Research ArticleHost-Microbe Interactions

Host mitochondria influence gut microbiome diversity: A role for ROS

See allHide authors and affiliations

Science Signaling  02 Jul 2019:
Vol. 12, Issue 588, eaaw3159
DOI: 10.1126/scisignal.aaw3159

Inheriting microbiome variation

In mice, the bacterial species that persist within the gut microbiome are maternally inherited. However, maternally inherited variations in mitochondrial DNA (mtDNA) sequence also correlate with gut microbiome diversity, as well as the production of reactive oxygen species (ROS). In mice with mtDNA variants associated with increased production of ROS, Yardeni et al. found reduced gut microbiome species diversity. When pups were cross-fostered to unlink inheritance of mtDNA and gut microbiota, after weaning, the gut microbiome species were reflective of inherited mtDNA variation. Both pharmacological and genetic reduction of mitochondrial ROS abundance increased microbiome diversity. These data suggest that microbiome diversity is genetically encoded, and further imply that antioxidants may improve the efficacy of cancer immunotherapy, which is sensitive to microbiome composition.


Changes in the gut microbiota and the mitochondrial genome are both linked with the development of disease. To investigate why, we examined the gut microbiota of mice harboring various mutations in genes that alter mitochondrial function. These studies revealed that mitochondrial genetic variations altered the composition of the gut microbiota community. In cross-fostering studies, we found that although the initial microbiota community of newborn mice was that obtained from the nursing mother, the microbiota community progressed toward that characteristic of the microbiome of unfostered pups of the same genotype within 2 months. Analysis of the mitochondrial DNA variants associated with altered gut microbiota suggested that microbiome species diversity correlated with host reactive oxygen species (ROS) production. To determine whether the abundance of ROS could alter the gut microbiota, mice were aged, treated with N-acetylcysteine, or engineered to express the ROS scavenger catalase specifically within the mitochondria. All three conditions altered the microbiota from that initially established. Thus, these data suggest that the mitochondrial genotype modulates both ROS production and the species diversity of the gut microbiome, implying that the connection between the gut microbiome and common disease phenotypes might be due to underlying changes in mitochondrial function.


The human gut contains trillions of microorganisms that coincide with an individual’s health status (1, 2). Changes in this gut microbiota community correlate with a variety of clinical phenotypes including diabetes mellitus (3), autism (4, 5), and Parkinson disease (6). A subset of these phenotypes is modified in animal models by gut microbiome transfer (5, 6). Concurrently, a wide range of human (7) and mouse (8) metabolic and degenerative diseases also correlates with mitochondrial genotypes, including diabetes mellitus (911), autism (9, 12, 13), and Parkinson disease (14). Moreover, changes in the microbiota correlate with changes in mitochondrial metabolism (15), human mitochondrial DNA (mtDNA) haplogroup variants (16), and a mouse mtDNA variant and haplogroup (17). The question then becomes does the microbiome cause the disease or do mitochondrial alterations determine both the microbiome composition and the animal phenotype?

Both mtDNA (18) and the microbiome community (19) are vertically inherited from the mother. In wild-trapped mice, the microbiome is stable over multiple generations, suggesting that vertical inheritance is the dominant mode of gut bacterial transmission (19). Because newborn pups receive their initial gut inoculum from their nursing mother (20), vertical transmission may result from repeated maternal microbiota inoculation or from the transmission of the mother’s mtDNA.

Each cell contains hundreds of mitochondria and maternally inherited mtDNAs. Mitochondria not only produce most of the cellular energy by the process of oxidative phosphorylation (OXPHOS) but also are integral to most of the cell’s metabolic pathways, regulate cytosolic Ca++, modulate cellular redox status, and generate much of the cell’s reactive oxygen species (ROS) (21). The mtDNA encodes 13 critical subunits of the OXPHOS complexes plus the 12S and 16S ribosomal RNAs (rRNAs) and the 22 transfer RNAs (tRNAs) for mitochondrial protein synthesis. Of the 13 mtDNA subunits, 7 encode polypeptides for complex I [reduced form of nicotinamide adenine dinucleotide (NAD+) (NADH) dehydrogenase], 1 for complex III (bc1 complex), 3 for complex IV (cytochrome c oxidase), and 2 for complex V [adenosine 5′-triphosphate (ATP) synthase]. All other mitochondrial polypeptide genes are encoded within the nuclear DNA (nDNA) (22, 23).

