Research ArticleGUT MICROBIOTA

Breast-fed and bottle-fed infant rhesus macaques develop distinct gut microbiotas and immune systems

See allHide authors and affiliations

Science Translational Medicine  03 Sep 2014:
Vol. 6, Issue 252, pp. 252ra120
DOI: 10.1126/scitranslmed.3008791


Diet has a strong influence on the intestinal microbiota in both humans and animal models. It is well established that microbial colonization is required for normal development of the immune system and that specific microbial constituents prompt the differentiation or expansion of certain immune cell subsets. Nonetheless, it has been unclear how profoundly diet might shape the primate immune system or how durable the influence might be. We show that breast-fed and bottle-fed infant rhesus macaques develop markedly different immune systems, which remain different 6 months after weaning when the animals begin receiving identical diets. In particular, breast-fed infants develop robust populations of memory T cells as well as T helper 17 (TH17) cells within the memory pool, whereas bottle-fed infants do not. These findings may partly explain the variation in human susceptibility to conditions with an immune basis, as well as the variable protection against certain infectious diseases.


The human and nonhuman primate gut microbiotas are extraordinarily diverse, vary between individuals, and can fluctuate over time in response to diet and environmental factors (13). The microbiota influences metabolic processes in the host, immune system development, responses to infection, and even behavioral abnormalities (410). For example, regulatory T cell (Treg) activity in mice is stimulated by the human commensal Bacteroides fragilis (11), by certain Clostridia bacterial species (12, 13), and by short-chain fatty acids produced by bacteria after digestion of fiber (14). Similarly, segmented filamentous bacteria have been found to induce T helper 17 (TH17) cells (15). More recently, McCune and colleagues have proposed that tryptophan catabolites produced by the gut microbiota are important regulators of TH17-Treg balance in HIV infection (16, 17).

TH17 cells are particularly important for host defense against mucosal pathogens such as HIV and Salmonella. HIV-positive individuals exhibit highly variable rates of disease progression, as measured by viral load and CD4+ T cell loss (18). This variability is only partly explained by host genetics (19, 20). Disease progression in humans is associated with TH17 cell loss; on the other hand, simian immunodeficiency virus (SIV) replication in the rhesus macaque is limited by the size of the preexisting TH17 cell compartment, possibly because of maintenance or enhancement of the gut mucosal barrier (21, 22). The progress of Salmonella infection is also influenced by TH17 cell numbers in the host, with greater dissemination of bacteria from gut tissue to lymph nodes under low TH17 conditions (23).

The gut microbiotas of infants are dynamic and highly dependent on a number of factors, including delivery method and diet (for example, breast-feeding versus formula-feeding). Numerous studies have found differences in the gut microbiotas of humans and nonhuman primates as a result of differences in early diet (2427). Moreover, these microbial differences are associated with differences in baseline concentrations of circulating cytokines as well as serum and urine metabolites, including increases in proinflammatory cytokines [interleukin-1β (IL-1β), IL-1 receptor antagonist, and tumor necrosis factor–α] and amino acid metabolites in bottle-fed (nursery-reared) animals (25). These differences may be linked to immunopathologic consequences, because bottle-fed/nursery-reared infant macaques were 7.5 times more likely than dam-reared infants to develop idiopathic chronic diarrhea later in life (28).

Despite this abundance of suggestive data, no published studies directly link different infant diets and gut microbiotas to altered microbiota-derived metabolites and durable changes in the developing immune system. Changes in the immune system could affect the host response to infections and diseases occurring beyond infancy. To test the possible effects of the gut microbiota on immune system development, we followed breast-fed and bottle-fed infant rhesus macaques (all housed indoors) between 5 and 12 months of age. We found that early differences in feeding regimens were associated with divergent development of the immune system over both the short term and long term, including durable changes in immune ontogeny persisting after identical diets had been imposed.


Infant macaques develop varying TH17 cell populations at disparate rates

We previously demonstrated that adult rhesus macaques have widely disparate fractions of TH17 cells among circulating and gut-resident CD4+ T cells (29). We next examined TH17 cell representation across the macaque life span to determine how quickly infant macaques generate normal adult frequencies of TH17 cells. We plotted data gathered through analysis of samples from 125 different uninfected macaques followed in various studies carried out at the California National Primate Research Center (CNPRC) (Fig. 1).

Fig. 1. Circulating TH17 cells throughout the macaque life span.

