Dietary bovine milk exosomes elicit changes in bacterial communities in C57BL/6 mice

Abstract

Exosomes and exosome-like vesicles participate in cell-to-cell communication in animals, plant, and bacteria. Dietary exosomes in bovine milk are bioavailable in nonbovine species, but a fraction of milk exosomes reaches the large intestine. We hypothesized that milk exosomes alter the composition of the gut microbiome in mice. C57BL/6 mice were fed AIN-93G diets, defined by their content of bovine milk exosomes and RNA cargos: exosome/RNA-depleted (ERD) versus exosome/RNA-sufficient (ERS) diets. Feeding was initiated at age 3 wk, and cecum content was collected at ages 7, 15, and 47 wk. Microbial communities were identified by 16S rRNA gene sequencing. Milk exosomes altered bacterial communities in the murine cecum. The abundance of three phyla, seven families, and 52 operational taxonomic units (OTUs) was different in the ceca from mice fed ERD and ERS (P < 0.05). For example, at the phylum level, Tenericutes had more than threefold abundance in ERS mice at ages 15 and 47 wk compared with ERD mice (P < 0.05). At the family level, Verrucomicrobiaceae were much less abundant in ERS mice compared with ERD mice age 47 wk (P < 0.05). At the OTU level, four OTUs from the family of Lachnospiraceae were more than two times more abundant in ERS mice compared with ERD at age 7 and 47 wk (P < 0.05). We conclude that exosomes in bovine milk alter microbial communities in nonbovine species, suggesting that exosomes and their cargos participate in the crosstalk between bacterial and animal kingdoms.

NEW & NOTEWORTHY This is the first report that exosomes from bovine milk alter microbial communities in mice. This report suggests that the gut microbiome facilitates cell-to-cell communication by milk exosomes across species boundaries, and milk exosomes facilitate communication across animal and bacteria kingdoms.

INTRODUCTION

Exosomes are nanoparticles that play essential roles in cell-to-cell communication (1, 54). Exosomes transfer diverse cargos from donor cells to adjacent or distant recipient cells. Cargos include various species of RNA, proteins, and lipids. In recipient cells, exosome cargos alter gene expression and metabolism. For example, miR-30d secreted by the endometrium is taken up by the preimplantation embryo and modifies the transcriptome in human embryos (46). Recently, we made the paradigm-shifting discovery that exosomes and their RNA cargos are not exclusively derived from endogenous synthesis but can also be obtained from dietary sources (3, 26, 52). We demonstrated that human and rat intestinal cells transport exosomes from bovine milk by endocytosis and secrete microRNA cargos across the basolateral membrane (52). A similar transport mechanism operates in human vascular endothelial cells (21). Exosomes accumulate in immune cells if transferred across species boundaries (3, 17, 18). We further demonstrated that bovine milk exosomes contribute to the body pool of microRNAs in human milk feeding and murine exosome depletion studies (3, 41, 48). MicroRNAs in dietary exosomes alter gene expression across species boundaries (3). These discoveries are consistent with observations by other investigators. Ongoing phenotyping studies in our laboratory suggest that dietary depletion of bovine milk exosomes alters plasma cytokine profiles in humans and mice and causes an aberrant flux of purine metabolites in mice (2).

Despite the compelling evidence in support of the theory that milk exosomes are bioavailable, concerns have been raised that the amount of cargos, particularly microRNAs, delivered by bovine exosomes to host organisms is exceedingly low in transgenic mouse models [(22, 45), reviewed in Ref. 56]. In a recent opinion article, we suggested that, while studies of dietary microRNAs are important and warrant investigation, the controversy surrounding the field of dietary microRNAs must not impede the rate of discovery in areas such as dietary exosomes, exosome cargos other than microRNAs, and noncanonical RNA signaling pathways (55). Here, we tested the hypothesis that bovine milk exosomes alter bacterial communities in the murine cecum. This hypothesis was based on the following rationale. First, prokaryotic and eukaryotic microbes communicate with their environment through exosome-like vesicles (51). This observation includes gram-positive bacteria, which use vesicles for communication despite the cell wall posing a barrier for vesicle transport (23, 37). Viruses may participate in exosome signaling through hijacking and modifying exosomes (29). Second, up to 20% and 40% of RNA sequence reads in plasma from healthy adults map to bacterial and fungal genomes, respectively (47). Third, evidence suggests that orally administered, fluorophore-labeled exosomes from bovine milk are delivered to peripheral tissues (33). While that study lacked important controls (unlabeled exosomes, free fluorophore), its findings are largely consistent with our ongoing studies, suggesting that endogenously and exogenously labeled milk exosomes accumulate in liver and spleen, but that a considerable fraction of orally administered exosomes escapes absorption and reaches the large intestine (26).

