To evaluate the impact of P. gingivalis infection on WD-induced NAFLD, mice fed either a CD or a WD for 4 weeks were infected with P. gingivalis or treated with vehicle for 3 weeks and sacrificed after a total of 16 weeks (Supplementary Figure 1). To evaluate the influence of P. gingivalis infection on the progression of WD-induced NAFLD, steatosis and hepatic inflammation were evaluated by H&E staining of the liver (Figure 1A). WD feeding of vehicle-treated mice resulted in hepatic steatosis when compared with mice fed a CD (Figure 1A). P. gingivalis oral infection induced higher total lipid surface area in liver samples when compared with vehicle-treated mice fed a WD (Figure 1B). The increased steatosis due to P. gingivalis infection of mice fed a WD when compared with their vehicle-treated counterparts was characterized by a higher abundance of medium and small lipid droplets (Figure 1C). Steatosis was accompanied by an increase in the number of inflammatory foci as compared with vehicle-treated, WD-fed mice (Blue arrows in Figures 1A,D). These results indicate that P. gingivalis oral infection further disrupts lipid metabolism in the livers of steatotic mice. Minimal staining was observed in the livers of vehicle-treated and P. gingivalis-infected mice fed a WD stained with Picrosirius Red (collagen, reticulin) or antibodies against CD4⁺ T cells and F4/80 (macrophages) (data not shown).

Periodontal disease has been tightly linked to insulin resistance in obese and non-obese human studies [24]. Insulin resistance correlates strongly with NAFLD severity and is believed to play a crucial role in the pathogenesis of NAFLD in humans [25, 26]. To evaluate the effect of P. gingivalis on glucose metabolism, we measured fasting blood glucose levels, glucose tolerance, and insulin tolerance. P. gingivalis infection of mice fed either a CD or a WD did not influence basal fasting blood glucose levels, though WD feeding of vehicle-treated mice did when compared with vehicle-treated mice fed a CD (Supplementary Table 1). P. gingivalis infection resulted in impaired glucose tolerance in mice fed a WD when compared with vehicle-treated mice fed a WD (Figures 2A,B). The impairment of glucose tolerance in P. gingivalis-infected mice was not associated with a difference in insulin sensitivity (Figures 2C,D). P. gingivalis infection reduced the insulin sensitivity of mice fed a CD (Figures 2C,D). These results indicate that P. gingivalis infection exacerbates glucose intolerance experienced by mice fed a WD.

To identify host processes involved with P. gingivalis-induced steatosis and insulin resistance, we measured mRNA expression levels of genes associated with metabolism and inflammation in the liver, adipose tissue, and ileum of P. gingivalis-infected mice (Figure 3; see all gene expression results in Supplementary Table 2 online). Vehicle-treated mice fed a WD experienced increased expression of genes involved in pro-inflammatory signaling, macrophage recruitment, and bacterial sensing when compared with vehicle-treated mice fed a CD (Figure 3; Supplementary Table 2). P. gingivalis infection partially reversed many of the effects of WD feeding. In contrast to vehicle-treated, WD-fed mice, P. gingivalis-infected mice fed a WD experienced increased hepatic expression of the pro-inflammatory cytokine Il-5 and reduced expression of the cytokines Il-6 and Il-10, macrophage marker Adgre1, Tlr4, as well as Cd5l, which controls apoptosis and immune cell differentiation via regulating lipolysis (Figure 3). Together, these findings suggest that P. gingivalis infection suppresses some WD-induced inflammation-associated pathways while also inducing other inflammatory signaling pathways.