(A) Percentage of TH17 cells among CD4+ T cells in peripheral blood, from birth to more than 9 years of age. Data points are connected by lines wherever they represent longitudinal samples from one animal. The curve shown is a loess curve plotted in R using a span value of 0.6. (B) Absolute numbers of circulating TH17 cells in rhesus macaques aged 17 to 62 months. The slight decrease seen is not significant in mixed-effects regression (P = 0.06). (C) Development of TH17 cells throughout the first 18 months of life. As in (A), the percentage of TH17 cells among CD4+ T cells in peripheral blood is shown. The curve shown is a loess curve; the increase shown is highly significant in mixed-effects regression (P < 0.00005).

These data confirmed our previous report of disparate TH17 cell representation in adult animals, with such cells ranging between 0.9 and 8.5% of circulating cells in animals over 18 months (547 days) old, without any apparent significant change beyond that age (Fig. 1A). More limited data on absolute counts of circulating TH17 cells also suggested no significant change after 18 months (Fig. 1B). Quantitation of gut tissue–resident TH17 cells was not attempted, but in previous studies, we have found that the percentage of circulating TH17 cells correlated with the percentage of TH17 cells found in colon tissue (29).

Most importantly, we found that newborn macaques harbor few TH17 cells and that these cells develop progressively throughout the first 18 months of life (Fig. 1C). Surprisingly, the rate of development of TH17 cells in infancy was markedly variable, with some individuals rapidly developing a TH17 cell population in the adult range and others generating very few cells even by 1 year of age (Fig. 1C). These data, along with published literature demonstrating stimulation of TH17 cell development by intestinal microbes (15), suggested the possibility that different infant gut microbiotas stimulated these different developmental trajectories.

Infant diet (breast-feeding versus bottle-feeding) influences the intestinal microbiome

To determine if different TH17 cell development could be related to different microbial colonization of the intestine, we studied microbial colonization and immune cell development in six infant macaques raised indoors by their mothers [dam-reared and breast-fed (DR)] versus six raised in the nursery [nursery-reared and bottle-fed (NR)].

The microbiotas of stool samples taken at 6 and 12 months were assessed by 16S ribosomal RNA (rRNA) profiling on third-generation PhyloChips. Comparison of gross microbial community metrics showed divergence between the rearing groups as early as 6 months, with a trend to greater richness, evenness, and Shannon index diversity in dam-reared animals (Fig. 2A). Despite the fact that identical diet and housing conditions were established by 1 year of age, the differences in these metrics grew larger and were statistically significant at the later time point (Fig. 2A; P = 0.04). Furthermore, principal component (PC) analysis of taxon abundance data clearly separated the two rearing groups at both 6 and 12 months of age (Fig. 2B). Permutational multivariate analysis of variance indicated a significant relationship between rearing conditions and divergent microbiota at both time points (adonis; P = 0.002 and P = 0.024 at 6 and 12 months, respectively).

Fig. 2. Different gut microbiotas in dam-reared and nursery-reared animals at 6 and 12 months of age.

(A) Richness, evenness, and Shannon index diversity are higher in dam-reared (DR) compared with nursery-reared (NR) animals. (B) Principal component analysis shows clear separation of NR and DR gut microbiotas at 6 and 12 months. Note that interindividual distances in these plots are generally reflective of Bray-Curtis dissimilarities. (C) Waterfall plot showing Δ abundance of taxa that are statistically different between rearing groups. Δ abundance is calculated by subtraction of average abundance in one rearing group from average abundance in another. (D) Abundance of genera based on pooled data from taxa assigned to each.

At the level of abundance of individual taxa, rearing groups showed significant differences at both time points (that is, 6 months, 68 taxa; 12 months, 484 taxa having q value <0.05, with those of largest change shown in Fig. 2C and table S1; see Materials and Methods). By 12 months of age, DR animals had microbiota enriched in members of Prevotella and Ruminococcus (Fig. 2, C and D). NR animals had microbiota enriched in members of Clostridium. Many of these differences had been established by 6 months of age and grew more extreme over time between rearing groups, suggesting a lasting impact of initial microbial colonization on the juvenile gut microbiota (Fig. 2D). For example, Prevotella and Lactobacillus increased in relative abundance over time in the DR group, whereas these genera decreased over the same period in the NR group (Fig. 2D, top left and bottom right panels). In contrast, Clostridium representation decreased in both DR and NR animals over the period examined, remaining significantly higher in the NR group (P = 0.008 and P = 0.002 at 6 and 12 months by Wilcoxon rank sum test).

