Eight-week-old male TLR4^-/- (KO) and TLR4^+/+ littermates (CON) were obtained from The Jackson Laboratory (Bar Harbor, ME) and acclimated for two weeks with ad libitum access to a purified standard diet (SD; Harlan Laboratories, TD 08485) consisting of 13% fat, 67.9% carbohydrate, and 19.1% protein (% kcal). After two weeks, animals were assigned to either a standard low-fat diet (SD) or a high fat diet (HFD; Harlan Laboratories, TD 06414) consisting of 60.3% fat, 21.4% carbohydrate, and 18.3% protein (% kcal), resulting in the following 4 experimental groups (n=12 per group): 1) TLR4^+/+ on SD diet (CON+SD); 2) TLR4^+/+ on HFD diet, (CON+HFD); 3) TLR4^-/- on SD diet (KO+SD); and 4) TLR4^-/- on HFD diet (KO+HFD). Mice remained on assigned diets through 6 months and were cohoused two per cage in a temperature and humidity-controlled environment on a 12:12-h light-dark cycle. Food intake and body weight were measured weekly at the time of cage and water changes. All animal procedures were reviewed and approved by the Colorado State University Institutional Animal Care and Use Committee protocol #1369.

Aortic pulse wave velocity (aPWV) was used to determine in vivo arterial stiffness using a Doppler Flow Velocity System (Indus Instruments, Webster, TX) via methods described previously by our laboratory (Battson et al., 2018; Lee et al., 2018). aPWV (in cm/s) was calculated as aPWV = (distance between the 2 probes)/(Δtime_abdominal − Δtime_carotid).

Following a 6h fast, mice received an intraperitoneal injection of 2g/kg dextrose, and blood glucose was assessed at 0, 15, 30, 45, 60, 90, and 120 minutes post-injection using tail-vein blood and AlphaTRAK 2 glucometers (Abbott Laboratories, Chicago, IL). Fasting blood was spun for 5 minutes at 2,000rpm and resultant serum was used for fasting insulin levels via ELISA. For insulin tolerance tests, mice received an intraperitoneal injection of 1.5g/kg insulin after a 6h fast, and blood glucose was assessed at 0, 15, 60, and 120 minutes post-injection as described above.

Animals were water-fasted for 12 hours prior to oral gavage administration of fluorescein isothiocyanate (FITC)-dextran (4,000 mol wt, no. 46944; Sigma-Aldrich; 125 mg/ml for SD fed mice and 150 mg/ml for HFD fed mice). Food was removed immediately after oral gavage and tail vein blood samples were collected after 4 hours for quantification of serum FITC-dextran concentrations. Serum samples were diluted 1:2 in 1X PBS, and fluorescence was measured on a BioTek Synergy 2 Multi-Detection Microplate Reader (BioTek Instruments, Winooski, VT) at 485/20 (excitation) and 528/20 (emission). Serum concentrations were quantified using a standard curve of known FITC-dextran concentrations prepared in a control serum sample from an untreated mouse.

Mice were anesthetized with isoflurane and euthanized by exsanguination via cardiac puncture. Blood was collected in a 0.5 M EDTA-coated syringe (no. 15575020; Invitrogen). Collected blood was added to 2% volume EDTA and plasma was obtained by centrifugation at 2,000 rcf for 10 min at 4°C, and immediately flash frozen in liquid nitrogen. The gastrointestinal tract was excised, and colon length and cecum weight were recorded. Cecal content and colon content were removed from the tissue and flash frozen. Adipose tissue depots (perivascular, subcutaneous, epididymal, and mesenteric depots) were isolated and weighed. Sections of distal colon and proximal small intestine were obtained and flash frozen for further analysis.

Circulating biomarkers associated with intestinal permeability were evaluated in plasma collected by cardiac puncture. Specifically, circulating levels of LPS binding protein (LBP; ALX-850-304/1, Enzo Life Sciences) and zonulin-family peptides (NC1314884, Fisher Scientific) were measured by ELISA. LBP is a soluble acute phase protein that binds and presents LPS to TLR4 and its co-receptor CD14. Zonulin-family peptides are potent regulators of intestinal tight junctions and circulating levels have been associated with “leaky gut”, inflammation, and gut microbiota perturbation (Tajik et al., 2020). Additionally, secretory IgA (sIgA; 50154425, Fisher Scientific) and intestinal alkaline phosphatase (IPA; NC9945170, Fisher Scientific) were measured by ELISA in collected ileal tissues. sIGA and IAP were normalized to total protein, measured by BCA. Intestinal sIgA is secreted in response to luminal pathogens or enterotoxins and IAP dephosphorylates bacterial LPS, reducing its immunogenicity and reduced IAP is associated with loss of gut homeostasis.

