Propionic acid (PPA), an antifungal agent and common dietary additive, has been shown to cause abnormal neurodevelopment in mice accompanied by gastrointestinal dysfunction, which may be caused by gut dysbiosis. A link between dietary PPA exposure and gut microbiota dysbiosis has been suggested, but has not been directly investigated. Here, we investigated PPA-associated changes in gut microbiota composition that may lead to dysbiosis. Gut microbiomes of mice fed an untreated diet (n=9) and a PPA-enriched diet (n=13) were sequenced using long-range metagenomic sequencing to assess differences in microbial composition and bacterial metabolic pathways. Dietary PPA was associated with an increase in the abundance of significant taxa, including several Bacteroides, Prevotella, and Ruminococcus species, members of which have previously been implicated in PPA production. The microbiomes of PPA-exposed mice also had more pathways related to lipid metabolism and steroid hormone biosynthesis. Our results indicate that PPA can alter the gut microbiota and its associated metabolic pathways. These observed changes highlight that preservatives classified as safe for consumption can influence the composition of the gut microbiota and, in turn, human health. Among them, P, G or S is selected depending on the classification level being analyzed. To minimize the impact of false positive classifications, a minimum relative abundance threshold of 1e-4 (1/10,000 reads) was adopted. Prior to statistical analysis, the relative abundances reported by Bracken (fraction_total_reads) were transformed using the centered log-ratio (CLR) transformation (Aitchison, 1982). The CLR method was chosen for data transformation because it is scale-invariant and sufficient for non-sparse datasets (Gloor et al., 2017). The CLR transformation uses the natural logarithm. The count data reported by Bracken were normalized using the relative log expression (RLE) (Anders and Huber, 2010). Figures were generated using a combination of matplotlib v. 3.7.1, seaborn v. 3.7.2 and sequential logarithms (Gloor et al., 2017). 0.12.2 and stantanotations v. 0.5.0 (Hunter, 2007; Waskom, 2021; Charlier et al., 2022). Bacillus/Bacteroidetes ratio was calculated for each sample using normalized bacterial counts. Values reported in the tables are rounded to 4 decimal places. Simpson diversity index was calculated using the alpha_diversity.py script provided in the KrakenTools v. 1.2 package (Lu et al., 2022). Bracken report is provided in the script and Simpson index “Si” is provided for the -an parameter. Significant differences in abundance were defined as mean CLR differences ≥ 1 or ≤ -1. A mean CLR difference of ±1 indicates a 2.7-fold increase in the abundance of a sample type. The sign (+/-) indicates whether the taxon is more abundant in the PPA sample and the control sample, respectively. Significance was determined using the Mann-Whitney U test (Virtanen et al., 2020). Statsmodels v. 0.14 (Benjamini and Hochberg, 1995; Seabold and Perktold, 2010) was used, and the Benjamini-Hochberg procedure was applied to correct for multiple testing. An adjusted p-value ≤ 0.05 was used as the threshold for determining statistical significance.
The human microbiome is often referred to as “the last organ of the body” and plays a vital role in human health (Baquero and Nombela, 2012). In particular, the gut microbiome is recognized for its system-wide influence and role in many essential functions. Commensal bacteria are abundant in the gut, occupying multiple ecological niches, utilizing nutrients, and competing with potential pathogens (Jandhyala et al., 2015). Diverse bacterial components of the gut microbiota are capable of producing essential nutrients such as vitamins and promoting digestion (Rowland et al., 2018). Bacterial metabolites have also been shown to influence tissue development and enhance metabolic and immune pathways (Heijtz et al., 2011; Yu et al., 2022). The composition of the human gut microbiome is extremely diverse and depends on genetic and environmental factors such as diet, gender, medications, and health status (Kumbhare et al., 2019).
Maternal diet is a critical component of fetal and neonatal development and a putative source of compounds that may influence development (Bazer et al., 2004; Innis, 2014). One such compound of interest is propionic acid (PPA), a short-chain fatty acid by-product obtained from bacterial fermentation and a food additive (den Besten et al., 2013). PPA has antibacterial and antifungal properties and is therefore used as a food preservative and in industrial applications to inhibit mold and bacterial growth (Wemmenhove et al., 2016). PPA has different effects in different tissues. In the liver, PPA has anti-inflammatory effects by affecting cytokine expression in macrophages (Kawasoe et al., 2022). This regulatory effect has also been observed in other immune cells, leading to downregulation of inflammation (Haase et al., 2021). However, the opposite effect has been observed in the brain. Previous studies have shown that PPA exposure induces autism-like behavior in mice (El-Ansary et al., 2012). Other studies have shown that PPA can induce gliosis and activate pro-inflammatory pathways in the brain (Abdelli et al., 2019). Because PPA is a weak acid, it can diffuse through the intestinal epithelium into the bloodstream and thus cross restrictive barriers including the blood-brain barrier as well as the placenta (Stinson et al., 2019), highlighting the importance of PPA as a regulatory metabolite produced by bacteria. Although the potential role of PPA as a risk factor for autism is currently under investigation, its effects on individuals with autism may extend beyond inducing neural differentiation.