Breast-fed but not bottle-fed macaque infants develop robust TH17 cell populations

To examine the possible effects of different gut microbiota composition on immune cell development, we isolated peripheral blood mononuclear cells (PBMCs) from each study animal at 5, 6, 9, and 12 months of age. Extensive immunophenotyping was then performed using a set of four flow cytometry panels defining more than 115 phenotypes of interest, including antigen-presenting cell and T cell surface, activation, and homing markers. These data were then used for unsupervised clustering of the samples just as with bacterial abundance data, above (Fig. 2B). When this process was performed using all cell populations at 5 and 6 months after birth, the algorithm identified clusters that were unrelated to feeding regimen (Fig. 3, A and B). Nevertheless, lasso regression (30) identified a set of incipient differences that did separate the feeding groups at these early ages, most prominently including TH17 cells as a percentage of all CD4+ T cells or of memory CD4+ T cells.

Fig. 3. Dam-reared animals harbor robust TH17 cell populations.

(A to D) Principal component analysis of immune cell abundances at (A) 5, (B) 6, (C) 9, and (D) 12 months, all plotted using loadings from the 12-month immune cell profile. (E) Heat map showing scaled values for various T cell subsets that were found to be statistically different between rearing groups. Animals (shown as columns) were clustered using hclust. (F) Populations of various immune cell subsets over time. Lines represent averages for rearing groups with SE brackets.

Breast-fed and bottle-fed macaque immune systems were radically different by 12 months of age, being clearly separated into groups by the Partitioning Around Medoids (PAM) algorithm (Fig. 3D with adonis P = 0.002). Immune profiles were also significantly different at 9 months (Fig. 3C with adonis P = 0.004). At 12 months, the groups were entirely separate when plotted along the axes of greatest variation (principal components, Fig. 3D) or when clustered by hierarchal clustering (Fig. 3E), indicating that immune cell differences associated with rearing method dominated all the variability detected in peripheral blood. Differences between breast-fed and bottle-fed infants were concentrated in TH17 cells and markers of immune cell activation (Fig. 3E), with less prominent differences in Treg phenotypes. Among breast-fed infants, TH17 cells were overrepresented both among all CD4+ T cells and among memory cells, indicating a true TH17 polarization (Fig. 3, E and F). The percentage of CD4+ T cells expressing interferon (IFN) and IL-17 upon stimulation was increased an average of 15-fold in breast-fed infants as compared to bottle-fed infants (Fig. 3F).

Dam-reared and nursery-reared macaque infants differ at the single-metabolite level

We searched for possible metabolic connections between altered gut microbiota and immune cell development using gas chromatography–mass spectrometry (GC-MS) analysis of chemicals in stool and plasma. Combined stool and plasma metabolite profiles differed between the two rearing groups at both 6 and 12 months, despite convergence in diet by the latter time point (adonis: P = 0.05 and P = 0.04 at 6 and 12 months, respectively; Fig. 4, A and B). NR animals showed increased amino acids such as asparagine in stool, a finding that is concordant with the results of previous studies (25) (Fig. 4B). Plasma samples also showed differences between rearing groups, principally in lipid metabolites such as oleic, linoleic, icosenoic, and palmitoleic acids and short-chain fatty acid metabolites such as 2- and 3-hydroxybutanoic acid, which all increased in DR animals (Fig. 4B).

Fig. 4. Metabolomic profiles differ between rearing groups.

(A to B) Waterfall plots of statistically significantly different metabolites at (A) 6 and (B) 12 months, defined by P < 0.05 and fold change > 2. (C) Scatterplots with regression lines for Treg populations and short-chain fatty acids found in plasma at 12 months. All metabolites are plotted using absorbance units.

We also investigated possible connections between specific metabolites and immune cell subsets using longitudinal regression analysis. Despite the small sample size available, linear mixed-effect regression models associated some plasma concentrations of short-chain fatty acids with Treg cell subsets [Fig. 4C; see (14)]. Specifically, 2-hydroxybutanoic acid was associated with CD4+CD25+127low Treg among memory T cells (P = 0.0098 for the longitudinal model; 12-month data shown in Fig. 4C). 2-Hydroxyvaleric acid also was associated with Treg among memory T cells (P = 0.0115), and 3-hydroxybutanoic acid was associated with both Treg among memory T cells (P = 0.0012) and CD3+CD4+CD25+FOXP3+ T cells (P = 0.021).