There are three phenotypically relevant classes of mtDNA mutations: maternally transmitted ancient adaptive polymorphisms, materially inherited recent deleterious mutations, and somatic mtDNA mutations that accumulate with age. The adaptive polymorphisms arose over thousands of years on radiating maternal mtDNA lineages. Those functional variants that were beneficial in a particular environment were enriched along with the linked mtDNA variants to generate a group of related haplotypes called a haplogroup. For example, a functional mutation that increased mitochondrial ROS could enhance innate immunity and be selected because of resistance to certain infections (2325). This situation has been observed in the mouse by comparing the mitochondrial ROS production of mice with the same nucleus that harbors mtDNAs from NZB and 129 or Balb/c mice lineages (2628).

Recent deleterious mutations arise along material lineages and can result in disease. Deleterious mutations can either be homoplasmic (pure mutant) or heteroplasmic (mixed mutant and normal) (23). As an example, a heteroplasmic missense mutation in the mtDNA ND6 gene at m.14600G>A (P25L) results in neurological disease (29), which is recapitulated in the equivalent mouse ND6 nt 13997 m.G A (P25L) mutation (30). mtDNA damage and mutations accumulate throughout life, eroding energy metabolism and increasing ROS production (3133). Last, mutations in nDNA-coded mitochondrial genes can also result in mitochondrial dysfunction and phenotypic manifestations (23).

Deleterious mtDNA or nDNA variants that partially impede mitochondrial OXPHOS result in redirection of electrons from the electron transport chain that would normally reduce O2 to 2H2O to add a single electron to O2, thus generating superoxide anion. Superoxide anion is the first of the ROS species, which can be converted to hydrogen peroxide (H2O2) by Mn superoxide dismutase located in the mitochondrial matrix (21). Addition of mitochondrially targeted catalase (mCAT) can remove the mitochondrial H2O2, protect the mtDNA, and extend the lifespan (31).

The mtDNAs of 129 and NZB mice differ in multiple nucleotide positions, but the most notable difference has been reported to be a polymorphism in the mtDNA DHU loop of the mtDNA tRNAArg gene. This polymorphism at nucleotide (nt) 9821 involves a homopolymer of A’s, whose length ranges from 8 to 10 A’s. The NZB mtDNA tRNAArg has 10 A’s, but 129 mtDNA has 9 A’s. The addition of extra A’s has been observed to increase mitochondrial ROS production (26, 27).

We now report the use of these mice to determine the role of the mitochondrial function on the gut microbiome composition. We found that the mitochondrial genotype of mice correlates with the gut microbiota species diversity. Even when the initial microbiota community was obtained from a foster mother with different microbiota community by cross-fostering, the microbiota community progressed toward that characteristic of the pup’s genotype within 2 months. Last, our data suggest that one important factor in the mitochondrial modulation of the gut microbiome is mitochondrial redox status and associated ROS production.


Mitochondrial genotypes correlate with gut microbiota composition

In these studies, we analyzed the gut microbiome species diversity in C57BL/6J mice harboring mtDNA variants (Table 1). To determine the gut bacterial composition of our mouse strains, we sequenced the bacterial 16S rRNA gene from fecal samples of 2-month-old male mice and used this information to determine the operational taxonomic units (OTUs) found in the gut of each strain (34). First, we determined the microbiome composition of four different mtDNA genotypes on the C57BL/6J background (Table 1): 129, NZB, NZB/129 (Fig. 1) (35), and ND6P25L (Fig. 2) (30). Analysis of the gut microbiota of these mice revealed that the Shannon diversity, a measure of gut microbiota species diversity, decreased progressively with the proportion of the NZB mtDNAs from 129 to NZB/129 to NZB mice (Fig. 1A). The differences in the microbial communities of the 129, NZB/129, and NZB mtDNA mice were confirmed using the weighted and unweighted UniFrac distances (Fig. 1, B and C). As the proportion of NZB mtDNAs increased the abundance of the Bacteroidales family, the most abundant mouse bacterial taxa increased and, concomitantly, the taxa belonging to the Firmicutes phylum (Clostridiales, Lachnospiraceae, Oscillospira, and Ruminococcaceae) decreased (Fig. 1, D and E).