Thoracic aorta (TA) was excised and placed in an ionized solution for removal of perivascular adipose tissue (PVAT). After PVAT removal, a 1-1.5mm section distal to the aortic arch was cut and embedded vertically in optimal cutting temperature (OCT) gel that was flash frozen. Sections were cut at 10μm using a cryostat set to -20°C. 4-5 sections were adhered to each slide and 3-5 slides per mouse were stored at -80°C until staining. 50ml fresh 5 μM DHE (D11347, Thermo Fisher Scientific) staining solution prepared and poured in slide box. 1mg DHE dissolved into 317 μL dimethyl sulfoxide (DMSO) for 10mM stock solution. 25μL DHE stock diluted into 50 mL Milli-Q pure water for 5μM staining solution. Slides were rinsed then placed in DHE staining solution and incubated for 15 min at room temperature, with limited light exposure. Slides were washed and fluorescent images were acquired. Two to three technical replicates were collected on 2-3 biological replicates for each mouse and replicates were averaged for one mean grey intensity value/area (μm^2).

Fecal DNA was extracted from samples excised from the colon at termination using the FastDNA Stool kit (Catalog 6540400, MP Biomedicals). PCR amplification of the V4 16S rRNA region was completed following the Earth Microbiome Project protocol using the 515F-806R fusion primers with 12bp Golay barcodes and sequencing adaptors (https://earthmicrobiome.org/protocols-and-standards/16s/). Pooled libraries were sent to the Next Generation Sequencing Facility at Colorado State University and sequenced using 2x250 chemistry on a MiSeq (Illumina Inc, San Diego, CA).

Forward and reverse reads were imported into QIIME2 version 2021.2. Reads were demultiplexed, concatenated, and trimmed to 200 base pairs. Reads were binned into ASVs using the DADA2 pipeline and taxonomic assignments were made using GreenGenes version 13.8. Count tables and taxonomy were imported into MicrobiomeAnalyst (www.microbiomeanalyst.ca) and further curated by removing reads that had fewer than 4 counts and that were present in less than 10% of the samples, which resulted in the removal of 16 low abundance features.

Data are presented as either mean ± standard deviation (SD), mean ± standard error of the mean (SEM), or as median (interquartile range (IQR)). Data analysis and visualization was conducted in GraphPad Prism (version 9). Before statistical analysis, identification and removal of outliers was performed using the ROUT test or Grubbs’ test available in GraphPad Prism. To test for differences in outcomes over time, we fit two-way repeated measures ANOVA, followed by Tukey’s multiple comparisons tests, if indicated. If observations were missing at random time points, we fit linear mixed effect models with individual mice as a random effect and week and group as fixed effects, followed by Tukey’s multiple comparisons tests if indicated. To test for differences in outcomes not measured over time, we fit one-way ANOVAs follwed by Tukey’s multiple comparisons tests if indicated. Either Welch ANOVA models followed by Dunnett’s T3 multiple comparisons tests or Kruskal Wallis tests followed by Dunns multiple comparisons tests were fit if data violated the assumption of homoscedasticitiy or normality, respectively. Microbiome data were statistically analyzed using Kruskal-Wallis test with FDR-corrected (Benjamini, Krieger, Yekutieli method) post-hoc analysis when indicated. Beta-diversity ordination ellipses show the 95% confidence interval for each group and statistical evaluation was conducted using PERMANOVA and PERMDISP to examine differences in centroids and heterogeneity between groups, respectively (Anderson, 2001). Heat trees provided between group comparisons and analyzed using non-parametric Wilcoxon test. LEfSe was used to identify taxa that serve as biomarkers for the different experimental conditions (Segata et al., 2011). A p-value ≤ 0.05 and LDA>2.0 was considered statistically significant.

By 4 weeks into the intervention, HFD-fed animals weighed significantly more than SD-fed animals (p<0.01 for all pairwise comparisons between HFD and SD; Figure 1A). At termination, animals on the HFD weighed significantly more than mice fed SD, regardless of genotype (p<0.01 for all pairwise comparisons between HFD and SD groups; Figure 1B); however, the KO+HFD animals weighed significantly less than the CON+HFD animals (52.5 ± 0.8 g. vs. 57.3 ± 1.3g.; p=0.01) while weighing significantly more than both SD groups (p<0.001) (Figure 1B). Changes in body weight could not be explained by changes in food intake, as there were no significant differences in food intake between any two groups (Figure 1C). In general, tissue weights and fat depots differed between diet groups, with no differences observed between CON+HFD compared to KO+HFD (Table 1). Of note, the KO+SD animals had significantly reduced cecum weight and colon length compared to the CON+SD animals, as well as slightly higher mesenteric adipose tissue weight (Table 1).

We used 16S rRNA sequencing to characterize diet and genotype-associated differences in the microbiota. Using Kruskal-Wallis tests with FDR <0.05, there were significant differences in all six phyla detected (Figure 2A; Table 2). Specifically, Pseudomonadota (formerly Proteobacteria) were associated with genotype and were higher in CON mice compared to KO animals. Bacillidota (formerly Firmicutes) were associated with diet and were higher in HFD-fed animals. Verrucomicrobiota were specifically associated with KO+HFD, and both Actinomycetota (formerly Actinobacteria) and Mycoplasmatota (formerly Tenericutes) were associated with the CON+SD group. Bacteriodota (formerly Bacteroidetes) differences were influenced by both diet and genotype.