We then focused on metabolites associated with TH17 cell subsets. We found that arachidonic acid concentrations were positively associated with TH17 cell subsets in linear mixed effect models using 6- and 12-month data (P = 1.3 × 10−7 and P = 0.001 for association with TH17 cells in all CD4 or in memory CD4 cells, respectively; 12-month data shown in Fig. 5, A and B). We then used regression to investigate possible association of specific bacterial genera with stool arachidonic acid. Association with Prevotella spp. was significant (P = 0.03; Fig. 5C), and bacteria of this genus have previously been shown to release arachidonic acid in vitro (31). Indeed, we found a network of significant correlations between stool concentrations of arachidonic acid, TH17 cells, and bacterial genera such as Prevotella and Campylobacter, concordant with regression results (Fig. 5D).

Fig. 5. Relationship between arachidonic acid concentrations and TH17 cell abundance.

(A to B) Scatterplot and longitudinal regression show a significant association between stool arachidonic acid and (A) TH17 cells or (B) memory TH17 cells. (C) Significant association between arachidonic acid concentrations and Prevotella. All panels show 12-month data. (D) Significant correlations between stool concentrations of arachidonic acid, TH17 cells, and bacterial genera plotted in network form.

The association between Campylobacter and arachidonic acid concentrations in stool was suggestive but not significant (P = 0.088; fig. S1A); previous literature has demonstrated arachidonic acid release by Campylobacter (32). The amount of Campylobacter at 6 and 12 months was correlated with TH17 cells and at 12 months with memory TH17 cells (P = 0.004 and P = 0.005, respectively; fig. S1, B and C). Both Prevotella and Campylobacter were significantly more abundant in DR than in NR animals at 12 months (P = 0.0008 and P = 0.007, respectively).

Finally, we assessed plasma kynurenine/tryptophan (kyn/trp) ratios to determine if they might account for TH17 cell abundances in healthy infants because they appear to do so in HIV-infected humans (16). Kyn/trp ratios did not differ significantly by rearing group (fig. S2A). Furthermore, kyn/trp ratios were not associated with various TH17 cell subsets in longitudinal regression models. We also assessed the relationship between bacteria that express enzymes in the tryptophan catabolism pathway and TH17 cell development, using methods similar to those of Vujkovic-Cvijin and colleagues (16). With UniProt, bacterial genera were categorized according to the number of enzymes involved in tryptophan catabolism that they expressed. Individual eOTUs (empirical operational taxonomic units; see Materials and Methods) were then correlated with kyn/trp ratios, and genera were assigned correlation coefficients on the basis of the highest correlated eOTUs within the genus. Here, among healthy infant macaques, we found no significant relationship between the mean correlation rank and number of pathway enzymes expressed (fig. S2B).


We demonstrate that infant diet has profound and durable effects on the gut microbiota of macaques, development of their immune system, and metabolite profiles in plasma and stool. These effects may prove important in understanding variable immune responses to vaccination and infection as well as different propensities for the development of autoimmune disease. Furthermore, the longitudinal data we collected proved useful for hypothesis generation about the mechanisms underlying microbiota-driven immune modulation (for example, the relationship between arachidonic acid and TH17 cells), which may allow for design of interventional experiments and eventually for pharmacologic manipulation of the developing immune system.

Differences between gut microbiotas were pronounced, both in community-level indices and individual taxa, and remained pronounced 6 months after all animals in the study began receiving an identical diet. NR (bottle-fed) animals had decreased community richness, evenness, and diversity. Previous studies have suggested a connection between decreased richness and inflammation, insulin resistance, and overall adiposity (33, 34).

Surprisingly, immune system profiles remained significantly different between rearing groups at 12 months of age, especially in various TH17 cell subsets. These included both the percentage of TH17 cells among CD4+ T cells and among memory cells, indicating a true change in immune cell polarization, rather than a global increase in memory cell production. Our finding that differences were most pronounced in these subsets is perhaps not surprising because TH17 cells are generated as a result of microbial interactions occurring in the gut mucosa and the cells traffic to that tissue (35, 36). DR animals additionally demonstrated increases in CD4+ and CD8+ T cells making IFN-γ (that is, TH1 cells and cytotoxic T lymphocytes). Indeed, DR animals demonstrated a profile of immune cell activation and proliferation that is consistent generally with a greater ongoing stimulation by bacteria. This difference in immune systems based on diet is congruent with differences in proinflammatory cytokine production documented by O’Sullivan et al. (25).