WD feeding increased the masses of all adipose tissue depots in vehicle-treated mice compared with their CD-fed counterparts (Supplementary Table 1). Mice fed a WD demonstrated an epididymal adipose tissue gene expression profile characterized by reduced insulin sensitivity and lipid oxidation but increased lipid accumulation, inflammation, macrophage recruitment, and adipose tissue remodeling (Supplementary Table 2). P. gingivalis infection resulted in an increase in the weight of the metabolically active mesenteric adipose tissue of mice fed a WD as compared with mice which were vehicle-treated (Supplementary Table 1). The increase in mesenteric adipose tissue weight corresponded with decreased expression of the lipid oxidation gene Acox1 and aryl hydrocarbon receptor (Ahr), a highly central regulatory gene involved in processes including antimicrobial peptide production, detoxification, and cell metabolism (Figure 3). The effect of P. gingivalis on the epididymal adipose tissue of WD-fed mice was limited to suppressed Cd5l expression when compared to vehicle-treated mice fed a WD (Figure 3).

To evaluate intestinal inflammation, we examined gene expression in the ileum, a site highly active in generation of host-interactive metabolites and the primary site of lipid absorption [27, 28] (Figure 3). When compared with vehicle-treated mice fed a WD, P. gingivalis infection of WD-fed mice induced expression of genes associated with inflammation (Ccl2, Il-5) but suppressed expression of genes associated with glucose transport and lipid storage (Glut4, Plin2) (Supplementary Table 2; Figure 3). In mice fed either a WD or a CD, P. gingivalis infection suppressed ileal Cd5l when compared with their vehicle-treated counterparts (Supplementary Table 2; Figure 3). WD feeding of vehicle-treated mice induced expression of genes involved in metabolic regulation, lipid storage, and inflammation and reduced expression of the glucose-sensitive transporter gene Glut4 when compared with vehicle-treated mice fed a CD (Supplementary Table 2; Figure 3).

Structural disruption of gut microbiota and associated inflammation are considered important etiological factors in diet-induced insulin resistance and steatosis [29]. Since P. gingivalis infection exacerbated diet-induced steatosis, we next investigated whether P. gingivalis had an additional impact on the dysbiotic gut microbiota of mice fed a WD by performing 16S rRNA gene sequencing of the cecal contents collected from mice fed a WD or a CD and either infected with P. gingivalis or treated with vehicle. A total of 281 amplicon sequence variants or ASVs were identified from 5,994,592 total reads assigned to 490 non-chimeric sequences and filtered by presence in more than 20% of samples (Supplementary Figure 2).

There was no significant difference in alpha diversity due to P. gingivalis infection of WD-fed mice or CD-fed mice when compared with their vehicle-treated counterparts (Figure 4A; Supplementary Figure 3). To determine the dissimilarity between the experimental groups' community membership and structure, beta diversity was assessed by unweighted and weighted UniFrac distances. These indices use phylogenetic relatedness and measure ASV presence or abundance, respectively [30]. Among mice fed a WD, the microbiota of vehicle-treated and P. gingivalis-infected mice were distinct from one another when assessed for ASV abundance rather than ASV presence (Figure 4B; Supplementary Figure 4). This was also evident when the microbiota samples were measured by Euclidean distance of sample relatedness; within each diet-specific cluster, the microbiota of P. gingivalis-infected mice clustered separately from the microbiota of vehicle-treated and untreated mice (Supplementary Figure 4B). Variation between the microbiota of P. gingivalis-infected mice fed a WD was less than those of vehicle-treated mice fed a WD. WD feeding alone affected the gut microbiota resulting in increased alpha diversity as measured by the Shannon index (Supplementary Table 3), which summarizes both taxonomic richness and evenness, and caused diet-specific clustering of the microbiota from the four experimental groups by both weighted and unweighted UniFrac (Figure 4B; Supplementary Figure 4). In contrast to WD-fed mice, P. gingivalis infection of mice fed a CD was associated with distinct community presence but not abundance when compared with vehicle-treated mice fed a CD (Figure 4B; Supplementary Figure 4).