In addition to compositional differences between the diet groups, there were also significant differences in several parameters of a-diversity. Specifically, when comparing groups at the genus level, there were significant differences in ASV richness (p-value: 0.041, H= 8.264; Figure 2B), Shannon’s diversity (p<0.001, H=21.191; Figure 2C), and Faith’s Phylogenetic indices (p=0.027; H=9.207; Figure 2D), but not in Pileu’s Evenness (p=0.195, H=4.705; data not shown). In general, the most striking differences was in Shannon’s diversity and was driven by higher diversity in the microbiota of the CON+SD group relative to all other groups (Figure 2C).

When visualizing the Bray Curtis distances between the four groups on a Principal Coordinates Analysis (PCoA; PERMANOVA F=11.7, R2 = 0.494, p<0.001), the KO+SD forms a cluster that is distinctly separate from the other groups along the first axis, which represents 41.4% of the total diversity (Figure 3A). Introduction of a HFD produces similar shifts in the microbiota of both the CON and KO mice and there is considerable overlap between the two HFD groups, suggesting that although the SD microbiota differed between the CON and KO mice, microbiota disruption was induced in both groups receiving the HFD. There was also a significant difference in the within group variance (PERMDISP, F=3.4905; p= 0.025), suggesting that the variability in microbial profiles between animals was reduced by the HFD.

Heat trees comparing diet groups (Figure 3B) or genotypes (Figure 3C) identified microbes associated with specific conditions. Relative to HFD, the SD-fed mice were enriched in Bifidobacteriacae, Anaeroplasmataceae, Bacteriodales, Christensenellaceae, and Erysipelotrichaceae, but had reduced Lactobacillaceae, Verrucomicrobiaceae, and Peptostreptococcacaceae (Figure 3B). In comparison, the CON mice were enriched in Actinobacteria, Lactobacillaceae, Alcaligenaceae, Bacteriodaceae, and Rickenellaceae relative to KO mice, which were enriched in Peptococcaceae and Mogibacteraceae (Figure 3C).

Using LefSE, nine genera were identified as biomarkers that distinguish between the treatment groups (Figure 4). Four of these genera: Ocillospora, Allobaculum, Ruminococcus, and Bifidobacterium were mostly influenced by diet group, while Bacteroides, Sutterella, and RC4-4 were linked to genotype (Figure 4). Akkermansia and Anaeroplasma were specifically enriched in KO+HFD relative to other treatment groups (Supplementary Table S1).

Both CON+HFD and KO+HFD displayed increased intestinal permeability compared to SD-fed animals, indicated by higher circulating levels of FITC after oral gavage (CON+HFD: 1.6 ± 0.1 ug/mL; KO+HFD: 1.3 ± 0.1 ug/mL; CON+SD: 0.4 ± 0.01 ug/mL; KO+SD: 0.4 ± 0.01 ug/mL; p<0.001 for all pairwise comparisons between HFD and SD groups; Figure 5A). Circulating lipopolysaccharide binding protein (LBP), a proxy measure for circulating endotoxin (Schumann and Zweigner, 1999), was also significantly higher in HFD-fed animals regardless of genotype (CON+HFD: 14.7 ± 1.7 ng/mL; KO+HFD: 14.8 ± 1.5 ng/mL; CON+SD: 7.9 ± 0.9 ng/ml; KO+SD: 9.6 ± 1.1 ng/ml; p<0.05 for all pairwise comparisons between HFD and SD groups; Figure 5B). Plasma zonulin (ZO) (more accurately, zonulin-family peptides (Fasano, 2021)), and intestinal alkaline phosphatase (IAP), an enzyme that dephosphorylates LPS (Bates et al., 2007), did not significantly differ between any of the groups (ZO: p=0.08, Figure 5C; IAP: p=0.2, Figure 5D). In addition, secretory immunoglobin (sIgA) measured in ileum was also the same across diet and genotype (p=0.8; Figure 5E).

Arterial stiffness was assessed via arterial pulse wave velocity (aPWV). After the 6-month dietary intervention, CON+HFD animals displayed significantly higher aPWV than CON+SD animals (CON+HFD: 525.4 ± 16.5 cm/sec; CON+SD: 455.2 ± 16.5 cm/sec; p=0.02, Figure 5F). However, TLR4 deletion exerted a protective effect against the HFD such that aPWV was not elevated in KO+HFD mice compared to CON+SD or KO+SD, Figure 5F). Thoracic aorta DHE intensity was similar in all groups (p=0.6) (Figure 5G). Glucose response (Figure 6A) and area under the (AUC) (Figure 6B) were significantly higher in all groups compared to the CON+SD cohort. Similarly, insulin responses (Figures 6C, D) were increased in HFD-fed animals compared to CON+SD, while KO+SD was not significantly different from any of the groups.