Our study of rhesus macaques uncovered some associations also seen in studies of human infants. Formula-fed human infants demonstrated greater abundance of Clostridia (24) and increased numbers of CD4+ naïve cells (37). Breast-fed human infants also had higher serum concentrations of the anti-inflammatory cytokine transforming growth factor–β2, which we did not test in our rhesus cohort (38).

Metabolomics of plasma and stool revealed differences at the single-metabolite level between rearing groups that were suggestive when compared to previous literature. At both 6 and 12 months, stool levels of amino acid metabolites were generally higher in NR animals, a finding that is congruent with another study of differently reared rhesus macaques (25). Previous studies have demonstrated that short-chain fatty acids regulate Treg induction and homeostasis, at least in mice (12, 14, 39). Indeed, we observed that derivatives of short-chain fatty acids in plasma were associated with Treg frequencies (12, 14, 39). Whereas previous studies have associated higher concentrations of TH17 cells with reduced tryptophan catabolism, specifically by the kynurenine pathway (16), we found no significant evidence of this effect in our study of uninfected animals.

Longitudinal sampling and comprehensive analysis of samples (microbiome composition by PhyloChip, immunophenotyping by flow cytometry, and metabolomics by GC-MS) allowed us to form testable hypotheses about the mechanisms of microbiota-driven immune modulation via mixed-effects regression. For example, we observed a longitudinal correlation between arachidonic acid and induction of TH17 cells. Arachidonic acid is present in rhesus macaque breast milk; however, weaning of macaques typically occurs by about 6 months of age, indicating that differences in arachidonic acid concentrations at later time points were not caused by differences in ingestion of the chemical. We therefore ascribe differences in arachidonic acid concentrations to the metabolic activities of the gut microbiota or to host metabolic activities stimulated by the microbiota. Release of arachidonic acid from phospholipid membranes is controlled by phospholipase A2 (PLA2) (40). Numerous mammalian cell types and many commensal bacteria express PLA2, which is regulated by phosphorylation or Ca2+-induced translocation from the cytosol to the cell membrane (41). Previous studies have shown that many bacteria (often Gram-negative) express forms of PLA2 that can release arachidonic acid (42, 43). Furthermore, bacteria such as Prevotella are known to activate cytosolic PLA in macrophages (31, 44). Higher concentrations of some eicosanoids have previously been associated with TH17 cell expansion (45, 46). For example, prostaglandin E2 stimulates the production of TH17-inducing cytokines in cultured macrophages (47), whereas phospholipase blockade in mice led to decreased circulating IL-17 (48). TH17 cell expansion caused by elevated arachidonic acid could be beneficial for future responses to pathogenic challenge. For example, studies have shown that infants receiving greater amounts of arachidonic acid were less likely to contract HIV from their HIV+ mothers (49), and higher numbers of TH17 cells have been associated with lower acute and plateau SIV viral loads (22).

Overall, our findings show that profound differences in macaque infant gut microbiotas are associated with equally profound differences in the immune system and that this association may be mediated by differing metabolomic environments. The immune differences observed were sufficient to completely separate DR and NR animals on the basis of certain parameters, most notably those relevant to activation and differentiation of T cell subsets. Such differences could explain wide variations among humans in vaccine responsiveness or the tendency to develop immune-mediated diseases (50).


Study design

Preliminary data indicated that six animals were required in each group to detect a 20% difference in key immunologic parameters with 80% power and an α of 0.05. No animals were excluded from analysis. This was an observational study designed to search for possible differences in the stool microbiota, stool metabolomics, and/or immune systems of breast-fed and bottle-fed rhesus macaques. Twelve rhesus macaque infants were used for the longitudinal study of immune cell phenotypes, microbiota, and metabolomics. Six study animals were raised indoors by their mothers; six were raised indoors in the CNPRC nursery; both infant groups were studied at the same period. All infants were delivered vaginally. Antibiotics were considered to be a confounding variable (51, 52). Medical histories were carefully examined before inclusion to ensure that no animal had been antibiotic-treated or had experienced any apparent enteric illness or abnormality. The study was not blinded.