To evaluate the significance of P. gingivalis's influence on the gut microbiota, ASVs were pooled by their taxonomic assignment (Supplementary Figure 5) and differences in the taxa were measured using linear discriminant analysis effect size (LEfSe). When compared with vehicle-treated mice fed a WD, the gut microbiota of P. gingivalis-infected mice fed a WD demonstrated reductions in the abundance of unclassified members of the family Lachnospiraceae and the genera Anaerosporobacter, the flavonoid degrader Flavonifractor, and Eisenbergella (LDA > 2.0, α < 0.05) (Figure 5A). WD feeding of vehicle-treated mice produced expected differences in the gut microbiota composition when compared with vehicle-treated mice fed a CD (LDA > 2.0, α < 0.05) (Supplementary Figures 5, 6).

P. gingivalis infection induced more changes in the composition of the gut microbiota of mice fed a CD (Figure 5B; Supplementary Figure 5). The phylum Bacteroidetes was decreased in the ceca of P. gingivalis-infected mice fed a CD, primarily due to the genus Bacteroides, though the Bacteroidetes genus Alistipes was increased when compared with vehicle-treated, CD-fed mice. The effect of P. gingivalis on the gut microbiota of CD-fed mice was also characterized by increased abundance of the Firmicutes genus Robinsoniella and decreases in the genera including Flavonifractor, the lactate-producing Coprobacillus, and the butyrate-producing Clostridium cluster XIVa, a key resident of the colonic mucous layer.

To evaluate the potential functional significance of the observed microbiota changes, PICRUSt2 was used to predict metagenome content and contribution of the metagenome to MetaCyc pathways. No pathways were differentially abundant (adjusted p-value < 0.01 by DESeq2) in their inferred abundance when comparing P. gingivalis-infected, WD-fed mice with vehicle-treated, WD-fed mice. Comparing vehicle-treated, WD-fed mice with their CD-fed counterparts substantially changed the predicted metagenome content of the cecal microbiota (Supplementary Figure 7). The gut microbiota of P. gingivalis-infected mice fed a CD demonstrated predicted enrichment of multiple functions (Figure 6; Supplementary Figure 7). The largest fold differences included enrichment of some vitamin K and L-Lysine biosynthesis pathways and reduction of some carbohydrate degradation pathways. P. gingivalis infection of CD-fed mice was also associated with increased predicted potential of the Bifidobacterium shunt, which was in large part due to an increase in Bifidobacterium that did not quite achieve significance by LEfSe due to sample variability. Half of the pathways impacted by P. gingivalis infection in mice fed a CD were influenced by WD feeding in the same direction (Supplementary Figure 7). The only pathways affected in the opposite direction were the two naphthanoate and one menaquinol biosynthesis pathways, which were predicted to be enriched slightly in P. gingivalis-infected, CD-fed mice compared to vehicle-treated, CD-fed mice but highly reduced in vehicle-treated, WD-fed mice compared to vehicle-treated, CD-fed mice. Collectively, these results indicate that the combination of P. gingivalis and WD feeding results in the expression of distinct functional pathways by the gut microbiota.

To evaluate the influence of the microbiota community on NAFLD, Pearson correlation coefficients for P. gingivalis- and vehicle-treated mice fed a WD (Figure 7) were calculated correlating taxa and MetaCyc pathways from the chronic timepoint with host gene expression and phenotype. In mice fed a WD, glucose tolerance and mesenteric adipose tissue weight were correlated with the differential abundance of a collection of pathways. This collection was enriched for cobalamin production and cobalamin-dependent pathways including (S)-propane-1,2-diol degradation, deoxyribonucleotide biosynthesis, methanogenesis, and multiple fermentation pathways. These pathways were primarily associated with the abundances of Lachnospiraceae, Ruminococcaceae, and Clostridiales. These findings suggest that elements of the microbiota composition and function are associated with glucose metabolism dysfunction and adiposity in mice fed a WD during chronic P. gingivalis infection.