MATERIALS AND METHODS

Mouse diets.

Exosome and RNA-depleted (ERD) and exosome and RNA-sufficient (ERS) diets are based on the AIN-93G formulation (3, 36). The milk added to the diets provided the equivalent of 0.5 liters milk consumed by a human adult per day, adjusted by body weight in mice. The milk used to prepare the powder for the ERD diet was ultrasonicated, which led to a >98% depletion of microRNA cargos in exosomes, 20% decreased in exosome count (9.1 × 10¹² ± 7.1 × 10¹¹ exosomes/mL in ERS milk vs. 7.3 × 10¹² ± 3.5 × 10¹¹ exosomes/mL in ERD milk] and a >60% decrease in intestinal exosome transport rates [(3, 24), S. Sukreet and J. Zempleni, unpublished data]. Diet ingredients other than milk were not ultrasonicated, i.e., nutrients other than exosomes and their RNA cargos were the same in ERD and ERS diets. Diets were dried and fed in pelleted form. Diets did not affect food and water consumption, feeding frequency, physical activity, and variables of liver and kidney disease (24).

Mouse feeding studies.

C57BL/6 mice (stock number 000664, Jackson Laboratories) were obtained at age 3 wk when ERS and ERD feeding was initiated. Mice were housed in groups of four mice per cage in eight cages, separated by diet and sex. Males and females were studied. True randomization of group assignment was achieved by labeling mice with numbers and randomly assigning numbers to groups. At timed intervals (ages 7, 15, and 47 wk), mice were sampled from different cages to avoid cage effects (n = 8 for each sex and age), and euthanized for sample collection. Body weight and body fat were measured by using a Scout Pro digital scale (Ohaus, Melrose, MA) and a PIXImus mouse densitometer (LUNAR, Madison, WI), respectively. Cecum content was collected, flash frozen in liquid nitrogen, and stored at −80°C. The study was approved by the Institutional Animal Care and Use Committee at the University of Nebraska-Lincoln (protocol 1229).

Bacterial growth ex vivo.

Cecum content was collected from three male and three female C57BL/6 mice under anaerobic conditions. Aliquots were suspended in medium containing M9 minimum salts (Gibco BRL, Gaithersburg, MD), 100 µM CaCl₂, and 2 mM MgSO₄ in a sterile, anaerobic work bench. Bovine milk exosomes were isolated from milk and authenticated by transmission electron microscopy, nanosight size analysis, and immunoblot analysis, as previously described (21, 26, 52). Half of the cultures were supplemented with sterile-filtered (0.22 µm) bovine milk exosomes (1.2 × 10¹² exosomes/mL) on a daily basis, whereas the other cultures received solvent (PBS). The number of exosomes added was the equivalent of 0.5 liters milk in the total gastrointestinal water in a human adult, adjusted by the body weight of mice. Contamination controls were prepared by culturing exosomes in the same media without cecum content. All samples were incubated in an anaerobic chamber at 37°C for 7 days (n = 3 per treatment). Bacterial growth was measured using the absorbance at 600 nm in a Synergy H1 Hybrid multimode microplate reader (Bio-Tek Instruments, Winooski, VT).

Analysis of bacterial communities.