Ethics statement

The UC Davis animal care program is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International (AAALAC). All animal procedures were approved prior to implementation by the UC Davis Institutional Animal Care and Use Committee and were consistent with the requirements of the Animal Welfare Act. Activities related to animal care were performed as per standard operating procedures at the California National Primate Research Center.


Dam-reared animals had a diet consisting primarily of breast milk and high-protein monkey chow (LabDiet; present in the cage for dam’s diet and normally sampled by infant 1 to 4 weeks after birth) and were in contact with their mothers until the age of about 6 months. The gross compositions of human and rhesus macaque breast milk are similar (53, 54). Both contain numerous components not found in formula, including certain small molecules (53), biological polymers (5557), microbes (53), and viable whole maternal cells (58); however, rhesus macaque breast milk lacks certain oligosaccharides found in human milk (59). Nursery-reared infants were fed formula (Enfamil Lipil + Iron) and had no contact with their mothers. They received high-protein chow as a supplement to the formula. The diets in both groups converged over time to high-protein chow at the age of 6 months. All animals received seasonal fruits as part of the CNPRC feeding enrichment program during the study; time of fruit introduction and frequency of provision were approximately equal for both groups because all animals were born in the same period.

Sample collection

Stool and blood samples were taken from each rhesus macaque infant at 5, 6, 9, and 12 months after birth. Fresh stool samples were collected from the cage pan in the morning and immediately frozen at –70°C for later DNA extraction for PhyloChip analysis. Plasma samples were stored at –70°C. PBMCs were isolated by gradient density purification using Lymphocyte Separation Medium (MP Biomedicals, LLC), then washed and cryopreserved in liquid nitrogen.

Stool sample processing and microbiota analysis with PhyloChip

PhyloChip analyses were performed using samples taken at 6 and 12 months only. Fecal specimens were stored at –70°C until processing. Total DNA was isolated using the QIAamp DNA Stool Mini Kit (Qiagen) with slight modifications. Briefly, fecal samples were resuspended in Qiagen’s ASL buffer and transferred into Lysing Matrix E tubes (MP Biomedicals). Samples were homogenized in a FastPrep-24 instrument (MP Biomedicals) at 6.0 m/s for 30 s. Beads were pelleted, and the supernatant was transferred into sterile 2-ml microcentrifuge tubes. ASL buffer was added to each LME tube, followed by homogenization at 6 m/s for 30 s. Beads were pelleted, and the supernatant was combined with the supernatant from the first extraction step. DNA isolation was completed as outlined in the QIAamp DNA Stool Mini Kit manual. DNA was eluted in 150 μl of distilled water and stored at –20°C. All samples yielded sufficient total DNA for downstream processing. The bacterial 16S rRNA genes were amplified in almost full length in a temperature gradient using the primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) (60) and 1492R (5′-GGTTACCTTGTTACGACTT-3′) (60, 61).

Polymerase chain reactions (PCRs) for PhyloChip analysis were performed in 25-μl reactions containing 0.02 U/μL of Ex Taq (Takara Mirus Bio Inc.), 1× Takara buffer with MgCl2, 27F and 1492R primers (0.3 pmol/μl) (60), 0.8 mM deoxynucleotide triphosphates, bovine serum albumin (0.8 mg/ml) (Roche Applied Science), and 30 ng of DNA template. A total of 12 reactions per sample were performed in an Eppendorf Mastercycler gradient thermocycler across a gradient (48° to 58.4°C) of annealing temperatures to maximize diversity recovered. Reaction conditions were as follows: initial denaturation (95°C for 3 min) followed by 25 cycles of 95°C (30 s), annealing (30 s), and extension at 72°C (2 min) and a final extension of 72°C (10 min). PCR reactions were pooled and purified using the QIAquick PCR Purification Kit (Qiagen). Amplification was verified by 1% tris-borate EDTA agarose gel electrophoresis. Two hundred fifty nanograms of amplicon per sample plus quantitative standards (consisting of 14 non-16S rRNA genes that permit data normalization) were fragmented, biotinylated, and applied to each PhyloChip Array, version G3. PhyloChip arrays were washed, stained, and scanned using a GeneArray scanner (Affymetrix) (61, 62). A pixel image was captured for each scan, and intensities were determined using standard Affymetrix software (GeneChip Microarray Analysis Suite). Hybridization values and the fluorescence intensity (proportional to RNA levels) for each taxon were determined by using a trimmed average. To calculate the eOTUs, probe sets (groups of probes) were determined empirically on the basis of both (i) taxonomic relatedness of the probes and (ii) their correlation in fluorescence intensity throughout the experiment. The empirical probe set was taxonomically annotated with a Bayesian method directly from the oligomers within the probes.