Table 1 Mouse models.

For each of the mouse models used in these studies, the “MGI (Mouse Genome Informatics) strain name,” “Nuclear background,” and “Mitochondrial mutation/SNP” mtDNA sequence compared to Bl6 mtDNA reference (NC_005089) are listed. SNP, single-nucleotide polymorphism.

View this table:
Fig. 1 mtDNA genotypes correlate with gut microbiota composition.

(A to E) 16S rRNA marker gene sequencing analysis of microbiota in fecal samples from C57BL/6J mice with the indicated mtDNA genotypes. Shannon diversity index (A) data and linear regression fit across groups of at least nine mice per group are from at least three independent breeding pairs per group. Principal coordinates analysis (PCoA) of (B) weighted UniFrac and (C) unweighted UniFrac distance between samples from all experiments. The percent total variation between samples is indicated on each axis, and ellipses represent the 80% confidence interval for samples in each group. The relative abundance of five major bacterial taxa (D) and Bacteroidetes/Firmicutes (B/F) ratio (E) with linear regression fit across all samples. *P < 0.05, **P < 0.01, and ***P < 0.001 by linear mixed-effects models with ordered factors (A), permutational multivariate analysis of variance (PERMANOVA) test (B and C), generalized linear mixed-effects models with ordered factors (D), or linear mixed-effects models on log-transformed ratios with ordered factors (E).

Fig. 2 The ND6P25L mtDNA mutation reduces gut microbiota diversity in C57BL/6J mice.

(A to C) 16S rRNA marker gene sequence analysis of microbiota in fecal samples from mice harboring the mtDNA ND6P25L on the C57BL/6J nuclear background. Shannon diversity index (A) data and medians ± interquartile ranges of 16 mice per group are from at least three independent breeding pairs per group. Data were analyzed using linear mixed-effects models. Principal coordinates analysis of (B) weighted UniFrac distance between samples from all experiments visualizing community amount differences between samples, where percent total variation captured by each axis is indicated. Ellipses represent the 80% confidence interval for samples in each group. The relative abundance of five major bacterial taxa (C) in mice with the indicated mtDNA with medians ± interquartile ranges are from all samples. *P < 0.05 and **P < 0.01 by linear mixed-effects models (A), PERMANOVA test (B), or generalized linear mixed-effects models (C).

Analysis of C57BL/6J mice without and with the ND6P25L mtDNA mutation revealed that the presence of the ND6P25L mtDNA also showed a decreasing trend in the Shannon diversity (Fig. 2A). The decline in the Shannon diversity of the mtDNA ND6P25L mice was also confirmed by significant change in the weighted UniFrac distances (Fig. 2B). This involves an increase in abundance of Bacteroidales S-24 family bacteria and a decrease in the abundance of the Firmicutes phylum bacteria (Fig. 2C). Thus, these results demonstrate that the gut microbiome community can be influenced by the mtDNA genotype.

Host genotype determines the composition of gut microbiota communities

To clarify the differential role of the maternal inoculation of the pups versus the pup’s mitochondrial genotype in determining the gut microbiota, we took advantage of the inherent differences in the genotypes of the C57BL/6EiJ and C57BL/6J mouse strains, which have diverged over many generations and evolved markedly different microbiota communities (Fig. 3, A to C; outside boxes in Fig. 3, B and C, the nonfostered pups). By cross-fostering newborn pups between the C57BL/6EiJ and C57BL/6J dams, we broke the linkage between the maternal microbiota inoculation and the transmission of the mother’s genotype. We then determined the microbiota OTUs of the fostered pups after 3 weeks of nursing and then maturation to 2 months of age (Fig. 3A).

Fig. 3 Inherited host genotype determines the gut microbiome composition.