Gastrointestinal symptoms such as diarrhea and constipation are common in patients with neurodevelopmental disorders (Cao et al., 2021). Previous studies have shown that the microbiome of patients with autism spectrum disorders (ASD) differs from that of healthy individuals, suggesting the presence of gut microbiota dysbiosis (Finegold et al., 2010). Similarly, the microbiome characteristics of patients with inflammatory bowel diseases, obesity, Alzheimer’s disease, etc. also differ from those of healthy individuals (Turnbaugh et al., 2009; Vogt et al., 2017; Henke et al., 2019). However, to date, no causal relationship has been established between the gut microbiome and neurological diseases or symptoms (Yap et al., 2021), although several bacterial species are thought to play a role in some of these disease states. For example, Akkermansia, Bacteroides, Clostridium, Lactobacillus, Desulfovibrio and other genera are more abundant in the microbiota of patients with autism (Tomova et al., 2015; Golubeva et al., 2017; Cristiano et al., 2018; Zurita et al., 2020). Notably, member species of some of these genera are known to possess genes associated with PPA production (Reichardt et al., 2014; Yun and Lee, 2016; Zhang et al., 2019; Baur and Dürre, 2023). Given the antimicrobial properties of PPA, increasing its abundance may be beneficial for the growth of PPA-producing bacteria (Jacobson et al., 2018). Thus, a PFA-rich environment may lead to changes in the gut microbiota, including gastrointestinal pathogens, which may be potential factors leading to gastrointestinal symptoms.
A central question in microbiome research is whether differences in microbial composition are a cause or symptom of underlying diseases. The first step toward elucidating the complex relationship between diet, the gut microbiome, and neurological diseases is to assess the effects of diet on microbial composition. To this end, we used long-read metagenomic sequencing to compare the gut microbiomes of offspring of mice fed a PPA-rich or PPA-depleted diet. The offspring were fed the same diet as their mothers. We hypothesized that a PPA-rich diet would result in changes in gut microbial composition and microbial functional pathways, particularly those related to PPA metabolism and/or PPA production.
This study used FVB/N-Tg(GFAP-GFP)14Mes/J transgenic mice (Jackson Laboratories) that overexpress green fluorescent protein (GFP) under the control of the glia-specific GFAP promoter following the guidelines of the University of Central Florida Institutional Animal Care and Use Committee (UCF-IACUC) (Animal Use Permit Number: PROTO202000002). After weaning, mice were housed individually in cages with 1–5 mice of each sex per cage. Mice were fed ad libitum with either a purified control diet (modified open-label standard diet, 16 kcal% fat) or a sodium propionate-supplemented diet (modified open-label standard diet, 16 kcal% fat, containing 5,000 ppm sodium propionate). The amount of sodium propionate used was equivalent to 5,000 mg PFA/kg total food weight. This is the highest concentration of PPA approved for use as a food preservative. To prepare for this study, parent mice were fed both diets for 4 weeks prior to mating and continued throughout the dam’s pregnancy. Offspring mice [22 mice, 9 controls (6 males, 3 females) and 13 PPA (4 males, 9 females)] were weaned and then continued on the same diet as the dams for 5 months. Offspring mice were sacrificed at 5 months of age and their intestinal faecal contents were collected and initially stored in 1.5 ml microcentrifuge tubes at -20°C and then transferred to a -80°C freezer until host DNA was depleted and microbial nucleic acids were extracted.
Host DNA was removed according to a modified protocol (Charalampous et al., 2019). Briefly, fecal contents were transferred to 500 µl InhibitEX (Qiagen, Cat#/ID: 19593) and stored frozen. Process a maximum of 1-2 fecal pellets per extraction. The fecal contents were then mechanically homogenized using a plastic pestle inside the tube to form a slurry. Centrifuge the samples at 10,000 RCF for 5 min or until the samples have pelleted, then aspirate the supernatant and resuspend the pellet in 250 µl 1× PBS. Add 250 µl 4.4% saponin solution (TCI, product number S0019) to the sample as a detergent to loosen eukaryotic cell membranes. The samples were mixed gently until smooth and incubated at room temperature for 10 min. Next, to disrupt eukaryotic cells, 350 μl nuclease-free water was added to the sample, incubated for 30 s, and then 12 μl 5 M NaCl was added. The samples were then centrifuged at 6000 RCF for 5 min. Aspirate the supernatant and resuspend the pellet in 100 μl 1X PBS. To remove host DNA, add 100 μl HL-SAN buffer (12.8568 g NaCl, 4 ml 1M MgCl2, 36 ml nuclease-free water) and 10 μl HL-SAN enzyme (ArticZymes P/N 70910-202). Samples were mixed thoroughly by pipetting and incubated at 37 °C for 30 min at 800 rpm on an Eppendorf™ ThermoMixer C. After incubation, centrifuged at 6000 RCF for 3 min and washed twice with 800 µl and 1000 µl 1X PBS. Finally, resuspend the pellet in 100 µl 1X PBS.