Immune cell phenotyping by flow cytometry

Flow cytometric analysis. Frozen rhesus macaque PBMCs were thawed and stained for flow cytometry. Predetermined optimal concentrations of the following antibodies were used: anti-CD3–Alexa 700 (clone SP34-2), anti-CD3–Pacific Blue (PacBlue) (clone SP34-2), anti-CD95–APC (allophycocyanin) (clone DX2), anti-CD95–FITC (fluorescein isothiocyanate) (clone DX2), anti-CD28–APC-H7 (clone CD28.2), anti-CCR5–APC (clone 3A9), anti-Ki67–Alexa 488 (clone B56), anti–IFN-γ–phycoerythrin (PE)-Cy7 (clone B27), anti-CD127–PE (clone HIL-7R-M21), anti-CD25–PE-Cy7 (clone M-A251), anti-CD196–PE (clone 11A9), anti-CD194–Alexa 647 (clone 1G1), anti-CD123–peridinin chlorophyll protein–Cy5.5 (clone 7G3), anti–HLA (human leukocyte antigen)-DR–PE-Cy7 (clone 646.6), anti-CD16–PacBlue (clone 3G8), anti-CD83–PE (clone HB15e), anti-CD80–FITC (clone L307.4), anti-CD86–APC (clone 2331 FUN-1) (all from BD); anti-CD8–PE-Cy5.5 (clone 3B5), anti-CD14–Qdot 605 (clone TUK4), anti–CD20–electron-coupled dye (ECD) (clone HI47), Aqua Live/Dead amine dye–AmCyan (Invitrogen); IL-17–PE (clone eBio64CAP17), anti-CD11c–Alexa 700 (clone 3.9) (eBioscience); anti-FOXP3–PacBlue (clone 206D), anti-CD28–Alexa 700 (clone CD28.2) (BioLegend); anti-CD4–Qdot 655 (19Thy5D7), anti-CD38–PE (clone OKT10) (NHP reagents); and anti–HLA ECD (clone Immu-357) (Beckman Coulter). A minimum of 300,000 events were collected for each sample. Compensation controls were prepared with antibody capture beads (BD Biosciences) stained separately with individual monoclonal antibodies. Analysis was performed using FlowJo version X (10.0.6 release).

Intracellular cytokine staining. Levels of cytokine-producing cells were determined by cytokine flow cytometry after in vitro stimulation with phorbol 12-myristate 13-acetate (PMA) and ionomycin. Rhesus macaque PBMCs were resuspended to 106 cells/ml in complete RPMI 1640 medium. Cells were then incubated for 6 hours at 37°C in medium containing PMA (50 ng/ml) (Sigma-Aldrich), ionomycin (1 μg/ml) (Sigma-Aldrich), and GolgiPlug (5 μg/ml) (BD Biosciences). After incubation, cells were washed and stained with surface markers. Cells were then washed, permeabilized using a Cytofix/Cytoperm kit (BD Biosciences) according to the manufacturer’s instructions, intracellularly stained, washed, fixed in phosphate-buffered saline containing 1% paraformaldehyde, and acquired on an LSR II cytometer.

Metabolomic analysis

Plasma and stool were analyzed for the presence and relative quantity of metabolites using GC–time-of-flight MS at 6 and 12 months (West Coast Metabolomics Center at UC Davis). Additionally, concentrations of kynurenine and tryptophan were quantified by liquid chromatography–tandem MS, as previously described (63). Morning plasma cortisol levels (samples collected between 8:30 and 9:30 AM) were tested by enzyme-linked immunosorbent assay and were not higher among nursery-reared and formula-fed infants.

Statistical analysis

Statistical analysis was performed in the R programming environment. Unless otherwise noted, all plots were created using ggplot2.