(A) The experimental design for cross-fostering between C57BL/6EiJ (red) and C57BL/6J (purple) mice. During the first 24 hours after birth, the pups were transferred between the mothers of the opposite genotype, the pups were weaned at 3 weeks, and fecal samples were collected from the pups at 2 months of age. (B and C) 16S rRNA marker gene sequence analysis of microbiota in fecal samples from pups that fostered with mothers of different mtDNA genotypes. Shannon diversity index (B) data represent medians ± interquartile ranges of at least 10 mice per group in fostered (middle boxes) and nonfostered (outside boxes) pups from at least three independent breeding pairs per group. The relative abundances of five major bacterial taxa (C) in fostered (middle boxes) and nonfostered (outside boxes) pups with medians ± interquartile ranges are from all samples. (D) Mitochondrial ROS abundance detected by Amplex Red analysis of freshly isolated liver mitochondria from the indicated mice. Data are means ± SD of three mice per group. (E) Microscopy analysis of NADH fluorescence lifetime in cross sections of mouse intestine from the indicated strains. Data are means ± SD of at least five mice per group. *P < 0.05, **P < 0.01, and ***P < 0.001 by linear mixed-effects model (B), generalized linear mixed-effects model (C), or Student’s t test (D and E).

As expected, the foster mother’s microbiota inoculum has a substantial effect on the microbiota of the fostered pups. However, by 2 months, the pup’s genotype significantly affects the microbiota community, causing it to shift back toward that characteristic of the pup’s genotype (Fig. 3, B and C). When weaned by the same genotype mother, the microbiota composition of 2-month-old C57BL/6J mice is 40% Bacteroidales S24-7, while that of C57BL/6EiJ mice is 78% Bacteroidales S24-7. When the C57BL/6J pups are fostered with C57BL/6EiJ mothers, their microbiota evolves to 68% Bacteroidales S24-7 by 2 months, and when C57BL/6EiJ pups are fostered with the C57BL/6J mothers, their microbiota communities shift to 60% Bacteroidales S24-7 (Fig. 3C, Bacteroidales S24-7). These data suggest that the animal’s genotype affects the composition of the gut microbiota community.

Mitochondrial ROS and redox status shapes the gut microbiota community

Cells harboring NZB versus 129 mtDNAs (26, 27) and mice harboring ND6P25L mtDNAs (30) have increased mitochondrial ROS production. Our data indicated that mice with NZB or ND6P25L mtDNAs have decreased Shannon diversity (Figs. 1 and 2). Thus, the gut microbiota may be modulated by the mitochondrial redox status and associated ROS production.

The first prediction of this hypothesis is that the C57BL/6EiJ mice with lower Shannon diversity should have higher mitochondrial ROS production, while the C57BL/6J mice with high Shannon diversity should have lower mitochondrial ROS production (Fig. 3B). When we assayed the mitochondrial liver H2O2 production by Amplex Red, we confirmed that the C57BL/6EiJ mitochondria did produce more ROS than C57BL/6J mitochondria, the C57BL/6J H2O2 production being a significant 12% less than that of C57BL/6EiJ mitochondria (Fig. 3D).

To determine whether the differential C57BL/6EiJ and C57BL/6J mitochondrial ROS production is related to altered cellular redox status, we analyzed intestinal tissue from the two strains using NADH fluorescence lifetime imaging microscopy (FLIM). This revealed that the NADH FLIM intensity was markedly higher in the C57BL/6EiJ small intestinal cells than in C57BL/6J cells, indicating a more oxidative redox state in the C57BL/6EiJ cells (Fig. 3E). Because H2O2 exposure of cells increases the NADH FLIM (36), we can conclude that the intestinal cells of the C57BL/6EiJ strain with a higher NADH lifetime have a higher ROS production than those of the C57BL/6J strain, thus confirming that host ROS abundance modulates the gut microbiota composition and associated Shannon diversity.

To further document that increased mitochondrial ROS production decreases the species diversity in the gut microbiota community represented by a decrease in the Shannon diversity, we aged C57BL/6J mice from 2 to 5 months to permit the accumulation of age-related oxidative damage (31). As predicted, the Shannon diversity of the C57BL/6J microbiota declines (Fig. 4A, Aging) in association with the rise in Bacteroidales S24-7 and decline in Clostridiales (Fig. 4B, Aging).

Fig. 4 Changes in host mitochondrial ROS abundance reshape the microbiota community.