DNA was extracted from cecum content using the PowerSoil DNA isolation kit following the manufacturer’s instructions (Mo Bio Laboratories, Carlsbad, CA). DNA purity and integrity were confirmed by using the 260-to-280 nm ratio (Nanodrop ND-1000, Nanodrop Technologies, Wilmington, DE) and agarose gel (0.8%) electrophoresis. The V4 region in the 16S rRNA gene was amplified and sequenced, as described previously (19). The sequencing reads were quality filtered and analyzed, as described previously (38). Contigs were generated from paired-end reads and were screened using MOTHUR v.1.38.1 (38) to exclude low-quality sequences and reads containing ambiguous bases or homopolymers longer than eight base pairs (bp). The resulting reads were trimmed to retain only reads between 245 bp and 275 bp. The UPARSE pipeline (USEARCH, v7.0.1090) (11) was used to cluster quality-filtered sequences into operational taxonomic units (OTUs) by using a 97% identity cut-off after removal of chimeras using UCHIME (12). ChimeraSlayer gold.fa was used as the reference database for chimera detection. Sequence alignment was performed using the SILVA v123 reference and was used to build a phylogenetic tree using Clearcut (40). Taxonomy assignment (Greengenes database: gg_13_8_otus) (28) was performed using assign_taxonomy script of QIIME v.1.9.1 (5) and a cutoff value set to 97%. Genera were not annotated due to short sequencing reads; therefore, OTUs (proxy for species) were used for the lower taxonomic rank analysis. We removed 10 samples that had sequencing read counts lower than 4,000. Eighty-two samples with an average read count of 40,468 reads and a range of 4,112–144,788 reads were used for downstream analysis. Reads were normalized by using the cumulative sum of counts (34). Alpha diversity metrics were used to evaluate richness (Chao1), diversity (Shannon-Weiner index), and coverage (Good’s coverage) (5). Rarefaction curves were constructed using Chao1 values. A core measurable microbiome was identified on the basis of the factors diet, sex, and age to reduce mouse-to-mouse variation. The core measurable microbiome was defined as the group of OTUs that are present in at least 80% of the samples within each factor. Differences in bacterial communities were assessed using permutational multivariate analysis of variance (PERMANOVA), utilizing the weighted UniFrac distance matrix. Additionally, the weighted UniFrac distance matrix was used for principal coordinate analysis. The linear discriminant analysis (LDA) of effect size (LEfSe) algorithm with default parameters (α = 0.05 for Kruskal-Wallis and Wilcoxon tests) was used to identify the differences in abundance between ERS and ERD feeding groups at different ages, phyla, family, and OTU levels (39). Differences were considered statistically significant if P < 0.05 and LDA score was >2.0. Multivariate analysis by linear models (MaAsLin) with default parameters was performed to test for associations between microbial community abundance and experimental factors, such as age, sex, and diet (31). Association was considered statistically significant at an FDR corrected P < 0.05. Sequence data were deposited in the NCBI-BioProject database under accession number PRJNA413623.

Statistical analysis.

Student’s unpaired t-test was used to compare body weight and fat in mice on ERS and ERD diets and growth of murine gut bacteria in exosome-defined media ex vivo. Differences were considered statistically significant if P < 0.05.

RESULTS

Effects of ERS and ERD diets on body weight and fat.

Body weight and body fat increased transiently in ERS males compared with ERD males at age 15 wk, which was no longer observed at age 47 wk (Supplemental Table S1 on https://doi.org/10.6084/m9.figshare.9639287.v4; all supplemental material can be found on this site). Diets did not affect body weight and body fat in young males, age 7 wk, and in females.

Growth of murine gut bacteria ex vivo.

Supplementation of culture media with milk exosomes conferred a growth advantage to bacteria from male mice ex vivo compared with control cultures (Supplemental Fig. S1, P < 0.01). The apparent growth advantage of exosome-supplemented bacteria from female mice compared with solvent control was not statistically significant. Bacterial communities were not assessed in ex vivo studies, because the sole purpose of this experiment was to assess whether milk exosomes confer a growth advantage to some bacteria.

Bacterial communities.