We previously demonstrated that P. gingivalis infection has distinct effects on host inflammatory and metabolic processes during acute vs. chronic infection [31]. However, associated changes to gut microbiota of mice fed a WD during acute vs. chronic P. gingivalis infection has not been investigated. To assess the impact of P. gingivalis acute infection on changes in the gut microbiota we collected feces 1 day after the final P. gingivalis exposure in mice fed either a WD or a CD (see Supplementary Figure 1). A total of 171 amplicon sequence variants or ASVs were identified from 771,769 total reads assigned to 392 non-chimeric sequences and filtered by presence in more than 20% of samples (Supplementary Figure 8).

Acute P. gingivalis infection reduced alpha diversity as measured by the Inverse Simpson index, which preferentially considers the most common species when evaluating both evenness and richness [32], reversing the effect of WD feeding (Figure 8A; Supplementary Figure 9). Acute P. gingivalis infection did not influence the beta diversity of the fecal microbiota collected from mice fed either a WD or a CD (Supplementary Figure 10). Acute infection of mice fed a WD resulted in increased abundance of the phylum Actinobacteria (due to the genus Bifidobacterium) and the genus Clostridium sensu stricto, which represents the truest Clostridium species [33], when compared with vehicle-treated mice fed a WD (Figure 8B; Supplementary Figure 11). Decreases of genera including Turicibacter, Lactococcus, Christensenella, the Clostridium clusters XIVa and XIVb, and Anaerotruncus were also observed (Figure 8B).

During P. gingivalis acute infection we observed substantial inferred enrichment of predicted pathways including the cob (II) yrinate ac-diamide biosynthesis pathway, the anaerobic pathway for the production of B vitamin and cofactor cobalamin, and the urea cycle (Figure 8C). WD feeding during the acute phase was associated with a predicted enrichment of fermentation pathways, altered ratios of energy metabolism and TCA cycle pathways, and was also associated with increased inferred urea cycle potential (Supplementary Figure 12). P. gingivalis acute infection did not significantly alter the gut microbiota of mice fed a CD at the taxonomic level. Our findings suggest that P. gingivalis oral infection minimally disrupted the gut microbiota of mice fed a CD. However, P. gingivalis infection immediately perturbs the already dysbiotic gut microbiota of mice fed a WD, leading to persistence of a more moderate form of dysbiosis characterized by reduced community diversity and increased variation between samples during chronic infection.

Male C57BL/6J mice (Jackson Laboratory) of 8 weeks of age were randomly assigned to either a high-fat, high-sucrose Western diet (WD) (Teklad Global 08811; 23% fat w/w) or a refined, matched low-fat, low-glycemic index control diet (CD) (Teklad Global 120455; 6.2% fat). After 4 weeks on either diet, half the mice (n = 5–8) on each diet were inoculated with 100 μL of P. gingivalis (1 × 10⁹ CFU) suspended in vehicle (2% carboxymethylcellulose in phosphate-buffered saline), topically applied to the buccal surface of the maxillary gingiva 5 times a week for 3 weeks (Supplementary Figure 1). Vehicle-treated mice (CD-Vehicle, WD-Vehicle in Supplementary Figure 1) received the same treatment with vehicle alone. One additional group per diet was included to evaluate the effect of sham infection with vehicle, since carboxymethylcellulose is part of a standard protocol for P. gingivalis infection but has also been recently identified as capable of modulating the gut microbiome in mice [63]. Fecal pellets were collected 1-day post-infection. Animals were euthanized after a total of 16 weeks to allow sufficient time for NAFLD development. At sacrifice, weights were recorded for the cecum, liver, mesenteric, inguinal, and epididymal adipose tissue, and whole body (Supplementary Table 1). The ileum, half the liver, mesenteric adipose tissue, and cecal contents were flash frozen and stored at −80°C. The remaining portion of the liver was fixed overnight in 10% formalin and stored in 70% ethanol for paraffin embedding.

P. gingivalis was cultured and administered as previously described [23]. Briefly, P. gingivalis strain 381 was grown anaerobically on blood agar plates and used to inoculate brain heart infusion broth supplemented with yeast extract, hemin, and menadione.