(A and B) 16S rRNA marker gene sequencing analysis of microbiota in fecal samples from mice after aging, NAC treatment, or mCAT transgene expression. Shannon diversity index (A) data and the relative abundance of five major bacterial taxa (B) are medians ± interquartile ranges of at least 12 mice per group from at least three independent breeding pairs per group. (C) ROS abundance detected by Amplex Red assay in lysates of liver tissue from NAC treatment of wild-type (WT) C57BL/6J mice or WT and mCAT transgenic mice. Data are means ± SD from three independent experiments. (D) Quantitative reverse transcription polymerase PCR (qRT-PCR) analysis of human mCAT mRNA expression in the liver and small intestine tissues of WT and mCAT transgenic C57BL/6J mice. Data are means ± SD of at least three mice per group. (E) Summary of the relative abundance trends in the gut microbiota community and ROS abundance changes. *P < 0.05, **P < 0.01, and ***P < 0.001 by linear mixed-effects models (A), generalized linear mixed-effects models (B), or Student’s t test (C and D).

To test whether decreased mitochondrial ROS increases Shannon diversity, we treated C57BL/6J mice in two ways. First, we increased the amount of reduced glutathione (GSH) by treatment of the C57BL/6J mice with N-acetylcysteine (NAC) in the drinking water for 10 weeks. NAC treatment causes a significant 30% decrease in ROS production in the mouse livers (Fig. 4C, NAC). This was associated with an increase in Shannon diversity (Fig. 4A, NAC) involving a 31% decline in the Bacteroidales S24-7 and a 2.1-fold increase in Firmicutes abundance (Fig. 4B, NAC).

Next, we introduced into the C57BL/6J mice the mCAT transgene, which removes mitochondrial H2O2. To accomplish this, C57BL/6J females were crossed with C57BL/6J males heterozygous for the mCAT transgene transcribed from the cytomegalovirus promoter (31). Half of a dam’s pups from this cross received the mCAT transgene, whereas the other half did not. Because mCAT is transcribed from a constitutive promotor, the transgene is expressed in all tissues (Fig. 4D) (31). As a result, pups that inherited the mCAT transgene showed a significant 70% decrease in ROS abundance in the liver compared to the littermates without the mCAT transgene (Fig. 4C, mCAT). However, the mCAT-positive intestine expresses the human catalase mRNA 10-fold higher than the liver. At 2 months, the mCAT-positive mice manifested a significant increase in the Shannon diversity (Fig. 4A, mCAT) involving a 25% decrease in Bacteroidales S24-7 and a reciprocal increase in Firmicutes abundance (Fig. 4B, mCAT). Because the mice with and without mCAT were littermates born at the same time to the same dam, other variables could be ruled out. Thus, decreasing mitochondrial ROS production is consistently associated with increased Shannon diversity (Fig. 4E).


The gut microbiome composition correlates with clinical presentations such as diabetes mellitus (22, 23), autism (4, 5), and Parkinson disease, as does mtDNA variation in these same diseases (9, 1214, 37). We now show that the mtDNA genotype correlates with the gut microbiota and by cross-fostering experiments that the offspring’s mitochondrial redox state and ROS abundance can restructure the gut microbiome, even when initially derived from a fostering mother who transmits a different gut microbiota community. From these observations, we can conclude that modification of the mitochondrial redox status and associated ROS production is associated with changes in the gut microbiota community, implicating the mitochondrial function in controlling the composition of the microbiome community. The question remains: How does the mitochondrial redox state and ROS production modulate the gut microbiome community?

One possibility is that the mitochondrial redox status might be modulating the gut microbiome through mitochondrial metabolites. A marked difference was observed in the NADH lifetime of C57BL/6 versus C57BL/6J, demonstrating a clear difference in the NAD+/NADH ratios in their intestinal tissues. Because the NAD+/NADH ratio is central to regulating metabolic flux through the mitochondrion, this could alter mitochondrial intermediates that could diffuse into the gut and select for altered microbiome populations. We have found that changes in the nuclear and mitochondrial NADH lifetimes correlate with changes in mitochondrial metabolites, which, in turn, modulate the epigenome (38).

Another possibility is that altered mitochondrial redox signaling could affect the function of the intestinal epithelial, neuronal, and smooth muscle cells. Because mitochondrial genetic defects are commonly associated with intestinal dysmotility, this could also affect the gut microbiota community.