Total bacterial DNA was isolated using the New England Biolabs Monarch Genomic DNA Purification Kit (New England Biolabs, Ipswich, MA, Cat# T3010L). The standard operating procedure provided with the kit is slightly modified. Incubate and maintain nuclease-free water at 60°C prior to operation for final elution. Add 10 µl Proteinase K and 3 µl RNase A to each sample. Then add 100 µl Cell Lysis Buffer and mix gently. Samples were then incubated in an Eppendorf™ ThermoMixer C at 56°C and 1400 rpm for at least 1 hour and up to 3 hours. Incubated samples were centrifuged at 12,000 RCF for 3 minutes and the supernatant from each sample was transferred to a separate 1.5 mL microcentrifuge tube containing 400 µL of binding solution. The tubes were then pulse vortexed for 5–10 seconds at 1 second intervals. Transfer the entire liquid content of each sample (approximately 600–700 µL) to a filter cartridge placed in a flow-through collection tube. The tubes were centrifuged at 1,000 RCF for 3 minutes to allow initial DNA binding and then centrifuged at 12,000 RCF for 1 minute to remove residual liquid. The sample column was transferred to a new collection tube and then washed twice. For the first wash, add 500 µL of wash buffer to each tube. Invert the tube 3–5 times and then centrifuge at 12,000 RCF for 1 minute. Discard the liquid from the collection tube and place the filter cartridge back into the same collection tube. For the second wash, add 500 µL of wash buffer to the filter without inverting. Samples were centrifuged at 12,000 RCF for 1 minute. Transfer the filter to a 1.5 mL LoBind® tube and add 100 µL of pre-warmed nuclease-free water. Filters were incubated at room temperature for 1 minute and then centrifuged at 12,000 RCF for 1 minute. Eluted DNA was stored at -80°C.
DNA concentration was quantified using a Qubit™ 4.0 Fluorometer. DNA was prepared using the Qubit™ 1X dsDNA High Sensitivity Kit (Cat. No. Q33231) according to the manufacturer’s instructions. DNA fragment length distribution was measured using an Aglient™ 4150 or 4200 TapeStation. DNA was prepared using Agilent™ Genomic DNA Reagents (Cat. No. 5067-5366) and Genomic DNA ScreenTape (Cat. No. 5067-5365). Library preparation was performed using the Oxford Nanopore Technologies™ (ONT) Rapid PCR Barcoding Kit (SQK-RPB004) according to the manufacturer’s instructions. DNA was sequenced using an ONT GridION™ Mk1 sequencer with a Min106D flow cell (R 9.4.1). Sequencing settings were: high accuracy base calling, minimum q value of 9, barcode setup, and barcode trim. Samples were sequenced for 72 hours, after which base call data were submitted for further processing and analysis.