Bacterial communities varied by age with a significant diet × age × sex interaction (P = 0.039; Table 1). Sex alone had no effect on microbial communities, and therefore, sexes were combined in subsequent analyses (P = 0.215; Table 1). At the phylum level, exosome-defined diets significantly affected bacterial communities in mice at ages 15 and 47 wk (Fig. 1A). The abundance of Tenericutes was more than three times greater in ERS mice at ages 15 and 47 wk compared with ERD mice, whereas the abundance decreased in both ERS and ERD mice at age 47 wk compared with age 15 wk (P < 0.05, Fig. 1B). Also, Firmicutes were less abundant in ERD mice than ERS mice at age 47 wk (P = 0.025). The abundance of Verrucomicrobia was greater in ERD mice compared with ERS mice at age 47 wk (P = 0.025). No significant differences at the phylum level were detected at age 7 wk (Supplemental Table S2).

Table 1.

Effect of diet, age, and sex on bacterial communities in the cecum of mice

	DF	SumsOfSqs	MeanSqs	F Model	P
Diet × age × sex	2	0.004	0.002	2.376	0.039
Diet × age	2	0.003	0.002	1.950	0.064
Diet × sex	1	0.001	0.001	1.047	0.326
Age × sex	2	0.002	0.001	1.345	0.215
Diet	1	0.002	0.002	2.342	0.075
Age	2	0.065	0.032	40.297	0.001
Sex	1	0.001	0.001	1.341	0.215
Residuals	70	0.056	0.001
Total	81	0.134

DF, degree of freedom; MeanSqs = SumsOfSqs/df, mean squares; F Model, F-tests; SumsOfSqs, sum of squares; n = 8/diet group/sex/time point.

Fig. 1.

Bacterial phyla in ceca from mice fed exosome/RNA-sufficient (ERS) or exosome/RNA-depleted (ERD) diets beginning at age 3 wk. A: linear discriminant analysis (LDA) of effect size (LEfSe) analysis of phyla in ERS and ERD mice. Bars represent the LDA score. Diet effects are statistically significant for LDA scores greater than 2. B: relative abundance of the phyla in A. Values are expressed as means and SD. MaAsLin, multivariate analysis by linear models; n = 16 per diet group and time point. *P < 0.05, **P < 0.01.

A total of 19 families were identified by 16S rRNA sequencing. Lachnospiraceae, Ruminococcaceae, and an unclassified family in the order Clostridiales (phylum Firmicutes), were the three most abundant families (Supplemental Fig. S2). Exosome-defined diets altered bacterial communities at the family level (Fig. 2). For example, two unclassified families from phylum Firmicutes were more abundant in ERD mice than in ERS mice at age 7 wk (P < 0.05). At age 15 wk, one of the two unclassified families from the phylum of Firmicutes and another unclassified family from phylum of Tenericutes were more abundant in ERS mice compared with ERD mice (P < 0.05). At age 47 wk, the family of Verrucomicrobiaceae was more abundant in ERD mice compared with ERS mice, whereas Lachnospiraceae and two unclassified families from phyla Firmicutes and Tenericutes were more abundant in ERS mice than in ERD mice (P < 0.05).

Fig. 2.

Bacterial families in ceca from mice fed exosome/RNA-sufficient (ERS) or exosome/RNA-depleted (ERD) diets beginning at age 3 wk. A: linear discriminant analysis (LDA) of effect size (LEfSe) analysis of families in ERS and ERD mice. Bars represent the LDA score. Diet effects are statistically significant for LDA scores greater than 2. B: relative abundance of the families in A. Values are expressed as means and SDs. MaAsLin, multivariate analysis by linear models; n = 16 per diet group and time point; *P < 0.05, **P < 0.01.