Intraperitoneal glucose and insulin tolerance tests were performed during experimental week 13 or 14, respectively (Supplementary Figure 1). Mice were fasted for 5 h then injected intraperitoneally with either 1 g/kg of 20% glucose in PBS or 0.75 U/kg of insulin in PBS (Humulin, Eli Lilly and Company, Indianapolis IN). Blood glucose was measured at time points 0, 15, 30, 45, 90, 120 min post-injection. Glucose concentration was determined with a glucose meter (One Touch® Ultra® 2, LifeScan, Millpitas, CA). The area under the curve (AUC) (0–120 min) was calculated for each group of mice.

RNA was extracted from frozen samples of the ileum, liver, and mesenteric adipose tissue using a bead-based tissue disrupter (TissueLyser II, Qiagen, Germantown, MD) and with phenol-chloroform extraction (Trizol, Invitrogen) followed by purification using a RNeasy Mini kit (Cat. no. 74104, Qiagen). RNA was made into cDNA using a High-Capacity RNA-to-cDNA™ Kit (Cat. no. 4387406, ThermoFisher Scientific, Waltham MA). Standard Taqman probes were ordered from ThermoFisher Scientific for 30 genes selected for their relevance to immune or metabolic function in the tissue of interest (probes listed in Supplemental Table 2). cDNA was probed in multiplex with the housekeeping gene B-2-Microglobulin (Taqman probe Mm00437762_m1) in 384-well plates using an Applied Biosystems™ QuantStudio™ 6 Flex Real-Time PCR System (ThermoFisher Scientific). Fold change was calculated using the ddCt method. Results were reported as fold relative to an average across all four primary groups in Supplementary Table 2.

Liver samples fixed in 10% formalin were embedded in paraffin. 5-μm-thick sections were processed and stained with haematoxylin and eosin (H&E). Lipid content was quantified using ImageJ software and R.

Five mice per group were randomly selected for 16S rRNA gene sequencing during acute and chronic infection. Control groups (CD-None, WD-None in Supplementary Figure 1) were only sequenced during chronic infection.

Fecal samples collected the day after the final infection were stored at −80°C. DNA was extracted from samples in DNA/RNA Shield^TM (Zymo Research, Irvine CA) using the ZymoBIOMICS®-96 MagBead DNA Kit (Zymo Research). Library preparation and sequencing were performed by the Zymo Research Corporation (Irvine, CA). For each sample, an amplicon library of the 16S rRNA gene was constructed by performing PCR using primers flanking the variable regions 3 and 4 (Quick-16S™ Primer Set V3-V4; Zymo Research) using the Quick-16S™ NGS Library Prep Kit (Zymo Research). Positive (ZymoBIOMICS® Microbial Community Standard) and negative controls were included for quality assurance. 16S rRNA amplicon libraries were then pooled in equal molar concentration, cleaned with the Select-a-Size DNA Clean & Concentrator™ (Zymo Research), and sequenced on an Illumina® MiSeq™ with a ZymoBIOMICS® Microbial Community Standard 3 reagent kit (600 cycles, paired-end format).

Cecal contents collected at the final timepoint were stored at −80°C and processed as previously described [23]. Briefly, frozen cecal samples were homogenized using a bead-based tissue disrupter (TissueLyser II, Qiagen) followed by phenol-chloroform extraction (Trizol, Invitrogen) and ethanol precipitation. Isolated DNA was purified (AllPrep DNA/RNA Mini Kit, Qiagen). Library preparation and sequencing were performed by the Tufts University Core Facility Genomics Core (Boston, MA). For each sample, an amplicon library of the 16S rRNA gene was constructed by performing PCR using primers flanking the variable region 4, followed by a nested PCR to introduce Illumina adaptors. Positive and negative controls were included for quality assurance. 16S rRNA amplicon libraries were then pooled in equal molar concentration and sequenced with an Illumina® MiSeq™ using MiSeq V2 chemistry.