Bioinformatics processing was performed using previously described methods (Greenman et al., 2024). The FASTQ files obtained from sequencing were divided into directories for each sample. Before bioinformatics analysis, the data were processed using the following pipeline: first, the FASTQ files of the samples were merged into a single FASTQ file. Then, reads shorter than 1000 bp were filtered using Filtlong v. 0.2.1, with the only parameter changed being –min_length 1000 (Wick, 2024). Before further filtering, read quality was controlled using NanoPlot v. 1.41.3 with the following parameters: –fastq –plots dot –N50 -o
For taxonomic classification, reads and assembled contigs were classified using Kraken2 v. 2.1.2 ( Wood et al., 2019 ). Generate reports and output files for reads and assemblies, respectively. Use the –use-names option to analyze reads and assemblies. The –gzip-compressed and –paired options are specified for read segments. Relative abundance of taxa in metagenomes was estimated using Bracken v. 2.8 ( Lu et al., 2017 ). We first created a kmer database containing 1000 bases using bracken-build with the following parameters: -d
Gene annotation and relative abundance estimation were performed using a modified version of the protocol described by Maranga et al. (Maranga et al., 2023). First, contigs shorter than 500 bp were removed from all assemblies using SeqKit v. 2.5.1 (Shen et al., 2016). The selected assemblies were then combined into a pan-metagenome. Open reading frames (ORFs) were identified using Prodigal v. 1.0.1 (a parallel version of Prodigal v. 2.6.3) with the following parameters: -d
Genes were first grouped according to Kyoto Encyclopedia of Genes and Genomes (KEGG) ortholog (KO) identifiers assigned by eggNOG to compare gene pathway abundances. Genes without knockouts or genes with multiple knockouts were removed before analysis. The average abundance of each KO per sample was then calculated and statistical analysis was performed. PPA metabolism genes were defined as any gene that was assigned a row ko00640 in the KEGG_Pathway column, indicating a role in propionate metabolism according to KEGG. Genes identified as associated with PPA production are listed in Supplementary Table 1 (Reichardt et al., 2014; Yang et al., 2017). Permutation tests were performed to identify PPA metabolism and production genes that were significantly more abundant in each sample type. One thousand permutations were performed for each gene analyzed. A p-value of 0.05 was used as a cutoff to determine statistical significance. Functional annotations were assigned to individual genes within a cluster based on the annotations of representative genes within the cluster. Taxa associated with PPA metabolism and/or PPA production could be identified by matching contig IDs in the Kraken2 output files with the same contig IDs retained during functional annotation using eggNOG. Significance testing was performed using the Mann-Whitney U test described previously. Correction for multiple testing was performed using the Benjamini-Hochberg procedure. A p-value of ≤ 0.05 was used as a cutoff to determine statistical significance.
The diversity of the gut microbiome of mice was assessed using the Simpson diversity index. No significant differences were observed between the control and PPA samples in terms of genus and species diversity (p-value for genus: 0.18, p-value for species: 0.16) (Figure 1). Microbial composition was then compared using principal component analysis (PCA). Figure 2 shows the clustering of samples by their phyla, indicating that there were differences in the species composition of the microbiomes between the PPA and control samples. This clustering was less pronounced at the genus level, suggesting that PPA affects certain bacteria (Supplementary Fig. 1).
Figure 1. Alpha diversity of genera and species composition of the mouse gut microbiome. Box plots showing Simpson diversity indices of genera (A) and species (B) in PPA and control samples. Significance was determined using the Mann-Whitney U test, and multiple correction was performed using the Benjamini-Hochberg procedure. ns, p-value was not significant (p>0.05).
Figure 2. Results of principal component analysis of the mouse gut microbiome composition at the species level. The principal component analysis plot displays the distribution of samples across their first two principal components. Colors indicate sample type: PPA-exposed mice are purple and control mice are yellow. Principal components 1 and 2 are plotted on the x-axis and y-axis, respectively, and are expressed as their explained variance ratio.
Using RLE transformed count data, a significant decrease in the median Bacteroidetes/Bacilli ratio was observed in control and PPA mice (control: 9.66, PPA: 3.02; p-value = 0.0011). This difference was due to higher abundance of Bacteroidetes in PPA mice compared to controls, although the difference was not significant (control mean CLR: 5.51, PPA mean CLR: 6.62; p-value = 0.054), while Bacteroidetes abundance was similar (control mean CLR: 7.76, PPA mean CLR: 7.60; p-value = 0.18).
Analysis of the abundance of taxonomic members of the gut microbiome revealed that 1 phylum and 77 species differed significantly between PPA and control samples (Supplementary Table 2). The abundance of 59 species in PPA samples was significantly higher than that in control samples, while the abundance of only 16 species in control samples was higher than that in PPA samples (Figure 3).
Figure 3. Differential abundance of taxa in the gut microbiome of PPA and control mice. Volcano plots display differences in the abundance of genera (A) or species (B) between PPA and control samples. Gray dots indicate no significant difference in taxa abundance. Colored dots indicate significant differences in abundance (p-value ≤ 0.05). The top 20 taxa with the largest differences in abundance between sample types are shown in red and light blue (control and PPA samples), respectively. Yellow and purple dots were at least 2.7 times more abundant in control or PPA samples than in controls. Black dots represent taxa with significantly different abundances, with mean CLR differences between -1 and 1. P values were calculated using the Mann-Whitney U test and corrected for multiple testing using the Benjamini-Hochberg procedure. Bold mean CLR differences indicate significant differences in abundance.
After analyzing the gut microbial composition, we performed a functional annotation of the microbiome. After filtering out low-quality genes, a total of 378,355 unique genes were identified across all samples. The transformed abundance of these genes was used for principal component analysis (PCA), and the results showed a high degree of clustering of sample types based on their functional profiles (Figure 4).
