consuming the unrefined diets. Furthermore, mice fed refined (10% and 45%) also had no differences in physical activity (Supplementary Figure 2F), food intake (Supplementary Figure 2G), or water intake (Supplementary Figure 2H).

When fed unrefined diets, mice on the higher fat version (26% kcal) had greater RQ values across the week which cycled by light/dark cycle (Figure 2A). The cumulative RQ was higher in the 26% fat group versus the 13% fat group (Figure 2B), which was consistent with greater RQ in both the light and dark cycles in mice fed these unrefined diets (Figure 2C). Alternatively, RQ was greater in mice given the low-fat (10% kcal) compared with the high-fat (45% kcal) refined diet (Figure 2D). These data were consistent when observed cumulatively (Figure 2E) and when parsed out by light and dark cycles (Figure 2F).

Serum insulin is influenced by fat content in the diet with mice eating the higher fat unrefined diet displaying greater insulin quantities prior to disease onset (Figure 3A; NG, normoglycemic). Once mice became hyperglycemic, serum insulin levels were detectably reduced in mice consuming the unrefined diet with higher fat content (Figure 3A; HG, hyperglycemic). Similar results were obtained with mice consuming the refined low- and high-fat diets (Figure 3B). Within the refined diets, the greater incidence of hyperglycemia occurred in the high fat group (Figure 1); therefore, we next measured immune cell infiltration in and near islets as a possible explanation for these findings. Insulitis scoring revealed similar immune cell infiltration patterns in both low-fat and high-fat fed mice (Figure 3C). Staining for specific immune cell types, such as regulatory T-cells (FoxP3+), T-lymphocytes (CD3+), and macrophages (IBA1+) were also congruent with insulitis scoring patterns for each diet (Figure 3D). We noted that hyperglycemic mice had more severe insulitis scores when compared with normoglycemic mice. Finally, we observed that ICAM-1, a protein involved in cell-cell contacts and immune cell activation ^2,26, did not appear to be altered by high-fat feeding (Supplementary Figure 3).

NOD mice fed the refined high-fat diet show reduced hepatic expression of the Acaca and Fasn genes and increased expression of Dgat1 and Dgat2 (Figures 4A–D). This gene expression pattern is consistent with reduced hepatic de novo lipogenesis, while maintaining triglyceride storage capability, phenotypes documented in other strains of mice fed matched low- and high-fat diets ²⁷. Moreover, there was increased expression of Slc2a2 and Pck1, which encode GLUT2 and PEPCK, respectively, in NOD mice fed a refined high-fat diet (Figures 4E & F). These latter changes in gene expression could support increased hepatic glucose production as one possible mechanism explaining an increased onset of hyperglycemia in mice fed a refined high-fat diet. Consistent with similar onset of diabetes in the unrefined (chow) diets (Figure 1), there were no major gene expression changes between the unrefined diets in the livers of these mice (data not shown).

With no major changes in body mass, body composition, or energy expenditure between the mice on each respective type of diet (Supplementary Figures 1 and 2), we next tested the premise that diet influences microbiome changes. After filtering out low expression and low variance operational taxonomic units (OTUs), the remaining OTUs represented 38 unique bacterial genera across five phyla (Actinobacterial, Bacteroidetes, Firmicutes, Tenericutes, and Verrucomicrobia; Supplementary Figure 4A). At the genus level, organismal diversity clustered by type of diet (13% unrefined similar to 26% unrefined; 10% refined similar to 45% refined; Supplementary Figure 4B).

Within the unrefined diet groups, a much higher compositional diversity (alpha diversity, measured by Shannon diversity index) was observed compared to the stool samples from the refined diets (Fig. 5A), resulting in a highly statistically significant difference across the four dietary groups (Kruskall-Willis H=25.3, p-value = 1.31E-05; Fig 5A). This significance was primarily driven by the diversity indices of the unrefined groups compared to the indices for the refined diet groups. The alpha-diversity indices were more similar between unrefined groups (within group comparison) than to refined 10% and 45% fat diets (across group comparison). These results argue for a qualitative difference in the richness of taxa in the unrefined-fed animals compared to the mice fed the refined 10% and 45% fat diets. The beta-diversity index further estimated the differences in microbiota composition across the four groups. As seen in the principal component analysis plot (Fig. 5B), samples belonging to the unrefined groups clustered together, whereas a separate cluster was observed for the refined diet samples. Testing via the PERMANOVA method showed an overall statistically significant difference across the four groups (F=21.2, p-value <0.001). Post-hoc testing showed the greatest pairwise differences between unrefined 13% and refined 45% diet (F=40.03, p<0.001) and unrefined 26% versus refined 10% diet (F=22.1, p<0.001). Comparing defined LF- to HF-diet driven microbiota communities directly, we observed 52 OTUs that were deemed statistically significantly different (Supplementary Table 1). Community members with the highest representation in this group include members of the genera Staphylococcus, Clostridium senso stricto, and Turicibacter; all these are significantly reduced in the HF diet-shaped communities. Conversely, members of the phylum proteobacteria – belonging to the family Desulfovibrionaceae – show increased representation in the HF-shaped community. While Desulfovibrio sp. have been linked to metabolic syndrome inflammation ²⁸, direct links to autoimmunity are unclear at this time. In summary, we conclude that microbiome changes are influenced by type of diet (i.e., unrefined versus refined) but do not necessarily contribute to increased or decreased outcome measures used in this study.