Raw data were converted to fastq format using bcl2fastq from Illumina. Unique amplicon sequences were inferred and chimeric sequences removed from raw reads using the Dada2 pipeline version 1.0.3 in R [64]. Briefly, for the 34 cecal samples collected during chronic infection, 6,523,979 reads were filtered, trimmed, and denoised using the filterAndTrim(), learnErrors(), and dada() functions down to a total of 5,994,592 sequences, 91.1% of which were successfully merged for a total of 490 non-chimeric amplicon sequence variants or ASVs (Supplementary Figure 1). Those ASVs were filtered down to 281 ASVs which were present in at least 20% of samples with a read count of more than 3 reads per sample. For the 20 fecal samples collected during acute infection, 771,769 reads were filtered, trimmed, and denoised for a total of 438,752 sequences, 99.1% of which were successfully merged. With the same criteria as above, the 392 resulting ASVs were filtered down to 171. Samples were not rarified since the smallest and largest library sizes only differed by 3-fold for cecal samples and 2-fold for fecal samples [65].

Taxonomy was assigned using the assign Taxonomy() function and the Ribosomal Database Project (RDP) [66] training set 16 (available from https://benjjneb.github.io/dada2/training.html) and a phylogenetic tree was produced using the phangorn version 2.5.5 and DECIPHER version 2.14.0 R packages.

Computational analyses and visualization of alpha and beta diversity were performed using using the phyloseq package version 1.30.0 in R and Microbiome Analyst (www.microbiomeanalyst.ca). Alpha diversity of filtered reads was assessed by the Shannon index. To measure beta diversity, the 10% of ASVs with the lowest interquartile range were removed and the remaining data was scaled using the total sum method. Beta diversity was calculated using the phylogeny-based unweighted UniFrac and weighted UniFrac distances based on the ASV distribution across samples, visualized by Principal Coordinate Analysis, and significance assessed by Permutational ANOVA (PERMANOVA) [30]. The unweighted UniFrac distance matrices measures ASV presence, while the weighted UniFrac distance matrices account for ASV abundance. Taxonomic abundance calculated in R. To elucidate community biomarkers of treatment effects, differential abundance of taxa across experimental groups was assessed using Linear discriminant analysis (LDA) of effect size (LEfSe) in the online Galaxy workflow framework (http://huttenhower.sph.harvard.edu/galaxy/).

Microbial gene content was inferred from ASV abundance using Phylogenetic investigation of communities by reconstruction of unobserved states 2 (PICRUSt2) (https://github.com/picrust/picrust2; v2.3.2-b) [67]. PICRUSt2 is a significant expansion of PICRUSt with a reference genomes database over ten times the size and provides MetaCyc pathway predictions comparable with typical shotgun metagenomics datasets [67]. the DESeq2 package in R with an adjusted p-value cutoff for significance of 0.05. Differentially abundant inferred pathways were visualized in R. Despite the improvements in PICRUSt2, a major limitation of PICRUSt inferences is that it depends on accurate gene annotations. Previous research has shown that microbial gene annotations are notoriously inaccurate, making biological interpretations of microbiome community function uncertain [68]. In addition, these genes may not be transcribed or translated, limiting the impact of their annotated function. Conclusions about microbiome function derived either from metagenomics or from PICRUSt should be treated as hypotheses that require further in-depth validation through functional assays.

Student's t-test was applied to compare two groups. One-way analysis of variance followed by a t-test with Bonferroni correction or a two-way, repeated measures analysis of variance followed by a t-test with Tukey correction were performed for multiple group comparisons using RStudio or GraphPad PRISM. Graphical inspection using a Q-Q plot was employed to test normality. Data that violated normality was log-transformed. R, GraphPad PRISM 8, and microbiomeanalyst.ca were used for statistics and graphics. Unless otherwise noted, an α of 0.05 was considered to reject null hypothesis.