Figure 4. PCA results using the functional profile of the mouse gut microbiome. The PCA plot displays the distribution of samples across their first two principal components. Colors indicate sample type: PPA-exposed mice are purple and control mice are yellow. Principal components 1 and 2 are plotted on the x-axis and y-axis, respectively, and are expressed as their explained variance ratio.
We next examined the abundance of KEGG knockouts in different sample types. A total of 3648 unique knockouts were identified, of which 196 were significantly more abundant in control samples and 106 were more abundant in PPA samples (Figure 5). A total of 145 genes were detected in control samples and 61 genes in PPA samples, with significantly different abundances. Pathways related to lipid and aminosugar metabolism were significantly more enriched in PPA samples (Supplementary Table 3). Pathways related to nitrogen metabolism and sulfur relay systems were significantly more enriched in control samples (Supplementary Table 3). The abundance of genes related to aminosugar/nucleotide metabolism (ko:K21279) and inositol phosphate metabolism (ko:K07291) was significantly higher in PPA samples (Figure 5). Control samples had significantly more genes related to benzoate metabolism (ko:K22270), nitrogen metabolism (ko:K00368), and glycolysis/gluconeogenesis (ko:K00131) ( Figure 5 ).
Fig. 5. Differential abundance of KOs in the gut microbiome of PPA and control mice. The volcano plot depicts the differences in the abundance of functional groups (KOs). Gray dots indicate KOs whose abundance was not significantly different between sample types (p-value > 0.05). Colored dots indicate significant differences in abundance (p-value ≤ 0.05). The 20 KOs with the largest differences in abundance between sample types are shown in red and light blue, corresponding to control and PPA samples, respectively. Yellow and purple dots indicate KOs that were at least 2.7-fold more abundant in control and PPA samples, respectively. Black dots indicate KOs with significantly different abundances, with mean CLR differences between -1 and 1. P values were calculated using the Mann-Whitney U test and adjusted for multiple comparisons using the Benjamini-Hochberg procedure. NaN indicates that the KO does not belong to a pathway in KEGG. Bold mean CLR difference values indicate significant differences in abundance. For detailed information on the pathways to which the listed KOs belong, see Supplementary Table 3.
Among the annotated genes, 1601 genes had significantly different abundances between sample types (p ≤ 0.05), with each gene being at least 2.7-fold more abundant. Of these genes, 4 genes were more abundant in control samples and 1597 genes were more abundant in PPA samples. Because PPA has antimicrobial properties, we examined the abundances of PPA metabolism and production genes between sample types. Among the 1332 PPA metabolism-related genes, 27 genes were significantly more abundant in control samples and 12 genes were more abundant in PPA samples. Among the 223 PPA production-related genes, 1 gene was significantly more abundant in PPA samples. Figure 6A further demonstrates the higher abundance of genes involved in PPA metabolism, with significantly higher abundance in control samples and large effect sizes, while Figure 6B highlights individual genes with significantly higher abundance observed in PPA samples.
Fig. 6. Differential abundance of PPA-related genes in the mouse gut microbiome. Volcano plots depict the differences in the abundance of genes associated with PPA metabolism (A) and PPA production (B). Gray dots indicate genes whose abundance was not significantly different between sample types (p-value > 0.05). Colored dots indicate significant differences in abundance (p-value ≤ 0.05). The 20 genes with the largest differences in abundance are shown in red and light blue (control and PPA samples), respectively. The abundance of yellow and purple dots was at least 2.7 times greater in control and PPA samples than in control samples. Black dots represent genes with significantly different abundances, with mean CLR differences between -1 and 1. P values were calculated using the Mann-Whitney U test and corrected for multiple comparisons using the Benjamini-Hochberg procedure. Genes correspond to representative genes in the non-redundant gene catalog. Gene names consist of the KEGG symbol denoting a KO gene. Bold mean CLR differences indicate significantly different abundances. A dash (-) indicates that there is no symbol for the gene in the KEGG database.
Taxa with genes related to PPA metabolism and/or production were identified by matching the taxonomic identity of the contigs with the contig ID of the gene. At the genus level, 130 genera were found to have genes related to PPA metabolism and 61 genera were found to have genes related to PPA production (Supplementary Table 4). However, no genera showed significant differences in abundance (p > 0.05).
At the species level, 144 bacterial species were found to have genes associated with PPA metabolism and 68 bacterial species were found to have genes associated with PPA production (Supplementary Table 5). Among the PPA metabolizers, eight bacteria showed significant increases in abundance between sample types, and all showed significant changes in effect (Supplementary Table 6). All identified PPA metabolizers with significant differences in abundance were more abundant in PPA samples. Species-level classification revealed representatives of genera that did not differ significantly between sample types, including several Bacteroides and Ruminococcus species, as well as Duncania dubois, Myxobacterium enterica, Monococcus pectinolyticus, and Alcaligenes polymorpha. Among the PPA-producing bacteria, four bacteria showed significant differences in abundance between sample types. Species with significant differences in abundance included Bacteroides novorossi, Duncania dubois, Myxobacterium enteritidis, and Ruminococcus bovis.