Similarly, mitochondrial function is central to the function of the immune system (24). The mild mitochondrial dysfunction associated with the ND6P25L mtDNA mutation preferentially impairs the more oxidative regulatory T cells, leading to loss of inhibition of the more glycolytic effector T cells (39). Reductions in T regulatory cells would increase the inflammatory response, which could affect the gut microbial community. Mitochondria are also known to be central to the innate immune system as well. Release of mtDNA in stressed cells can activate the cGAS-Sting pathway, which regulates interferon production (24). Similarly, mtDNA oxidation within macrophages can activate the NLRP3 inflammasome, causing nuclear factor κB (NFκB) activation and stimulation of systemic inflammation (25). Perhaps mitochondrial variation might mediate the differential effects of gut microbiota inoculation on the therapeutic efficacy in metastatic melanoma patients treated with anti–PD-1 (programmed cell death protein-1) therapy (4043).

Mitochondrial NADH also regulates mitochondrial H2O2. NADH is converted to NADPH (reduced form of nicotinamide adenine dinucleotide phosphate) in the mitochondrion by the nicotinamide nucleotide transhydrogenase using the mitochondrial inner membrane potential as the added source of energy. NADPH is used to reduce oxidized glutathione (GSSG) to 2GSH, and GSH is used to reduce H2O2 to 2 H2O by glutathione peroxidase. Hence, changes in NAD+/NADH ratio could also modulate H2O2 amounts.

Conversely, mitochondrial H2O2 might act directly on specific gut microbiota. This possibility is supported by the fact that simply expressing mCAT at the mitochondrion had the single greatest impact on the gut microbiome. The function of mCAT is to convert mitochondrial H2O2 to H2O and O2. Because H2O2 tends to react with other molecules rapidly, if it is acting as a mitochondria-to-microbiota messenger, it would need to be generated in close proximity to the gut microbiota. This expectation is consistent with our observation that the expression of mCAT mRNA was 10 times higher in the gut than in other somatic tissues in the mCAT mice. Because mitochondrial H2O2 detoxification requires both NADPH and GSH, alterations in H2O2 detoxification by mCAT could also account for the observed changes in gut NADH lifetimes and the effect of adding NAC to generate more GSH.

Undoubtedly, there are other potential mitochondrial–gut microbiota signal pathways, perhaps even harkening back to the use of peptides and small molecules in quorum sensing carried over from the bacterial origin of the mitochondrion. Regardless of the molecular basis of the specific mitochondrial molecular signals that are regulating the gut microbiome that correlate with H2O2 production and NAD+/NADH redox status, our data provide substantial support for the hypothesis that the correlation between the gut microbiome composition and the range of clinical manifestations reported is because both the clinical phenotypes and the gut microbiota are regulated by the mitochondrial genotype and associated functions. This conclusion suggests new approaches for treating both complex diseases and gut microbiota dysbiosis.



The Institutional Animal Care and Use Committee from the Children’s Hospital of Philadelphia approved all protocols, and the protocols comply with all relevant ethical regulations regarding animal research. The mice were fed a 5LOD diet from PicoLab Laboratory and were maintained on a 13-hour/11-hour light-dark cycle. Seven mouse models were used for this study (Table 1); six were on C57BL/6J background (4446): B6 mtDNA, ND6 m.13997 G>A (ND6P25L) (30), 129 mtDNA homoplasmic, NZB mtDNA homoplasmic, NZB/129 mtDNA heteroplasmic (35), and transgenic (Tg)-mCAT (31). The last strain was C57BL/6EiJ. All mice, except for the Tg-mCAT, were bred and maintained in the Wallace laboratory colony for more than 10 years. The Tg-mCAT strain was purchased from The Jackson Laboratory and bred to the Wallace laboratory C57BL/6J strain. To determine the effect on NAC, 2-month-old C57BL/6J mice were divided into two groups: one group was treated with NAC (2 g/kg body weight per day) (Sigma-Aldrich, St. Louis, MO) in the drinking water for 10 weeks, and the second group served as control.

Feces collection

Fecal samples were collected from male mice at 2 to 3 months or 5 months of age in the morning. Individual mice were moved to an empty clean cage, and fresh fecal pellets were collected into a 1.5-ml Eppendorf tube and placed on dry ice. All samples were stored at −80°C until DNA extraction.