Microbial populations. Community indices, including richness, evenness, and Shannon diversity, were calculated using the Vegan package in R. The function adonis was used to assess correlations between dissimilarity of samples and factors such as rearing within the same software package. Principal component analysis was performed using the ade4 package. A two-tailed unpaired t test was used to identify bacteria eOTUs found to be significantly different between rearing groups. These significance values were then adjusted for false discovery using the q value package. The q value of a test measures the proportion of false positives incurred (called the false discovery rate) when that particular test is called significant. In the case of dam-reared animals, the 50 statistically significant eOTUs with the highest Δ (increased in DR with Δ >3.5 log2 fluorescence intensity) were included (Fig. 2C), and the 38 statistically significant eOTUs with the highest Δ (increased in NR with Δ >0.94) were included (Fig. 2C). Fewer NR-increased eOTUs were included because only 38 eOTUs met both P and q value criteria for significance.

Immune cell subsets. The same procedures were followed for statistical analysis of immune cell data for all time points. Wilcoxon rank sum tests were used to identify immune cell subsets found to be significantly different between the two rearing groups. To create the heat map, those immune cell subsets that were found significantly different between rearing groups were rescaled using the scales package. For each immune cell subset, values were rescaled from percentage units to a range from 0 to 1. These new values were then plotted into a heat map. To assess progression of immunophenotypic profile, we used the ade4 package to perform principal component analysis based on data from 12-month samples. Using loadings from this time point, we then applied these loadings to immune data from 5, 6, and 9 months. After centering the data using a custom script, we plotted the data on uniform axes.

Metabolomics data. Each metabolite was subjected to a two-tailed unpaired t test, assuming unequal variance and correcting for false discovery using the q value package (done on total metabolites − stool and plasma samples). Those found significantly different between rearing groups (P < 0.05) were grouped by type of metabolism [based on assignments in KEGG (Kyoto Encyclopedia of Genes and Genomes) Pathway for each metabolite] and displayed in waterfall plots.

Analysis of plasma kyn/trp ratios. Spearman correlation coefficients (R values) for each eOTU abundance were calculated in relation to plasma kyn/trp. For fig. S2B, each genus was ranked on the basis of R values for the eOTUs in that genus with the highest R value. These genera were then ranked by R value and divided by the number of enzymes in the kynurenine pathway for tryptophan catabolism that each genus was inferred to express. Inferences were made by browsing the UniProt Consortium database (64) ( by taxonomy for each of the enzymes in the pathway (Enzyme Commission nos. EC, EC, EC, and EC

Data analysis. To assess possible relations between microbial genera, immune cell subsets, and metabolites, we performed hierarchical clustering on a combined data set of all variables. To evaluate possible relationships between microbial genera, immune cell subsets, and metabolites, we performed linear mixed-effects regression using the lmer command in the lme4 package. Significance of random effects was assessed on the basis of likelihood ratio test between nested models.


Fig. S1. Associations of Campylobacter with arachidonic acid concentrations and TH17 cells.

Fig. S2. No significant differences found in tryptophan catabolism between rearing groups.

Table S1. Significant differences in taxa between rearing groups.


  1. Acknowledgments: We thank J. Lee and P.-M. Sosa for assistance with sample collection; S. Iwai, K. E. Fujimura, and J. Mar for assistance with PhyloChip experiments and analysis; I. Vujkovic-Cvijin and J. M. McCune for their assistance in interpretation of kynurenine-tryptophan results; the West Coast Metabolomics Center at UC Davis, directed by O. Fiehn, for expert services; and A. Tarantal for editorial suggestions. Funding: Research reported in this publication was supported by the National Institute of Allergy and Infectious Diseases, NIH under award no. K23AI081540 to D.J.H.-O., by the Bill and Melinda Gates Foundation under a “Grand Challenges Exploration” award (#52094) to D.J.H.-O., and by the Office of the Director of NIH under grant P51-OD011107 to the CNPRC. Author contributions: A.A., D.J.H.-O., and S.V.L. designed the study. A.A. and M.R. performed PhyloChip assays. A.A., N.R.N., M.R., and D.J.H.-O. performed the statistical analysis. A.A., N.R.N., G.M.-L., and D.J.H.-O. wrote the manuscript. G.M.-L. performed the flow cytometry assay and analysis. Y.H. performed the kyn/trp assay. D.L. and K.K.A.V.R. collected the samples. Competing interests: S.V.L. consults for Janssen Pharmaceuticals and Theravance Biopharma and is on the Scientific Advisory Board of Second Genome. The other authors declare no competing interests.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article