In this study, we examined the effects of PPA exposure on the gut microbiota of mice. PPA can elicit different responses in bacteria because it is produced by certain species, used as a food source by other species, or has antimicrobial effects. Therefore, its addition to the gut environment via dietary supplementation may have different effects depending on tolerance, susceptibility, and the ability to utilize it as a nutrient source. Sensitive bacterial species may be eliminated and replaced by those that are more resistant to PPA or able to utilize it as a food source, leading to changes in the composition of the gut microbiota. Our results revealed significant differences in microbial composition but no effect on overall microbial diversity. The largest effects were observed at the species level, with over 70 taxa significantly different in abundance between PPA and control samples (Supplementary Table 2). Further evaluation of the composition of PPA-exposed samples revealed greater heterogeneity of microbial species compared to unexposed samples, suggesting that PPA may enhance bacterial growth characteristics and limit bacterial populations that can survive in PPA-rich environments. Thus, PPA may selectively induce changes rather than cause widespread disruption of gut microbiota diversity.
Food preservatives such as PPA have previously been shown to alter the abundance of gut microbiome components without affecting overall diversity (Nagpal et al., 2021). Here, we observed the most striking differences between Bacteroidetes species within the phylum Bacteroidetes (previously known as Bacteroidetes), which were significantly enriched in PPA-exposed mice. Increased abundance of Bacteroides species is associated with increased mucus degradation, which may increase the risk of infection and promote inflammation (Cornick et al., 2015; Desai et al., 2016; Penzol et al., 2019). One study found that neonatal male mice treated with Bacteroides fragilis exhibited social behaviors reminiscent of autism spectrum disorder (ASD) (Carmel et al., 2023), and other studies have shown that Bacteroides species can alter immune activity and lead to autoimmune inflammatory cardiomyopathy (Gil-Cruz et al., 2019). Species belonging to the genera Ruminococcus, Prevotella, and Parabacteroides were also significantly increased in mice exposed to PPA (Coretti et al., 2018). Certain Ruminococcus species are associated with diseases such as Crohn’s disease through the production of proinflammatory cytokines (Henke et al., 2019), while Prevotella species such as Prevotella humani are associated with metabolic diseases such as hypertension and insulin sensitivity (Pedersen et al., 2016; Li et al., 2017). Finally, we found that the ratio of Bacteroidetes (previously known as Firmicutes) to Bacteroidetes was significantly lower in PPA-exposed mice than in control mice due to a higher total abundance of Bacteroidetes species. This ratio has previously been shown to be an important indicator of intestinal homeostasis, and disturbances in this ratio have been associated with various disease states (Turpin et al., 2016; Takezawa et al., 2021; An et al., 2023), including inflammatory bowel diseases (Stojanov et al., 2020). Collectively, species of the phylum Bacteroidetes appear to be most strongly affected by elevated dietary PPA. This may be due to a higher tolerance to PPA or the ability to utilize PPA as an energy source, which has been shown to be true for at least one species, Hoylesella enocea (Hitch et al., 2022). Alternatively, maternal PPA exposure may enhance fetal development by rendering the gut of mouse offspring more susceptible to Bacteroidetes colonization; however, our study design did not allow such an assessment.
Metagenomic content assessment revealed significant differences in the abundance of genes associated with PPA metabolism and production, with PPA-exposed mice exhibiting a higher abundance of genes responsible for PPA production, whereas non-PPA-exposed mice exhibited a higher abundance of genes responsible for PAA metabolism (Figure 6). These results suggest that the effect of PPA on microbial composition may not be solely due to its use, otherwise the abundance of genes associated with PPA metabolism should have shown a higher abundance in the gut microbiome of PPA-exposed mice. One explanation is that PPA mediates bacterial abundance primarily through its antimicrobial effects rather than through its use by bacteria as a nutrient. Previous studies have shown that PPA inhibits the growth of Salmonella Typhimurium in a dose-dependent manner (Jacobson et al., 2018). Exposure to higher concentrations of PPA may select for bacteria that are resistant to its antimicrobial properties and may not necessarily be able to metabolize or produce it. For example, several Parabacteroides species showed significantly higher abundance in PPA samples, but no genes related to PPA metabolism or production were detected (Supplementary Tables 2, 4, and 5). Furthermore, PPA production as a fermentation byproduct is widely distributed among various bacteria (Gonzalez-Garcia et al., 2017). Higher bacterial diversity may be the reason for the higher abundance of genes related to PPA metabolism in control samples (Averina et al., 2020). Furthermore, only 27 (2.14%) of 1332 genes were predicted to be genes associated exclusively with PPA metabolism. Many genes associated with PPA metabolism are also involved in other metabolic pathways. This further demonstrates that the abundance of genes involved in PPA metabolism was higher in the control samples; these genes may function in pathways that do not result in PPA utilization or formation as a byproduct. In this case, only one gene associated with PPA generation showed significant differences in abundance between sample types. In contrast to genes associated with PPA metabolism, marker genes for PPA production were selected because they are directly involved in the bacterial pathway for PPA production. In PPA-exposed mice, all species were found to have significantly increased abundance and capacity to produce PPA. This supports the prediction that PPAs would select PPA producers and therefore predict that PPA production capacity would increase. However, gene abundance does not necessarily correlate with gene expression; thus, although the abundance of genes associated with PPA metabolism is higher in control samples, the expression rate may be different (Shi et al., 2014). To confirm the relationship between the prevalence of PPA-producing genes and PPA production, studies of the expression of genes involved in PPA production are needed.