Fifty-two OTUs were different in the ceca from mice fed ERD and ERS as per LEfSe analysis (P < 0.05, Supplemental Table S3). Five of these were confirmed by MaAsLin analysis (Table 2). At age 7 wk, two OTUs from the family of Lachnospiraceae (phylum Firmicutes) were twofold higher in abundance in ERS mice compared with ERD mice (P = 0.009 and P = 0.044, respectively; Fig. 3). In contrast, the abundance of OTUs from the family Verrucomicrobiaceae (phylum Verrucomicrobia) was lower in ERS mice compared with ERD at both age 15 and 47 wk (P < 0.05). Another two OTUs from family Lachnospiraceae were significantly more abundant in ERS mice compared with ERD mice at age 47 wk. Analysis by MaAsLin suggested that 8 and 199 OTUs were associated with sex and age, respectively (Supplemental Table S4).

Table 2.

Associations between diet and OTU abundance in murine ceca

OTU	Family	Diets	Coefficient	P
307	Verrucomicrobiaceae	ERD vs. ERS	0.005	0.003
288	Lachnospiraceae	ERD vs. ERS	−0.005	0.012
179	Lachnospiraceae	ERD vs. ERS	−0.002	0.023
554	o Clostridiales	ERD vs. ERS	0.006	0.035
350	Lachnospiraceae	ERD vs. ERS	−0.007	0.038

A positive coefficient indicates higher relative abundance in exosome and RNA-depleted (ERD) diet and a negative coefficient indicates a lower relative abundance in ERD. ERS, exosome and RNA-sufficient; o, order; OTU, operational taxonomic unit; P, FDR adjusted P value; n = 16/diet group/time point. Multivariate analysis was performed using linear models (MaAsLin).

Fig. 3.

Bacterial operational taxonomic units (OTUs) in ceca from mice fed exosome/RNA-sufficient (ERS) or exosome/RNA-depleted (ERD) diets beginning at age 3 wk. A: linear discriminant analysis (LDA) of effect size (LEfSe) analysis of OTUs associated with diets. The bar column represents the LDA score. Diet effects are statistically significant for LDA scores greater than 2. B: relative abundance of OTUs in A. Values are means and SDs. MaAsLin, multivariate analysis by linear models; n = 16 per diet group and time point; *P < 0.05, **P < 0.01.

DISCUSSION

This paper provides novel insights in the following areas of research. First, this paper provides strong evidence that dietary exosomes, at least those in milk, might be responsible for some of the effects of diet on the gut microbiome (25, 53). To the best of our knowledge, the only evidence that bovine milk compounds alter the microbiome comes from a study reporting that consumption of casein protein alters bacterial communities in the gut of rats (58). A previous paper suggests that exosome-like nanoparticles from ginger alter microbial communities in mice, but it is safe to assume that the consumption of milk by adults and milk and formulas by infants is quantitatively greater than the consumption of ginger in most societies (8, 44, 50). Also, in this study, we administered a nutritionally relevant number of exosomes to mice, whereas the dose of ginger-derived nanoparticles given to mice was the equivalent of 39 kg of ginger consumed by a human adult (44).

Second, this paper suggests that gut microorganisms might act as transmitters or amplifiers of dietary exosome signals. Following early reports that microRNA cargos in dietary exosomes might be bioavailable, it was proposed that dietary microRNAs achieve tissue concentrations that are too low to elicit biological effects and that the bioavailability of milk exosomes might be low (10, 22, 42, 45). Here we provide evidence that exosome-defined diets alter bacterial communities in the murine cecum, and the effects are particularly strong if studied at the OTU level. Importantly, this article suggests that milk exosomes that escape absorption by mucosal cells may still elicit major biological effects, facilitated by the gut microbiome. It is now widely accepted that prokaryotic and eukaryotic microbes communicate with their environment through exosome-like vesicles (23, 37, 51). It is reasonable to speculate that changes in bacterial communities are paralleled by changes in the production of microbial metabolites, which may transmit and amplify milk exosome signals (49).

Third, this paper provides evidence to suggest that sex influences bacterial growth ex vivo in cultures from male mice, but not in cultures from female mice. The factors mediating sex effects are unknown. We speculate that milk exosomes conferred a growth advantage to a quantitatively minor species of bacteria that escaped detection in 16S rRNA sequencing of gut content in vivo, which then outgrew other bacteria ex vivo. We interpret this observation with due caution and consider the possibility that sex of the donor may not affect microbial growth ex vivo.