Cross-fostering experiments were performed using C57BL/6EiJ and C57BL/6J mice. Breeding pairs of C57BL/6EiJ and C57BL/6J mice were set up at the same time. Fourteen days after introduction, females were monitored daily for pregnancy stage and the males were removed from pregnant females. The entire litter of the newly born pups was transferred during the first 24 hours of life to a nursing mother from the opposite strain. After 3 weeks, all pups were weaned. Feces were collected when pups were 2 months old (20).

DNA extraction

DNA was extracted from frozen mouse pellets using the MO BIO PowerSoil HTP DNA Isolation Kit (MO BIO Laboratories, Carlsbad, CA, USA). Sequencing libraries were generated by polymerase chain reaction (PCR) amplifying the V1 and V2 regions of the 16S rRNA gene using barcoded universal primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 338R (5′-3TGCTGCCTCCCGTAGGAGT-3′) (47, 48). For each sample, PCRs were performed in quadruplicate with AccuPrime Taq High Fidelity (Invitrogen, Carlsbad, CA, USA). Individual reactions were combined, purified using AMPure XP beads (Beckman Coulter, Brea, CA, USA), and then pooled with other samples in equimolar quantities. Extraction blanks and DNA-free water were subjected to the same extraction and amplification procedures to allow for empirical assessment of environmental and reagent contamination. Positive controls, consisting of eight artificial 16S gene fragments synthesized in gene blocks (Integrated DNA Technologies, Coralville, IA, USA) and combined in known abundances, were also included. Libraries were sequenced on the Illumina MiSeq to obtain 250-base pair (bp) paired-end reads.

Bioinformatics processing

Sequence data were processed using QIIME version 1.9.1 (49). Read pairs were quality-filtered and joined using default parameters. Sequences were grouped into OTUs using UCLUST, with a 97% sequence identity threshold setting (50). Taxonomic assignments were generated using the default method in QIIME, which selects among top matches in the Greengenes reference database (51). Representative sequences from each OTU were aligned using PyNAST (49), and a phylogenetic tree was inferred from the multiple sequence alignment using FastTree (52). The Shannon diversity of each sample was calculated with the formula H=i=1Npilnpi where pi is the relative abundance of the ith organism in the sample. Similarity between samples was assessed by weighted and unweighted UniFrac distance (53, 54).

Statistical analysis

Data files from QIIME were analyzed in the R environment. Because mice sharing the same cage harbor similar microbiota (55), the cage of each mouse was modeled as a random effect in the statistical analysis. The alpha diversity differences between the genotypes were assessed with linear mixed-effects models. The PERMANOVA method was used to test for differences in bacterial community composition, as quantified by unweighted and weighted UniFrac distance between samples (56). Relative abundance was assessed using the most specific taxonomic assignment available for each OTU. Bacteroidetes/Firmicutes ratios were calculated by summing up the relative abundance at the phylum amounts. Differences between groups were tested using linear mixed-effects models on log-transformed ratios. Taxa were selected for testing if the mean abundance exceeded 1% among the samples to be analyzed. Taxon abundance was tested using generalized linear mixed-effects models. To account for multiple comparisons of taxon relative abundance, P values were adjusted to control for a false discovery rate of 5%. Each filled or empty circle in a graph represents an individual mouse.

Measuring ROS production from mouse strain livers

Mitochondrial H2O2 production was assayed by Amplex Red. Whole liver homogenates, clarified at 20,000g for 10 min in radioimmunoprecipitation assay buffer and protease inhibitor, were assayed for comparison between C57BL/6J mice with and without NAC treatment and with and without the mCAT transgene. Isolated liver mitochondria were used and assayed for the C57BL/6EiJ versus C57BL/6J comparison. Mice were euthanized by cervical dislocation, and the livers were quickly removed and placed on ice. Livers were washed with the isolation buffer [215 mM mannitol, 75 mM sucrose, 0.1% bovine serum albumin (BSA), 1 mM EGTA, and 20 mM Hepes (Na+) (pH 7.2)] and then transferred to 10 ml of isolation buffer for trimming and glass Dounce homogenization. For mitochondrial isolation, unbroken cells and nuclei were removed by centrifugation at 1000g for 10 min, and the mitochondria were pelleted from the supernatant by centrifugation at 20,000g for 10 min. The mitochondrial pellet was washed in isolation buffer without EGTA (10,000g for 10 min) and resuspended in isolation buffer without EGTA and BSA. Protein amounts were determined using bicinchoninic acid (BCA) assay (Pierce).