Functional annotation of the PPA and control metagenomes revealed some differences. PCA analysis of gene content revealed discrete clusters between PPA and control samples (Figure 5). Within-sample clustering revealed that control gene content was more diverse, while PPA samples clustered together. Clustering by gene content was comparable to clustering by species composition. Thus, differences in pathway abundance are consistent with changes in the abundance of specific species and strains within them. In PPA samples, two pathways with significantly higher abundance were related to aminosugar/nucleotide sugar metabolism (ko:K21279) and multiple lipid metabolism pathways (ko:K00647, ko:K03801; Supplementary Table 3). Genes associated with ko:K21279 are known to be associated with the genus Bacteroides, one of the genera with a significantly higher number of species in the PPA samples. This enzyme can evade the immune response by expressing capsular polysaccharides (Wang et al., 2008). This may account for the increase in Bacteroidetes observed in PPA-exposed mice. This complements the increased fatty acid synthesis observed in the PPA microbiome. Bacteria utilize the FASIIko:K00647 (fabB) pathway to produce fatty acids, which may influence host metabolic pathways (Yao and Rock, 2015; Johnson et al., 2020), and changes in lipid metabolism may play a role in neurodevelopment (Yu et al., 2020). Another pathway showing increased abundance in PPA samples was steroid hormone biosynthesis (ko:K12343). There is growing evidence that there is an inverse relationship between the ability of gut microbiota to influence hormone levels and to be influenced by hormones, such that elevated steroid levels may have downstream health consequences (Tetel et al., 2018).
This study is not without limitations and considerations. An important distinction is that we did not perform physiological assessments of the animals. Therefore, it is not possible to directly conclude whether changes in the microbiome are associated with any disease. Another consideration is that the mice in this study were fed the same diet as their mothers. Future studies may determine whether switching from a PPA-rich diet to a PPA-free diet improves its effects on the microbiome. One limitation of our study, like many others, is the limited sample size. Although valid conclusions can be drawn, a larger sample size would provide greater statistical power when analyzing the results. We are also cautious about drawing conclusions about an association between changes in the gut microbiome and any disease (Yap et al., 2021). Confounding factors including age, gender, and diet can significantly influence the composition of microorganisms. These factors may explain the inconsistencies observed in the literature regarding the association of the gut microbiome with complex diseases (Johnson et al., 2019; Lagod and Naser, 2023). For example, members of the genus Bacteroidetes have been shown to be either increased or decreased in animals and humans with ASD (Angelis et al., 2013; Kushak et al., 2017). Similarly, studies of gut composition in patients with inflammatory bowel diseases have found both increases and decreases in the same taxa (Walters et al., 2014; Forbes et al., 2018; Upadhyay et al., 2023). To limit the impact of gender bias, we tried to ensure equal representation of the sexes so that differences were most likely driven by diet. One challenge of functional annotation is the removal of redundant gene sequences. Our gene clustering method requires 95% sequence identity and 85% length similarity, as well as 90% alignment coverage to eliminate false clustering. However, in some cases, we observed COGs with the same annotations (e.g., MUT) (Fig. 6). Further studies are needed to determine whether these orthologs are distinct, associated with specific genera, or whether this is a limitation of the gene clustering approach. Another limitation of functional annotation is potential misclassification; the bacterial gene mmdA is a known enzyme involved in propionate synthesis, but KEGG does not associate it with the propionate metabolic pathway. In contrast, the scpB and mmcD orthologs are related. The large number of genes without designated knockouts may result in an inability to identify PPA-related genes when assessing gene abundance. Future studies will benefit from metatranscriptome analysis, which can provide a deeper understanding of the functional characteristics of the gut microbiota and link gene expression to potential downstream effects. For studies involving specific neurodevelopmental disorders or inflammatory bowel diseases, physiological and behavioral assessments of animals are needed to link changes in microbiome composition to these disorders. Additional studies transplanting the gut microbiome into germ-free mice would also be useful to determine whether the microbiome is a driver or characteristic of disease.