This study does not establish a causal relation between RNA cargos in milk exosomes and changes in bacterial communities in the murine gut. Ever since a publication in 2012 suggested that microRNAs in rice are bioavailable and alter gene expression in animals, much of the attention has focused on microRNA cargos in exosomes [(57), reviewed in Ref. 56]. It is tempting to give in to this speculation, because microRNAs regulate more than 60% of human genes, milk exosomes contain up to 700 species of microRNAs with nucleotide sequences identical to those in humans, and microRNAs encapsulated in milk exosomes are bioavailable (3, 16, 20, 26, 43). Crude estimates of the bioavailability of microRNAs in milk exosomes suggest that 25% the microRNAs are absorbed, suggesting that the unabsorbed fraction may interact with gut microbes (26). This being said, there is experimental evidence that both microRNAs and exosomes may affect cell signaling. For example, microRNAs elicit biological effects through binding to Toll-like receptors, whereas mere exosome-cell surface interactions activate signaling by mitogen activated kinases in human myeloma cell lines and human T-cell subsets (6, 14, 32, 35). Future studies will need to determine whether effects of dietary exosomes on bacterial communities are caused by exosomes, microRNA cargos, or cargos other than microRNAs.

The bacterial communities altered by milk exosome-defined diets are related to pathological and physiological conditions, as evidenced by the following examples. A loss of Lachnospiraceae in the ileal mucosa has been implicated in inflammatory bowel disease (4, 15). The proportion of Firmicutes and Tenericutes was lower in cecal content of obese diabetic mice compared with controls (13). Dysbiosis in the gut microbiome may cause liver disease due to microbial metabolites altering the metabolism in hepatic cells (27), e.g., a decreased abundance of Verrucomicrobia has been linked with nonalcoholic fatty liver disease (30). Inflammatory bowel disease, obesity, and nonalcoholic fatty liver disease are major health concerns in the United States (7, 9).

The functional consequences of exosome-induced changes in bacterial communities in the gut, if any, and the mechanisms by which these changes contribute phenotypes of health and disease are unknown. We have recently completed a one-year mouse feeding trial to assess phenotypes associated with feeding milk exosome-defined diets to mice. To date, we have published the following phenotypes. First, hepatic concentrations of purine metabolites were higher in ERD mice compared with ERS mice, and similar increases were observed in the urine from infants fed soy formula compared with human milk or milk formula, and in adult dairy avoiders compared with dairy consumers (2). Note that pathways in purine metabolism were among those identified in the metagenomics studies described above. Second, the muscle grip strength was moderately lower in ERD mice compared with ERS mice (24). It will be interesting to test which of these, and other phenotypes depend on the presence of gut microbiota by comparing germ-free and conventional mice or conducting fecal transplant studies.

GRANTS

This work was supported by the National Institute of Food and Agriculture, US Department of Agriculture (USDA), under award numbers 2015-67017-23181 and 2016-67001-25301; National Institutes of Health Grant 1P20GM104320; the University of Nebraska Agricultural Research Division (Hatch Act) and USDA multistate groups W3002 and W4002 (all to J. Zempleni); a Food for Health grant by the University of Nebraska President’s Office (J. Cui and J. Zempleni); a grant from the University of Nebraska Agricultural Research Division (Hatch Act), USDA Agricultural Research Station, United States Meat Animal Research Center (collaborative funding program); and USDA multistate group Grant W2010 (all to S. Fernando).

DISCLOSURES

J. Zempleni is a consultant for PureTech Health in Boston, MA. S. Fernando is a co-owner of NuGUT LLC. The other authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS

J.Z. conceived and designed research; F.Z., H.A.P., and M.S. performed experiments; F.Z., H.A.P., J.C., S.D.K., and S.C.F. analyzed data; J.Z. interpreted results of experiments; J.Z. prepared figures; F.Z. drafted manuscript; S.D.K., S.C.F., and J.Z. edited and revised manuscript; F.Z., H.A.P., M.S., J.C., S.D.K., S.C.F., and J.Z. approved final version of manuscript.