One hundred micrograms of isolated liver homogenate or isolated mitochondria was assayed using the Amplex Red Hydrogen Peroxide/Peroxidase Assay Kit (Invitrogen, catalog no. A22188). Amplex Red fluorescence was measured with an excitation wavelength of 530 nm and a 590-nm emission filter.

FLIM of NADH autofluorescence from mouse small intestine

Cross sections of freshly isolated small intestine were imaged in Tyrodes buffer (pH 7.4) using LSM710 (Zeiss) equipped with a time-correlated single-photon counting module (HPM-100-40 and SPCM 9.81, Becker and Hickl) (57). NADH was excited by a femtosecond-pulsed two-photon laser (Coherent) at 730 nm, and its autofluorescence signal was detected through a 680-nm short-pass and 460/50-nm band-pass emission filter. To cover the whole cross section, a 20× lens (numerical aperture, 0.8) was used in combination with the tile imaging function (4 × 4 tiles, 2121.2 μm by 2121.2 μm). FLIM images were analyzed in SPCImage 7.4 using a biexponential decay model with T1, T2, and the shift not specified. Non-NADH autofluorescence was reduced by setting the minimum lifetime to 200 ps, and the maximum Chi2 to 5. The average mean NADH lifetime (Tmean) of the image sections was quantified.

RNA isolation and real-time qRT-PCR of human catalase mRNA

Total RNA was isolated using TRIzol reagent (Thermo Fisher Scientific) according to the manufacturer’s protocols. Briefly, mouse was sacrificed by cervical dislocation, and tissues (small intestinal and liver) were flash-frozen on liquid nitrogen and kept at −80°C. The tissues were homogenized using a motorized blade homogenizer (Polytron) in 1.0 ml of TRIzol. Total RNA was resuspended in 50 μl of ribonuclease-free H2O. Contaminating DNA was removed from RNA using the TURBO DNA-free Kit from Ambion. Approximately 3.0 μl of total RNA was treated in a 10-μl reaction according to the manufacturer’s protocol. First-strand complementary DNA (cDNA) was created from total RNA using SuperScript IV First-Strand Synthesis System (Thermo Fisher Scientific). The entire deoxyribonuclease-treated RNA (10 μl) was used for first-strand synthesis with oligo-dT primers. TaqMan-based qRT-PCR was performed using TaqMan Expression Assays (Thermo Fisher Scientific) on the Viaa7 platform. Assay was Hs00937387_m1 specific human catalase mRNA, and the control (housekeeping) assay was mouse HPRT Mm00446968_m1. qRT-PCRs were 5 μl of 2× Universal Master Mix (no UNG), 1 μl of first-strand cDNA, 0.5 μl of TaqMan assay, and 3.5 μl of H2O. All reactions were run in triplicate on a 384-well plate.


Acknowledgments: We thank A. Butic, S. L. Royston, R. Morrow, K. L. Mitchell, and J. A. Tintos (Wallace laboratory) and C. Hofstaedter, D. Kim, M. Moraskie, and H. Zhang (CHOP Microbiome Center, Children’s Hospital of Philadelphia) for technical assistance and B. Boursi (Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine at the University of Pennsylvania, Philadelphia) for fruitful discussions. All authors acknowledge the essential contribution of mouse models. Funding: This work was supported by the PennCHOP Microbiome Pilot Grant and NIH grants MH108592, NS021328, MN110285, and DO W81XWH-16-1-0401 (to D.C.W.) and the PennCHOP Microbiome program (to G.D.W.). Author contributions: T.Y. and D.C.W. directed the project. T.Y., D.G.M., and D.C.W. wrote the manuscript. T.Y., C.E.T., K.B., D.G.M., and D.C.W. were responsible for experimental design. T.Y., L.M.M., and C.E.T. performed the experiments. T.Y., C.E.T., K.B., P.M.S., L.N.S., G.D.W., and D.C.W. performed data analysis. Competing interests: The authors declare that they have no competing financial interests. Data and materials availability: The DNA sequence datasets generated during and/or analyzed during the current study are available in the NCBI SRA repository (BioProject ID PRJNA423318). All other data needed to evaluate the conclusions in the paper are present in the paper.

Stay Connected to Science Signaling

Navigate This Article