In summary, we demonstrated that dietary PPA acts as a factor in altering the composition of the gut microbiota. PPA is an FDA-approved preservative widely found in various foods that, upon long-term exposure, can lead to disruption of the normal gut flora. We found changes in the abundance of several bacteria, suggesting that PPA can influence the composition of the gut microbiota. Changes in the microbiota can lead to changes in the levels of certain metabolic pathways, which can lead to physiological changes that are relevant to host health. Further studies are needed to determine whether the effects of dietary PPA on microbial composition can lead to dysbiosis or other diseases. This study lays the foundation for future studies on how PPA effects on gut composition may impact human health.
The datasets presented in this study are available in online repositories. The repository name and accession number are: https://www.ncbi.nlm.nih.gov/, PRJNA1092431.
This animal study was approved by the University of Central Florida Institutional Animal Care and Use Committee (UCF-IACUC) (Animal Use Permit Number: PROTO202000002). This study complies with local laws, regulations, and institutional requirements.
NG: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing (original draft), Writing (review & editing). LA: Conceptualization, Data curation, Methodology, Resources, Writing (review & editing). SH: Formal analysis, Software, Writing (review & editing). SA: Investigation, Writing (review & editing). Chief Judge: Investigation, Writing (review & editing). SN: Conceptualization, Project administration, Resources, Supervision, Writing (review & editing). TA: Conceptualization, Project administration, Supervision, Writing (review & editing).
The authors declared that they received no financial support for the research, authorship, and/or publication of this article.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. not applicable.
All opinions expressed in this article are solely those of the authors and do not necessarily reflect the views of their institutions, publishers, editors, or reviewers. Any products evaluated in this article, or any claims made by their manufacturers, are not guaranteed or endorsed by the publisher.
Supplementary material for this article can be found online: https://www.frontiersin.org/articles/10.3389/frmbi.2024.1451735/full#supplementary-material
Abdelli LS, Samsam A, Nasser SA (2019). Propionic acid induces gliosis and neuroinflammation by regulating PTEN/AKT pathway in autism spectrum disorders. Scientific reports 9, 8824–8824. doi: 10.1038/s41598-019-45348-z
Aitchison, J. (1982). Statistical analysis of compositional data. J R Stat Soc Ser B Methodol. 44, 139–160. doi: 10.1111/j.2517-6161.1982.tb01195.x
Ahn J, Kwon H, Kim YJ (2023). Firmicutes/Bacteroidetes ratio as a risk factor for breast cancer. Journal of Clinical Medicine, 12, 2216. doi: 10.3390/jcm12062216
Anders S., Huber W. (2010). Differential expression analysis of sequence count data. Nat Prev. 1–1, 1–10. doi: 10.1038/npre.2010.4282.1
Angelis, M. D., Piccolo, M., Vannini, L., Siragusa, S., Giacomo, A. D., Serrazanetti, D. I., et al. (2013). Faecal microbiota and the metabolome in children with autism and pervasive developmental disorder not otherwise specified. PloS One 8, e76993. doi: 10.1371/journal.pone.0076993
Averina O.V., Kovtun A.S., Polyakova S.I., Savilova A.M., Rebrikov D.V., Danilenko V.N. (2020). Bacterial neurometabolic characteristics of the intestinal microbiota in young children with autism spectrum disorders. Journal of Medical Microbiology 69, 558–571. doi: 10.1099/jmm.0.001178
Baquero F., Nombela K. (2012). The microbiome as a human organ. Clinical Microbiology and Infection 18, 2–4. doi: 10.1111/j.1469-0691.2012.03916.x
Baur T., Dürre P. (2023). New insights into the physiology of propionic acid-producing bacteria: Anaerotignum propionicum and Anaerotignum neopropionicum (formerly Clostridium propionicum and Clostridium neopropionicum). Microorganisms 11, 685. doi: 10.3390/microorganisms11030685
Bazer FW, Spencer TE, Wu G, Cudd TA, Meininger SJ (2004). Maternal nutrition and fetal development. J Nutr. 134, 2169–2172. doi: 10.1093/jn/134.9.2169
Benjamini, Y., and Hochberg, J. (1995). Controlling the false-positive rate: A practical and efficient approach to multiple testing. J R Stat Soc Ser B Methodol. 57, 289–300. doi: 10.1111/j.2517-6161.1995.tb02031.x
Post time: Apr-18-2025